Dokumen - Pub - Perspectives On Causation Selected Papers From The Jerusalem 2017 Workshop 1st Ed 9783030343071 9783030343088

Jerusalem Studies in Philosophy and History of Science
Elitzur A. Bar-Asher Siegal

Nora Boneh Editors
Perspectives on
Causation
Selected Papers from the Jerusalem
2017 Workshop
Jerusalem Studies in Philosophy and History
of Science
Series Editors
Orly Shenker, The Hebrew University of Jerusalem, The Sidney M. Edelstein
Center for the History and Philosophy of Science, Technology and Medicine
Nora Boneh, The Hebrew University of Jerusalem, Language, Logic
and Cognition Center, The linguistics Department
Jerusalem Studies in Philosophy and History of Science sets out to present state of
the art research in a variety of thematic issues related to the fields of Philosophy of
Science, History of Science, and Philosophy of Language and Linguistics in their
relation to science, stemming from research activities in Israel and the near region
and especially the fruits of collaborations between Israeli, regional and visiting
scholars.
More information about this series at http://www.springer.com/series/16087

Elitzur A. Bar-Asher Siegal • Nora Boneh
Editors
Perspectives on Causation
Selected Papers from the Jerusalem 2017
Workshop
Editors
Elitzur A. Bar-Asher Siegal Nora Boneh
Language, Logic and Cognition Center, Language, Logic and Cognition Center,
The Department of Hebrew Language The Linguistics Department
Hebrew University of Jerusalem Hebrew University of Jerusalem
Jerusalem, Israel Jerusalem, Israel
ISSN 2524-4248 ISSN 2524-4256 (electronic)

Jerusalem Studies in Philosophy and History of Science
ISBN 978-3-030-34307-1 ISBN 978-3-030-34308-8 (eBook)
https://doi.org/10.1007/978-3-030-34308-8
© Springer Nature Switzerland AG 2020

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Causation stands at the heart of all sciences, and as such, philosophers, linguists,
and cognitive scientists seek to understand the exact nature of this concept and how
causal structures are represented in the human cognitive systems.
The philosophical models have been a central motor and a constant point of
reference in how thought and to some extent methodology in other disciplines
have been shaped. For example, linguists often borrow philosophers’ analyses
of causation and assume that the relevant linguistic expressions denote such
concepts. Similarly, psychologists and cognitive scientists put to the test models of
causation in investigating central cognitive competencies such as causal learning
and reasoning. The connections between the disciplines, however, are definitely
not unidirectional. Philosophers, for example, occasionally seek insights from the
linguistic literature in understanding what yields certain interpretations of causal
statements. Similarly, other types of interactions can be sought: cognitive psycholo-
gists may benefit from being informed by linguistic analyses in their explorations of
specific human behavior involving language. And of course, linguists may benefit
from cognitive investigations that can be brought to bear on questions pertaining to
domain generality of language, taking causation and its linguistic encoding to be a
study case.
These broad considerations served as the framework for an interdisciplinary
encounter held in June 2017 at the Language, Logic and Cognition Center at the
Hebrew University of Jerusalem, where scholars from the three disciplines attended
the workshop Linguistic Perspectives on Causation. This workshop aimed to bring
together cognitive psychologists, linguists, and philosophers in order to explore
further how the different disciplines can be beneficial and instructive to one another.
The selection of papers grouped in this volume stems from the talks presented at
that workshop, representing a wide range of angles on the study of causation in the
three abovementioned disciplines. To reflect this, the papers are organized in five
parts. In what follows, we present the structure of the book, briefly describing the
papers constituting it.
Part one, titled Perspectives on Causation, concentrates on points of junction
between philosophical and linguistic studies on causation. It consists of papers by
v
vi Preface
Bar-Asher Siegal & Boneh and by Hitchcock. The adoption of central concepts
from classic philosophical accounts to causal relations by linguists stands at the
heart of Bar-Asher Siegal & Boneh’s paper. This paper scrutinizes to what extent
the philosophical concepts are applicable for linguistic analyses of various causative
constructions. In turn, it also critically evaluates cases in which philosophical
discussions seek insights from judgments that are primarily linguistic when dealing
with the metaphysics of causation. In its panoramic perspective on causation and
causative constructions, and with its consideration of the meeting points between
disciplines, this first chapter can also be read as an introduction to the volume, since
it locates the other papers of this book in the discussions it surveys.
Hitchcock’s paper points to the discrepancy between what looks like the binary
representation of causation in language and the way causal relations are modelled in
the framework of the structural equation model, where such relations are sensitive to
multiple variables. He asks how we successfully communicate about causal relations
given this discrepancy.
The papers of the second part, grouped under the title Methodology: Uncov-
ering the Representation of Causation, propose novel methodologies for study-
ing representations of causation. Bellingham, Evers, Kawachi, Mitchell, Park,
Stepanova & Bohnemeyer’s paper presents preliminary findings of the project
Causality Across Languages. Whereas, usually, linguistic studies presuppose some
implicit semantic criterion to what should be included under the category of
“causative constructions,” this study proposes to begin from a systematic observa-
tion of how speakers of different communities communicate about various cognitive
concepts. It proposes several methodologies for exploring production, comprehen-
sion, and conceptualization of causation across a sample of languages. Their studies
pay particular attention to cultural influences and crosslinguistic differences, when
subjects are presented with various visual scenarios, and judge what they have
been shown. The preliminary results are relevant for inquiries interested in causal
pluralism, subcategories of causation (e.g., physical vs. abstract), and crosslinguistic
differences between causative constructions and issues pertaining to lexicalization
vs. pragmatic enrichment in the linguistic representation of causation.
In turn, the paper by Hagmayer & Engelmann traces the way people ask
questions in order to get or give explanations. The goal of their experiments is to
gain insights into the validity of two groups of cognitive-psychological theories of
causal explanations, dependency-related and mechanistic, the assumption being that
the different theories require different types of knowledge for causal explanation.
This paper provides a good overview of current cognitive-psychological theories
for how people explain facts, and its originality lies in the methodology: the authors
allow participants to ask unguided questions seeking explanations, which are in turn
the object of a quantitative analysis, unlike the standard methodology of presenting
subjects with information and then asking them to judge or evaluate.
The next two parts of the book are dedicated to linguistic analyses of causative
constructions. The papers in part three revolve around the topic of Meaning Com-
ponents of Causation. Each of the four papers in it tackles phenomena pertaining to
central inquiries in lexical semantics and in so doing deal with a variety of essential
Preface vii
questions in the literature, among them, event causation, direct causation, internal
causation, zero-change and defeasible causation, and the agent/causer distinction.
The paper by Croft & Vigus is couched in a force dynamics framework. It extends
the first author’s seminal theory of argument realization, where causation serves
as an organizational factor in lexical semantics, to cases in which one finds event
nominals instead of individual participants as arguments of the predicate. Based on
a crosslinguistic investigation, this paper argues that event nominals correspond to
participant sub-events, which are in turn realized according to the same rules as
participants in the causal chain.
Levin’s paper provides support for the prototypical conception of direct causa-
tion in the literature by examining resultative predicates in transitive constructions,
both when the direct object NP is selected by the main verb and in cases where it is
not. It shows that the notion of direct causation, in terms of absence of intervening
participant that applies in the case of simplex causative verbs, also holds here. The
constructions are of interest since they represent concealed causatives, and at the
same time, they behave similarly to sentences with lexical causative verbs, with
respect to direct causation. This observation raises a fundamental question regarding
causative constructions: what is the source of the causative component in them? – a
question that can be of interest to scholars outside of linguistics as well.
Next, Rappaport Hovav’s paper undermines the linguistic validity of the widely
accepted division between internally and externally caused change of state verbs.
This division relies on the assumption that the so-called internally caused verbs
appear as intransitives only – lacking an external cause. The author demonstrates
that what has been accepted in the literature as rigid generalizations is, in fact,
merely a tendency. She, consequently, claims that it does not reflect any grammatical
property of change of state verbs. Instead, the data propose various general princi-
ples that govern lexical causatives and the (non)appearance of cause arguments,
which shape this tendency.
In the last paper of this part, Martin elucidates, on the basis of experimental
studies in Mandarin, French, and English, the crosslinguistic tendency for zero-
change use of causative predicates to occur with an agentive subject contrary to a
cause subject, or an intransitive verb, where zero-change does not arise. It proposes
two types of arguments introducing heads and considers in detail how they combine
with the VPs in languages with weak perfectives and in cases where the verb has a
sub-lexical modal component, yielding defeasible causatives. This paper introduces
different ways in which causal relations are represented in the syntax and how it
affects the semantics of such constructions.
The last point regarding Martin’s paper can also serve to introduce the fourth
part of the book, titled Syntactic and Semantic Aspects of Causation, as the first
two papers by Alexiadou & Anagnostopoulou and Ahdout, as well as Doron’s,
deal with the distinction between agent and causer and its adequate linguistic
representation. All papers in this part argue that, at least at the syntactic level, causal
relations are represented in more than one way.
Alexiadou & Anagnostopoulou discuss the syntactic properties of subjects of
a subclass of psychological predicates (e.g., interest) and claim that there is a
viii Preface
syntactic distinction between the types of causers they license: agents introduced
by Voice and causers introduced in the specifier position of vP, assimilating the
latter to internally caused causative verbs. Contrary to Martin’s semantic account
that distinguishes agent from causers, Alexiadou & Anagnostopoulou claim that
causers form one syntactic domain with the result state constituent, whereas agents
do not. This difference in structure, according to them, also explains the patterns
observed with defeasible causatives with coerced psychological predicates. Their
account also advocates in favor of syntactic indistinctness in encoding causation in
the physical and psychological domains.
Ahdout in turn describes a phenomenon known as agent exclusivity effect in
nominalizations of causative verbs. It has been shown, mainly on the basis of data
drawn from English, that agents in this syntactic environment are licit, whereas
causes are not. Previous work has provided syntactic analyses to account for this
effect, claiming that agent and cause are attached in constructions of different
sizes and therefore can or cannot fit in nominalizations. Other accounts sought
the difference in the type of Voice head available. On the basis of new data from
Hebrew, Ahdout shows that like in Greek, Romanian, and German, the agent
exclusivity effect can be overridden with cause-PPs, therefore casting doubt on
previous analyses. This paper, like the two previous ones, makes clear that at some
level of representation, agents and causes are different. An interesting question
raised here is whether causation can be taken to be a meaning primitive or rather
is read off the structure post-syntactically.
Next, Nash’s paper is concerned with the syntactic and lexical semantic prop-
erties of embedded causees in Georgian. Her central claim is that in neither of
the constructions, the causee is realized as an agent, even if it is an agent in the
simple, unembedded, verb. The paper surveys ways in which the agent argument
is “demoted” when it surfaces as the causee in these causative constructions.
In particular, the paper unveils subtle differences between types of causativized
transitive verbs and provides a novel discussion of causativized unergatives. The
investigation of the syntactic and lexical semantic properties of these constructions
proposes a take on the issue of direct vs. indirect causation by analyzing the
structural and semantic properties of an intervening event participant, between the
causer and the effect. Interestingly, in relation to the main discussion in the previous
three papers, Georgian, at least, does not distinguish between agents and causers at
the structural level.
Returning to psychological predicates, Doron distinguishes between two sub-
classes of verbs, realizing differently the causer component, taken to be an argument,
rather than a relational element. One subclass consists in a two-place relation
between the experiencer argument and the T/SM argument, where the cause
brings about the relation; the other subclass is a one-place property predicate, the
experiencer argument being the subject. In this analysis, the cause argument varies
in its interpretation according to its broader environment. The paper goes on to show
that these two subclasses are not particular or special to the psychological domain;
rather, they pattern like stative physical predicates.
Preface ix
Lastly, Charnavel departs from the other authors in this part in focusing on the
connectives because and since rather than on the lexical properties of verbs. She
proposes that these connectives constitute attitude contexts introducing a judge from
whose perspective the causal relation between the content of the main clause and
that of the adjunct clause is evaluated. The paper argues that the causal judge is
syntactically present. It is shown, on the basis of data collected in experiments, that
the causal judge is introduced as an argument of the connective and is identified
through exhaustive binding by the closest relevant attitude holder in the sentence,
which is either the speaker alone or the speaker together with a relevant animate
event participant. This depends on the site of adjunction of the because and since
phrase, allowing in the first case, but not in the second, an animate event participant
to be the attitude holder controlling the judge.
The closing fifth part contains two papers concerned with Philosophical
Inquiries on Causation by Statham and Kment. Statham’s paper surveys recent
advances in philosophical thinking about causation and causal reasoning, paying
particular attention to those models construing causation reasoning as deviation
from the norm. Similarly to Hitchcock, this paper also considers and evaluates the
structural equation model as a powerful system for representing causal systems.
Considering causal relations through deviation from the norms leads the author
to break from the tradition that bases the metaphysics of causation on insights
from the physical and natural world of laws, independent of human concerns. One
consequence of this is the enrichment of the traditional classification of types of
clausal claims customarily distinguishing type and token claims and taking only
tokens to be deviant, whereas types are always normal. The novel proposal in the
paper is that these categories of claims are orthogonal, and therefore, one can also
encounter deviant types. The paper invites further investigation of the question how
the typological abundance of causal relations made available by the recent models
can inform linguistic research and more generally the issue of sub-types of causal
locutions.
Kment’s paper criticizes the standard view, attributed to Lewis, according to
which, causal relationships are defined by counterfactual dependency. Instead, he
argues that counterfactual dependence provides evidence for causal connections
but does not constitute them. That is, counterfactual reasoning is only useful for
establishing causal claims, and natural laws and past history are needed to establish
a new claim about relationships of (actual token) causation. This paper is in line with
the literature in philosophy and in linguistics, according to which, counterfactual
statements are accounted for by causal relations, since prior knowledge is required
for establishing such claims. In this sense, it elucidates that one is not reducible to
the other.
While this preface provides one way of grouping the papers thematically, various
other ways could be thought of, according to several recurrent topics throughout this
book, regardless of the discipline of each chapter. We will briefly mention some, so
as to propose ideas for other possible inquires across disciplines.
Many discussions in this volume can be read with the fundamental question in
mind of whether and how causality can be reduced to other noncausal terms (Croft
x Preface
& Vigus, Hagmayer & Engelmann, Kment and Rappaport Hovav). Another central
question is whether it is advisable to consider causal pluralism instead of one all-
encompassing causative account for causation (Bellingham et al. and Hagmayer
& Engelmann). As noted earlier, this question can be extended to the syntactic
representation of the causal relations, inquiring whether it is better to assume a
single syntactic structure or multiple ones. Another relevant question that received
different treatments is whether causal relations are different when the effect pertains
to the mental realm and whether, linguistically, such descriptions are grammatically
marked (Alexiadou & Anagnostopoulou, Bellingham et al., Croft & Vigus, and
Doron).
Turning to the lexico-syntactic representations of causal relations, many authors
indirectly deal with the basic question of what categories constitute causative
constructions, in terms of types of arguments implicated in them and their selectors
or introducers. More specifically, under discussion is the question whether, on
the one hand, it is necessary that such constructions denote causal relations, as
some authors consider constructions which do not entail the effect took place,
and on the other hand, whether it is sufficient that such relations are entailed in
order to be analyzed as causative constructions, as is the case with, for example,
concealed causatives, when causation is not marked overtly (Ahdout, Alexiadou &
Anagnostopoulou, Charnavel, Croft & Vigus, Levin, Martin and Nash).
There are also questions across disciplines, which at least at first sight seem
similar, but one is left to wonder how exactly the different types of discussions
should or can interact. We have in mind issues pertaining to the relata in the causal
relations (Bellingham et al., Croft & Vigus, Doron, Hitchcock and Levin) and the
issue of causal selection, and under this category, we also include the restriction
of direct causation in different constructions (Bellingham et al., Hitchcock, Levin,
Rappaport Hovav and Statham).
This is only a sample of topics that one repeatedly encounters when reading the
papers in this volume. In our own chapter (Bar-Asher Siegal & Boneh), we elaborate
more on these themes and reflect on how the contributions of the papers in this
volume are relevant to them.
∗
We would like to conclude by noting that this book represents a community

effort. Notably, all authors dedicated time and energy to participate in a true
interdisciplinary conversation and sought ways in which less familiar knowledge
and methods can inform and improve their own scholarship. Additionally, all papers
in this volume were read and reviewed by two to three scholars, most of the
reviewers participated in the conference, which initiated this volume. It is our
privilege to thank the reviewers for the hard work they have put into providing
helpful and constructive reviews. At the same time, we wish to acknowledge the
willingness of the authors to go outside of their comfort zone and implement insights
from other disciplines. It is our hope that more interdisciplinary contributions in the
study of causation will follow this volume.
Preface xi
We are grateful to Orly Shenker for intellectually and materially supporting this
endeavor, in inviting us to inaugurate the linguistic part of the series Jerusalem
Studies in Philosophy and History of Science. We would like to thank Padmapriya
Ulaganathan and Malini Arumugam from Springer for their hard work in bringing
this book to publication.
Finally, with an ache in our hearts, we reserve a special thought to our mentor and
colleague, Edit Doron, who passed away at the end of March 2019, just a few weeks
after submitting her paper for this volume. Her relentless quest for knowledge and
intellectual breadth had an important role in the journey that led to this volume and
will continue to be a source of inspiration.
Jerusalem, Israel Elitzur A. Bar-Asher Siegal

August 2019 Nora Boneh
Contents
Part I Perspectives on Causation

1 Causation: From Metaphysics to Semantics and Back . . . . . . . . . . . . . . . . . 3
Elitzur A. Bar-Asher Siegal and Nora Boneh
2 Communicating Causal Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Christopher Hitchcock
Part II Methodology: Uncovering the Representation of Causation

3 Exploring the Representation of Causality Across
Languages: Integrating Production, Comprehension and
Conceptualization Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Erika Bellingham, Stephanie Evers, Kazuhiro Kawachi,
Alice Mitchell, Sang-Hee Park, Anastasia Stepanova and Jürgen
Bohnemeyer
4 Asking Questions to Provide a Causal Explanation – Do
People Search for the Information Required by Cognitive
Psychological Theories?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
York Hagmayer and Neele Engelmann
Part III Meaning Components of Causation

5 Event Causation and Force Dynamics in Argument Structure
Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
William Croft and Meagan Vigus
6 Resultatives and Constraints on Concealed Causatives . . . . . . . . . . . . . . . . 185
Beth Levin
7 Deconstructing Internal Causation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Malka Rappaport Hovav
xiii
xiv Contents
8 Aspectual Differences Between Agentive and Non-agentive

Uses of Causative Predicates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Fabienne Martin
Part IV Syntactic and Semantic Aspects of Causation

9 Experiencers and Causation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
Artemis Alexiadou and Elena Anagnostopoulou
10 “Agent Exclusivity” Effects in Hebrew Nominalizations. . . . . . . . . . . . . . . 319
Odelia Ahdout
11 Causees are not Agents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Léa Nash
12 The Causative Component of Psychological Verbs . . . . . . . . . . . . . . . . . . . . . 395
Edit Doron
13 Linguistic Perspectives in Causation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Isabelle Charnavel
Part V Philosophical Inquiries on Causation

14 Causes as Deviations from the Normal: Recent Advances in the
Philosophy of Causation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
Georgie Statham
15 Counterfactuals and Causal Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
Boris Kment
Contributors
Odelia Ahdout Humboldt Universität zu Berlin, Berlin, Germany

Artemis Alexiadou Humboldt Universität zu Berlin, Berlin, Germany
Leibniz-Zentrum Allgemeine Sprachwissenschaft, Berlin, Germany
Elena Anagnostopoulou University of Crete, Rethymno, Greece
Elitzur A. Bar-Asher Siegal Department of Hebrew Language; The Language,
Logic and Cognition Center, The Hebrew University of Jerusalem, Jerusalem, Israel
Erika Bellingham Department of Linguistics, University at Buffalo, Buffalo, NY,
USA
Jürgen Bohnemeyer Department of Linguistics, University at Buffalo, Buffalo,
NY, USA
Nora Boneh Department of Linguistics; The Language, Logic and Cognition
Center, The Hebrew University of Jerusalem, Jerusalem, Israel
Isabelle Charnavel Department of Linguistics, Harvard University, Cambridge,
MA, USA
William Croft University of New Mexico, Albuquerque, NM, USA
Edit Doron† Department of Linguistics and Language, Logic and Cognition Center,
The Hebrew University of Jerusalem, Jerusalem, Israel
Neele Engelmann Department of Cognitive and Decision Sciences, Institute of
Psychology, University of Göettingen, Göttingen, Germany
Stephanie Evers Department of Linguistics, University at Buffalo, Buffalo, NY,
USA
York Hagmayer Department of Cognitive and Decision Sciences, Institute of
Psychology, University of Göettingen, Göettingen, Germany
xv
xvi Contributors
Christopher Hitchcock Division of Humanities and Social Sciences, California

Institute of Technology, Pasadena, CA, USA
Kazuhiro Kawachi National Defense Academy of Japan, Yokosuka, Japan
Boris Kment Department of Philosophy, Princeton University, Princeton, NJ, USA
Beth Levin Department of Linguistics, Stanford University, Stanford, CA, USA
Fabienne Martin Humboldt-Universität zu Berlin, Berlin, Germany
Alice Mitchell Institute for African Studies and Egyptology, University of Cologne,
Cologne, Germany
Léa Nash Department of Language Sciences, Université Paris Lumières-Saint
Denis/CNRS, Paris, France
Sang-Hee Park Department of Linguistics, University at Buffalo, Buffalo, NY,
USA
Malka Rappaport Hovav Department of Linguistics; Language, Logic and Cogni-
tion Center, The Hebrew University of Jerusalem, Jerusalem, Israel
Georgie Statham Polonsky Academy Fellow, The Van Leer Jerusalem Institute,
Jerusalem, Israel
Anastasia Stepanova Department of Linguistics, University at Buffalo, Buffalo,
NY, USA
Meagan Vigus University of New Mexico, Albuquerque, NM, USA
Part I
Perspectives on Causation
Chapter 1
Causation: From Metaphysics
to Semantics and Back
Elitzur A. Bar-Asher Siegal and Nora Boneh
Abstract This paper examines reciprocal connections between the discussions

on causation in philosophy and in linguistics. Philosophers occasionally seek
insights from the linguistic literature on certain expressions, and linguists often
rely on philosophers’ analyses of causation, and assume that the relevant linguistic
expressions denote philosophical concepts related to causation. Through the study
of various semantic aspects of causative constructions, mainly targeting the nature
of the dependency encoded in various linguistic constructions and the nature of the
relata, this paper explores interfaces between the discussions in the two disciplines,
and at the same time points to significant differences in their objects of investigation,
in their methods and in their goals. Finally, the paper attempts to observe whether
the disciplinary line is maintained, i.e. whether or not it is the case that metaphysical
questions are examined as linguistic ones and vice versa.
Keywords Cause · Effect · Dependency · Counterfactuality · Causal Selection ·

Negation · Relata · Metaphysics · Causative constructions
E. A. Bar-Asher Siegal
Language, Logic and Cognition Center, The Department of Hebrew Language, Hebrew
University of Jerusalem, Jerusalem, Israel
e-mail: ebas@mail.huji.ac.il
N. Boneh ()
Language, Logic and Cognition Center, The Linguistics Department, Hebrew University of
Jerusalem, Jerusalem, Israel
e-mail: nora.boneh@mail.huji.ac.il
© Springer Nature Switzerland AG 2020 3

E. A. Bar-Asher Siegal, N. Boneh (eds.), Perspectives on Causation,
Jerusalem Studies in Philosophy and History of Science,
https://doi.org/10.1007/978-3-030-34308-8_1
4 E. A. Bar-Asher Siegal and N. Boneh
1.1 Introduction: Philosophical and Linguistic Discussions

on Causation
Discussions about the nature of causal relations stood at the heart of philosophical
inquiries since the days of the ancient Greek philosophers, most notably in the work
of Aristotle. Although, for Aristotle causality was not defined as a unitary notion, as
he developed the doctrine of the four causes,1 at least since the days of the British
empiricist David Hume, philosophers attempt to provide a unified account for what
stands behind the attribution of the terms “cause” and “effect” to two things.
For various philosophers, deliberations on the nature of causal relations, is
an attempt to characterize the intuition, broadly described as “the folk theory of
causation”, implicitly entertained by many (inter alia Lewis 2000; Menzies 2009).
Consequently, among the objects of their investigation are linguistic expressions
that seem to underlie these relations. In other words, such philosophers attempt to
provide a conceptual account, in non-causal terms, to all and only cases in which
people have an intuition to assert correctly that: “c is the cause of e” (or other causal
judgments).2 From a linguistic point of view, de facto such inquiries aim to identify
the semantics of such expressions.3
Putting it more broadly, one can identify reciprocal connections between the
discussions on causation in philosophy and in linguistics. Philosophers, on the one
hand, are often interested in the language of causal judgments and occasionally seek
insights from the linguistic literature on certain expressions, and linguists, on the
other hand, often borrow philosophers’ analyses of causation, and assume that the
relevant linguistic expressions denote such concepts. This paper explores various
interfaces between the discussions in the two disciplines, and at the same time
points to significant differences in their objects of investigation, in their methods
and in their goals. Finally, it attempts to observe whether the disciplinary line
is maintained, i.e. whether it might be the case that metaphysical questions are
examined as linguistic ones and vice versa.
Considering first the object of investigation, most philosophers take it to be “the
world” – as causal relations are between entities in the world. The metaphysics of
causation, generally speaking, depicts the structure of the world itself, so that it
will be one that hosts such causal relations (inter alia Hall & Paul 2013). Thus, a
prominent question is what the relata are in a causal relation. Approaches differ
1 Aristotle, in all likelihood, did not provide an account for causality in the sense that causation
was analyzed in the philosophical literature since Hume. For Aristotle causes are whatever answers
the question “why” and therefore his causes are various types of because-answers (see inter alia
Hocutt 1974). For a somewhat parallel approach from recent literature, see Skow (2016).
2 It is sufficient to mention examples from the last decade, such as Schaffer (2013: 49), Skow (2016:
26–27), and Hitchcock’s contribution to this volume.

3 An even more radical claim is that human knowledge of causality derives from “the linguistic
representation and application of a host of causal concepts.” (Anscombe 1981: 93; see also Psillos
2009).
1 Causation: From Metaphysics to Semantics and Back 5
with respect to the kinds of things the relata in causal relations (events, facts, tropes,
attributes etc.) are.4 Another central issue in philosophical accounts of causation,
which has some bearing on various issues that will be discussed in this paper, is
the question whether causation can be reduced to other more basic relations.5 For
some philosophers, each causal judgment has some suitable description in which
it is an instantiation of some lawful regularity (Davidson 1967), or they argue
that an account of causation must determine the logical dependencies between the
participants in such relations, such as e.g. necessity and sufficiency (Mackie 1965),
other types of dependencies such as counterfactuality (Lewis 1973a, b), probability
(Kvart 2004), or by revealing the physical events that stand behind such claims
(Dowe 2000).
In contrast, for linguists, the object of investigation is, for the most part, linguistic
expressions, which we will henceforth refer to as causative constructions (to be
defined below).6 These span overt causative verbs such as cause but also make,
allow, enable, let; connectives such as because (of), from, by, as a result of ; and
change of state verbs such as open, boil, which may or may not include what
are thought to be dedicated causative morphemes, and constructions involving
affected participants. The specific concern in each of these types of constructions
varies: whereas the goal of formulating the truth conditions of connectives and
overt causative verbs is fairly straightforward, pinpointing a presumed causative
component in change of state verbs is less trivial. With respect to these verbs, one
central point is to understand the regularity of derivation between a stative-like
expression and change of state verbs. The aim of such a discussion is to reveal
the role of the causative meaning component in the derivation (Haspelmath 1993;
Haspelmath et al. 2014; Doron 2003; Lundquist et al. 2016 among many others).7,8
4 For Davidson (1969), for example, the individuation of events derives from their participation in
causal relations.
5 See, Woodward (2003) and Carroll (2009) for non-reductionist approaches to causation.
6 Some linguists emphasize that causal expressions are not about actual causation in the world but
rather, about how it is psychologically construed. For example, based on this assumption Levin
and Rappaport Hovav (=LRH) propose a distinction between internal and external causation,
which cannot be accounted for in terms of classical analyses of causation (see inter alia Levin
& Rapaport Hovav 1994, 1995 et seq. and Rappaport Hovav’s contribution to this volume). It
is unclear, however, in a model-based approach to semantics, how the truth values of causative
sentences are determined, according to those who claim that these types of judgments should not
be evaluated against causal relations in the world.
7 In certain languages, in pairs of inchoatives and causatives, the former are marked. These are
cases, known in the literature as anticausatives (see in this book Alexiadou & Anagnostopoulou,
Ahdout, Rappaport Hovav).
8 Linguists’ concerns in causation cover other levels of analyses besides the semantic one. One
central topic, where the relevance of causation became significant is with respect to issues
pertaining to argument realization mostly in dealing with the following two questions: A. Is
causation an or the organizing factor in the grammatical relations of the basic predication (Croft
1991 et seq., see also Croft & Vigus this book)? B. Is it reflected in specific types of the predicates’
arguments: whether there is a thematic role of CAUSER (e.g. Pesetsky 1995; Reinhart 2000; Doron
Importantly, also within linguistics, the issue of the relata comes up, and views
on their nature diverge a great deal.9 It is not always clear what the criteria are in
linguistics for determining the nature of the relata, and, in fact, different approaches
derive from different motivations: some linguists motivate their choice by referring
to a philosophical conceptual analysis of causation (see Pylkkänen 2008, or the
contribution of Levin this volume). Others, especially those who take individuals to
be part of the causal relation, point to linguistic manifestations of causal judgments,
where more often than not nominal expressions (NPs/DPs) are the participants in
the actual linguistic expressions (see Doron 1999 and this volume; Reinhart 2000,
2002; Neeleman & van de Koot 2012).10 This approach, very often, comes with
a claim that linguistic causative expressions do not correlate with the way causal
relations are perceived from a philosophical perspective.
Crucially, a non-trivial assumption underlying the question of the relata in the
philosophical discussion is the issue of it embodying a binary relation between cause
and effect. Philosophers committed to the framework of the Structural Equation
Model (such as Pearl 2000; Yablo 2004; Woodward 2003 and Hitchcock this
volume) do not take the binary relation to hold metaphysically; Hitchcock goes on
to claim that the binary relation pertains to or stems from linguistically influenced
causal judgments. In the rest of the paper, we will not refer to this framework
directly, since much of the existent linguistic literature does not incorporate insights
stemming form it.11,12
In contrast, within linguistics, various scholars argue that, while conceptually,
causation involves a binary relation, it is not necessary for the linguistic expression
this volume); or whether there is, at the syntactic level, a designated functional head of CAUSE
(see discussions by Ahdout and Alexiadou & Anagnostopoulou this volume).
9 In a superficial way, it is possible to mention the following options:
Cause Effect
Proposition Proposition (Dowty 1979)
Event Event (Pylkkänen 2008)
Individual Proposition (McCawley 1976)
Individual Event (Doron 2003; Neeleman & van de Koot 2012;
Reinhart 2000; Pesetsky 1995)
Individual Individual (Talmy 1976; Croft 1991)
10 Since Dowty (1979), it is acknowledged that there is a discrepancy between the grammatical
realization of the causer as a nominal phrase and the semantic facet. Accordingly, the individual
syntactically realized is seen as part of a causing event (see Croft & Vigus this volume)
11 For recent linguistic work building on this framework consider inter alia Bjorndahl & Snider
(2015), Baglini & Francez (2016), Nadathur & Lauer (2020) and Baglini & Bar-Asher Siegal
(forthcoming).
12 As noted by Hitchcock (this volume), one can identify the inspiration for the SEM approach
already in Mill’s observation that causality is always held between a set of conditions and an effect.
In this respect we will also be engaged in the current paper with this approach in the discussion in
Sect. 1.4 regarding Causal Selection. Another reason for not engaging with this approach is that
it is not a trivial matter what the principles are in constructing the relevant models (see inter alia
Halpern & Pearl 2005a, b; Hall & Paul 2013).
to represent the cause.13 At the same time, issue is taken with cases where there
seem to be more than two parts to the relation.14
With this background in place, this paper critically traces points of interaction
between the two disciplines, focusing on ways in which philosophical ideas were
brought to bear on linguistic work. At the same time, we seek to expand our
understanding of what in the philosophical discussion pertains to the linguistic realm
(in line with Hitchcock’s & Statham’s papers, in this volume).
We will illustrate this type of inquiry by exploring several facets of the inter-
pretative properties of linguistic constructions, some overtly encoding causation
via the verb cause and its kin, or the connective because, others covertly – such
as lexical causative verbs (change of state verbs, and caused activity verbs) or
Affected Participant constructions. In order to have a common denominator for
the discussion, we take linguistic Causative Constructions to be divided into three
parts:15
(i) a cause (c);

(ii) the effect of the cause (e); and
(iii) the dependency (D) between c and e
(1) [c] D [e]
Using this working definition, we examine the nature of the relation in (1) in
various constructions, by answering the questions that will be laid out in the next
section. It must be emphasized that “cause” (c) and “effect” (e) are used here loosely
in a pre-theoretical manner. Accordingly, the use of the term “causative” or the
division of the components to “cause” and “effect” neither indicates an assumption
that a construction denotes causal relations, nor does it commit to the nature of (c)
and (e). In fact, it is quite the opposite: we will use (c), (e) and D, in an uncommitted
manner, as it is our goal to understand their nature. We would like to examine to
what extent the nature of (c) and (e) is similar to what philosophers think about the
relata of the causal relation, and whether the philosophical accounts for causality
can provide better insights as to the nature of the D in these constructions.
13 Itwas argued that there is a set of intransitive verbs, designated anticausative verbs, that denote
an event affecting its subject, without a syntactic representation of the cause (Alexiadou et al. 2006,
and subsequent work; see also early work by Levin & Rappaport Hovav 1995 for similar ideas).
14 This is particularly relevant for the analysis of psychological predicates and the distinction
between cause and Target/Subject Matter (Pesetsky 1995; Doron this volume, among others); but
also cases where agents and instruments appear together and bring about the effect (these cases are
extensively discussed by Croft 1991, also Croft & Vigus, this volume).
15 Cf. Bellingham et al. in this volume, who also compare between causative constructions. They
propose, however, a different approach as to what should be considered as a causative construction,

without holding received semantic preconceptions.
In Sect. 1.2, we lay out the questions to be explored in the subsequent sections of
this paper; to anticipate, these questions seek to identify philosophical concepts rel-
evant for the linguistic analysis, the way they should be defined truth conditionally,
and to see whether all causative constructions underlie one and the same causative
concept. In turn, we also explore what in the philosophical metaphysical inquiry
pertains to the linguistic one. In the second part of the section, we provide a general
survey of the various causative constructions to be analyzed in the paper. Then,
in Sects 1.3, 1.4 and 1.5, we move to consider specific interpretative components
of D relating c & e. The focus of Sect. 1.3 is counterfactuality, central to the
philosophical discussion, also prevalent in linguistic treatments. In Sect. 1.4, we
put to the test the question of Causal Selection in linguistic constructions, and
compare how the various linguistic constructions pattern in this respect, observing
that besides counterfactuality, D in each type of construction has different properties
in singling out, or not, The Cause. Sect. 1.5 takes issue with negation, and through
this further examines the semantic properties of D and the relata: whether D is
asserted or not (1.5.1), and whether the relata (c) and (e) can be independently
negated, opening a discussion on whether the relata are event-like or individual-
like (Sects. 1.5.2, 1.5.3 and 1.5.3.1). Finally, Sect. 1.6 applies insights from the
previous sections to an additional causative construction – the Affected Participant
construction, where causation is not overtly encoded by any particular linguistic
material. Sect. 1.7 concludes the discussion.
1.2 Setting the Scene
1.2.1 Theoretical Questions and Their Background
As we explore the flow of ideas about causation between philosophy and linguistics,
we will focus on the following set of broad questions:
A. Can philosophical accounts of causation be relevant for linguistic analyses of
causal constructions? Taking a semantic point of view, we ask whether such
accounts can be “translated” to truth-conditions examining whether they provide
the accurate truth conditions to these expressions. From a syntactic point of
view, one may ask whether metaphysical accounts should put constraints on
the syntactic analysis of the relevant constructions, for example, by determining
the categorical nature of the relata.
B. Is there one all-encompassing causative meaning component underlying the
diverse linguistic phenomena, regardless of whether the marker of the causal
dependency is overt (e.g. cause, because) or covert (such as in lexical causative
verbs); or should there be different ones for the various constructions, possibly
correlating with the type of linguistic form?
C. As for the philosophical discussions on causation, we inquire whether they are

sensitive to the linguistic data they rely on; or whether the disciplinary borderline
between metaphysical questions and semantic ones is blurred.
A consequence of answering A positively often leads to answering Question B
by claiming, or at least assuming, that there is only one type of causative meaning
component for the diverse linguistic phenomena. Dowty (1979) is a good example
of an influential linguist who followed this path, as he adopted Lewis’ (1973a, b
et seq.) analysis of causation, and consequently took it almost for granted that
counterfactuality underlies the semantics of the various causal constructions. In
contrast, many linguists observe a strict disciplinary borderline, and assume that
although the concept of a causal relation is indeed relevant for the linguistic analysis,
its particular semantic nature can remain opaque. Accordingly, the component
CAUSE, either in the syntax or in the morphology, is taken to be an unanalyzable
semantic primitive (e.g. Morgan 1969; Lakoff 1970; Jackendoff 1972: 39; Levin
& Rappaport Hovav 1995 et seq.; Pylkkänen 2008). Yet a different approach is
represented by such scholars as Talmy (2000) and Marantz (2005), who argue, in
the context of verbs, each within a different framework, that causation is not part of
their lexical properties, or that other concepts are more relevant (see also Neeleman
& van de Koot 2012).
In the discussion bellow, we follow those who advocate semantic analyses
that are informed by the philosophical literature, and assume that Question A is
answered positively.
To set the stage, we turn now to introduce, in a somewhat simplified manner,
two prominent approaches to causal relations: the dependency account and the
production account (for a philosophical introduction of the two approaches to cau-
sation see Dowe 2000, and also Copley & Wolff 2014 for application in linguistics
and psychology). According to the former, a basic conception of causation was
to perceive Cause and Effect as related according to the following (from the 70s
to now: Shibatani 1976b and see also Comrie 1981; Dixon 2000; Talmy 2000;
Escamilla 2012):16
(a) Dependency between events – the causal relation is held between two events.
(b) Temporal precedence – the cause must precede effect.17,18
(c) Counterfactuality – the dependency is defined in the following way: “had the
cause not occurred, the effect would not have occurred either.”
16 Cf. Neeleman & van de Koot (2012) who argue that although this is indeed the conceptual
representation of causal relations, languages do not encode such a relation. It is unclear, however,
how their alternative concept of Crucial Contributing Factor (CCF) can be established without
recourse to some notion of causation. See also Martin (this volume) for the possibility that
languages syntactically represent causal relations in different ways.
17 For Lewis (1979) the temporal asymmetry of causal dependence derives from his counterfactual
analysis in terms of closeness between possible worlds (cf. Anscombe 1981).

18 The assumption that the cause must precede the effect goes back to Hume (A Treatise of Human
Nature, §1.3.14).
This understanding of causation, to a large extent, follows Lewis’ (1973a, b)

counterfactual theory of causation (see Sect. 1.3), and was adopted whole-sale from
philosophers without engaging in a fundamental discussion (but see Dowty 1979:
106–109 and Eckardt 2000).
The latter way to conceptualize causation assumes that some quality of the cause
produces the effect. This approach emphasizes the intuition that the cause brings
about the effect. While, since Hume, there is skepticism about theories of production
as they seem to entail an unanalyzable causal primitive, various philosophers,
linguists and psychologists developed such theories, according to which causation
conceptually derives from people’s representations of transfer of force and spatial
relations. Within linguistics, this approach can be traced back to Talmy’s (1976,
2000) work as well as to Croft’s (1991 et seq.), see also the representation of
this approach in this volume in the following papers: Bellingham et al., Croft and
Vigus and Hagmayer and Engelmann. Causation, accordingly, is viewed from a
conceptual or cognitive perspective, where the purpose in the linguistic literature
is to understand how it is reflected in the grammar, or serves as an organizational
mechanism for argument realization (for more recent literature see Wolff 2007;
Copley & Harley 2015; Copley et al. 2015 and Wolff & Thorstand 2016).19
This paper, for the most part, examines different aspects of the dependency
approach, with occasional notes to the literature from the production approach,
when it will be directly relevant for the examined topics. The main reason for this
choice is that the three respects in which causation is examined in this paper –
counterfactuality, causal selection and negation of causation – are more easily
applicable within the dependency approach, than in the force-dynamic one.
Question B, regarding the unitary concept, is quite complex. Indeed, in the
history of the linguistic literature, one can repeatedly identify the underlying
assumption of a unitary analysis, just to mention a few examples: An early stab on
the question of causation in linguistics was provided in the framework of Generative
Semantics. In this framework, an attempt was made to claim that underlyingly
the semantic primitive CAUSE and the overt verb cause are in fact one and the
same thing. They have the same entailed propositions and the same conditions of
temporality, dependency and counterfactuality hold for both (see also van Valin
2005: 38). Syntactically, evidence was adduced in favor of event decomposition
(McCawley 1968 and Morgan 1969). Similarly, when Pesetsky (1995) introduced
CAUSER as a thematic role, he assumed that its underlying syntax is identical to
that of the overt preposition because of ; Alexiadou et al.’s (2006, and later work)
propose a similar structure to verbs with a NP/DP causer as their subject and the
participant with preposition from. In a different context, recently, Copley et al.
19 Withinthe same line of thought, various philosophers provide accounts for causation that do not
reduce causation to some dependency defined merely by logical relations. Among those there are
production accounts (Hall 2004), which aims to capture the notion of “bringing about” affiliated
with causation, and causal processes which focus on the role of physical processes as those that
define causal relations (Salmon 1997; Dowe 2000).
(2015) attempt to provide a unified analysis of verbs and connectives through force-
dynamic theories.
As noted earlier, one can identify this assumption concerning the unitary analysis
for causal relation as an inheritance from the philosophical tradition. Recently,
however, philosophers proposed various theories of causal pluralism (Hitchcock
2003; Hall 2004; Psillos 2009). Similarly, within cognitive studies, Waldmann &
Hagmayer (2013), inter alia, indicate that people have a pluralistic conception of
causation, and different judgments rely on different types of concept of causal
relations. Traces of this tendency can be observed also in recent linguistic studies.
Copley & Wolff (2014) suggest that different types of causative constructions should
be analyzed in light of different approaches to causation (e.g. causal connectives
are best captured as a dependency, whereas the semantics of causal verbs is best
captured in the framework of production based theories). Similarly, Lauer (2010),
Martin (2018), Bar-Asher Siegal & Boneh (2019) and Nadathur & Lauer (2020)
argue that the semantic content of D is different in various constructions, tracing
whether the main verb encodes a necessary and/or a sufficient condition.
Finally, we wish to conclude this section with an example for how philosophical
analyses can fruitfully inform linguistic ones. We, pre-theoretically, characterized
causative constructions by the D that stands between (c) and (e). However, linguists
do not always distinguish between causation and other types of dependencies,
such as grounding,20 logical dependence, teleology21 or reasoning, which are
kept distinct in philosophy. Nevertheless, several studies did point out that not
all causative constructions are dedicated to the expression of just and only causal
relations. For example, connectives as well as the verb cause give rise to situations
where temporal precedence and counterfactuality do not simultaneously hold with
dependency:22
20 For an introduction of the notion of grounding see Correia & Schneider (2012). Schaffer (2016:
96) lists the following differences between causation and grounding:
• causation can be non-deterministic, grounding must be deterministic;
• causation can only connect distinct (grounding-disconnected) portions of reality; and
• causation can be non-well-founded, grounding must be well-founded.
21 In discussions on the philosophy of action, for various philosophers, such as Davidson (1963, and
more broadly in 1980), teleological explanations are themselves analyzable as causal explanations.
Others, such as Taylor (1964), argue that they should be analyzed in non-causal terms.
22 Another use of because is when it is used to indicate the source of the speaker’s knowledge, as
in sentences like They are getting married, because I saw an engagement ring on her finger. We
wish to thank Larry Horn for mentioning this type of because; we do not refer to such cases as they
may involve a different kind of causal relations. Cf. Charnavel (this volume and related work) on
similar uses of since.
(2) a. A kangaroo is a marsupial because it has a pouch. (Dowty 1979: 132b)

b. Mary’s living nearby causes John to prefer this neighborhood. (Dowty
1979: 132c)
c. The floor is black because of the ants that might infest it. (adopted from
Maienborn & Herdtfelder 2015)
This paper proposes a preliminary study that attempts to critically consider
points of meeting between the discussions in philosophy and linguistics, and also
where they part ways. We will do so by exploring differences between various
causative constructions, as we propose a preliminary semantic characterization of
some of them. Throughout Sects. 1.3, 1.4, 1.5 and 1.6 we will explore differences
in the semantics of various causative constructions, and examine the source for
these differences. More specifically, we will ask whether the differences in the
semantics indicate that the various constructions encode different causal concepts
(cf. Thomason 2014) or whether they can be accounted in other ways such as
different syntactic structures.
The questions evoked in C are general in their nature, and require a vast and
careful investigation. In the paper, we will refer to C mainly in Sect. 1.4, and also in
the concluding discussion.
The next section introduces several types of causative constructions that we then
compare in the subsequent Sects. 1.3, 1.4 and 1.5.
1.2.2 Causative Constructions
We center on three central types of causative constructions in English and Hebrew.

Hebrew is useful as it enables to widen the discussion of lexical causatives
(Sect. 1.2.2.3), with its overt morphology absent in English. Our categorization
is classified according to a basic syntactic characterization, and it is purely for
presentational purposes (for typologies of cross-linguistic causative constructions
see Shibatani 1976a; Comrie 1981: 158–177; Song 1996; Dixon 2000, among
others). In Sect. 1.6, we add another construction to the discussion: the Affected
Participant construction, available both in Hebrew and in English. This will enable
us to examine further the semantic properties of causal constructions in the absence
of an overt D.
1.2.2.1 Overt Causative Verbs
Under this category fall verbs such as cause, make, enable, allow, let, that seemingly
express causal relations, where the subject is the cause and the complement of the
verb is the effect.
(3) a. [c The neighbor/the music] caused / made / enabled [e the kids

(to) dance].
b. [c ha-šxena/ha-musika] garma / ifšera [e la-yeladim lirkod].
The-neighbor/the-music made / let the children dance
Such overt verbs are used most often in philosophical discussions about causal
relations, assuming that they are true in a given circumstance only when (c) is the
cause of (e) (inter alia Anscombe 1981; Hitchcock & Knobe 2009; Schaffer 2013,
also Statham this volume).
A few linguistic analyses of these verbs, focusing mostly on the verb cause,
provide a semantic analysis of a counterfactual dependency (inter alia Abbott
1974; Eckardt 2000; Lauer 2010). Others have noted on the role of causation in
the meaning of other verbs, such as implicative verbs (Nadathur 2015; Baglini
& Francez 2016). Recent accounts of such verbs, assuming semantic analyses of
causation as forces, argue for two interacting forces or tendencies. They propose
that verbs vary with respect to whether the force is associated with the agent, as
is the case with the verb cause, or with the patient, as is the case with the verb
enable (Talmy 2000; Wolff & Song 2003; Wolff 2007; Copley et al. 2015). They
take the availability of such distinctions to be a theoretical advantage for a force-
dynamics analysis of causation. It seems necessary, however, to examine whether
these are indeed differences in the semantics of the verbs, or whether the differences
between the semantics of these verbs should be relegated to a variety of pragmatic
implications associated with them.23 The focus in this paper is mostly on the verbs
cause in English, and its Hebrew rough equivalent garam.
1.2.2.2 Connectives
Connectives are conjunctions such as because, since, for; and prepositions such
as because (of), from-PPs, by-PPs. Some of them come as complex nominal
expressions, such as as a result of, out of, added as adjuncts introducing the cause
to a main clause, expressing the effect. Whereas the latter two introduce a nominal
expression, because and since can also connect two clauses. These elements have
been studied from various perspectives (inter alia Alexiadou et al. 2006; Charnavel
23 According to Wolff (2003), ENABLE is associated with the tendency of the patient for the result
and with lack of opposition between the effector and the patient, while this tendency is absent in
the case of CAUSE, as there is an inherent opposition between the effector and the patient. Such
a dichotomy must assume that these two verbs are in a complementary distribution, and therefore
cannot describe the same state-of-affairs. However, it seems to be the case that often the distinction
is merely with respect to the way speakers favor the result. Thus, one can imagine the following two
sentences describing the same situation, (i) by supporter of the strike and (ii) by its opponent:
(i) The decision of the party enabled the strike.
(ii) The decision of the party caused the strike.
2018 et seq. and this volume; Copley et al. 2015; Degand 2000; Johnston 1994;
Kadmon & Landman 1993: 389–398; Maienborn & Hertfelder 2015, 2017; Solstad
2010; Sweetser 1990).
(4) a. [e The kids danced] because of [c the music].

b. [e The door opened] because of / from [c the wind].
c. [e She lost this case] because of [c the witness’ death].
d. [e She died] from [c drinking too much water].
e. [e The kids danced] because [c they were happy].
f. [e You are biting your thumb at me] because [c you want to insult me]
(Davidson 1963: 688).
(5) a. [e ha-delet niftexa] biglal / me- [c ha-ruax].

The-door opened because / from the-wind
b. [e hi meta] biglal / me- [c štiyat mayim].
She died because / from drinking water
The conjunction because, as noted earlier, can also indicate reasoning, as is
the case in (4f). An explanation of an intentional action in terms of its motives
and reasons is different from expressing a causal relation. Together with what has
been exemplified in (2), clearly the connective because does not denote a causal
relation stricto sensu. Indeed, various philosophers have noted that because is the
preliminary way to convey grounding dependencies (see Schneider 2011; Correia &
Schnieder 2012: 22–24, Schaffer 2016: 84, Skow 2016).24 However, the linguistic
literature often includes it among the causal expressions and analyzes it as such (see,
for example, Charnavel this volume and related work). Nevertheless, the preposition
from has been often taken to be the ultimate linguistic means to introduce the
cause in a relation between entities (Alexiadou et al. 2006, 2015; also this volume;
Ahdout this volume). Presumably, this is related to the more restricted distribution
of from-PPs, in comparison to the connective because, being mainly attested with
verbs lacking an agentive or causative external argument such as unaccusatives and
statives.
Differences in meaning between the two connectives have been discussed by
linguists (Maienborn & Herdtfelder 2015, 2017), and are nicely revealed by the
asymmetry in the inference relations they give rise to, as demonstrated in (6):
(6) a. Maria is tired from the trip. ⇒ Maria is tired because of the trip.
b. Maria is tired because of the trip. Maria is tired from the trip.
24 It
is worth noting that Aristotle’s so-called four causes belong to various notions of reasoning
and explanation, and it has been noted that in fact he spoke about four becauses (see Vlastos 1969:
293ff.)
While with the connective because the tiredness of Maria can be related to a trip
she helped her partner prepare for, with from she must have participated in the actual
trip. These inference patterns are extendable to other languages as well.
A semantic analysis should account for these differences, and others to be
discussed throughout this paper. In light of question B, it is reasonable to entertain
the possibility that these differences have a bearing on the question of the unitary
concept of causal relations expressed by linguistic causative constructions, namely
on the nature of D in the various constructions. This issue will be systematically
considered in Sects. 1.3, 1.4 and 1.5.
1.2.2.3 Lexical Causatives
This category consists in constructions with verbal predicates, in which the subject
is perceived as (part of) the cause responsible for bringing about the state-of-affairs
denoted by the VP, which in turn is conceived as the effect.
This type of constructions primarily features change of state verbs such as
open, kill, boil (Jackendoff 1972; Croft 1991; Rappaport Hovav & Levin 1991
et seq. among many others), together with change of location verbs and ditransitive
verbs: put, send (e.g. Gropen et al. 1989; Beavers 2011). Another relevant type of
constructions is resultatives such as hammer the metal flat in English (extensively
discussed by Levin & Rappaport Hovav 1991 et seq.; Bittner 1998; Kratzer 2005
and Levin this volume). This latter sub-group will not be taken up here.
Alongside verbs of change of state (7), we will consider also caused activity
verbs (8). Caused activity verbs are attested, to a limited degree, in English as well
(cf. Cruse 1972 for a brief discussion),25 but in this context, Modern Hebrew adds
another dimension with its so-called causative templatic morphology (see Doron
this volume), where a root can appear in a pair of templates, one of which increases
valency by adding a participant that may be conceived as CAUSE or implicated in
the CAUSE (rakad ‘dance’ vs. hirkid ‘make.dance’).26
(7) a. [c John / the wind / the key] [e opened the door].
b. [c ha-šaxen / ha-ruax / ha-mafteax] [e patax et ha-delet].
The-neighbor / the-wind / the-key opened ACC the-door
25 Here are the examples provided by Cruse 1972 (exx. 4–7) for caused activity verbs:
(i) John galloped the horse around the field.
(ii) John flew the falcon.
(iii) John worked the men hard.
(iv) John marched the prisoners.
26 In
this paper we set aside causation involving psychological predicates (Dowty 1979; Belletti
& Rizzi 1988; Pesetsky 1995; Arad 1999; Doron 2012, this volume; Ahdout 2016; Gaulan 2016;
Alexiadou and Anagnostopoulou this volume and related work).
(8) a. John (#the music) danced the kids to the other side of the room.
b. [c ha-šxena / ha-musika] [e hirkida et ha-yeladim].
the-neighbor.F / the-music dance.CAUSE ACC the-kids
As will become clear in the following sections, change of state verbs and caused
activity verbs should be analyzed separately, and we will examine in what sense
the addition of CAUSE entails a causal relation in each. As was clarified in the
introduction, we use the denotation (c) in an uncommitted manner. Similarly, in
the glosses, the templatic morpheme CAUSE indicates an operation on the verb’s
valency which is associated with the addition of (c). In fact, the morphological and
syntactic literature contains numerous discussions of whether there are morphemes
or syntactic heads whose role is to introduce a CAUSE(R), (c) in our terms, whereas
the piece of structure below it in the syntactic tree constitutes (e).
Here and in the next sections, our goal is to have a better understanding of the
nature of D, also when it is covertly expressed. Analogously to connectives, it has
been noted that assertions of sentences with change of state verbs entail the truth of
an equivalent sentence with the overt causative cause (9a), but an entailment in the
other direction does not necessarily hold (9b):
(9) a. John broke the window. ⇒ John caused the breaking of the window.
b. John caused the breaking of the window. John broke the window.
This asymmetry was accounted for by the observation that lexical change of state
causative verbs, unlike overt verbs, have an additional constraint of a direct causal
link between (c) and (e).27 This additional requirement can be the reason for the
contrast between the constructions, as demonstrated in (10a) and (10b) (Fodor 1970;
Katz 1970; Ruwet 1972; Shibatani 1976b; Levin & Rappaport Hovav 1995):
(10) a. *Sue broke the glass on Sunday, by heating it on Saturday.
b. Sue caused the glass to break on Sunday, by heating it on Saturday.
Several studies have recently shown that this dichotomy is not as strict as it was
believed to be, and in certain contexts lexical causation expresses indirect causation
as well (Bittner 1998; Danlos 2001; Neeleman & van de Koot 2012). How to capture
this additional requirement that creates the direct causation effect and whether it is
semantically or pragmatically encoded is an ongoing discussion (for a recent survey
and a novel account see Baglini & Bar-Asher Siegal forthcoming).
Moreover, change of state verbs occasionally describe state of affairs with zero-
change (or failed-attempt), especially with agent subjects. It has been observed that
in some languages this is a more widespread phenomenon than in others (Martin
2015, et seq. and see the review of the literature on this in Martin’s contribution to
this volume).
27 For a survey of the various characterisations for direct causation in the literature see Wolff (2003).
Typological studies often seek correlations between the type of the construction and the level of
directness of the causation (see Nedjalkov & Silnitsky 1973; Dixon 2000; Shibatani & Pardeshi
2002; see also Levin’s contribution this volume).
(11) John taught Mary how to iron sheets, but despite of the fact that she watched
him do it, she still doesn’t know how to. (adapted from Oehrle 1976)
Finally, lexical causative verbs, which realize their external argument as the
causer, have fueled a debate within linguistics as to the nature of the relata. At
this preliminary stage, we abstract away from the issue of whether the cause is an
individual, an event or a proposition, namely whether the causer must be conceived
as a “representative” of some underlying event or proposition at the level of the
semantic analysis of the causal relation (Fodor 1970; McCawley 1976; Dowty 1979;
Levin & Rappaport Hovav 1991 et seq.; Reinhart 2000, 2002; Doron 2003, this
volume; Pylkkänen 2008; Neeleman & van de Koot 2012). We return to this in Sect.
1.5.3.1.
In the next three subsections we turn to directly tackle the questions presented in
Sect. 1.2.1, by observing how selected meaning components in D are manifested in
the three causative constructions introduced in this section.
1.3 Counterfactuality
In what is probably the most famous comment on causal relations, Hume proposed
the following definition for causation:
We may define a cause to be an object, followed by another, and where all the objects
similar to the first are followed by objects similar to the second. Or in other words where, if
the first object had not been, the second never had existed (An Enquiry Concerning Human
Understanding, Of the Idea of Necessary Connexion, Part II).
Much attention was paid to the fact the Hume proposed here two different
definitions, and to why he believed them to be two formulations of the same one (“or
in other words”). Since Lewis (1973a, b), the second, counterfactual, definition “if
the first object had not been”, became the central component in the conceptualizing
of the causal dependency.28 It was taken to define the dependency relation between
(c) and (e) when the former is a cause of the latter. In other words, in such cases it
can be stated that (e) could not have occurred without (c) – causa sine qua non.
Despite several known problems, such as cases of transitivity and preemption,
counterfactuality still stands as a major component of most contemporary depen-
dency approaches (see for instance Hall 2004; Kment this volume). This definition
was well established for centuries, and Lewis’ (1973a, b) main contribution is the
proposal to conceptualize counterfactuality with possible worlds, and the relation of
comparative similarity between them (cf. Von Wright 1968: 43–45).
28 While for Lewis, causation should be reduced to counterfactual terms, there is a strong
philosophical and linguistic opinion that the relation holds in the opposite direction, as causal
notions should figure in a semantic account of counterfactuals (see inter alia Veltman 2005; Schulz
2011; Bjorndahl & Snider 2015 and Kment this volume).
As observed earlier, semantic analyses for causative constructions often take for
granted that counterfactuality is a component in their meaning. This is probably
the most prevalent influence of the philosophical literature on formal studies of
these constructions. It is, therefore, only natural to begin our semantic journey in
examining whether counterfactuality indeed emerges as a meaning component in
causative constructions.
Phrasing this formally, when we mark the various construction as pcDe , we ask
the following questions with respect to each one of them:
(12) a. When the relevant pcDe is true, is the counterfactual claim
necessarily true?
Dc : pcDe ⇒ (∼c → ∼e)
b. Does counterfactuality exhaust the semantic content of D?
This section will be dedicated to question (12a), and subsequent sections will
take issue with answering various aspects of (12b).
Consider first the connective because. Causal statements with because do not
necessarily convey counterfactuality, as in example (13a), where (e) is negated.
(13) a. I did not go to France because of the rain / because it rained.
b. Had it not rained I would have gone to France.
In a situation where the speaker chose a destination for her vacation among a list
of cities, she can state (13a) as a reason for removing France from the list. In such
a scenario, (13a) does not entail (13b) as it is not the case that had it not rained, the
speaker would have gone to France, she may have not gone to France anyhow, as
her final choice was independently motivated. It must be noted, that (13a) can be
stated also in cases where there is a counterfactual relation (i.e., that if there was no
rain the speaker would have gone to France). The point here is that counterfactuality
is not necessarily entailed by the use of this connective, but it can be. We return to
this point in Sect. 1.5.3.2.
Because differs radically from overt causatives in this respect, where counterfac-
tuality necessarily holds with the verb cause.29
(14) a. The rain caused him not to go to France. ⇒
b. Had it not been raining she would have gone to France.
The causing part of overt causative verb, is a necessary condition,30 and counter-
factuality holds for the effect (cf. Eckardt 2000). Here are additional examples:
(15) a. The heat caused me to open the door. ⇒
b. If it weren’t hot, I would not have opened the door.
29 Nadathur & Lauer (2020), argue that the verb make is preferred in cases of preemption (in which
counterfactuality does not hold). It is beyond the scope of this paper to discuss the validity of their
proposal (see Baglini & Bar-Asher Siegal Forthcoming)
30 But it doesn’t have to be a sufficient condition (Lauer 2010; Nadathur & Lauer 2020).
(16) a. The recession caused Jerry to lose his home. ⇒

b. (Other things being equal,) if the recession had not happened,
Jerry would not have lost his home. (Lauer 2010: ex. 10)
Interestingly, there is no correlation between the grammatical category of the
construction and whether it has counterfactuality as a component of the D it encodes.
Namely, it is not a verb vs. connective distinction, since from-PPs contrast with
because of. Consider examples (17)–(18), where (e) is negated.
(17) She is not functioning from stress.

(18) hi lo metafked-et me-ha-laxac
She NEG function-F.SG from-the-stress
Here, contrary to example (13) above, (17)–(18) indeed entail that had she not
been under stress, she would have been functioning.
This lack of correlation can also be demonstrated in the realm of lexical causative
verbs: as they seem to pattern differently in this respect, according to whether they
encode a caused change of state or a caused activity. In the case of change of state
verbs, counterfactuality obtains (Von Wright 1968: 43–45; Dowty 1979):
(19) a. The baby opened the door. His mom pushed his hand over
the button that opens the door. ⇒
b. Had the baby not pushed the button, the door would not have
opened.
This entailment arises when we compare the state of affairs in the actual world
to a very similar world differing only by the fact that the mother did not push the
baby’s hand; it definitely does not entail that the mother would have not sought for
other ways to open the door, or in the case of (14), for example, that there could not
have been other motivations to go to France.
In contrast, counterfactuality does not necessarily hold when verbs express a
caused activity, even with a close similarity between worlds. Consider example (20):
(20) ha-zamar hirkid et ha-yeladim.

The-singer dance.CAUSE ACC the-kids
≈‘The singer made the kids dance.’
(20) can very well be uttered in a context where the kids started to dance
prior to the singing, and may imply various additional contextual meanings such
as the singer adding motivation, or intensity or the time of the actions. However,
crucially, none of these additional meanings are entailed by the content of the verb
itself, similarly to the connective because. The sentence in (20) therefore does not
necessarily entail that “had it not been for the singer, the kids would have not
danced”, as (20) can be stated if they where dancing before. It must be noted that,
here as well, (20) can be used in situations where counterfactuality is assumed to
hold. Our point is that this is not necessary, see Sect. 1.5.3.2.
Similarly, in English, (21) can describe a situation in which counterfactuality

does not obtain, specifically when the prisoners were marching in the prison’s
courtyard, and then John came and marched them some more by commanding them
to do so.
(21) John marched the prisoners. (Cruse 1972: ex. 7)

Lexically, what sets the two subclasses of verbs apart is (i) their lexical aspectual
properties: telic verbs encoding a result state, in the case of change of state
causatives, and activity or process verbs, without an encoded result state (see Martin,
Alexiadou & Anagnostopoulou this volume; see also Neeleman & van de Koot
2012, Levin & Rappaport Hovav 1991), and (ii) animacy of the direct object, namely
in the case of caused activity verbs, the direct object/causee is agent-like (see Nash
this volume, cf. Nadathur & Lauer 2020). The latter property has to do with the
possibility for a causa sine qua non to hold when volitionality is implicated. For an
elaboration on this contrast see Bar-Asher Siegal & Boneh (2019).
To summarize, the answer to question (12a) is that not all causative constructions
entail counterfactual dependencies between their (c) and (e). Sentences of both the
connective because and caused activity verbs, can be true even when their relation
cannot be paraphrased in counterfactual terms, although they often involve such a
relation. For discussion, see Sect. 1.5.3.2 (Table 1.1).
In terms of the broad goals of the paper, this section already makes clear that
not all causative constructions pattern alike, and that the presence or absence of
the counterfactual entailment cannot be correlated with a particular linguistic form
or type of marking. It also clarifies that the meaning components inherited from
philosophical analyses for causal relations, should be scrutinized more closely by
linguists, rather than automatically assuming their viability for semantic analyses of
causative constructions.
As for the question in (12b), whether counterfactuality exhausts the semantic
content of D, when we recall the differences between connectives exemplified
in Section (2.2.2), as well as the asymmetric entailment relation between lexical
causatives and overt causative cause - the former exhibiting an additional require-
ment of direct causation - evidently counterfactuality does not exhaust the content
of D.
Armed with these observations, we extend further the investigation of question
B, and turn in the next section to examine the relevance of various philosophical
accounts of Causal Selection to the semantics of the causative constructions. We
examine what type of cause is (c), or alternatively, how D is construed such that it
Table 1.1 Counterfactuality Dc : pcause cDe ⇒ (∼c → ∼e)

in causal constructions
Dc : pfrom cDe ⇒ (∼c → ∼e)
Dc : pchange-of-state cDe ⇒ (∼c → ∼e)
Dc : pcaused-activity cDe (∼c → ∼e)
Dc : pbecause cDe (∼c → ∼e)
establishes the nature of the relation of (c) to (e). In so doing, Sect. 1.4 continues to
focus on answering the question raised in (12b).
1.4 Singling out the Cause
When seeking to characterize the metaphysics associated with “the folk theory
of causation”, philosophers often rely on linguistic judgments. As we explore
additional semantic differences between causative constructions, this section sets
out to reflect on this methodology. It makes the point of considering whether some
of the data of the philosophical observations could have been different, had they
resorted to a different causative construction. The focus of this discussion, will
revolve around the topic of Causal Selection, to be introduced hereafter.
In practice, a standard methodology in the philosophical literature is to describe
a given scenario, and to ask, with respect to potential c(ause) and e(ffect), whether
it is possible to assert that “c is the cause of e”. Some discussions are careful to
distinguish between this type of judgment, with a definite article, and its indefinite
counterpart: “c is a cause of e”. Lewis (1973a, b: 162), for example, emphasizes
that his analysis of causation in terms of counterfactuality provides an account for a
cause and not for the cause. Similarly, for Mackie (1965), an INUS (=Insufficient
but Necessary/Non-redundant part of an Unnecessary but Sufficient) condition is
the characterization of a cause.31 The intuition behind the version of the causal
judgment with the definite article aims to further capture Causal Selection.
Causal Selection consists in teasing apart real causes and mere back-
ground/enabling conditions. Taking as an illustration the classic case of a burned
down house: while a house would not have caught fire if there were no oxygen in
the relevant space, as well as some flammable material, in this toy example, only a
discarded cigarette butt was The Cause of the fire.
Mill (1884, Volume I, Chapter 5, §3) introduced this distinction and it stood at the
heart of numerous discussions of philosophers, historians and legal theorists, who
tried to motivate the signaling of a condition as The Cause among various causal
factors (Einhorn & Hogarth 1986; Hart & Honoré 1959; Hesslow 1983; 1984; Hilton
1990; Mackie 1965, 1974; White 1965; Cheng & Novick 1991, inter alia). These
accounts made it clear that such selections cannot be motivated by characterizing the
dependency between (c) and (e) in terms of necessity and sufficiency, as causes and
conditions hold similar logical relationships to the effect. Therefore, the choices are
accounted for via other types of criteria, such as the normality of the potential causal
factors (for an overview, see the chapters of Statham & Hitchcock this volume), or
based on conversational principles, given assumptions about the state of knowledge
31 This is true for any account that emphasizes the intuition that causal relation is a transitive
relation.
and interests of the seeker of a causal judgment (Beebee 2004: 296 and Hitchcock
& Knobe 2009).
In light of the broader goals of this paper, we turn now to examine whether
linguistic causative constructions, with their binary division into (c) and (e), are
sensitive to select The Cause, and not mere background/enabling conditions or
causal factors.32
Given this background, the current section has a twofold goal:
a. To examine whether the constructions under discussion involve the selection
of The Cause.
b. To characterize the philosophical discussion on Causal Selection in linguistic
terms.
The first issue delves on question (12b) in Sect. 1.3, as to whether counterfactu-
ality exhausts the semantic content of D. The second targets question C presented at
the outset of the paper. In demonstrating this, we will concentrate on the following
issues:
First, we will use this discussion to clarify the semantic scope of the various
constructions, and check whether they describe other types of dependencies besides
causation (such as grounding, teleology and reasoning).
Second, Causal Selection involves a choice of The Cause among a set of
conditions. From a linguistic point of view, when these analyses focus on selection
among causes, they de facto aim to formulate the truth conditions of sentences of
the form: “C is the cause of E”. This is relevant regardless of what the right analysis
of causation is, and which philosophical analysis captures best what lies behind
people’s intuitions to see a causal relation in the world. In light of this, in order to
examine whether the semantics of a given causative construction involves a choice
of a salient cause, it is sufficient to test whether the proposition in this construction
entails an equivalent proposition, with the same (c) and (e), in the form of “C is the
cause of E”. This can be formally represented as follows:
(22) pcDe ⇒ “C is the cause of E”

The discussion hereafter demonstrates that only rarely is (c) of causative
constructions also necessarily The Cause.
32 Notably,previous semantic accounts, struggled with cases where these constructions are used,
but when (c) is not The Cause. They considered such cases as empirical reasons to doubt the
assumption that causative constructions encode causal relations (Abbott 1974; Dowty 1979;
Eckardt 2000). However, it is possible that it is only an indication that these constructions do
not involve the selection of The Cause.
1.4.1 Overt Causative Verbs
Consider first the following example with the overt verb cause, exemplifying that
(22) fails to hold:
(23) Context: John left the door open, a gust of wind came in and
shattered the window.
a. John caused the window to shatter.
b. John is the cause of the shattering of the window.
Clearly (c) in (23a), John, is not necessarily what we will intuitively designate as
The Cause of (e). Instead, given (23b), (c) is to be perceived as one of the conditions
that brings about (e), with often an additional flavor of responsibility attributed to
the selected cause - John.
A preliminary survey indicates that this account holds true of all the other overt
causative verbs, albeit with varying semantic nuances for (c) (cf. Wolff 2007).
The possible identification of subjects of sentences with overt causative verbs
as The Cause presents yet a more puzzling challenge. Let us consider a sentence
like (24), taken from Eckardt (2000). Pat cooks spaghetti every day when he returns
from work. On the specific evening described by (24), he cooked spaghetti late,
rather than at the regular time, and the reason for the late hour of the cooking was a
traffic jam.
(24) The traffic jam caused Pat’s cooking spaghetti late.

Note that while the cause of Pat’s cooking spaghetti is whatever set him off
on this daily custom, the cause for his late cooking of spaghetti is the traffic
jam. However, from a metaphysical point of view, and for theoretical linguistic
considerations, a reasonable assumption is that the event of cooking spaghetti and
the event of cooking spaghetti late are one and the same. The latter description of
the event adds a qualification. It is, therefore, puzzling that the entailment, expressed
in (24’), does not hold.
(24') a. The traffic jam caused Pat’s cooking spaghetti late.
b. The traffic jam is the cause of Pat’s cooking spaghetti, which was late.
These are cases known in the literature as containing fragile events (see Paul
2000 for an introduction of the topic in philosophical terms). In light of such
cases, Eckardt (2000) proposes to distinguish between two different uses of the verb
cause: Those which, in our terms, pass the test in (22), dubbed by her real causal
statements, whereas those which fail are termed pseudocausal statements. The latter
involve focus on a certain syntactic constituent. Thus, (24a) must be interpreted as:
the traffic jam is the cause of Pat’s cooking spaghetti late rather than on time.
If we wish to avoid polysemy for the verb cause, we can follow the proposal
that interpreting sentences with this verb always involves a contextual contrast
(cf. Achinstein 1976; Dretske 1977; Woodward 2003; Maslen 2004; Schaffer 2005,
2016; Northcott 2008). Such proposals argue that contrast is part of what defines
causal relations. One is left to wonder how contrasts constitute the metaphysical
notion of causation, in the sense that this is a characteristic of causal relations in
the world. Indeed, van Frassen (1980, Chapter 5) and others, as Woodward (1984),
relate resorting to contrasts to explanations and not to causal relations in the world.
van Frassen emphasizes the pragmatic factors in explanations, and stressing that
contrasts are determined by context. However, as noted by Hitchcock (1996), the
relationship between explanatory and causal claims are rather complicated, and it
is not trivial in what sense contrasts can be relevant only for explanations and not
for the causal claims themselves. This is, however, beyond the scope of the current
paper.33
This leads us to consider the nature of the discussion concerning Causal Selec-
tion. If we take the test in (22) seriously, selection of a cause is reflected in the
semantics of the sentence: “C is the cause of E”. The semantics of the focused
definite article, as noted by Eckardt, on the one hand, involves a choice of a
salient condition, as The Cause, and on the other hand, it triggers the denial of the
other conditions from being salient causes.34 This is a well-established linguistic
phenomenon and should be analyzed as such.
Beyond explaining the fact that sentences with the verb cause do not entail sen-
tences with The Cause, the significance of these observations may lead to identifying
a disciplinary confound, relevant for dealing with question C. The absence of infer-
ences between causative constructions challenges the naïve methodology of how
“the folk theory of causation” can be drawn from intuitions about causal judgments,
since there are meanings that might be associated with a specific construction, and
accordingly different “folk theories of causation” might be derived from different
constructions. Therefore, if causative constructions vary with respect to their truth
conditions, then intuitions about causal judgments differ from one constructions to
33 We thank Arnon Levy for bringing up this issue.

34 This discussion assumes that we agree with Eckardt (2000) that it is possible to distinguish
between causal- and pseudocausal statements, and in our account only the former pass the test
proposed in (22). The following, for example, is a causal statement:
(i) Dr. Spock’s first aid caused Joe’s heart to start beating again.
As it can be paraphrased:
(ii) Dr. Spock’s first aid is the cause for Joe’s heart to start beating again.
According to Eckardt, this is a causal statement, since it does not involve a denial of contextual
alternatives under focus. However, since even in her analysis actual phonological focus is not
required, it is possible to consider this sentence as also involving some denial of alterna-
tive/contrast, as is assumed by some philosophical accounts (Schaffer 2005, 2013):
(iii) Dr. Spock’s first aid caused Joe’s heart to start beating again, rather than not beating anymore.
Therefore, it is necessary to find some consistent way to distinguish between the two types of
contrasts. This seems to be related to what determines the identity of events, an issue which
is beyond the scope of this paper. In the context of comparing between the linguistic and the
philosophical literature, it is interesting to note on the similarity between van Frassen’s (1980)
discussion on the contrast-class, as a set of alternatives, and the work of Rooth (1992) regarding
the semantics of focal elements and the role of the contextual alternatives in their interpretations.
another, and thereby they do not necessarily reflect the conception of causation per
se, rather indicate the meaning of the specific types of constructions.
1.4.2 Connectives
As noted already in Sect. 1.2.2.2, unlike overt causative verbs, some connectives
can convey propositions that do not indicate causation at all, as in (25):
(25) Fractions are not even numbers or odd numbers, because they are not
whole numbers.
There is no temporal ordering possible between the two relata, since (25) states
a mathematical explanation. Thus clearly, such a construction does not necessarily
involve the selection of The cause. The question is, therefore, when it does express
causal relation, whether it slelcts such a salient cause. For this purpose, we examine
the application of the test proposed in (22). As a matter of fact, similarly to what we
saw with overt causatives (24), sentences with the connective because (26a) do not
entail sentences of the type illustrated in (26b):
(26) a. Pat is cooking spaghetti late because of the traffic jam.
b. The traffic jam is the cause of Pat’s cooking spaghetti, which was late.
To stress this point further, the following use of this connective emphasizes how
sentences with because do not mark the choice of a salient condition. Assume that
the doctor told Ann that, for her health, she should eat foods containing vitamin
C. Prior to the doctor’s appointment, Ann never ate such foods on a regular basis,
but due to the doctor’s recommendation, she decided to eat every day a different
type of food with vitamin C: Sunday – Guava; Monday – Broccoli; Tuesday – Kale;
Wednesday – Oranges. In this context, it is reasonable to say (27a), which does not
entail (27b), since The Cause is the requirements of vitamin C for her well-being.
(27) a. Ann ate broccoli today because it’s Monday.

b. The fact that today is Monday is the cause for her eating of
broccoli today.
We turn now to the connective from. Although it too, similarly to because,
covers cases included under the category of grounding as in (28) (cf. Maienborn
& Hertfelder’s 2015, 2017 stative reading of causation), interestingly, when from-
PPs express a causal relation as in (29), as far as we could observe, sentences with
this type of PP pass the test in (22). The entailment described in (29) seems to hold
firmly:
(28) The table is black from the ants. (Maienborn & Hertfelder 2015)
(29) A British woman died from drinking too much water while hiking. ⇒
Drinking too much water was the cause of her death.
In order to stress further the difference in this respect between from and because
further, let us consider the toy example with the burning house above, and the three
possible causes, all necessary, enumerated for the result to take place: the presence
of oxygen in the air, the house’s construction from flammable material and the
discarded cigarette butt, each with a different connective.
(30) a. #The house burned down from the oxygen in the air.
b. #The house burned down from the flammable material.
c. The house burned down from the discarded cigarette butt.
(31) a. The house burned down because of the oxygen in the air.
b. The house burned down because of the flammable material.
c. The house burned down because of the discarded cigarette.
It is therefore plain to see that because differs from from-PP in allowing just any
causal condition to appear in its complement position, whereas from is restricted
to the one condition that entails (22). We return to other peculiarities in the
construction with from-PPs in Sect. 1.5.
1.4.3 Lexical Causatives
In the case of lexical causatives with change of state verbs, the participant denoted
by the subject of the clause is intuitively qualified as The Cause. For example,
sentence (32a), stated without a specific context, seems to entail (32b):
(32) a. The baby opened the door. ⇒

b. The baby is the cause for the opening of the door.
However, when implemented in a broader context, (32a) does not entail (32b).
(33) a. After the mother pushed his hand over the button, the baby
opened the door.
b. The baby is the cause for the opening of the door.
While the baby is definitely a causal factor, or an enabling condition, it is still
not The Cause for the opening of the door. The standard judgment would be that the
action of the mother is The Cause, and the baby is more like an instrument. Consider
also (34):
(34) a. The key opened the door.

b. The key is the cause for the opening of the door.
Thus, the apparent entailment in (32) does not derive from the meaning of the
lexical causative. It is simply the case that often the subject of such sentences is also
the salient cause.
Finally, in the case of caused activity verbs a similar picture obtains. Considering
(35a), this sentence can be stated to describe a party in which the kids started to
dance as soon as there was music, and where at some stage of the party, there was
such a rhythm that made them jump even more. Under such circumstances, (35a)
does not entail (35b), since The Cause for the dancing can be argued to be the party
and the music in general:
(35) a. ha-kecev hikpic et ha-yeladim
the-rhythm jump.CAUSE ACC the-kids
‘The rhythm caused/made the kids (to) jump.’
b. ha-kecev haya ha-siba še-ha-yeladim kafcu
the-rhythm was the-cause that-the-kids jumped
‘The was the cause that the kids jumped.’
This last observation is not surprising. If the dependency of such verbs does not
necessarily involve counterfactuality, as was demonstrated in Sect. 1.3, then (c) in
this type of construction is not even a cause, as it cannot be construed as a necessary
condition, all the more so it would not be The Cause.
1.4.4 Summary
While causative constructions often describe situations in which (c) can be depicted
also as The Cause of (e), it is not necessarily so. Thus, for most constructions,
selection of the salient cause is not part of the truth conditions of D and are not
associated with its implicatures as well. Among our observations in this section, it
is worth repeating the following:
I. In some of the cases, (c) is merely an enabling condition or a causal factor; for
example, change of state verbs in (33)–(34), and the connective because.
II. (c) does not always indicate The Cause of (e), but a reason for some qualifica-
tion of the event/state denoted by the (e), as we saw in example (24).
III. (c) in various constructions represents the ground or an explanation and
therefore lies outside of the scope of causation, as in example (25).
IV. Different causal constructions have different truth conditions, some of them
require that (c) be a cause, and others, like the connective from-PP, require that
it be a or the salient condition.
Finally, the discussion at the end of Sect. 1.4.1 suggests that Causal Selection
is part of the meaning of some of the causative constructions and not others. Thus,
whatever motivates such selections should not have a bearing on the metaphysical
characterization of causal relations. One crucial outcome of this discussion is
the need to carefully distinguish between causal relations and the features of the
linguistic expressions that describe them. Another outcome is that in relying on
linguistic intuitions as indicators for “the folk theory of causation”, one must be
Table 1.2 Causal Selection Dc : pcause cDe (22)

in causative constructions
Dc : pbecause cDe (22)
Dc : pchange-of-state cDe (22)
Dc : pcaused-activity cDe (22)
Dc : pfrom cDe ⇒ (22)
aware that not all constructions have the same truth conditions. This constitutes the
basis for answering Question C (Table 1.2).
1.5 Causation Under Negation
Previous sections took as their starting point ideas that were developed in the
philosophical literature and examined their relevance to the understanding of
linguistic causative constructions. This section takes the opposite direction, as it
revolves around the linguistic phenomenon of negation – considering the various
interpretations causative constructions may have when interacting with sentential
and constituent negation. Studying the interpretation of these constructions under
negation is another way to grasp their meaning, since only that which is asserted
can fall under the scope of negation.
It may come as no surprise that the outcome of this consideration reaches the
same result as in previous sections: causative constructions do not pattern alike.
As before, we pay attention to whether the differences between the constructions
are related to their distinctive syntactic features, or whether they reflect differences
between the dependencies each construction encodes.
1.5.1 Negating the Dependency: D Entailed or Not?
Taking p to represent the entire relevant linguistic expression, namely, the entire
proposition with the relevant verbs and their arguments, or the connectors and their
relata, the question to be answered is the following: what is the relation between
p and the construct [c] D [e] underlying p in each type of causative constructions:
Does p assert the relation D expressed by [c] D [e], or does it presuppose it? Can D
be not-at-issue?
Prima facie, sentential negation indicates that the root-proposition p, the one
without negation, is false. Consequently, there can be two types of “truth-makers”
that falsify the root-proposition of the form pcDe : either (i) both the (c) and the (e)
took place, but there is no dependency between them [(c&e&∼(cDe)) => ∼pcDe ] –
if this is the case, then clearly D is asserted; or (ii) such a negative statement can
be true due to the fact that one of the members of the relata did not occur and then
[(∼c) or (∼e) => ∼pcDe ]. If D is not asserted, the first option should, therefore, be
unavailable.
We set aside readings where negation operates on a focused constituent, e.g. THE
KEY didn’t open the door, the card did.35
In constructions realizing D overtly, such as overt causative verbs and connec-
tives, the dependency is part of the assertion, and can be targeted by negation, as the
following examples demonstrate:
(36) a. [c The neighbor / the music] didn’t cause / didn’t make [e the
kids (to) dance].
b. [c ha-šxena / ha-musika lo garma [e la-yeladim lirkod].
the-neighbor / the-music NEG made to.the-children to.dance
(36) is true in a situation where there was music and the kids danced, and the
claim is that one didn’t induce the other. Similarly, (37a)–(38a), with connectives,
can describe the same state of affairs:
(37) a. [e The kids were not afraid] because of [c the wind].
b. [e The door didn’t open because of / from [c the wind].
(38) a. [e ha-yeladim lo paxadu] biglal / me- [c ha-ruax].

the-kids NEG be.afraid because / from the-wind
b. [e ha-delet lo niftexa] biglal / me- [c ha-ruax].
the-door NEG opened because / from the-wind
Since negation can capture any overt element in the sentence, it is not surprising
that in all of these constructions the morphologically represented D can be negated.
It is therefore interesting to examine whether this is true also when the expression
of the dependency is covert as in lexical causatives.
And indeed, in lexical causatives denoting a change of state, D cannot be targeted
by negation, namely ∼p cannot be interpreted as c&e&∼[cDe], rather ∼p only
entails ~(e) without reference to the D. Consider (39):
(39) a. John didn’t open the door.
b. The wind didn’t open the door.
Both sentences never describe a state of affairs in which John did the relevant
action or the wind blew, and the door is open, but nevertheless the door was open
35 In constituent negation, or in negation with focus, the causal relation can be de facto negated. This
is the outcome of several factors: (i) the definite expression, the key, comes with a presupposition
of existence; (ii) focus contributes the negation of the alternative propositions, with different (c)s to
the same (e), i.e. [not (card not open the door) = the card opened the door]. Since both (c) and (e)
hold, it is indeed only D that doesn’t. Moreover, it is possible to get contrastive readings without
focus (as Larry Horn informed us). This might be related to the fact that causality is often asserted
in the context of negating the contextual alternative-set (see the discussion earlier in Sect. 1.4.1).
For our purposes, we seek cases in which negation does not involve a clear case of affirmation of
one or more contextual alternatives.
due to some other factor or condition. The only available reading (again, without
focus in this sentence) is one where the result state is negated, namely, the effect
does not hold. In this case the door must be closed. Crucially, negation never targets
the claim that the dependency holds [cDe].
This is an interesting result. As we saw in Sect. 1.3, pchange-of-state cDe entails
counterfactuality (19), and thus D is part of its meaning. However, given that
D cannot be captured by negation it is then not part of the assertion, nor is it
presupposed, since it does not project under negation. It seems therefore plausible
to suggest that in this construction counterfactuality arises as a Conventional
Implicature, since it is entailed but is non-cancelable. In Sect. 1.5.2, we elaborate
further on negating caused change of state verbs.
As for verbs expressing caused activity, we saw earlier, in (20), that they do not
entail counterfactuality.
(40) ha-zamar lo hirkid et ha-yeladim.

The-singer NEG dance.CAUSE ACC the-kids
≈‘The singer did not cause the kids to dance.’
This sentence can be true either when the kids did not dance, or when there is no
dependency relation between the singer preforming the relevant action and the kids’
dancing, namely, the singer’s actions did not lead to the kids’ dancing.
In Bar-Asher Siegal & Boneh (2019), we propose a detailed analysis of the
causal semantics of the two sub-classes of lexical causative verbs and the way
they pattern under negation. Relying on the interplay between necessary and
sufficient conditions and the notion of potentially sufficient relevant for capturing
their meanings, they demonstrate that in fixed contexts, sentences such as (39)–(40)
actually presuppose some knowledge about the dependency relation (cDe), and that
this knowledge projects under negation, even if D, as described here, does not (see
below Sect. 1.5.3.2).
Several conclusions emerge from this short discussion:
I. Overt causative expressions assert [cDe], namely a dependency relation. This
is so even if this dependency is not strictly causal, i.e. when the connectives
because and from-PP realize a D that expresses other dependencies such as
grounding.
II. In change of state verbs the dependency D of the relation [cDe] cannot be
negated, thus it is not asserted. The counterfactuality meaning component of
D is a Conventional Implicature.
III. In caused activity verbs, negation may capture [cDe], where D does not entail
counterfactuality.
In Sect. 1.6, we present an additional type of causative construction – the
Affected Participant construction – in which D will be shown to be presupposed.
This section reaffirms what has started to emerge in Sect. 1.3, that the nature of
D is not monolithic, and varies in different ways from one construction to another
(Table 1.3).
Table 1.3 D is asserted (one pcause cDe ⇒ c&e&[cDe]

of the truth makers of ~p is
pbecause cDe ⇒ c&e&[cDe]
(c&e& ~[cDe]))
pfrom cDe ⇒ c&e&[cDe]
pchange-of-state cDe ⇒ [c&e]
c&e&~[cDe] ⇒ ~pcaused activitity cDe
1.5.2 Negating the Dependents
In this section, we set out to examine dependencies in which at least one of the
participants of the relata is negative [(~c)D(e)] or [(c)D(~e)].
(41) a. [c NEG taking the medicine] D [e her death]

b. [c breaking the key in the lock] D [e NEG the door open]
Assuming that the possibility of negating (c) and (e) is indicative of predication at
some level of the linguistic representation, (c) and (e) should then denote an event /
a proposition / an instantiation of a property, rather than an individual. Accordingly,
at least prima facie, the possibility to negate the constituents that are taken to
instantiate the relata allows us to advance the discussion of their nature.
In this respect, it is necessary to note that philosophers disagree as to the
availability of absence as a cause. However, even philosophers who deny that
absence can be a cause, admit that we often explain causal relations with the non-
occurrence of certain events (cf. Lewis 2004; Beebee 2004; McGrath 2005).
We hypothesize that the possibility to negate either (c) or (e) suggests that
causation necessarily holds between events, conceptually, but in many cases also
linguistically.
In what follows, we start examining this hypothesis, by passing under review the
various causative constructions, observing whether (c) or (e) fall under the scope of
negation, be it constituent negation or clausal negation, completing the picture from
Sect. 1.5.1. We substantiate this hypothesis further in Sects. 1.5.3.1 and 1.6.
Starting with overt causatives, it can be observed that when (c) or (e) denote
events, constituent negation is available:
(42) a. His not standing still caused the window to open. [~c] D [e]
b. Her not drinkig water caused her to die.
c. i-kibuy ha-eš garam la-mayim lirtoax.
NEG-turning.off the-fire caused/made to.the-water boil.INF
‘The non-turning off of the fire caused / made the water (to) boil.’
(43) a. His standing still caused the window not to open. [c] D [~e]
b. Her drinking water caused her not to die.
c. kibuy ha-eš garam la-mayim lo lirtoax.
turning.off the-fire caused/made to.the-water NEG boil.INF
The turning off of the fire caused / made the water (to) boil.
In comparison, to a certain extent, the application of constituent negation is

possible also with lexical causatives, but it is not as freely available as with
overt ones. Examples (44a-b) illustrate the variability of application of constituent
negation on (c).
(44) a. (*i-)kibuy ha-eš hirtiax et ha-mayim.
NEG-turning.off the-fire boiled ACC the-water
b. i-kibuy orot me’ir yeladim ba-lyla.
NEG-turning.off light wakes.up children in.the-night
‘The non-turning off of the lights wakes up children at night.’
c. His not giving-up smoking killed him.
It seems, therefore, reasonable to seek for a semantic characterization to account
for the availability of constituent negation with lexical causatives, this is, however,
beyond the scope of the current paper.
Now, while (c) and (e) can be negated via constituent negation, only in causative
constructions with connectives, relata can fall under the scope of sentential negation,
without also negating D. In the case of the connective because (of), for example,
clausal negation regularly induces two possible readings. In one of them, it can
apply to (e) alone (cf. Jespersen 1917: 47; Lakoff 1970; Johnston 1994; Kadmon &
Landman 1993):
(45) She didn’t lose this trial because of the witness’ death.
i. ‘It’s not the case that she lost this trial because of the ~[[c] D [e]]
witness’ death.’
ii. ‘The witness’ death is the cause of her not losing this [c] D [~e]
trial.’
(46) hi lo meta biglal ha-trufot.
She NEG died.F.SG because (of) the-medicines
i. ‘It is not the case that she died because of the medicine.’ ~[[c] D [e]]
ii. ‘Due to the medicine she did not die. (had she not
taken the medicine should would have died)’ [c] D [~e]
In contrast to because, in the case of from, the following examples from English
(47) and Hebrew (48), demonstrate that negation has only wide scope over the
proposition.
(47) She didn’t die from the medicine.
(48) hi lo meta me-ha-trufot.

She NEG died.F.SG from-the-medicines
i. ‘It is not the case that she died from the medicine’. ~[[c] D [e]]
ii. ‘#The medicine was the cause of her survival.’ #[c] D [~e]
However, the following sentences are fine with negation scoping under from-PP:
(49) hi lo barxa me-ha-paxad
She NEG run.away.F.SG from-the-fear
i. ‘It is not that case that she ran away out of fear’.
ii. ‘Fear caused her not to run away, to stay put.’ [c] D [~e]
(50) hi lo metafkedet me-ha-laxac

She NEG function.F.SG from-the-stress
i. ‘It is not the case that she is functioning due to stress’.
ii. ‘Stress causes her to be dysfunctional.’ [c] D [~e]
(51) The window didn’t open from the wind.

i. ‘It is not the case that the wind opened the window.’
ii. ‘The wind prevented the opening of the window’ [c] D [~e]
We contend that the possibility for negation to scope low in from-PP construc-
tions, targeting (e), depends, at least in some cases, on what the normal state of
affairs is. In (48), one is normally taken to be alive, and dying is the deviation from
the norm, but in (49) the normal state of affairs is not to run away, and in (50) the
normal state of affairs is to function.36 This can also be the case with windows that
in their default position are closed (51). We leave this issue for further research.
At this point, it is enough to repeat what has been mentioned in Sect. 1.4, that a
causal factor in Causal Selection has often something to do with deviation from the
norm. This is another case in which the causative construction with from-PP exhibits
additional requirements with respect to what can be part of the relata.
Syntactically speaking, the make-up of causative constructions with connectives
is such that negation can be interpreted with two different scopes.37 This is
presumably due to the fact that the PP with because/from, is an adjunct that can
36 Anadditional factor for the availability of a local negation with connectives seems to be lexical.
Consider the following pair in Hebrew featuring the connectives merov vs. mitox:
(i) ha-delet lo niftexa me-rov laxac. [c] D [~e]

The-door NEG opened.F.SG from-abundance pressure
i. ‘The door did not open due to the pressure on it.’
ii. ‘It is not the case that the door opened from the pressure.’
(ii) ha-delet lo niftexa mi-tox laxac.
The-door NEG opened.F.SG from-within pressure
‘It is not the case that the door opened out of pressure.’
37 The conjunction since has only the narrow scope reading. Iatridou (1991: 81–90) relates this to
the fact that the content of the since-clause is presupposed. See also Charnavel’s paper this volume
about the differences between since-clause and because-clause.
have two attachment sites (Johnston 1994).38 Schematically, there are two scopal
options for the interaction between negation, a sentence and its adjunct component:
(52) a. [(because of X ) ~(P)]
b. ~[(P) (because of X)]
In contrast, sentential negation in overt causatives does not target only the
occurrence of (e) or (c). In (53), as illustrated in the previous section, D necessarily
falls under the scope of negation.
(53) He didn’t cause the opening of the window.
i. It is not the case that he caused the opening of the window ~[[c] D [e]]
ii. # He is the cause of the non-opening of the window. [c] D[~e]
iii. # He didn’t engage in an activity and as a result the [~c] D [e]
window opened.
As noted earlier, sentential negation, indicates that the root-proposition, which
asserts for a dependency (pcDe ) is false, and there can be two types of “truth-makers”
that falsify the root-proposition: either both the (c) and the (e) took place, but there
is no dependency between them [(c&e&~(cDe)) => ~pcDe ], or one of the members
of the relata did not occur [(~c) or (~e) => ~pcDe ]. Note that these two options
correlate with the contrast between sentences with a definite and an indefinite (e):
(54) a. He didn’t cause the opening of the window. [(c&e&~(cDe)) => ~pcDe ]
b. He didn’t cause an opening of the window. [(~e) => ~pcDe ]
This contrast results from the existential presupposition that comes with definite
expressions (Strawson 1950).
Lastly, we turn to consider lexical causatives negated by sentential negation, first
with change of state verbs. The discussion in Sect. 1.5.1 demonstrated that the
dependency in this causative construction is not asserted, hence there is only one
type of “truth-maker” that can falsify the root-proposition, the non-occurrence of
the cause or of the effect [(~c) or (~e) => ~pcDe ], in (55) it is (~e); in (56) it can
also be (~c).
(55) The baby didn’t open the door.
[possible state-of-affairs: The baby’s action did not lead to the
door to be open]
(56) Fire did not boil the water.

[possible state-of-affairs: There was no fire, and therefore it must not
have been the cause for boiling the water]
In verbs denoting caused activity, falsifying the root sentence can also be due to
the fact that the effect or the cause did not take place:
38 Horn(2018), following Jespersen (1917: 47), related the availability of the second reading (the
wide scope reading of the negation) to the broader phenomenon of Neg-first.
(57) hu lo hirkid et ha-yeladim.

He NEG dance.CAUSE ACC the-kids
‘It is not the case that he made the kids to dance, (as they didn’t dance).’
As we saw in Sect. 1.3, the root-sentences of caused activity verbs do not entail a
counterfactual relation between the (c) and the (e). The root sentence in (57) can
describe a state of affairs in which the subject merely did an action that could
have been sufficient to bring about the effect (Bar-Asher Siegal & Boneh 2019).
Thus, (57) can negate such a state of affairs, and can mean “he did not do whatever
was sufficient to make the kids dance.” De facto, since such claims assume the
occurrence of both (c) and (e), this state of affairs amounts to the negation of the
causal dependency itself [(c&e&~(cDe) => ~pcDe ].
Thus, contrary to connectives, where the syntax overtly allows putting two events
in relation, with overt causatives and most lexical causative verbs, negation cannot
directly target one of (c) or (e) disjoint from D. Oddly, in this respect caused activity
verbs and overt causatives pattern alike. Change of state causatives stand in stark
contrast to these two since they readily render available a negated (e) or (c).
The observations from this section raise once again the question of the degree
of uniformity underlying causative constructions. However, following the proposed
analysis, the main difference between the constructions is accounted for by the
way sentential negation interacts with various syntactic components of the causative
constructions. Other differences stem from semantic factors such as those pointed
out with respect to negating (e) in from-PP constructions. This discussion reveals
the delicate interplay between semantic differences among constructions, and the
Ds encoded in them.
Furthermore, these observations pave the way to examining the hypothesis
formulated at the beginning of this section, whereby:
(58) The possibility to negate one of (c) or (e) suggests that causation
necessarily holds between events.
We turn to this in the next sub-section.
1.5.3 Discussion and outlook

1.5.3.1 The nature of the Relata
Following the observation that only adjuncts can scope above negation, we would
like to demonstrate that not all adjuncts interact similarly with sentential negation,
and in this way provide support for hypothesis (58).
Consider (59), where again, as with negation in connectives, either she wasn’t
the reason for Dani’s flying abroad, or she was the reason why he did not fly
abroad. In this respect, causative connectives pattern exactly like purpose clauses,
or beneficiary clauses. This is exemplified in (60), where, either Dani flew abroad
but not for her sake, or it was for her sake that he didn’t fly abroad. Thus, in both
examples, (at least) two possibilities exist to interpret the sentence: either she was
the cause / purpose of a negative event, or she wasn’t the cause / purpose of a positive
event.
(59) dani lo tas bigla-la lexul.
Dani NEG flew because.of-her abroad
i. ‘It was not the case that Dani had a flight abroad whose reason
was her (Dani either didn’t fly abroad, or she was not the reason
for his flying abroad).’
ii. ‘She was the reason he didn’t fly (he didn’t fly & it was because
of her).’
(60) dani lo tas bišvi-la lexul.

Dani NEG flew for-her abroad
i. ‘It was not the case that Dani had a flight abroad for her (Dani either
didn’t fly, or he flew but not for her).’
ii. ‘It was for her that Dani didn’t fly abroad (i.e. he stayed home for her
sake).’
These adjuncts will be dubbed Group 1 adjuncts, which are bi-eventive. Group
1 adjuncts radically differ from another group of adjuncts that cannot interact in
the same way with negation – Group 2. In example (61), the comitative phrase
cannot be severed from the underlying eventuality when negation is present; namely,
a reading where the adjunct escapes the scope of negation and only the underlying
eventuality is negated is not possible. Similarly, in (62), an instrumental adjunct
cannot escape the scope of negation. In both (61) and (62) the adjunct cannot be
added to the negated eventuality, and the ambiguity attested in (59)-(60) is not
available.
(61) dani lo tas it-a lexul.
Dani NEG flew with-her abroad
i. ‘It is not the case that Dani flew with her abroad (either he did not
fly, or he flew without her).’
ii. ‘#Dani’s not flying was with her.’
(62) dani lo axal suši be-mazleg.

Dani NEG ate sushi with-fork
i. ‘It is not the case that Dani ate sushi with a fork (either he didn’t
eat, or he ate sushi without a fork).’
ii. ‘#Dani’s not eating sushi was with a fork.’
Group 2 adjuncts pattern like Patients/direct objects. They too cannot escape the
scope of negation. In (63), the patient cannot be added to a negative event.
(63) dani lo pagaš / ra'a / hikir ota b-exul.
Dani NEG met / saw / knew her in-abroad
i. ‘It is not the case that Dani met / saw / knew her abroad.’
ii. ‘#Dani’s not meeting / seeing / knowing abroad was of her.’
In summary, the two groups of adjuncts and the possibility to scopally interact
with negation can be schematized as follows:
(64) a. Group 1: cause, purpose, beneficiary

i. ~[P + ADJUNCT]
ii. ADJUNCT ~P
b. Group 2: comitative, instrument, also patient
i. ~[P + ADJUNCT]
ii. #ADJUNCT ~P
It is reasonable to suggest that the two groups differ as to how the adjuncts
interact with the eventuality of the main predication: those of Group 2 add informa-
tion about the eventuality of the main predication, and instantiate relations between
individual denoting arguments; therefore, negation can only have wide scope over
the adjunct. In contrast, adjuncts of Group 1 introduce another eventuality and
express a dependency between the eventuality of the main predication and other
eventualities.
In other words, adjuncts like with-PP (61) behave similarly to an argument of
a predicate (63), as they add a participant to the same event, while “because of
someone” and “for someone” add either the (c) for the event described by the main
predication (59), or the (e) of the main predication (60).
Importantly, the idea underlying this set of facts is that while it is possible
to assert a dependency relation with the non-occurrence of an event (hence the
possibility to be above the scope of negation), it is meaningless to add information
about an event which did not take place.
Moreover, the significance of this discussion is that even if, prima facie, there are
no clear linguistic indications that causative constructions involve events, there are
linguistic facts that require the assumptions that the relata of causative constructions
are events (cf. Neeleman & van de Koot 2012). Accordingly, it is possible that the
conceptual and syntactic representations of the dependency relation expressed by
the causative construction are in fact dissociated. While at the syntactic level, the
(c) does not include an event, the NP/DP in such position must be a participant, as
was already proposed by Dowty (1979).
1.5.3.2 A Note About pbecause cDe and pcaused-activity cDe
At this point, one may wonder why pbecause cDe and pcaused-activity cDe are included
among causative constructions, as they do not even entail counterfactuality. It should
be kept in mind though, as clarified in the introduction, that we defined causative
constructions in a schematically broad manner, enabling a multifaceted examination
of the part of the causative construction cDe, where D does not necessarily involve
one particular type of dependency.
Thus, although in many cases, pbecause cDe and pcaused-activity cDe do not obligatorily
entail typical causal relations, they do describe such relations, including counter-
factuality, which is assumed to hold between (c) and (e) in many contexts. In this
paper, we did not provide a full account of the semantics of the constructions, but
we would like to elaborate somewhat on them, and on why they most often express
counterfactual dependencies.
As noted in the philosophical literature, pbecause cDe answers all types of “why
questions” (inter alia van Frassen 1980), thus it provides answers in the broad sense
of “reasoning”. Among the means at our diposal for reasoning about situations /
events / facts, it is very common to provide causal explanations. Therefore, it comes
with no surprise that this construction describes causal relations as well.
As for pcaused-activity cDe , clearly it often does not assert that it is de facto (c) which
brought about (e), as this construction can describe situations in which (e) precedes
(c). Bar-Asher Siegal & Boneh (2019) argue that in this construction, D triggers
the presupposition that (c) is a potential sufficient condition for (e), and propose the
definition of sufficient conditions in (65):
(65) {c | ~Oe ~Oc}

This definition expresses the fact that part of the lexical content of a given lexical
causative verb consists in the type of events which are sufficient for bringing about
the result described in the given (e). Thus, (65) states, for a causative verb in a
given context, the presupposed knowledge of types of events (c), such that the non-
occurrence of (e) necessitates their non-occurrence as well. In other words, (c) must
have the potential of being a sufficient condition for the (e). It is not surprising,
therefore, that more often than not, this construction is used when (c) is de facto
the (c) which brought about (e). For more details, see Bar-Asher Siegal & Boneh
(2019).
1.6 Covert Causation: Affected Participant Constructions
We wish to conclude this paper by adding another type of causative construction –

one which features an Affected Participant.
It has been claimed with respect to various constructions that they involve a
participant affected by the event described in the clause (for example, O’Connor
2007, and more broadly Beavers 2011). Since the idea that an event participant is
affected by some occurrence implicates some notion of causation, it is only natural
to consider this type of constructions as part of the current discussion. In addition,
this inquiry demonstrates the significance of the observations from the previous
sections to a broader range of linguistic phenomena.
1.6.1 General Properties
In the construction under discussion, the Affected Participant is added to a clause

either via a datival expression (the preposition l- in Hebrew, see also Hole 2005,
2006), or with a prepositional element like on in English. In this construction, we
claim, the dependency holds between the expressed eventuality and a contextually
determined one. This dependency has already been analyzed as having a causative
component of meaning (cf. Bosse et al. 2012; Bar-Asher Siegal & Boneh 2015).
Consider the following attested example from Modern Hebrew, where the
relevant clause containing the affected dative marked participant is found in an
embedded interrogative:
(66) od Eli Zohar lo yaxol liško'ax ex [c met [e l-o] pa'am ed
Att. E. Z. NEG can forget how died to-him once witness
be-'emca xakira negdit]
in-middle investigation cross
‘Attorney Eli Zohar cannot forget how a witness once died on him during a cross
investigation.’
This example conveys that the death of the witness during the cross investigation
caused Attorney Eli Zohar to lose the trial. He was unable to win the case due to
the witness’ premature death. Note that the particular effect of the concrete and
psychological damage caused by the witness’ death is accommodated and gathered
from context and world knowledge. Similar content is expressed in English with
the introduction of an additional participant following the preposition on (cf. Bosse
2015).
(67) [c The old bugger (went and) died] [e on me].

Schematically, the causal relation can be depicted in (68):
(68) [c The old bugger/the witness died] D [e-context Attorney Eli Zohar cannot win the
trial/the situation made it difficult for attorney Eli Zohar to win the case]
While the precise nature of the effect is determined contextually in both

languages, whether the effect is positive or negative is determined contextually in
Modern Hebrew, but lexically in English, as the preposition on seems only to give
rise to a negative effect on the added participant. In both cases, D is implicit.
Consider an additional example, where the relevant clause, schematized in (69’),
is overtly part of a conjunction:
(69) axarkax hu tas l-i le-šana la-mizrax,

Then he flew to-me to-year to.the-east,
ve-hiš'ir oti xareda ve-lexuca
and-left me anxious and-stressed
(69') [c He flew to the Far East for a year] D [e-context I am anxious and distressed]
The precise nature of the effect in (69) is clarified by the conjoined clause
indicating that the effect can be psychological, not only a material one.
In what follows, we examine the properties of D in this particular construction,
along the lines of the investigation for the other constructions presented in Sects.
1.3, 1.4 and 1.5.
1.6.2 Counterfactuality Under Negation
While in the other constructions, D was either asserted or implied (cf. Sect.
1.5.1 above), in this construction D is presupposed, as it projects under negation.
Interestingly, projection under negation, which is a prominent feature of presuppo-
sition, allows a clear access to the content of D, in this case. Furthermore, in this
construction, it can give rise to counterfactuality transparently. Let us consider in
turn (70) and (71), in Modern Hebrew and English, respectively:
(70) ha-ed lo met l-i.
the-witness NEG die to-me
‘It’s not the case that the witness died’. [implied: Had he died, I
would have been affected, e.g. by loss of reputation]
(71) The old bugger didn’t die on me.
‘It is not the case that the old bugger died.’ [implied: had he died I
would have been affected by e.g. sadness, sense of loss]
Under negation, the only contribution of adding the datival expression li, in
(70), and of the PP on me, in (71), is the counterfactual implication, which is not
explicitly phrased, but implied from the negative sentence. In comparison, consider
the following equivalent negated sentences without the affected participant, where
the counterfactual inference is absent:
(70') ha-ed lo met.
the-witness NEG die
‘The witness didn’t die / It’s not the case that the witness died.’
#Had he died, . . . .
(71') The old bugger didn’t die.
It is not the case that the old bugger died. #Had he died, . . . ..
This counterfactual element can be accommodated, if we assume the following
two components for this construction:
(72) a. D represents a counterfactual dependency.
b. D is presupposed, and as such projected under negation.
Bar-Asher Siegal & Boneh (2015), therefore, propose (73) as the semantic
representation of the Affected Participant in Modern Hebrew, and presumably also
on in English, which describes D in terms of a relation between events:
(73) [[AP]]= λe.λe’.λF.λG.λx:(~Fe → ~O Ge’)=1.Fe & Participant(x, e’)=139

In this formula:
a. Fe is an abbreviation for everything which the underlying sentence without the
datival expression states to be true about the event it describes (in (59), the
witness’ dying). This is (c) of the causal construction.
b. Ge’ is a description of the effect, which is the relevant state of affairs known or
given by the context (contextual knowledge is indicated with O ). The DP within
the PP (be it le- or on) is a participant in the eventuality that is the effect.
c. (~Fe → ~O Ge’) captures counterfactuality: if Fe takes place, Ge must take
place as well; moreover, if Fe does not hold, Ge must not either. Thus, when
it is asserted that Fe did not take place, the counterfactual claim surfaces as part
of the meaning. This dependency is presupposed and not asserted.
Given all this, even when the underlying clause is negated: ~Fe, (~Fe →
~O Ge’) holds true. This is the origin of the counterfactual reading under negation
demonstrated in (70)–(71). Therefore, Affected Participant constructions are an
example where counterfactuality is directly relevant for the meaning of the linguistic
expression in question.40
1.6.3 Negating the Relata
Still in the realm of negation, we move on to consider the properties of negating the
relata of D in this construction. It emerges that in Modern Hebrew, the construction
displays the mirror image of connectives, as clausal negation can target (c). Consider
the following examples, where (74) repeats (70) and shows that the Affected
Participant patterns like Group 1 adjuncts, allowing negation to scope under it:
(74) Context: said by a gangster facing imprisonment

lo met l-i ha-ed.
NEG die to-me the-witness
i. ‘It is not the case that the witness died on me.’ ~[[c] D [e]]
ii. ‘The witness’ not dying affects the speaker (e.g. he is not
acquitted).’ [~c] D [e]
Here is an additional example with its context:
39 Bar-Asher Siegal & Boneh (2015) also added the presupposition of precedence in time (e≤e’)
between the eventualities, a presupposition customarily related to causation (see also fn. 41 below).
40 Nothing in the morpho-syntactic constitution of this constructions indicates at first blush
that this is a causative construction, however one may want to consider in this context the
applicative/causative syncretism in Indonesian languages (see Kroeger 2007).
(75) Context: Danny is sitting in a speeding car, which does not make a stop when
it should, putting his life in danger:
hu lo acar le-dani be-adom.
he NEG stop to-Danny in-red (light)
‘He did not stop at the red light for/on Danny.’
i. ‘It is not the case that he stopped at the red ~[[c] D [e]]
light on/for Danny.’
ii. ‘The non-stopping at the red light affected [~c] D [e]
Danny.’ (e.g. he was scared)
Interestingly though, the availability of two scopes for negation in Hebrew
Affected Participant constructions is not shared by English on-PP construction,
where clausal negation cannot target (c):
(76) The old bugger didn’t die on me.

i. It’s not the case that the bugger died on me. Had he died, I would
have been affected by e.g. not knowing how to handle the estate.
ii. #The witness’ not dying affects the speaker [e.g. said by a
gangster facing imprisonment].
This cross-linguistic difference may stem from a syntactic difference between
the two constructions, regarding the height of attachment of the two PPs. We leave
this unexplored for the time being.
Crucially, though, a clear contrast emerges between the Hebrew Affected Partic-
ipant, which is a non-selected dative and ditransitives featuring selected datives:
(77) hu lo natan matanot la-yeladim.

He NEG gave presents to.the-kids
i. ‘It is not the case that he gave presents to the kids.’
ii. #He did not give presents and it was to the kids / his not giving
presents was to the kids.
Ditransitives are caused change of location verbs or caused possession verbs (cf.
Rappaport Hovav & Levin 2008; Beavers 2011), and therefore fall under lexical
change of state causatives, extensively discussed above.
The contrast between the possibility to negate (c) brings us back to the discussion
in Sects. 1.5.2 and 1.5.3.1: adjuncts that denote an additional eventuality can
scope out negation, whereas direct arguments of the verb and adjunct of the same
eventuality cannot be outside of the scope of the negation. For our purposes, it is of
significance that the Affected Dative patterns with Group 1 adjuncts, which supports
the claim that it introduces another eventuality.
1.6.4 Concluding Remarks
To conclude the discussion of the Affected Participant construction in the framework

of the current paper, we should consider the issue of Causal Selection (Sect. 1.4).
Interestingly, like the preposition from, discussed in Sect. 1.4.2, Affected Participant
constructions demonstrate the type of entailment mentioned earlier in (22), and
indicate that part of the meaning of the D involves a selection of a salient condition
to construe (c). Thus, (69) entails (78) and (67) entails (79):41
(78) His flight to the Far East for a year is the cause for my anxiety and stress.
(79) The death of the old bugger is the cause for (my contextually determined)
discontent / loss of the trial.
Sect. 1.6, then, provided us with further confirmation to issues discussed
throughout this paper:
• There is no unique causal construal encoded across linguistic causative construc-
tions.
• Counterfactuality is a significant component in the meaning of the dependency
encoded in linguistic expressions. Affected Datives provide an example where
counterfactuality is transparently entailed, as a presupposition.
• The fact that this construction lacks overt marking raises the issue of the linguistic
origin of the causal content of the various constructions – an issue definitely
worth pursuing.
The next section concludes the discussion, incorporating the findings regarding
the Affected Participant construction with the others.
41 AffectedParticipant constructions in Modern Hebrew also feature inanimate affected partici-

pants, where the verb’s object and the datival expression hold a part-whole relation (see Bar-Asher
Siegal & Boneh 2014):
(i) nišbar la-šulxan ha-regel.
broke to.the-table the-leg
‘The leg of the table broke’ / ‘The table had a leg broken off of it.’
In this case, there is no separation between the breaking of the leg and the “effect” on the table,
namely its lacking a leg; that is, (e) is part of (c), and they constitute one and the same event, where
no temporal separation between (c) and (e) is possible. Thus, this is not a causal relation in the
strict sense. However, interestingly, it is appropriate to refer to (c) as The Cause, in this case too:
(ii) švirat ha-regel hi ha-siba le.xax še-eyn la-šulxan regel.
breaking the-leg is the-reason to.so that-NEG to.the-table leg
‘The breaking of the leg is the reason that the table doesn’t have a leg.’
(iii) švirat ha-regel garma le.xax še-eyn la-šulxan regel.
breaking the-leg caused to.so that-NEG to.the-table leg
‘The breaking of the leg is the reason that the table doesn’t have a leg.’
We leave this puzzle for a future study.
1.7 Conclusions
The last four sections provided a survey regarding various semantic aspects of six
causative constructions, the essence of which is now summarized in Table 1.4 below.
The discussion began with the question regarding the relevance of philosophical
accounts of causation for linguistic analyses of causal constructions (Sect. 1.1
Question A). Our survey demonstrated various discussions informed by ideas first
developed as a metaphysical analysis for causal relations, motivating the assumption
that question A should be answered positively, despite the rather limited goals of this
paper, which do not provide a complete analysis of the construction’s semantics.
Turning to the question of whether there is a one all-encompassing causative
meaning component underlying the diverse linguistic phenomena (Sect. 1.1, Ques-
tion B), first it became clear that these constructions do not necessarily encode
causal relations. All of them, besides perhaps the verb cause, are not exclusive for
the description of causal relations. Some denote also grounding relations and/or
express other logical relations. Even in terms of the relation between the (c) and
the (e) in these constructions, not all of them necessarily entail counterfactuality
(Sect. 1.3).
The table below summarizes the properties of D studied in the various construc-
tions throughout Sect. 1.3, 1.4, 1.5 and 1.6.
How should we characterize the differences between the causative constructions?
We may ask the following: do these differences necessitate some sort of causal
pluralism? More specifically, do the differences indicate substantial differences with
respect to the content of the D they denote? As we saw in Sects. 1.5.2, 1.5.3,
and 1.5.3.1, some of the differences can be accounted for syntactically. But, it is
not always clear how to account for the differences in this way, as constructions
are grouped together according to some semantic feature, without any apparent
correlation to their syntactic type.
Furthermore, although not all constructions demonstrate similar requirements as
to what can be selected as the cause among the set of conditions, due to Causal
Selection (Sect. 1.4), or due to additional requirements such as direct causation,
described in Sect. 1.2.2, it is still possible to see that counterfactuality plays a
central role in determining what can be a cause. Moreover, it became clear that
constructions differ as to whether the dependencies they represent are asserted,
conventionally implied or presupposed. It will be, therefore, important to understand
why the constructions are not the same in this regard, and what can motivate such
differences.
With respect to the nature of the relata, the discussion in Sects. 1.5.2 and 1.6
reveals that, despite the differences between the constructions as to the possibility
to independently negate (e) or (c), the fact that this possibility exists both in the
case of connectives, where D is overt, as well as in the case of Affected Participant
constructions, where D is covert and presupposed, and (e) is realized as an individual
(within a PP), suggests that at some level of the linguistic representation, the relata
are event-like, and not individuals, in alignment with the philosophical conception.
Table 1.4 Summary
pcause cDe pbecause cDe pfrom cDe pchange-of-state cDe pcaused-activity cDe PAffected-Participant cDe
Counterfactuality entailed ˛ ˝ ˛ ˛ ˝ ˛
p ⇒ (∼c → ∼e) Asserted Asserted Conventional Presupposed
implicature
D designates The cause ˝ ˝ ˛ ˝ ˝ ˛

p ⇒ C is the cause of E
D entails p⇒c&e&[cDe]. Therefore, ˛ ˛ ˛ ˝ ˛ ˝

available reading under negation:
∼p⇒c&e&∼[cDe] Indirectly (projects under
entailed negation)
Additional meaning component Direct Direct causation
1 Causation: From Metaphysics to Semantics and Back
causation /
implies
deviation from
the norm
Possibility to negate the relata (under ˝ ˛ ˛/˝ ˝ ˝ ˛

sentential negaton)
semantic
∼p available readings: constraint at
[(∼c)D(e)], or [(c)D(∼e)] work
45
Lastly, as for the philosophical discussions on causation and their sensitivity

to the linguistic aspects of the data they rely upon (Sect. 1.1 Question C), a
crucial result of this paper is the need to tease apart causal relations and the
features of the linguistic expressions that describe them. Furthermore, in relying on
linguistic intuitions as indicators for “the folk theory of causation”, it is important to
acknowledge that not all constructions have the same truth conditions. The semantic
differences between causative constructions impose challenges to a methodology,
common among philosophers, to rely on linguistic judgments in recognizing the
intuitions that constitute this “folk theory of causation”. As noted in Sect. 1.4, one
is left to wonder whether philosophical accounts could have been different, had they
referred to causative constructions other than those including overt causatives.
Acknowledgements We first wish to thank the participants in the reading group on Causation,
held at the Language, Logic and Cognitive Center at the Hebrew University of Jerusalem during
the academic year of 2016–2017. Many of the ideas in this paper stem from these meetings. We are
especially grateful to our partner in organizing the reading group - Arnon Levy - with whom we
engaged in many fruitful and enlightening conversations, and who also commented on an earlier
draft of this paper. We would also like to extend our gratitude to the participants in the Workshop
Linguistic Perspectives in Causation held at the Language Logic and Cognition Center, the Hebrew
University of Jerusalem in the summer of 2017, for a stimulating and truly interdisciplinary
encounter. We are indebted to the participants of the seminar on Causative Constructions held
at the Hebrew University, in Fall 2018, for their important questions and feedback. For their
generosity in accepting to read earlier drafts of this paper and for their insightful comments, we
are ever thankful to Rebekah Baglini, Noa Bassel, Larry Horn and Anne Temme. Research on this
paper was supported by the Volkswagen Stiftung, “Forschungskooperation Niedersachsen-Israel”
for the project “Talking about causation: linguistic and psychological perspectives” given to the
authors and to Prof. Dr. York Hagmayer (U. of Göttingen), who acquainted us with the cognitive-
psychological research on causation.
References
Abbott, B. (1974). Some problems in giving an adequate model-theoretic account of CAUSE. In C.

Fillmore, G. Lakoff, & R. Lakoff (Eds.), Berkeley studies in syntax and semantics (BS3) (Vol.
I, pp. 1–14). Berkeley: Department of Linguistics and Institute of Human Learning, University
of California.
Achinstein, P. (1976). Causation, transparency, and emphasis. Canadian Journal of Philosophy, 5,
1–23.
Ahdout, O. (2016). The syntax-semantics interface in Hebrew Psychological Nominalizations. MA
thesis, The Hebrew University of Jerusalem.
Alexiadou, A., Anagnostopoulou, E., & Schäfer, F. (2006). The properties of anticausatives
crosslinguistically. In M. Frascarelli (Ed.), Phases of interpretation (pp. 187–211). Berlin/New
York: Mouton de Gruyter.
Alexiadou, A., Anagnostopoulou, E., & Schäfer, F. (2015). External arguments in transitivity
alternations: A layering approach. Oxford Studies in Theoretical Linguistics.
Anscombe, G. E. M. (1981). The collected philosophical papers of G.E.M. Anscombe: v. 2.
Metaphysics and the philosophy of mind. Minneapolis: University of Minnesota Press.
Arad, M. (1999). What counts as a class? The case of psych-verbs. MITWPL, 35, 1–23.
Baglini, R., & Bar-Asher Siegal, E. A. (Forthcoming). Direct causation: A new approach to an old
question (U. Penn Working Papers in Linguistics 26).
Baglini, R., & Francez, I. (2016). The implications of managing. Journal of Semantics, 33(3),
541–560. https://doi.org/10.1093/jos/ffv007.
Bar-Asher Siegal, E. A., & Boneh, N. (2014). Modern Hebrew non-core dative in their context.
L@šonénu, 76, 461–495. [in Hebrew]
Bar-Asher Siegal, E. A., & Boneh, N. (2015). Non-core datives in Modern Hebrew.
Proceedings of the 30th annual conference of the Israel Association for Theoret-
ical Linguistics. http://www.iatl.org.il/wp-content/uploads/2015/10/IATL30proceedings-01-
Bar-Asher-Siegal_Boneh-.pdf
Bar-Asher Siegal, E. A., & Boneh, N. (2019). Sufficient and necessary conditions for a non-unified
analysis of causation. In R. Stockwell, M. O’Leary, Z. Xu, & Z. L. Zhou (Eds.), Proceedings of
the 36th west coast conference on formal linguistics (pp. 55–60). http://www.lingref.com/cpp/
wccfl/36/index.html.
Beavers, J. (2011). An aspectual analysis of ditransitive verbs of caused possession in English.
Journal of Semantics, 28, 1–54. https://doi.org/10.1093/jos/ffq014.
Beebee, H. (2004). Causing and Nothingness. In J. Collins, N. Hall, & L. Paul (Eds.), Causation
and counterfactuals (pp. 291–308). Cambridge, MA: MIT Press.
Belletti, A., & Rizzi, L. (1988). Psych-verbs and theta-theory. Natural Language and Linguistic
Theory, 6, 291–352.
Bittner, M. (1998). Concealed causatives. Natural Language Semantics, 7, 1–78.
Bjorndahl, A., & Snider, T. (2015). Informative counterfactuals. Proceeding of Semantics and
Linguistic Theory (SALT), 25, 1–17.
Bosse, S. (2015). Applicative arguments: A syntactic and semantic investigation of German and
English. Peter Lang Academic Publishing.
Bosse, S., Bruening, B., & Yamada, M. (2012). Affected experiencers. Natural Language &
Linguistic Theory, 30, 1185–1230.
Carroll, J. (2009). Anti-reductionism. In H. Beebee, C. Hitchcock, & P. Menzies (Eds.), 279–298.
The Oxford Handbook of Causation: Oxford University Press.
Charnavel, I. (2018). Perspectives in causal clauses. Natural Language and Linguistic Theory, 36,
1–36.
Cheng, P. W., & Novick, L. R. (1991). Causes versus enabling conditions. Cognition, 40, 83–120.
Comrie, B. (1981). Language universals and linguistic typology. Oxford: Blackwell.
Copley, B., & Wolff, P. (2014). Theories of causation should inform linguistic theory and vice
versa. In B. Copley & F. Martin (Eds.), Causation in grammatical structures (Oxford studies
in theoretical linguistics) (Vol. 52, pp. 11–57). Oxford: Oxford University Press.
Copley, B., Wolff, P., & Shepard, J. (2015). Force interaction in the expression of causation. In S.
D’Antonio, M. Moroney, & C. R. Little (Eds.), Proceedings of the 25th semantics and linguistic
theory conference (pp. 433–451).
Copley, B., & Harley, H. (2015). A force-theoretic framework for event structure. Linguistics and
Philosophy, 38(2), 103–158.
Correia, F., & Schnieder, B. (2012). Grounding: An opinionated introduction. In C. Fabrice & B.
Schnieder (Eds.), Metaphysical grounding: Understanding the structure of reality (pp. 1–36).
Cambridge: Cambridge University Press.
Croft, W. (1991). Syntactic categories and grammatical relations: The cognitive organization of
information. Chicago: University of Chicago Press.
Cruse, D. A. (1972). A note on English causatives. Linguistic Inquiry, 3, 522–528.
Danlos, L. (2001). Event coreference in causal discourses. In P. Bouillon & F. Busa (Eds.), The
language of word meaning (pp. 216–242). Cambridge: Cambridge University Press.
Davidson, D. (1963). Actions, reasons and causes. Journal of Philosophy, 60, 685–700.
Davidson, D. (1967). Causal relations. The Journal of Philosophy, 64(21), 691–703.
Davidson, D. (1969). The individuation of events. In N. Rescher (Ed.), Essays in honor of Carl G.
Hempel (pp. 216–234). Dordrecht: D. Reidel.
Davidson, D. (1980). Essays on Actions and Events. Oxford: Oxford University Press.
Degand, L. (2000). Contextual constraints on causal sequencing in informational texts. Functions

of Language, 7, 33–56.
Dixon, R. M. W. (2000). A typology of causatives: Form, syntax and meaning. In R. M. W.
Dixon & A. S. Aikhenvald (Eds.), Changing valency: Case studies in transitivity (pp. 30–83).
Doron, E. (1999). The semantics of transitivity alternations. In P. Dekker (Ed.), Proceedings of
the Twelfth Amsterdam Colloquium (pp. 103–108). Universiteit van Amsterdam: Institute for
Logic, Language and Computation.
Doron, E. (2003). Agency and voice: The semantics of the Semitic templates. Natrual Language
Semantics, 11, 1–67.
Doron, E. (2012). The causative component of psych verbs. Paper Presented at the Roots III
Workshop, Jerusalem, June 2011. Also presented at Universitat Pompeu Fabra, Barcelona, May
2012.
Dowe, P. (2000). Physical causation. New York: Cambridge University Press.
Dowty, D. (1979). Word meaning and Montague grammar. Dordrecht: Reidel.
Dretske, F. I. (1977). Referring to events. Midwest Studies in Philosophy, 2, 90–99.
Eckardt, R. (2000). Causation, contexts, and event individuation. In J. Higginbotham, F. Pianesi,
& A. C. Varzi (Eds.), Speaking of events (pp. 105–121). New York/Oxford: Oxford University
Press.
Einhorn, J. H., & Hogarth, R. (1986). Judging probable cause. Psychological Bulletin, 99, 3–19.
Escamilla, R. M. Jr. (2012). An updated typology of causative constructions: Form- function
mappings in Hupa (Californian Athabaskan), Chungli Ao (Tibeto-Burman) and Beyond. PhD
dissertation, University of California, Berkeley.
Fodor, J. A. (1970). Three reasons for not deriving “kill” from “cause to die”. Linguistic Inquiry,
1, 429–438.
Gaulan, Y. (2016). The causative component in psychological verbs: Emotion and causation in
Modern Hebrew. MA thesis, The Hebrew University of Jerusalem.
Gropen, J., Pinker, S., Hollander, M., Goldberg, R., & Wilson, R. (1989). The learnability and
acquisition of the dative alternation in English. Language, 65, 203–257.
Hall, N. (2004). Two concepts of causation. In J. Collins, N. Hall, & L. A. Paul (Eds.), Causation
and counterfactuals (pp. 255–276). Cambridge, MA: A Bradford Book The MIT Press.
Hall, N., & Paul, L. A. (2013). Metaphysically reductive causation. Erkenntnis, 78, 9–41.
Halpern, J. Y., & Pearl, J. (2005a). Causes and explanations: A structural-model approach Part I:
Causes. British Journal of Philosophy of Science, 56, 843–887.
Halpern, J. Y., & Pearl, J. (2005b). Causes and explanations: A structural-model approach Part II:
Explanations. British Journal of Philosophy of Science, 56, 889–911.
Hart, H. L. A., & Honoré, A. M. (1959). Causation in the Law. Oxford: Oxford University Press.
Haspelmath, M. (1993). More on the typology of inchoative/causative verb alternations. In B.
Comrie & M. Polinsky (Eds.), Causatives and transitivity (pp. 87–120). Amsterdam: John
Benjamins.
Haspelmath, M. A. C., Spagnol, M., Narrog, H., & Bamyacı, E. (2014). Coding causal-noncausal
verb alternations: A form-frequency correspondence explanation. Journal of Linguistics, 50,
587–625.
Hesslow, G. (1983). Explaining differences and weighting causes. Theoria, 49, 87–111.
Hesslow, G. (1984). What is a genetic disease? On the relative importance of causes. In L.
Nordenfelt & B. I. B. Lindahl (Eds.), Health, disease and causal explanations in medicine
(pp. 183–193). Dordrecht: Reidel.
Hilton, D. (1990). Conversational processes and causal explanation. Psychological Bulletin, 107,
65–81.
Hitchcock, C. R. (1996). The role of contrast in causal and explanatory claims. Synthese, 107,
394–419.
Hitchcock, C. (2003). Of humean bondage. The British Journal for the Philosophy of Science, 54,
1–25.
Hitchcock, C., & Knobe, J. (2009). Cause and norm. Journal of Philosophy, 106, 587–612.
Hocutt, M. (1974). Aristotle’s four Becauses. Philosophy, 49, 385–399.

Hole, D. (2005). Reconciling ‘possessor’ datives and ‘beneficiary’ datives – Towards a unified
voice account of dative binding in German. In C. Maienborn & A. Wöllstein-Leisten (Eds.),
Event arguments in syntax, semantics, and discourse (pp. 213–242). Tübingen: Niemeyer.
Hole, D. (2006). Extra argumentality–Affectees, landmarks, and voice. Linguistics, 44, 383–424.
Horn, L. (2018). Negation and word order in the footsteps of Neg-first. Ms, a paper presented at
the University of Maryland, May 2018.
Iatridou, S. (1991). Topics in conditionals. PhD dissertation. MIT.
Jackendoff, R. (1972). Semantic interpretation in generative grammar. Cambridge MA: MIT
Press.
Jespersen, O. (1917). Negation in English and other languages. Copenhagen: Høst.
Johnston, M. (1994). The syntax and semantics of adverbial adjuncts: University of California.
University of California.
Kadmon, N., & Landman, F. (1993). Any. Linguistics and Philosophy, 16, 353–422.
Katz, J. J. (1970). Interpretive semantics vs. generative semantics. Foundations of Language, 6,
220–259.
Kratzer, A. (2005). Building resultatives. In M. Claudia & A. Wöllstein (Eds.), Event arguments:
foundations and applications (pp. 177–212). Tübingen: Max Niemeyer Verlag.
Kroeger, P. (2007). Morphosyntactic vs. morphosemantic functions of Indonesian –kan. In A.
Zaenen, J. Simpson, T. H. King, J. Grimshaw, J. Maling, & C. Manning (Eds.), Architectures,
rules, and preferences: Variations on themes of Joan Bresnan (pp. 229–251). Stanford: CSLI
Publications.
Kvart, I. (2004). Causation: Probabilistic and counterfactual analyses. In J. Collins, N. Hall, & L.
A. Paul (Eds.), Causation and Counterfactuals (pp. 359–386). Cambridge, MA: A Bradford
Book The MIT Press.
Lakoff, G. (1970). Irregularity in Syntax. New York: Holt, Rinehart & Winston.
Lauer, S. (2010). Periphrastic causative verbs in English: What do they mean? Ms., Stanford
University.
Levin, B., & Rappaport Hovav, M. (1991). Wiping the slate clean: A lexical semantic exploration.
In B. Levin & S. Pinker (Eds.). Special Issue on Lexical and Conceptual Semantics. Cognition,
41, 123–151. Reprinted as B. Levin & S. Pinker, eds. (1992) Lexical and Conceptual Semantics,
Blackwell, Oxford.
Levin, B., & Rappaport Hovav, M. (1994). A preliminary analysis of causative verbs in English.
Lingua, 92, 35–77.
Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity. Cambridge, MA: MIT Press.
Lewis, D. (1973a). Causation. Journal of Philosophy, 70, 556–567.
Lewis, D. (1973b), Counterfactual. Oxford/Cambridge, MA: Blackwell Publishers/Harvard Uni-
versity Press.
Lewis, D. (1979). Counterfactual dependence and Time’s arrow. In Philosophical Papers, 2, 32–66.
Lewis, D. (2000). Causation as influence. Journal of Philosophy, 97, 182–197.
Lewis, D. (2004). Causation as influence. In J. Collins, N. Hall, & L. A. Paul (Eds.), Causation
and counterfactuals (pp. 75–106). Cambridge, MA: A Bradford Book The MIT Press.
Lundquist, B., Corley, M., Tungseth, M., Sorace, A., & Ramchand, G. (2016). Anticausatives are
semantically reflexive in Norwegian, but not in English. Glossa, 1, 1–30.
Mackie, J. L. (1965). Causes and conditions. American Philosophical Quarterly, 2, 245–264.
Mackie, J. L. (1974). The cement of the universe. Oxford: Oxford University Press.
Maienborn, C., & Herdtfelder, J. (2015). A compositional account of the eventive/stative ambiguity
of German causal von-modifiers. In Proceedings of Semantics and Linguistics Theory (SALT),
25, 163–183.
Maienborn, C., & Herdtfelder, J. (2017). Eventive vs. stative causation: The case of German causal
von-modifiers. Linguistics and Philosophy, 40, 279–320.
Marantz, A. (2005). Objects out of the lexicon: Objects as event. Handout of Talk presented at the
University of Vienna. http://web.mit.edu/marantz/Public/Vienna/Vienna.pdf
Martin, F. (2015). Explaining the link between agentivity and non-culminating causation. In
Proceedings of Semantics and Linguistics Theory (SALT), 25, 246–266.
Martin, F. (2018). Time in probabilistic causation: Direct vs. indirect uses of lexical causative
verbs. In Proceedings of Sinn und Bedeutung
Maslen, C. (2004). Causes, contrasts, and the nontransitivity of causation. In Collins et al. (Eds.),
Causation and counterfactuals (pp. 341–357). Cambridge, M.A.: MIT Press.
McCawley, J. D. (1968). Lexical insertion in a transformational grammar without deep structure.
Chicago Linguistic Society, 4, 71–80.
McCawley, J. D. (1976). Remarks on what can cause what. In M. Shibatani (Ed.), The grammar
of causative constructions (Syntax and semantics) (Vol. 6, pp. 117–129). New York: Academic
Press.
McGrath, S. (2005). Causation by omission: A dilemma. Philosophical Studies, 123, 125–148.
Menzies, P. (2009). Platitudes and counterexamples. In H. Beebee, C. Hitchcock, & P. Menzies
(Eds.), The Oxford handbook of causation (pp. 341–367). Oxford: Oxford University Press.
Mill, J. S. (1884). A system of logic, ratiocinative and inductive: Being a connected view of the
principles of evidence and the methods of scientific investigation (Vol. 1). Longmans, Green,
and Company.
Morgan, J. L. (1969). On arguing about semantics. Research on Language & Social Interaction, 1,
49–70.
Nadathur, P. (2015). Implicative verbs and their presuppositions. Manuscript, Stanford.
Nadathur, P., & Lauer, S. (2020). Causal necessity, causal sufficiency, and the implications of
causative verbs. Glossa: A Journal of General Linguistics, 5, 1–37. https://doi.org/10.5334/
gjgl.497
Nedjalkov, V. P., & Silnitsky, G. G. (1973). The typology of morphological and lexical causatives.
In F. Kiefer (Ed.), Trends in Soviet theoretical linguistics (pp. 1–32). Dordrecht: D. Reidel
Publishing.
Neeleman, A., & van de Koot, H. (2012). The linguistic expression of causation. In M. Everaert, M.
Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the interface (pp. 20–51).
Oxford: Oxford University Press.
Northcott, R. (2008). Weighted explanations in history. Philosophy of the Social Sciences, 38(1),
76–96. https://doi.org/10.1177/0048393107311045.
O’Connor, M. C. (2007). External possession and utterance interpretation: A crosslinguistic
exploration. Linguistics, 45, 577–613.
Oehrle, R. (1976). The grammatical status of the English dative alternation. PhD thesis, MIT.
Pearl, J. (2000). Causality: Models, reasoning, and inference. Cambridge, MA: Cambridge
University Press.
Paul, L. A. (2000). Aspect causation. The Journal of Philosophy, 97 (Special Issue: Causation),
235–256
Pesetsky, D. (1995). Zero syntax. Cambridge, MA: MIT Press.
Psillos, S. (2009). Causal Pluralism. In V. Robrecht & B. Bart D’Hooghe (Eds.), Worldviews,
science and us (pp. 131–151). Singapore: World Scientific Publishing.
Pylkkänen, L. (2008). Introducing arguments (Linguistic inquiry monographs 49). Cambridge,
MA: MIT Press.
Reinhart, T. (2000). The theta system: Syntactic realization of verbal concepts (OTS Working
Papers). Utrecht University
Reinhart, T. (2002). The theta system – An overview. Theoretical Linguistics, 28, 229–290.
Rappaport Hovav, M., & Levin, B. (2008). The English dative alternation: The case for verb
sensitivity. Journal of Linguistics, 44, 129–167.
Rooth, M. E. (1992). A theory of focus interpretation. Natural Language Semantics, 1, 75–116.
Ruwet, N. (1972). Théorie syntaxique et syntaxe du français. Paris: Seuil.
Salmon, W. (1997). Causality and explanation: A reply to two critiques. Philosophy of Science, 64,
461–477.
Schaffer, J. (2005). Contrastive causation. Philosophical Review, 114, 297–328.
Schaffer, J. (2013). Causal contextualism. In M. Blaauw (Ed.), Contrastivism in philosophy (pp.

35–63). London: Routledge.
Schaffer, J. (2016). Grounding in the image of causation. Philosophical Studies, 173, 49–100.
Schnieder, B. (2011). A logic for ‘because’. The Review of Symbolic Logic, 4, 445–465.
Schulz, K. (2011). If you’d wiggled A, then B would’ve changed. Synthese, 179, 239–251.
Shibatani, M. (Ed.). (1976a). The grammar of causative constructions (Syntax and semantics 6).
New York: Academic Press.
Shibatani, M. (1976b). The grammar of causative constructions: A conspectus. In M. Shibatani
(Ed.), The grammar of causative constructions (Syntax and semantics 6) (pp. 1–40). New York:
Academic Press.
Shibatani, M., & Pardeshi, P. (2002). The causative continuum. In M. Shibatani (Ed.), The grammar
of causation and interpersonal manipulation (pp. 85–126). Amsterdam: John Benjamins.
Skow, B. (2016). Reasons why. Oxford: Oxford University Press.
Solstad, T. (2010). Some new observations on ‘because (of)’. In M. Aloni, H. Bastiaanse, T. de
Jager, & K. Schulz (Eds.), Logic, language and meaning: 17th Amsterdam colloquium (pp.
436–445). Berlin: Springer.
Song, J. J. (1996). Causatives and causation: A universal-typological perspective. London:
Longman.
Strawson, P. F. (1950). On referring. Mind, 59, 320–344.
Sweetser, E. (1990). From etymology to pragmatics. Metaphorical and cultural aspects of semantic
structure (Cambridge studies in linguistics 54). Cambridge: Cambridge University Press.
Talmy, L. (1976). Semantic causative types. In M. Shibatani (Ed.), The grammar of causative
constructions (Syntax and semantics 6) (pp. 43–116). New York: Academic Press.
Talmy, L. (2000). Towards a cognitive semantics II: Typology and process in concept structuring.
Cambridge, MA: MIT Press.
Taylor, C. (1964). The explanation of behavior. London: Routledge & Kegan Paul.
Thomason, R. H. (2014). Formal semantics for causal constructions. In B. Copley & F. Martin
(Eds.), Causation in grammatical structures (pp. 58–75). Oxford: Oxford University Press.
van Frassen. Bas C. (1980). The scientific image. Oxford: The Calrendon Press.
van Valin, R. D., Jr. (2005). Exploring the syntax-semantics interface. Cambridge: Cambridge
University Press.
Veltman, F. (2005). Making counterfactual assumption. Journal of Semantics, 22, 159–180.
Vlastos, G. (1969). Reasons and causes in the Phaedo. The Philosophical Review, 78, 291–325.
Von Wright, G. H. (1968). An essay in deontic logic and the general theory of action: With
a bibliography of deontic and imperative logic (Acta Philosophica Fennica. Fasc. 21.).
Amsterdam: North-Holland.
Waldmann, M. R., & Hagmayer, Y. (2013). Causal reasoning. In D. Reisberg (Ed.), Oxford
handbook of cognitive psychology (pp. 733–752). New York: Oxford University Press.
White, M. (1965). Foundations of historical knowledge. New York: Harper & Row.
Wolff, P. (2003). Direct causation in the linguistic coding and individuation of causal events.
Cognition, 88, 1–48.
Wolff, P. (2007). Representing causation. Journal of Experimental Psychology: General, 136, 82–
111.
Wolff, P., & Thorstand, R. (2016). Force dynamics. In W. Michael (Ed.), Oxford handbook of
causal reasoning. Oxford, UK: Oxford University Press.
Wolff, P., & Song, G. (2003). Models of causation and the semantics of causal verbs. Cognitive
Psychology, 47, 276–332.
Woodward, J. (1984). A theory of singular causal explanation. Erkenntnis, 21, 231–262.
Woodward, J. (2003). Making things happen: A theory of causal explanation. Oxford: Oxford
University Press.
Yablo, S. (2004). Advertisement for a sketch of an outline of a proto-theory of causation. In
J. Collins, N. Hall, & L. A. Paul (Eds.), Causation and counterfactuals (pp. 119–137).
Cambridge, MA: A Bradford Book The MIT Press.
Chapter 2
Communicating Causal Structure
Christopher Hitchcock
Abstract One common and useful tool for representing causal relationships in
philosophy and in the sciences is the structural equation model (SEM). A SEM
describes relations of functional dependence among several variables. On the other
hand, in ordinary language we normally express causation by describing a binary
relation among two events: “C causes E”. This essay describes several dimensions
along which a SEM describes a richer causal structure than seems possible using the
simple binary construction “C causes E”. This raises several questions about how
we successfully communicate about causal structure. First, how do we decide which
aspects of a causal structure to explicitly describe to our audience? And second,
what linguistic tools do we have at our disposal to communicate the richer structure
represented by a SEM?
Keywords Causation · Causative · Communication · Implication · Norm ·

Structural equation model · Variable
2.1 Introduction
As a philosopher who studies causation, the focus of my research is different from

that of the linguists who have contributed to this volume. I do not study language
for its own sake. Nonetheless, there are good reasons for philosophers who study
causation to pay attention to the nuances of the language we use to talk about
causal relations. First, philosophers are familiar with the warnings of Wittgenstein
(1958) and others that one should not simply read off our ontology from the
surface structure of language. But heeding these warnings is not always easy, and
being attuned to the deeper structure of language can make it easier to search for
C. Hitchcock ()
Division of Humanities and Social Sciences, California Institute of Technology, Pasadena, CA,
USA
e-mail: cricky@caltech.edu

https://doi.org/10.1007/978-3-030-34308-8_2
54 C. Hitchcock
alternative ontologies that still fit our discourse. Second, theories of causation are
often assessed by their agreement with intuitive judgments. One tells a story about
Billy and Suzy throwing rocks at a window, and asks whether Suzy’s throw caused
the window to shatter. As Swanson (2010) argues, our willingness to assent to a
causal statement may depend upon pragmatic factors that go beyond the causal facts
of the case, so it is important to be sensitive to these factors.
Similarly, I think there is reason for linguists interested in causal language
to pay attention to the kinds of models used by philosophers and others for
representing causal relations. While these models are not intended as models for
natural language, and do not stand or fall by their success in this domain, they may
nonetheless provide resources for understanding the semantics and pragmatics of
causal constructions. To the extent that these models accurately characterize the
structure of causal relations in the world, we would expect natural languages to be
equipped with resources for describing this structure.
In this chapter I will discuss a simple type of model that is often used to
represent causal structures, the structural equation model (SEM). The information
represented in a SEM is considerably more complex than the information that is
directly expressed by a simple causal claim of the form “C causes E”. This poses a
prima facie challenge for the communication of causal information. I will explore
some of the ways in which this challenge is met, drawing on research in philosophy,
including my own, as well as empirical psychology and linguistics.
2.2 Structural Equation Models
Consider a particular causal system: the gas grill in my back yard. We can
represent this system with a set of variables, whose values are given the following
interpretation:
Gas knob = 0 if the gas knob is set to “off”

= 1 if the gas knob is set to “low”
= 2 if the gas knob is set to “medium”
= 3 if the gas knob is set to “high”
Gas supply = 0 if there is no gas supply to the grill
= 1 if there is gas supply
Igniter = 0 if the igniter button is not pressed
= 1 if the igniter button is pressed
Battery = 0 if no live battery is installed
= 1 if a live battery is installed
Gas level = 0 if no gas enters the grill
= 1 if a low level of gas enters the grill
= 2 if a medium level of gas enters the grill
= 3 if a high level of gas enters the grill
2 Communicating Causal Structure 55
Spark = 0 if the igniter does not produce a spark

= 1 if the igniter produces a spark
Flame = 0 if there is no flame
= 1 if there is a low flame
= 2 if there is a medium flame
= 3 if there is a high flame
Chicken on = 0 if no chicken is put on the grill
= 1 if chicken is put on the grill
Chicken cooked = 0 if the chicken is raw.
= 1 if the chicken is undercooked
= 2 if the chicken is well cooked
= 3 if the chicken is burnt
Next, we represent the way in which some variables depend upon others with
structural equations:
Gas level = Gas knob × Gas supply
Spark = Igniter × Battery
Flame = Gas level × Spark
Chicken cooked = Flame × Chicken on
The variables Gas level, Spark, Flame and Chicken cooked are endogenous: their
values are determined by the values of other variables in the system. For example,
the equation for Flame tells us that the value of Flame is equal to the value of Gas
level times the value of Spark. In other words, if there is no spark (Spark = 0),
there will be no flame (Flame = 0); if there is a spark (Spark = 1), then the level
of the flame will depend upon the level of gas entering the grill: low gas (Gas
level = 1) produces a low flame (Flame = 1), high gas produces a high flame, etc.
Thus multiplication functions a bit like the logical and. The variables Gas knob, Gas
supply, Igniter, Battery, and Chicken on are exogenous variables: their values are not
determined by other variables in the system. The relations between the variables
can be represented qualitatively using a directed acyclic graph (Fig. 2.1). An arrow
from one variable to another indicates that the first variable figures (non-trivially)
Fig. 2.1 A directed acyclic Chicken cooked

graph representing the causal
structure of a gas grill
Flame Chicken on
Gas level Spark
Gas knob Gas supply Igniter Battery

56 C. Hitchcock
in the structural equation for the second variable. The graph does not indicate the
exact nature of the dependence (for example, whether the equation is additive or
multiplicative).
Given a setting of values for the exogenous variables, the values of the endoge-
nous variables are uniquely determined. For example, if the gas knob is set to “high”,
the gas supply is connected, the igniter is pressed, a battery is installed, and the
chicken is put on the grill, the values of the variables will be:
Gas knob =3
Gas supply =1
Igniter =1
Battery =1
Gas level =3
Spark =1
Flame =3
Chicken on =1
Chicken cooked =3
Sadly, the chicken gets burned.
The distinctively causal interpretation of the equations emerges from the way
in which interventions are represented. An intervention overrides the usual causal
structure and directly imposes a value on a variable. For example, if we intervene
to set Flame = 1, we sever the dependence of Flame on Gas level and Spark,
and directly set the value of Flame to 1. (We might do this, for example, by
flooding the grill with some substance that is barely inflammable and lighting it.)
We represent this by replacing the equation for Flame with the setting Flame = 1,
effectively turning Flame into an exogenous variable. This means that the changes
introduced by interventions propagate forward through the causal structure, but not
backward. Intervening to set Flame = 1 will result in Chicken cooked = 1, but
not in Gas level = 1. Thus, interventions function much like the non-backtracking
counterfactuals described by Lewis (1979). This way of representing interventions
also lets us answer counterfactual questions. For example, in the situation just
described, we can calculate that if the gas knob had been set to medium (Gas
knob = 2), the chicken would have been well cooked (Chicken cooked = 2). This
dependence of Chicken cooked on Gas knob indicates that the latter variable causally
influences the former.
This model is intended to represent a physical system: my gas grill. More
specifically, the model represents the causal dependence of certain states of the
grill (or parts of the system comprising the grill and the chicken) on other states.
The model is not intended as a description of my psychological representation of
the grill. If the model accurately describes the grill, and I understand how my grill
works, then I might have a mental model of the grill resembling this one. But the
adequacy of the model depends upon the grill, not upon my mental representation.
Analogously, the model is not intended to provide semantics for natural language
causal discourse about the grill. If the model accurately represents the grill, we
might hope that natural language is up to the task of describing it; but that is not a
criterion of adequacy for the model.
It is possible to add probability to the model, either directly or indirectly. We
can add probability directly by replacing the structural equations with conditional
probability distributions. We can add probability indirectly by adding additional
exogenous variables, incorporating them into the equations, and placing a probabil-
ity distribution over the values of these exogenous variables. We will not go into the
details here (see Hitchcock 2018 for discussion).
This model is oversimplified in a number of respects. For example, the gas and
spark will not produce a flame if there is no oxygen present, or if the grill is flooded
with water. We can represent these particular details easily enough by adding further
variables, but we may never capture all of the variables that are relevant. We may
acknowledge the omitted variables by saying that, e.g. the equation for Flame
describes how this variable depends upon Gas level and Spark, in the circumstances.
That is, given the actual presence of oxygen, absence of water, and so forth, Flame
will depend upon Gas level and Spark in the way described.
A more complicated issue concerns the role of time. If the chicken is on the
grill only briefly, it will remain undercooked, even if the flame is high; and the
chicken can be cooked well on a low flame if it cooks for long enough. However,
it seems wrong to conceive of time as causal variable akin to the setting of the gas
knob. Moreover, representing time as a variable can lead to technical problems (see
Hitchcock 2012). The way to represent the role of time is to include copies of the
causal variables, indexed by time. For example, we could have a variable Chicken
on at 18:00, representing whether the chicken is on the grill at 18:00. Then we can
represent the chicken being on the grill for 20 min by Chicken on at t taking the
value 1 for t = 18:01, 18:02, . . . , 18:20, and taking the value 0 for other values of
t. Expanding the model in this way would allow us to represent other possibilities
as well, such as cooking the chicken on a high flame for a short period of time,
and then turning the gas knob down to low. Obviously, this would make the model
considerably more complex.
A further concern is whether detailed mechanical information can be represented
using these kinds of structural equation models. For example, chemical interactions
(such as those that occur during the combustion of the gas, or within the chicken as
it cooks), depend heavily upon the geometry of molecules, and this kind of spatial
information is not easily represented by variables in a SEM.1
Despite these limitations, the type of SEM described above does go some way
toward describing the causal structure of a system such as a gas grill. These kinds
of models have been used successfully for a variety of purposes: inferring causal
relationships from statistical data, predicting the effects of interventions, developing
1 See, for example, the exchange between Menzies (2012) and Cartwright (2017) about whether
this is possible in principle. I will remain agnostic about this issue here.
58 C. Hitchcock
a logic of counterfactuals, and providing analyses of specific causal concepts. See

for example Spirtes et al. (2000) and Pearl (2009) for detailed treatments involving
a number of applications.
2.3 Causal Language
According to a fairly standard philosophical view (see e.g. Davidson 1967; Lewis
1973, 1986), causation is a binary relation between events. That is, causal relation-
ships have the form C causes E, where C and E are events. Events are happenings
within bounded spatiotemporal regions, often involving the realization of properties
by objects or individuals.2 Most philosophers allow that events can be standing
states, such as the battery being connected, as well as changes (the chicken being put
on the grill) and momentary happenings (the igniter button being pressed).3 Other
philosophers (e.g. Bennett 1988; Mellor 2004) have argued that facts are the relata
of causation. Facts correspond to true propositions. This view has the advantage that
it can easily countenance absences as causal relata. Consider the sentence.
(1) The absence of gas supply caused the flame to remain off.
“The absence of gas supply” is not readily interpreted as referring to an event; rather,
it indicates the absence of an event of a specific kind. On the other hand, it can be
naturally construed as referring to the fact that no gas was supplied to the grill.
Even on this view, however, the causal relata should still be conceived of as being
localized in space and time. It is the absence of gas on this particular occasion that
caused the flame to be off. Gas has been supplied to the grill on other occasions,
so it is not even a fact that no gas has (ever) been supplied to the grill. I will
remain neutral about the metaphysics of the causal relata, but for terminological
convenience I will use the word “event” to refer to the causal relata.
In the SEM described above, the specific values of variables correspond to events.
For example, in the specific realization of the model where I set the gas knob to high,
and the chicken burns, my setting the gas knob to high and the chicken burning are
both events. The value Igniter = 0 corresponds to an absence: my failure to press the
igniter button. This may not be an event in the strict metaphysical sense, but it is a
potential causal relatum, so it is an event in the sense in which I am using this term.
A variable such as Gas knob (as opposed to a specific value of that variable) does
not correspond to a single event; the variable ranges over a set of possible events:
the gas knob being set to ‘off’, the gas knob being set to ‘low’, etc.
Ordinary causal language seems to support this picture of causation as a binary
relation among events. After burning the chicken, I might say:
2 See Casati & Varzi (2015) for a survey of philosophical accounts of events.
3 Moore (2009) is one exception.
(2) My setting the gas knob to high caused the chicken to burn.
“My setting the gas knob to high” and “the chicken to burn” are linguistic
constituents that can be understood to denote events. As this example suggests,
many different grammatical forms may be used to denote events, including gerunds,
participles, infinitives, and specific nouns such as “party”, “flood”, and “birth”.4
Absences, such as those described in sentence (1), can be denoted by a similar range
of linguistic constituents.
Sentence (2), while grammatical, is a bit stilted. More idiomatically, I might
say:
(3) I caused the chicken to burn by setting the gas knob to high.
I might also say:
(4) The chicken burnt because I set the gas knob to high.
“Because” indicates an explanatory relation: the clause that follows “because”
provides an explanation of why the first clause is true. There is an extensive
philosophical literature on the nature of explanation, especially in the scientific
context.5 Not all explanations cite causal relationships,6 but causation is one kind
of explanatory relationship: causes explain why their effects occur.
Causal relations can also be described using causative verbs:
(5) I burnt the chicken by setting the gas knob to high.
The transitive verb “burn” here means, to a first approximation, “cause to burn”.
One can also say:
(6) I caused the chicken to burn.
or
(7) I burnt the chicken.
Read literally, these sentences seem to indicate that the first relatum is not an event,
but a person, namely me. Most philosophers would interpret these sentences as
elliptical, indicating that I participated in some event that caused the chicken to
burn (or more colloquially: I did something that caused the chicken to burn). Most
linguists would interpret the verbs in (6) and (7) as semantically bearing an implicit
event argument.7
It is apparent that ordinary causal claims like (1)–(7) underdescribe the actual
causal structure of a particular situation, even the simplified model of the causal
4 See Bennett (1988) for discussion of both the philosophical and linguistic dimensions of events.
5 See, for example, Woodward (2017) for an overview of this topic.
6 See, for example, Lange (2017) on non-causal explanation.
7 But see Neeleman & van de Koot (2012) for an exception.
60 C. Hitchcock
structure given above.8 These ordinary causal claims cite the values of two variables,
and say that there is some causal relationship between them. They seem to say
nothing about other possible values of those two variables, about other variables in
the model, about the relationships among the variables, or about the way in which
one variable depends upon another. Saying that I caused the chicken to burn by
setting the gas knob to high says nothing about the role of the battery or the igniter
button, it says nothing about the gas or the flame, it says nothing about the other
possible settings of the gas knob, and it does not tell us how well the chicken would
have been cooked if the gas knob had been set to one of these other levels.
The discrepancy between the complexity of the underlying causal structure and
the simplicity of ordinary causal claims suggests two things. First, when speakers
wish to communicate information about a causal structure, they must make choices
about which features to communicate. Second, there may be ways of indirectly
communicating further features of the structure. In the remainder of this chapter,
I will discuss these two aspects of communicating causal structure.
2.4 The Horizontal Dimension
For simplicity, let us assume that the speaker has already selected some event, such
as the burning of the chicken, and wishes to report on its causes. The selection
of a cause variable to report on can be divided into two dimensions, horizontal
and vertical, corresponding to the horizontal and vertical directions in Fig. 2.1. The
horizontal dimension concerns the choice of a causal pathway leading to the effect
of interest. When reporting on the causes of Chicken cooked, we can follow the
causal path backward to the right, leading to Chicken on, or to the left, leading to
Flame, and continuing on to other variables. The vertical dimension concerns the
choice of a variable along this causal pathway to report on. For example, having
selected the leftmost pathway in Fig. 2.1, we could describe Gas knob, Gas level, or
Flame. We will address each of these dimensions in turn.
A great deal of empirical and theoretical work has demonstrated that norms play
a prominent role in determining which of two interacting factors we tend to identify
as causes (McGrath 2005; Cushman et al. 2008; Knobe & Fraser 2008; Hitchcock
& Knobe 2009; Sytsma et al. 2010; Halpern & Hitchcock 2015; Kominsky et al.
2015; Icard et al. 2017). In our causal model, the setting of the gas knob, and the gas
supply jointly determine the level of gas in the grill, which in turn helps determine
the level of the flame. We are much more likely to cite the setting of the gas knob
as a cause of high flame than we are to cite the gas supply. This is because it is
normal for the gas supply to be connected to the grill. It is normal in two distinct
senses. First, the gas supply is connected most of the time. Occasionally, the gas
8 Neeleman & van de Koot (2012) stress a similar discrepancy between the content of ordinary
causal claims and the speaker’s complex mental model.
supply gets shut off; usually this happens if something knocks the seismic safety
valve (designed to turn off the gas supply in an earthquake). But I can reliably plan
to grill chicken without worrying that there will be no gas. Second, the gas supply
is supposed to be connected to the grill. When the grill is in good working order,
when it is set up as designed, there will be a gas supply to the grill.
As this example illustrates, the words norm and normal group together a variety
of different ideas. There is statistical normality: that which happens frequently is
normal; that which happens infrequently is abnormal. There is an epistemic dimen-
sion of normality: something is normal if it is expected, abnormal if unexpected.
The default state of a system might also be described as normal. In Newtonian
mechanics, a body will normally travel in a straight line with uniform velocity,
meaning that this is what it will do in the absence of external forces acting on it.
An action is normal if it conforms to a rule: this could be moral rule, a law, or an
institutional policy. Part of an organism or a machine is functioning normally if it
is contributing to the overall functioning of the system. In the case of machines,
this will usually mean that the part of the machine is operating as it was designed
to operate. In the case of biological organisms, it is harder to say what grounds
the assessment of proper functioning. Nonetheless, we talk naturally of a normally
functioning heart.
These ideas are clearly distinct, and they can pull in different directions. Most
bodies do have external forces acting on them; most drivers do exceed the speed
limit. Conflating these different senses of normality can cause harm. When my
grandmother (born in 1906) went to school, she was forced to write with her right
hand, even though she was naturally left-handed. This led to poor handwriting,
poorer performance on written tests, and so on. In retrospect, it seems a silly mistake
to think that it is wrong to write with the left hand, just because it is statistically
more common to write with the right hand. Nonetheless, this was once a common
educational practice, illustrating how easily we can slide between the different
senses of normality.
The hypothesis that I and others have developed is that these different senses of
normality influence causal attributions in similar ways. The experimental evidence
bears this out: see for example Knobe & Fraser (2008) on institutional rules,
Hitchcock & Knobe (2009) on norms of proper functioning, Cushman et al. (2008)
on moral norms, and Sytsma et al. (2010) on statistical norms.
Somewhat surprisingly, the precise way in which norms influence the selection
of a causal variable depends upon how the causes interact with one another. In our
running example, the setting of the gas knob and the gas supply are both necessary
(in the circumstances) for gas to enter the grill and for the flame to light. That is, in
order to have Gas level > 0, we must have both Gas knob > 0 and Gas supply > 0.
If we intervene to set either Gas knob = 0 or Gas supply = 0, then Gas level will
be 0. In causal structures of this kind, we tend to identify the variable that takes an
abnormal value as the cause. In this case, we would be more likely to identify the
setting of the gas knob as a cause of the flame than the gas supply. But in cases where
there are multiple causes, each one of which is sufficient for the effect, the reverse
happens. Icard et al. (2017) present subjects with vignettes such as the following:
62 C. Hitchcock
Billy and Suzy are both working on an important project. Suzy is told to show up for
work at 9 A.M. sharp, but Billy is told to stay home (perhaps he is sick). Both show
up simultaneously at 9 A.M. As they enter the building, they set off a motion sensor.
The motion sensor would have been triggered by either one arriving individually.
The SEM for this scenario would be:
Variables:
Billy = 0 if Billy does not come to work at 9 A.M.
= 1 if Billy comes to work at 9 A.M.
Suzy = 0 if Suzy does not come to work at 9 A.M.
= 1 if Suzy comes to work at 9 A.M.
Motion sensor = 0 if the motion sensor does not go off at 9 A.M.
= 1 if the motion sensor goes off at 9 A.M.
Equation:
Motion sensor = max{Billy, Suzy}.
The equation tells us that the motion sensor will go off (Motion sensor = 1) if
either Billy comes to work (Billy = 1) or Suzy does (Suzy = 1). (The max function
works like a logical or.) Thus Billy’s coming to work and Suzy’s coming are each
sufficient (in the circumstances) for the sensor to be triggered. In this scenario,
subjects were more likely to agree that Suzy caused the motion detector to go
off than that Billy did. In this case, people were more likely to identify the norm-
conforming event as the cause.
We can make sense of this apparent inconsistency by noting that people tend
to select the causal variable that makes a difference for the outcome in normal
conditions. In the gas grill, the setting of the gas knob makes a difference for the
flame in the normal condition when the gas supply is connected. The gas supply
makes a difference for the outcome in the less frequent condition when the gas knob
is set to some position other than “off”. In the vignette involving Billy and Suzy,
Suzy’s arrival makes a difference to the motion detector in the normal condition
when Billy is absent. In the abnormal condition when Billy is present (contrary to
orders), Suzy’s arrival makes no difference for whether the motion sensor goes off.
By contrast, Billy’s arrival makes no difference for whether the sensor is triggered
in the normal condition when Suzy is present. In the case where two events are each
necessary for the effect, each event makes a difference when the other is present.
In the case where two events are each sufficient for the effect, each event makes
a difference when the other is absent. So in one case, our causal assessment is
sensitive to the normality of the other cause being present; in the other, it is sensitive
to the normality of the other cause being absent.
Having chosen to report on one causal pathway, are there ways for a speaker
to provide indirect information about other causal pathways? Some causative verbs
as well as other expressions implicate the presence of additional causal pathways.
Consider “allow”, and near synonyms like “let”, and “permit”. I might say, for
instance:
(8) I allowed the chicken to burn by leaving it on the grill too long.
In philosophy, allowing is often discussed in the context of causation by omission.

Consider philosophers’ favorite example:
(9) The gardener allowed the flowers to die by failing to water them.
Philosophers have debated whether the gardener caused the flowers to die. After all,
the gardener did not do anything to the flowers. She might have been nowhere near
the flowers. How can the absence of an event cause something? See, e.g. Beebee
(2004), Dowe (2004), Lewis (2004), and Schaffer (2004) for arguments on both
sides of this debate. I will not try to answer this question here. All parties to the
dispute can agree that it would be appropriate to include a variable for the gardener’s
behavior in a causal model.
As Moore (2009, Ch. 3) points out, however, we can use “allow” in cases where
agents do not abstain from action. For example:
(10) I allowed the horse to escape by opening the barn door.
In this example, I performed an action: opening the door. (Perhaps this is still a case
of causation by omission in the sense that it is the absence of anything blocking the
horse’s path that permits him to leave the barn.)
Rather than indicating the absence or omission of action, “allow” indicates the
presence of a primary causal process that tends to produce the outcome in question.
In (8), the primary process is the flame under the chicken that causes it to burn. In
(10) it is the desire of the horse for freedom and its resulting behavior. The existence
of such a primary process is implicated, rather than asserted, by these sentences. We
see this by noting that the negation of these sentences also indicates the presence of
such a process:
(11) I did not allow the chicken to burn.
(12) I did not allow the horse to escape.
(11) implicates that there was some process at work that, left unchecked, would burn
the chicken. In order to avoid this outcome, I had to intervene. (11) would not be
felicitously asserted if I had simply left the chicken in the refrigerator, or if I had
absent-mindedly put the chicken on the grill without turning the grill on. (11) also
implicates that the chicken did not burn. In this respect, it behaves differently from
the negation of (6):
(13) I did not cause the chicken to burn.
This would be assertable if the chicken did burn, but I wished to absolve myself
of responsibility for the culinary disaster. This is a semantic difference between
“cause” and “allow” that is orthogonal to the issue of causation by omission.
“Prevent” has a similar implication to “allow”.
(14) I prevented the chicken from burning.
has a meaning similar to (11). It implicates that some process was underway which
had a tendency to burn the chicken, but that I intervened to save the chicken from
burning.
64 C. Hitchcock
Nadathur (2016), building on work by Baglini & Francez (2016), develops a

general theory of implicative verbs that appeals to causal structure. For example,
in
(15) I dared to insult the chef’s cooking,
“dare to” indicates that in order to insult the chef’s cooking, I had to overcome
pressures that would normally make one fearful of doing so. Thus (15) would
implicate that pressures of the relevant kind belong in the causal model of the
situation.
2.5 The Vertical Dimension
We will now consider the “vertical” dimension. There is comparatively little empir-
ical work addressing the question of which variable along a causal pathway people
tend to choose when supplying causal information. Swanson (2010) articulates a
pragmatic principle according to which speakers choose “good representatives” of
causal paths, but he says relatively little about what makes a representative good,
other than that this may vary by context. He argues that this pragmatic principle
may explain our reluctance to accept certain causal claims that may nonetheless be
true, and hence that it may be used to protect certain theories of causation against
potential counterexamples.
There has been considerably more literature about the ways in which causal
language can be used to convey information about parts of a causal path that are
not explicitly described. For example, it is a fairly standard view (see, e.g. Dixon
2000) that use of simple causative verbs, such as
(7) I burnt the chicken.
implies a fairly direct causal connection between my action and the outcome. By
contrast, more complex constructions like.
(6) I caused the chicken to burn.
are consistent with more indirect causal connections. For example, suppose that my
wife distracts me. As a result, I forget to take the chicken off the grill, and it burns.
We could say:
(16) My wife caused the chicken to burn by distracting me.
But we would not describe this scenario by saying:
(17) #My wife burnt the chicken by distracting me.
We would not say this, even if she distracted me by burning something. She was in
charge of cooking the potatoes while I was outside working the grill.
This view about causative verbs has been challenged, e.g. by Neeleman & van de
Koot 2012. The present analysis lends some support to this challenge. In the causal
model depicted in Fig. 2.1, setting the gas knob is not a direct cause of the chicken
burning. This causal relationship is mediated by other variables: the amount of gas
released into the grill, and the level of the flame. Indeed, it will almost always be
possible to find such mediating causes between any cause and effect. Moreover, we
can use causative verbs in cases where cause and effect are widely separated in space
and time. The California Supreme Court case of People v. Botkin (1901) concerned
the trial of Cordelia Botkin for murder. Botkin had mailed poisoned candy from San
Francisco to Elizabeth Dunning in Dover, Delaware, roughly 2500 miles (4000 km)
away. Dunning ate the candy and died.9 Although the causal process unfolded over
considerable distance and time, and involved many intermediate events (the package
being loaded and unloaded from trains, etc.), we would still accept any of:
(18) Botkin poisoned Dunning;
(19) Botkin killed Dunning;
(20) Botkin murdered Dunning.
This suggests that the felicity of sentences employing these causative verbs does
not depend upon the causal relationship being direct, or that the relevant notion of
“directness” is not really about the “distance” – in space, time, or number of causal
links – between the cause and the effect.
The connection between the semantics of causative verbs and the directness of
causation remains an open question. At the very least, it is apparent that there
are contexts where a sentence such as 16 employing “cause” is felicitous, but a
sentence such as (17) employing the corresponding causative verb is not (and vice
versa). This suggests that a richer causal model may be needed to make sense of
the difference in information conveyed. We might hope that SEMs could provide a
useful tool in further explorations of this issue.
2.6 Variables
A variable such as Gas knob has values corresponding to different, incompatible

events or states of the world: the gas knob is set to ‘off’; the gas knob is set to ‘low’;
the gas knob is set to ‘medium’; the gas knob is set to ‘high’. One of these will
correspond to the actual setting, the rest to merely possible settings. If I assert:
(3) I caused the chicken to burn by setting the gas knob to high,
the nominal “setting the gas knob to high” indicates that the gas knob was in fact set
to high. Likewise, the clause “the chicken to burn” indicates that the chicken was
burned. But sentence (3) does not tell us explicitly what the underlying variables are.
It does not tell us what other settings of the gas knob were available. It does not even
9 Oneof the legal issues was whether the state of California had legal jurisdiction over the case. It
was ruled that it did, since the murder at least partly took place in California.
66 C. Hitchcock
tell us that the variable ranges over different settings of the gas knob, as opposed to
different knobs that might be set to high. How might this further information be
communicated?
We can bring this problem into focus by considering an example of Dretske
(1977). Suppose that Susan steals a bicycle and is later arrested. The sentence
(21) Susan was arrested because she stole the bicycle.
can appear to be true while.
(22) #Susan was arrested because she stole the bicycle.
appears false. (Assume that the police diligently prosecute thefts of all kinds, and
don’t single out bicycle thieves.) How can the difference in emphasis affect the
truth values of these sentences? Dretske draws a metaphysical conclusion from this
phenomenon: causation is not a relation between events, as commonly thought, but
rather a relation between more fine-grained entities he called event-allomorphs.10
Emphasis determines which event-allomorph is denoted.
I find Dretske’s proposal ontologically extravagant, and have argued (1996a, b)
that a more plausible interpretation is that the emphasis in (21) and (22) picks out
different dimensions of variation.11 The event of Susan’s stealing the bicycle can
be represented by the value of different variables. One variable might range over
ways of acquiring the bicycle: stealing the bicycle, buying the bicycle, renting the
bicycle, borrowing the bicycle . . . A different variable might range over targets of
theft: stealing the bicycle, stealing the skis, stealing the soccer ball . . . The stress in
(21) indicates that the first variable is being described. The means by which Susan
acquires the bicycle affects her chances of being arrested, so (21) sounds true. The
stress in (22) indicates that the second variable is being described. Since Susan’s
chances of arrest do not depend upon what she steals, (22) sounds false.
In addition to focal stress, as illustrated in (21) and (22), the dimension of
variation can be communicated in other ways. The most direct is explicit mention
of the relevant alternatives:
(23) Susan was arrested because she stole the bicycle rather than buying it;
(24) Susan was arrested because she stole the bicycle rather than acquiring it in
some other way;
(25) #Susan was arrested because she stole the bicycle rather than the skis;
10 Itis common to distinguish between coarse-grained theories of events, such as Davidson (1967),
and fine-grained theories, such as Kim (1973) and Lewis (1986). According to the former, Susan’s
stealing and Susan’s stealing the bicycle would be the same event; according to the latter, they
are different events. But Dretske’s event-allomorphs would be even more fine-grained: Susan’s
stealing the bicycle would include as distinct event-allomorphs Susan’s stealing the bicycle and
Susan’s stealing the bicycle. The metaphysical nature of these event-allomorphs is left mysterious.
11 See also Eckardt (2000). Eckardt similarly proposes that focus is used to pick out a dimension of
variation in causal contexts, although her model of the underlying causal relationship is different.
(26) #Susan was arrested because she stole the bicycle rather than one of the other
sports items.
Sometimes, the dimension of variation can also be conveyed using cleft construc-
tions:
(27) #Susan was arrested because it was the bicycle that she stole.
(While an analog of (21) using a cleft construction is possible, the result is very
awkward.)
This phenomenon is related to the phenomenon of presupposition. The stress
in
(28) Susan stole the bicycle.
gives rise to the presupposition that Susan acquired the bicycle in some way. This
presupposition is inherited under negation. Thus, when we entertain counterfactual
possibilities in which (28) is false, we imagine situations in which she acquired the
bicycle in some other way.
Specific causative verbs can also indicate the relevant dimension of variation in
the effect variable. For example, “trigger” and “delay” indicate that the timing of
the effect depends upon the cause; “accelerate” indicates that the speed of the effect
depends on the cause; “amplify” indicates that the volume of the effect depends on
the cause, and so on. “Affect” indicates that relatively minor details of the effect
depend upon the cause, but that the named cause does not determine whether or not
the effect occurs at all. For example, we might say that a lightning strike caused a
forest fire, while selective logging in the area affected the fire (perhaps making a
difference to the speed and direction in which it spread).12
2.7 Patterns of Dependence
One final issue is that the word “cause” underdescribes the way in which one
variable depends upon another. When I say:
or
(5) I burnt the chicken by setting the gas knob to high.
I indicate that the value of the variable Gas knob was 3 (high), and that the value
of the variable Chicken cooked was 3 (burnt). I also indicate that the value of the
12 Lewis (2000) proposes the word “influence” for this type of relation. In Lewis’s terminology,
one event influences another if the specific time and manner in which the former occurs makes a
difference for the specific time and manner in which the latter occurs. This usage has now become
common in philosophy.
68 C. Hitchcock
latter variable depended upon the value of the former. We can cash out this notion
of dependence using counterfactuals: if the value of Gas knob had been different,
the value of Chicken cooked would have been different in some way. We can use
the SEM to calculate that if the gas knob had been set to medium (Gas knob = 2),
the chicken would have been well cooked (Chicken cooked = 2). If I had set the
gas knob to low, the chicken would have been undercooked; and if I had set the
gas knob to off, the chicken would have been raw. Lewis (1973) famously tried to
analyze causation in terms of counterfactuals.13 The problem I want to focus on here
is that there are many different ways in which Chicken cooked can depend upon Gas
knob. Each variable has four possible values, so there are 44 = 256 different ways
for the values of Gas knob to map onto the values of Chicken cooked. Sentences 3
and 5 eliminate some of these possibilities. They tell us that the value Gas knob = 3
(high) leads to Chicken cooked = 3 (burnt), since these are the values that were
actually realized. Moreover, the causal claim in 3 implies (modulo worries about
preemption) that some change in the value of Gas knob leads to some change in the
value of Chicken cooked: not all values of Gas knob lead to Chicken cooked = 3. The
same is true for the causal claim expressed using the causative verb in 5. But that still
leaves 63 different ways in which Chicken cooked can depend upon Gas knob. How
might the speaker convey further information about this pattern of dependence?
This problem does not arise if all the variables are binary. But as the example
of the grill illustrates, many of the causal variables we are interested in are not
binary. We are not only concerned with whether the chicken is raw or cooked, we
are interested in cooking the chicken thoroughly enough to kill bacteria, without
burning it. In the sciences, we almost always deal with quantitative relationships
between variables, rather than with cause and effect relations among binary
variables. In our example, the relationship between the variables is monotonic:
turning the gas knob to higher settings results in the chicken being cooked more
thoroughly. But not all examples are like this. For example, it may be that moderate
levels of alcohol consumption promote heart health while higher levels of alcohol
consumption cause heart damage.
One strategy for communicating information about the pattern of dependence
between variables is to report the values of the variables at a level of granularity
that tracks the underlying pattern of dependence. This proposal is closely related to
Yablo’s (1992a, 1992b) proportionality condition, which will be familiar to many
philosophers. Yablo proposed that causes and effects must be proportional to one
another. To illustrate this idea, suppose that the gas knob has settings marked with
numbers from 0 to 9. 0 is labeled “off”, 1–3 are labeled “low”, 4–6 “medium”, and
7–9 “high”. In fact, I had set the gas knob to 8, and the chicken burned. If instead
I had set the gas knob to 7 or 9, the chicken still would have burned. According to
Yablo, the claim
13 Thereare familiar problems concerning preemption (see e.g. Lewis 1973, 2000; Pearl 2009,
Chap. 10; Moore 2009; Halpern & Hitchcock 2015), but we will put these aside.
(29) #I caused the chicken to burn by setting the gas knob to 8.

is false. The putative cause, setting the gas knob to 8, is not proportional to the
putative effect, the chicken burning. This is because the former event can fail to
occur without the latter event failing to occur. The event that caused the chicken to
burn was setting the gas knob to “high” (7–9), not setting it to 8.
My own view is that (29) is not literally false. I did set the gas knob to 8, and
the state of the chicken did depend upon the setting of the gas knob. But there are
many contexts in which a claim such as (29) would be misleading. This follows
from Grice’s maxim of relation (Grice 1975). Since I specified the precise setting of
the gas knob, it is natural to infer that the precise setting of the gas knob is relevant
to the outcome. Thus (29) pragmatically implicates, although it does not literally
imply, that setting the gas knob to 7 would not have resulted in the chicken burning.
By contrast, (3) is more felicitous, since it supplies no more detail than necessary.
In making claim (3), I correctly suggest that settings other than “high” would have
produced different results for the chicken. In this way, the fineness of the grain with
which one describes the cause variable can communicate information about the way
in which the effect variable depends upon the cause variable.
In a different context, I might be interested in a more fine-grained value of the
effect variable. For instance, my neighbor and occasional host always manages to
grill the chicken perfectly: it is cooked through, slightly charred on the outside, but
still tender and juicy. How does he do it? Now it may be relevant that he sets the gas
knob to 5, rather than 4 or 6. In this case:
(30) #Setting the gas knob to medium caused the chicken to be perfectly cooked.
is not literally false, but it is misleading in suggesting that any medium setting (4–6)
would suffice.
2.8 Conclusion
Ordinary causal claims like:

suggest that causation is a specific, binary relation between two events. On the
other hand, our most successful tools for modeling causal relationships posit more
complex structures. The events described in 3 are the values of variables that are
embedded in a causal structure that includes further variables, and a variety of
patterns of dependence can hold among those variables. There is thus a prima facie
tension between our causal language and our causal models. However, a closer
examination reveals a variety of psychological and linguistic tools that can be used
to communicate additional information about causal structure. This suggests that
our mental models and linguistic practice at least tacitly acknowledge this richer
structure.
70 C. Hitchcock
Acknowledgments For comments, criticism, suggestions, and encouragement, I thank Nora

Boneh, Fabienne Martin, Elitzur Bar-Asher Siegal, Georgina Statham, three anonymous referees,
and the participants at the workshop on Linguistic Perspectives on Causation at the Hebrew
University of Jerusalem.
References
Baglini, R., & Francez, I. (2016). The implications of managing. Journal of Semantics, 33, 541–
560.
Beebee, H. (2004). Causing and nothingness. In J. Collins, N. Hall, & L. A. Paul (Eds.), Causation
Bennett, J. (1988). Events and their names. Indianapolis/Cambridge: Hackett.
Cartwright, N. (2017). Can structural equations explain how mechanisms explain? In H. Beebee,
C. Hitchcock, & H. Price (Eds.), Making a difference: Essays on the philosophy of causation
(pp. 132–152). Oxford: Oxford University Press.
Casati, R., & Varzi, A. (2015). Events. In E. Zalta (Ed.), The Stanford encyclopedia of philoso-
phy.https://plato.stanford.edu/archives/win2015/entries/events/
Cushman, F., Knobe, J., & Sinnott-Armstrong, W. (2008). Moral appraisals affect doing/allowing
judgments. Cognition, 108, 281–289.
Davidson, D. (1967). Causal relations. Journal of Philosophy, 64, 691–703.
Dixon, R. M. W. (2000). A typology of causatives: Form, syntax and meaning. In R. M. W. Dixon
& A. Y. Aikhenvald (Eds.), Changing valency: Case studies in transitivity (pp. 30–83). New
York: Cambridge University Press.
Dowe, P. (2004). Causes are physically connected to their effects: Why preventions and omitters
are not causes. In C. Hitchcock (Ed.), Contemporary debates in philosophy of science (pp.
189–196). Oxford: Basil Blackwell.
Dretske, F. (1977). Referring to events. Midwest Studies in Philosophy, 2, 90–99.
Eckardt, R. (2000). Causation, contexts, and event individuation. In J. Higginbotham, F. Pianesi, &
A. Varzi (Eds.), Speaking of events (pp. 105–122). New York/Oxford: Oxford University Press.
Grice, P. (1975). Logic and conversation. In D. Davidson & G. Harman (Eds.), The logic of
grammar (pp. 64–75). Encino: Dickenson.
Halpern, J., & Hitchcock, C. (2015). Graded causation and defaults. British Journal for the
Philosophy of Science, 66, 413–457.
Hitchcock, C. (1996a). Farewell to binary causation. Canadian Journal of Philosophy, 26, 267–
282.
Hitchcock, C. (1996b). The role of contrast in causal and explanatory claims. Synthese, 107, 395–
419.
Hitchcock, C. (2012). Events and times: A case study in means-ends metaphysics. Philosophical
Studies, 160, 79–96.
Hitchcock, C. (2018). Causal models. In E. Zalta (Ed.), Stanford encyclopedia of philosophy. https:/
/plato.stanford.edu/archives/fall2018/entries/causal-models/
Icard, T., Kominsky, J., & Knobe, J. (2017). Normality and actual causal strength. Cognition, 161,
80–93.
Kim, J. (1973). Causation, nomic subsumption, and the concept of event. Journal of Philosophy,
70, 217–236.
Knobe, J., & Fraser, B. (2008). Causal judgment and moral judgment: Two experiments. In W.
Sinnott-Armstrong (Ed.), Moral psychology, volume 2: The cognitive science of morality (pp.
441–447). Cambridge, MA: MIT Press.
Kominsky, J., Phillips, J., Gerstenberg, T., Lagnado, D., & Knobe, J. (2015). Causal superseding.
Cognition, 137, 196–209.
Lange, M. (2017). Because without cause. Oxford: Oxford University Press.

Lewis, D. (1973). Causation. Journal of Philosophy, 70, 556–567.
Lewis, D. (1979). Counterfactual dependence and time’s arrow. Noûs, 13, 455–476.
Lewis, D. (1986). Events. In D. Lewis (Ed.), Philosophical papers (Vol. II, pp. 241–270). Oxford:
Oxford University Press.
Lewis, D. (2000). Causation as influence. Journal of Philosophy, 97, 182–197.
Lewis, D. (2004). Void and object. In J. Collins, N. Hall, & L. A. Paul (Eds.), Causation and
counterfactuals (pp. 277–290). Cambridge, MA: MIT Press.
McGrath, S. (2005). Causation by omission. Philosophical Studies, 123, 125–148.
Mellor, H. (2004). For facts as causes and effects. In J. Collins, N. Hall, & L. A. Paul (Eds.),
Causation and counterfactuals (pp. 309–324). Cambridge, MA: MIT Press.
Menzies, P. (2012). The causal structure of mechanisms. Studies in the History and Philosophy of
Biological and Biomedical Sciences, 43, 796–805.
Moore, M. (2009). Causation and responsibility. Oxford: Oxford University Press.
Nadathur, P. (2016). Causal necessity and sufficiency in implicativity. Proceedings of SALT, 26,
1002–1021.
Neeleman, A., & van de Koot, H. (2012). The linguistic expression of causation. In M. Eraert, M.
Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the crossroads (pp. 20–51).
Pearl, J. (2009). Causality: Models, reasoning, and inference (2nd ed.). Cambridge: Cambridge
University Press.
People v. Cordelia Botkin, Supreme Court of California. (1901). Crim. No. 832; 132 Cal. 231; 64
P. 286.
Schaffer, J. (2004). Causes need not be physically connected to their effects: The case for negative
causation. In C. Hitchcock (Ed.), Contemporary debates in philosophy of science (pp. 197–
216). Oxford: Basil Blackwell.
Spirtes, P., Glymour, C., & Scheines, R. (2000). Causation, prediction and search (2nd ed.).
Swanson, E. (2010). Lessons from the context sensitivity of causal talk. Journal of Philosophy,
107, 221–242.
Sytsma, J., Livengood, J., & Rose, D. (2010). Two types of typicality: Rethinking the role
of statistical typicality in ordinary causal attributions. Studies in History and Philosophy of
Biological and Biomedical Sciences, 43, 814–820.
Wittgenstein, L. (1958). Philosophical investigations (G. E. M. Anscombe, Trans.). New York:
Macmillan.
Woodward, J. (2017). Scientific explanation. In E. Zalta (Ed.), Stanford encyclopedia of philoso-
phy. https://plato.stanford.edu/archives/fall2017/entries/scientific-explanation/
Yablo, S. (1992a). Cause and essence. Synthese, 93, 403–449.
Yablo, S. (1992b). Mental causation. Philosophical Review, 101, 245–280.
Part II
Methodology: Uncovering the
Representation of Causation
Chapter 3
Exploring the Representation of
Causality Across Languages: Integrating
Production, Comprehension and
Conceptualization Perspectives
Erika Bellingham, Stephanie Evers, Kazuhiro Kawachi, Alice Mitchell,

Sang-Hee Park, Anastasia Stepanova and Jürgen Bohnemeyer
Abstract We present three new studies into the representation of causality across
languages and cultures, drawing on preliminary findings of the project Causality
Across Languages (CAL; NSF Award BCS-1535846 and BCS-1644657). The first
is an examination of the strategies that speakers of different languages employ
when verbalizing causal chains in narratives. These strategies comprise the output of
decisions concerning which subevents to represent specifically, which to represent
in an underspecified manner, and which to leave to nonmonotonic inferences such
as conversational implicatures. The second study targets the semantic typology
of causative constructions. We implemented a multiphasic design protocol that
combines the collection of production data with that of comprehension data from
a larger number of speakers. Goodness-of-fit judgments were collected based on
an eight-point scale. We found a strong main effect of language and of domain
of causation (physical vs. psychological vs. speech act causation); in contrast, the
involvement of an intermediate event participant in the causal chain did not exert
a significant effect. The third study investigates whether culture modulates the
effect of intentionality on nonverbal attributions of responsibility. A linear mixed
effects regression model indicated a significant interaction between intentionality
and population, in line with previous findings by social psychologists. These studies
represent the first large-scale comparison of how speakers of different languages
categorize causal chains for the purposes of describing them.
E. Bellingham · S. Evers · S.-H. Park · A. Stepanova · J. Bohnemeyer ()

Department of Linguistics, University at Buffalo, Buffalo, NY, USA
e-mail: ebelling@buffalo.edu; saevers@buffalo.edu; sangheep@buffalo.edu; jb77@buffalo.edu
K. Kawachi
National Defense Academy of Japan, Yokosuka, Japan
A. Mitchell
Institute for African Studies and Egyptology, University of Cologne, Cologne, Germany
e-mail: alice.mitchell@uni-koeln.de

https://doi.org/10.1007/978-3-030-34308-8_3
76 E. Bellingham et al.
Keywords Causal chain · Directness · Domain of causation · Iconicity ·

Intentionality · Responsibility attribution · Semantic typology ·
Underspecification
3.1 Introduction
In this chapter we provide an overview of the goals and methodologies of the

international collaborative research project Causality Across Languages (CAL).
CAL investigates the extent to which the representation of causality in language
and thought is variable across languages. To this end, we have been gathering
data on how causality is represented in language and thought from a typologically,
genealogically and areally diverse range of populations. Eventually we also plan to
investigate to what extent these verbal and nonverbal datasets are predictive of one
another. Such an investigation can take one of two directions or perspectives, both
of which we believe should eventually be explored:
Perspective I: Look for naturally occurring data on the verbal and nonverbal
representation of causality in different communities. To compare such datasets,
one then requires a set of criteria for the diagnosis of representations of
causality and a standard of comparison: some idea of the properties in which
representations of causality might differ from one another.
Perspective II: Start out from a set of ideas of the dimensions along which rep-
resentations of causality in language and thought might vary across populations,
encode these in a set of scenarios, and study how members of different cultural
and speech communities encode these scenarios and reason about them when
given appropriate tasks.
We focus on the second perspective. However, either perspective introduces a
paradox: it presupposes assumptions regarding which causal chain properties are
relevant in the representation of causal chains across languages. To initiate this
investigation it is necessary to start with a set of properties we assume to be relevant,
yet at the same time, discovering such a framework of variables is one of the
principal goals of cross-cultural research on the representation of causality. We
believe that the only solution to this paradox is an approach that starts out from
a set of assumptions that is maximally informed by the available cross-cultural and
cross-linguistic literature and then revises these assumptions continuously on the
basis of the emerging evidence in the course of the investigation.
In this spirit, we present here a set of causal chain properties gleaned from
previous cross-linguistic research, along with three case studies whose design
manipulates these properties as independent variables in both linguistic research and
research on nonverbal cognition. It should be noted that the findings of these studies
are preliminary. Of greater significance is the innovative methodology discussed
here. While our findings are promising, the methodological contribution of this
3 Exploring the Representation of Causality Across Languages 77
research is the central concern of this chapter. Section 3.2 describes the causal
chain properties under investigation (a set of independent variables with two or more
possible values), and the representation of different combinations of variable values
in our video stimuli. In Sect. 3.3, we provide three case studies: cross-linguistic
experiments designed around these stimuli to investigate different aspects of the
conceptualization and verbal representation of causality. Section 3.4 reflects on the
causal chain properties and stimulus design in light of our experience in the three
experiments: we discuss the aspects which ran smoothly, the design limitations we
uncovered, and the improvements we will implement for future investigations.
3.2 A Study Design for Cross-Population Research on Causal

Language and Thought
In this section, we introduce the framework of independent variables (causal chain

properties) that the CAL studies have been designed around, and describe the stimuli
we have created to represent different combinations of these independent variables.
A comparison of multiple languages or cultures necessitates the use of concepts
that serve as standards of comparison. The first validity threat faced by any
cross-cultural or cross-linguistic research is the potential bias introduced by these
notions. The risk that these notions are biased toward the categories and concepts
of the cultural and linguistic communities most familiar to the researchers must
be minimized. For the purposes of cross-population research into representations
of causality, this means first of all that no notion of causality that is specific to
the members of certain cultural and linguistic communities – e.g., to speakers
of ‘Standard Average European’ languages – should be imposed on the study
populations. In addition, the same kind of bias must also be avoided in the definitions
of the independent variables of the study designs. The dilemma raised by this
requirement is that it is impossible to know whether certain notions are applicable
to particular languages and cultures without studying the representation of the
particular conceptual dimension in these languages and cultures and comparing the
results to those obtained from members of other populations.
There is to date no solution to this dilemma that is universally or at least
standardly accepted among cultural anthropologists, social psychologists, and
linguists. All proposed solutions continue to be subject to (sometimes intense)
controversy. However, we believe we can assume that at least this much is standardly
agreed upon in the social and behavioral sciences: that it is crucial to maintain a
careful distinction between the emic concepts of particular cultural and linguistic
communities and the etic concepts a given study treats as independent of individual
languages and cultures in terms of its design (e.g. Harris 2001). The terms ‘emic’
and ‘etic’, abstracted from ‘phonemic’ and ‘phonetic’, have been standard in
cultural anthropology and anthropological linguistics since Pike (1967). They are
used to distinguish two perspectives on cultural and linguistic phenomena: the
emic perspective, which classifies the phenomena in the way members of the
particular community do, and the etic perspective, which strives to classify the
same phenomena in a matter that is valid for cross-cultural and crosslinguistic
comparison.
In the remainder of this section, we lay out concepts of causality and the proper-
ties of causal chains that we treat as strictly etic notions. Specifically, we investigate
the cognitive and verbal representations of complex events in members of different
populations under the understanding that these complex events instantiate various
different types of causal chains in a purely etic sense. That is to say, we do not make
the claim that these complex events are conceptualized as causal chains emically
according to whatever folk theories of causality the members of the different
populations might have (if any). We do, however, hope that the research based on the
etic grid of variables laid out below can ultimately help discover emic differences in
the conceptualization of causality. The case study presented in Sect. 3.3.3 has indeed
uncovered results that are at least suggestive of such emic differences.
A causal chain is a complex event consisting of minimally a causing subevent
and a resulting subevent, with a causal relation between the two subevents.1,2
But what is a causal relation? The criteria that are used for inferring causality
have been the subject of much research in the social and behavioral sciences and
philosophy. We do not commit to a single monolithic concept of causality, but
consider the possibility that causal inferences are informed by a cluster of properties
(spatiotemporal contiguity; probabilistic dependence; counterfactual dependence;
beliefs about underlying regularities; etc.) that do not necessarily all co-occur
(‘causal pluralism’; cf. Anscombe (1971) and Heider & Simmel (1944), inter alia;
cf. Grimshaw (2000) for a summary). In order to simplify the present study, we
include only scenarios that meet all of these properties.3
The remainder of this section describes the particular dimensions of semantic
variation in causal chains (the independent variables) around which our studies
are designed, and the representation of different combinations of these variables
in video stimuli.
1 When we define a causal chain in terms of a series of causally related events (cf. Davidson
(1969), Parsons (1990), and Croft (1998)), we are well aware of an alternative perspective which
centers around force dynamic interaction (Talmy 1988, 2000). Both of these approaches have been
informing our work, although the complex event view has been more central.
2 The term ‘subevent’ refers to an event that is part of another event. We treat causal chains as
complex events that have proper parts that are events in their own right and thus subevents.
3 One exception to this is scenarios involving ‘letting dynamics’, which we explore as a variable in
a supplementary set of stimuli, as described in Sect. 3.2.1.

3.2.1 The Causal Chain Properties Under Consideration
Our study is designed to focus on four major dimensions of semantic variation

in causal chain types: ‘mediation’, ‘participant type’, ‘participant behavior’ and
‘resulting event type’. (We focus on ‘resulting event type’ rather than causing event
type for the reason that existing literature suggests that resulting event type matters
(cf. Smith 1978). Additionally, most causal constructions do not specify causing
events.) Each dimension of variation can be broken down into one or more variables,
and different combinations of these variables are represented in video stimuli. An
additional dimension (‘force dynamics’) is explored in a supplementary set of video
stimuli.
3.2.1.1 Mediation
Mediation is one dimension of causal chain complexity, measured in terms of the

number of causal chain participants. There is no real world limitation on the number
of participants or the number of events in a causal chain, however we restrict
the domain of our study to include only causal chains involving 2–4 participants
operating in distinct positions within the chain. One participant is the initiator of
the causal chain (the causer), one is the finally affected participant (the affectee),
and (in chains with 3–4 participants) the remaining participants are involved in
intermediate segments of the causal chain. Following Bohnemeyer et al. (2010), we
consider both ‘unmediated’ causal chains, which involve only two participants (a
causer and an affectee), as well as causal chains which also incorporate one or two
intermediate participants (‘mediated’ causal chains). An intermediate participant
does not initiate the causal chain, nor are they the finally affected participant
in the causal chain: the actions of the initial participant (the causer) affect the
intermediate participant in some way, and this in turn causes the resulting event,
in which the final participant in the causal chain is affected. The intermediate
participant could be human (in which case we call it the intermediator),4 or it
could be an inanimate instrument used by either the causer or the intermediator.
We have broken mediation down into two binary variables (features): PRESENCE
OF INTERMEDIATOR and PRESENCE OF INSTRUMENT (unmediated clips lack both
PRESENCE OF INTERMEDIATOR and PRESENCE OF INSTRUMENT ).
4 Note that this term is used by some authors to denote the finally affected participant in the causal
chain (human or inanimate), or the finally affected human participant in the causal chain. For
Dixon (2000), the intermediator is the original A argument in the pre-causativized version of the
clause. As we are defining our variables in terms of the etic properties of causal chains, rather
than the emic properties of causative descriptions, this definition is not appropriate. We distinguish
intermediator, an intermediate human participant, from affectee, the final participant (human or
non-human) in the causal chain.
Mediation is closely related to the concept of directness of causation. Directness

of causation is frequently cited as the contrasting semantic feature between two
different causative constructions within a language (e.g. Comrie (1981), Dixon
(2000), Shibatani & Pardeshi (2002), Wolff (2003)), however there is considerable
variation in how directness is defined (see Escamilla (2012), for an overview).
Bohnemeyer et al. (2010) propose that directness of causation can be divided
into three dimensions: mediation (as defined above), spatio-temporal contiguity of
causing and resulting subevents, and force dynamics (letting versus causing, cf.
Talmy 2000). Spatio-temporal contiguity was excluded as a dimension of variation
in the present study design, see Sect. 3.2.2 for discussion. Force dynamics (letting
versus causing) is included as a supplementary dimension of variation, discussed in
Sect. 3.2.1.6. In Sect. 3.3.1, we discuss the concept of directness further, proposing
an analysis whereby directness is conceptualized as a function of all semantic
predictors of causal chain complexity.
3.2.1.2 Participant Type
Any type of entity in the real world could potentially participate in a causal chain,
however we restrict the domain of our study to include only human, inanimate, or
natural force participants. Each causal chain participant in our study is filled by a
restricted set of participant types. We define control as the ability of initiate (partial
control) and terminate (total control) an action at will. We consider only human or
natural force causers: human causers are potentially controllers (although they do
not always have control over their actions, they have the potential for control), and
natural forces (e.g. the wind, a wave, fire) are non-controlling causers/instigators,
while inanimate participants lack the wherewithal for control or non-controlled
instigation of a causal chain. We do not consider animals or ‘animate objects’
(machines/robots) (see Wolff et al. (2009) for experimental evidence of variation
in the treatment of ‘energy generating’ inanimate causers cross-linguistically).
As described in Sect. 3.2.1.1, intermediate participants are already divided into
intermediators (human), and instruments (inanimate). Affectees can be human or
inanimate. We do not consider any instances of natural forces as intermediators,
instruments or affectees. Besides simplifying the design of the study, an additional
motivation for excluding other types of causal chain participants is the ability
to clearly and unambiguously represent the participants in video stimuli. The
dimension of participant type can be captured in two variables: causer type
(HUMAN or NATURAL FORCE), and affectee type (HUMAN or INANIMATE).
3.2.1.3 Degree of Participant Autonomy
This dimension incorporates notions of intentionality and control (as defined

above) into a fine-grained classification of the degree of participant autonomy. It
is intricately connected to other variables such as the ‘domain’ of causation (cf.
below).
Human causers prototypically possess the highest level of autonomy. They are
conceptualized as potentially acting not as a result of any external event/stimulus,
but as independently initiating the event in their own mind. We distinguish
two levels of causer autonomy: INTENTIONAL, and UNINTENTIONAL.5 Human
intermediators and human affectees on the other hand do not initiate the causal
chain, and by definition their involvement in the causal chain has a cause external
to themselves (i.e. the causing subevent). There are a number of different ways
they might interact with the preceding subevent, each of which generates a different
degree of autonomy for the intermediator/affectee.
The highest level of autonomy that a human intermediator/affectee can hold is
to respond intentionally to some external stimulus, which compels them to act by
some (variable) degree. In our stimuli, this often takes the form of a request or
directive (speech act causation) from a human causer. We recognize that the degree
to which a person is compelled to act by a speech act (i.e. the degree of autonomy
they possess in deciding whether or not to act) is highly variable, and presumably
depends on the power dynamics between the two individuals (and whether there are
perceivable consequences for not complying with the request/directive). However
we suggest that generally the level of autonomy an intermediator/affectee has in
intentionally responding to a directive/request is greater than that when responding
unintentionally to some other external stimuli/physical forces. Our stimuli also
include several scenarios in which the intermediator/affectee responds intentionally
but in response to a non-speech act event. These are listed in (1).
(1) a. It is rainingCR , and so a manI M opens an umbrellaAF .6
b. A huge waveCR is approaching, and so a manAF runs away.
c. A womanCR is singing very loudly and out of tune, and so a womanAF
covers her ears and leaves.
Human intermediators and affectees may also act reflexively, in response to some
external stimulus. This could potentially involve a huge range of different external
stimulus types (e.g. visual, auditory, tactile, olfactory. . . ), however we restrict these
possibilities to physical contact, unexpected loud noises, and visual stimuli which
generate an (at least partially uncontrolled) urge to act, e.g. laughing in response to
someone pulling a funny face, or yawning in response to someone else yawning.
The force with which physical contact is made varies across our stimulus scenarios:
in some cases it is so great that the intermediator/affectee is propelled purely by
the momentum of the causing event (and does not act in any additional way),
and in other cases the force is weaker and they are startled by it. We assume that
intermediators/affectees who are physically propelled have the least autonomy, less
than intermediators/affectees who act reflexively, or intentionally in response to
some external stimulus.
5 Cf.Sect. 3.3.3.1 for discussion of a further breakdown of causer intentionality into ‘intention to
action’ and ‘intention to outcome’.
6 CR, IM and AF stand for ‘causer’, ‘intermediator’ and ‘affectee’ respectively.
Natural force causers and inanimate affectees are each restricted to a single
possibility for this dimension: we assume that intentionality is not a relevant
dimension for a natural force causer, and that inanimate affectees can only be
involved in the causal chain by being physically impacted. Hafeez (2018) presents
a detailed analysis of intentionality, volitionality and control in Urdu (Indo-Aryan;
India and Pakistan) based on the CAL Clips. The clause structure of Urdu and other
Indo-Aryan languages is sensitive to these variables in two aspects: case alternations
on causer and intermediator NPs and light verb selection in complex predicates.
3.2.1.4 Domain of Causation
Related to both participant type and degree of participant autonomy, the domain
of causation variable is intended to capture the potential impact of domain-specific
knowledge and conceptualizations in representations of causality. Quite a few such
distinct domains have been suggested in the anthropological and psychological lit-
erature. However, in the design of the CAL Clips, we restricted ourselves to a broad
distinction between PHYSICAL CAUSATION and NON-PHYSICAL CAUSATION,
where the latter can be broken down further between PSYCHOLOGICAL CAUSATION
and SPEECH ACT CAUSATION. In the CAL Clips, all instances of PHYSICAL
CAUSATION involve force interactions in the sense of Classical Mechanics: pushing
events, ballistic collisions, falling events, events of separation in material integrity
(cutting and breaking), and throwing actions. We did not include thermodynamic,
electrodynamic, or chemical interactions, to name just the most obvious conceivable
additional subdomains.
In PHYSICAL CAUSATION, the intentionality of the affectee or intermediator is
generally irrelevant. This is potentially different in PSYCHOLOGICAL CAUSATION,
which we define as a causal chain one link of which is a cognitive state change in
the affectee or intermediator. The response may be largely an involuntary reflex, as
when the affectee/intermediator is startled or scared, or may involve a decision on
the affectee/intermediator’s part, e.g., a decision to leave in order to avoid continued
exposure to an unpleasant stimulus.
SPEECH ACT CAUSATION can be understood as a special case of PSYCHOLOG-
ICAL CAUSATION . Here, the causal link between causer and affectee/intermediator
is a communicative act. This entails that the affectee/intermediator carries out the
caused action with some autonomy: while their response may be involuntary in the
sense that they did not initiate the causal chain, it is nevertheless typically intentional
and controlled.
Beyond PHYSICAL CAUSATION, PSYCHOLOGICAL CAUSATION, and SPEECH
ACT CAUSATION , other domains in which the conceptualization of causality is
potentially subject to domain-specific knowledge and folk theories include social
causation (involving collective agency) and biological causation. We decided to
disregard these in the design of the CAL Clips in the interest of keeping the stimulus
set small.
3.2.1.5 Resulting Event Type
We distinguish three kinds of resulting events. The final event in the causal chain
can be a physical or psychological state change (e.g. an egg breaking, a human
sitting), a location change (e.g. a ball flying out the door, a person leaving the room),
or a process7 (e.g. a swing swinging back and forth). This can be captured in a
single categorical variable (resulting event type) with three levels: STATE CHANGE
versus LOCATION CHANGE versus PROCESS. Resulting event type is recognized by
Dixon (2000) as a parameter relevant to the applicability of causative constructions
in some languages. Note that this dimension interacts with degree of participant
autonomy. In the case of human affectees who are not physically propelled, the
resulting event (STATE CHANGE/LOCATION CHANGE/PROCESS) must be preceded
by some psychological change in the Affectee (e.g. a decision to act, or being
startled).
An additional resulting event type is considered in our supplementary stimuli:
projectile breaking. Here the affectee changes state (breaks) as a result of impact
with a surface following projectile motion (and the projectile motion occurred as
a result of the causer/intermediator’s action). In one example of a PROJECTILE
BREAKING clip, a woman (the causer) pushes a man (the intermediator), he drops
the plate he is holding to the floor, and the plate shatters upon contact with the
floor. In the initial stimulus design, we did not differentiate between change of
state and projectile breaking as distinct resulting event types. In piloting, however,
we observed that descriptions of these clips often patterned quite differently from
clips in which the affectee’s state change occurred as a direct result of contact with
the intermediator/instrument/causer. Descriptions of scenarios involving projectile
motion would typically encode more subevents, and the surface seemed to be treated
almost like an additional participant in the causal chain.
3.2.1.6 Letting Dynamics
Talmy’s force-dynamics framework (Talmy 1988, 2000) conceptualizes causa-

tion as one type of force-dynamic interaction between entities. Other types of
force-dynamic interactions (such as letting, helping, hindering, preventing, etc.)
differ from causation and from each other with respect to the amount and direc-
tion/tendency of (not necessarily physical) force exerted by each entity. The force-
dynamic approach focuses on interactions between two entities (an ‘antagonist’
acting upon an ‘agonist’: in our terms, a causer acting on a intermediator or affectee,
7 We use the term ‘process’ in the sense of von Wright (1963) and Mourelatos (1978), i.e., for
dynamic situations that do not involve state change. In this usage, it is more or less synonymous
with (Vendeler 2005) ‘activity’. We prefer ‘process’ to avoid misinterpretations to the effect of
controlled actions. All ‘processes’ in the CAL Clips are either externally caused (a swing swinging)
or, at least by default, conceptualized as involuntary and uncontrollable (a person sneezing,
yawning, or laughing).
or a intermediator acting on an affectee, depending on which link in the causal chain

is considered).
In the case of causation, the agonist and antagonist are exerting force in opposite
directions: the agonist has a tendency towards remaining in the same state or
location and the antagonist has a tendency towards (the agonist’s) motion/change.
The force exerted by the antagonist is greater than that of the agonist, and
so causation occurs. In the case of letting, the agonist has a tendency towards
movement/change, and the antagonist is impinging on the agonist and preventing
it from changing/moving. The antagonist then ceases to impinge on the agonist (by
removing a blockage or restriction), and the agonist fulfills its inherent tendency.
Letting versus causation was explored as a variable in Bohnemeyer et al. (2010),
although only scenarios involving gravity as the inherent tendency of the agonist
were considered. The importance of this variable was found to differ across the
sample languages (Dutch, Ewe, Japanese, Lao, and Yucatec): at least one Lao
construction was highly sensitive to this distinction (the construction could be used
to express situations with causation dynamics, but not letting dynamics).
In the present study, we capture the contrast between force-dynamic causation
and letting in the supplementary set of stimuli (in all of the core stimuli, all force-
dynamic interactions in the chain are of the causation type). We consider two
different types of inherent tendencies: gravity, and continued motion along a path.
Ten of the fifteen supplementary stimulus clips involve at least one letting type
interaction. These letting interactions either consist of dropping an item (initially
impinging on the item by preventing it from fulfilling its gravity-given tendency to
fall to the floor, then ceasing to impinge, allowing the object to fall to the floor),
or stepping away from a position where someone’s path was blocked (initially
impinging on the person by preventing them from fulfilling their inherent tendency
of walking along their chosen path, then ceasing to impinge by stepping aside and
allowing them through).
3.2.2 The Video Representation of Causal Chain Types
We captured different combinations of the variables of each dimension in video

stimuli. All of the video clip stimuli were live action videos of interactions among
humans, natural forces, and inanimate objects (or some subset of these) recorded by
and starring members of the University at Buffalo Semantic Typology Lab, or taken
from YouTube.8 We chose to use live action video rather than animation or static
representations (photos or line drawings), since the interpretation of animation and
static images relies on conventions that may be subject to cross-cultural variation in
ways that the interpretation of recorded video is not.
8 The field manual and stimuli for all CAL studies is available online at https://
causalityacrosslanguages.wordpress.com/ project-summary/ field-manual-and-stimuli.
Causation is a complex concept, with many different dimensions of variation,

and it would not be practically feasible to consider every possible dimension and
every possible combination of values from different dimensions, at least not in a
study involving primary data collection from speakers of a wide range of languages.
We constrained the design of the study by restricting the dimensions of variation we
considered. A major motivating factor for choosing some dimensions over others
is the ease to which differences in these dimensions could be represented in live
action video with obvious cues and unambiguous representation of causation. We
also aimed to produce scenes which are not culturally specific: we did not want
to show any actions which would be seen as either offensive or very unusual
or uninterpretable in some cultures. For example, spatio-temporal contiguity was
investigated by Bohnemeyer et al. (2010) using animations, however it was apparent
that some participants did not perceive the events in the stimuli as involving a causal
relation.
Among the dimensions we did consider, there are many combinations of variable
values which are not possible. For example, as discussed in Sect. 3.2.1.3, only
humans can behave intentionally or unintentionally (while humans and inanimate
objects can both be physically impacted). Below we describe several examples of
causal chain types as they are represented in our stimuli. For a full list of the core
and supplementary stimuli, see Appendices 1 and 2.
(2) a. HO5_cuptower (cf. Fig. 3.1 below):

A man slaps a tower of cups, which causes the tower to collapse.
Mediation: Unmediated
Participant type: Causer: Human; Affectee: Inanimate
Degree of participant autonomy: Causer: Intentional
Domain: Physical
Resulting event type: State change
Letting dynamics: No
Projectile breaking: No
b. HUO2_cups (cf. Fig. 3.2 below):
A woman sneaks up behind a man and yells loudly, startling him, and
causing him to knock over a tower of cups.
Mediation: Mediated (Intermediator)
Participant type: Causer: Human; Affectee: Inanimate
Degree of participant autonomy: Causer: Intentional;
Affectee: Reflexive reaction to noise
Domain: Intermediator: Psychological;
Affectee: Physical
Resulting event type: State change
c. UU2_sneeze:
A woman sneezes loudly behind another woman, causing her to jump.
Mediation: Unmediated
Participant type: Causer: Human; Affectee: Human

Degree of participant autonomy: Causer: Unintentional;
Affectee: Reflexive reaction to noise
Domain: Psychological
Resulting event type: Process
Having introduced this grid of variables, we now proceed to illustrate its applica-
tion in the design of three separate cross-population studies of causal language and
cognition. These studies primarily manipulate the variables ‘mediation’, ‘participant
type’, ‘participant autonomy’, ‘domain of causation’, and ‘resulting event type’
to investigate their relationships with different aspects of the conceptualization
and verbal representation of causality. These aspects are tightly interrelated. Con-
sequently, while the specific domains of each study presented here differ, each
provides data for or must be evaluated based on the results of others. The first study,
presented in Sect. 3.3.1, examines patterns of linguistic descriptions of causal chains
across causal chain types and linguistic populations. This data was used extensively
as the basis for the production of verbal stimuli in our second study, presented in
Sect. 3.3.2, which examined participant judgments of descriptions of causal chains.
Our final study, described in Sect. 3.3.3, examines assignment of responsibility to
members of a causal chain, and uses a non-verbal task to explore conceptualization
of causality at a cultural level. Differences in responsibility assignment observed in
this task will ultimately be compared to the production and speaker judgment data
collected in Studies 1 and 2 to determine if a link may be present between causal
cognition and a community’s linguistic practices. Where possible, all three studies
were conducted with each speaker population. As a result, some participants in each
population participated in all three studies, but it is not the case for any population
that complete participant overlap occurred for all three. In order to minimize
the impact that participation in any given study may have had on the results of
subsequent experiments, studies were sequenced such that if participants were
involved in multiple tasks, they completed the non-verbal responsibility assignment
task first (Case Study 3), followed by the discourse production task (Case Study 1),
and concluding with the sentence ratings task (Case Study 2).
3.3 Applications: Three Case Studies
3.3.1 Case Study 1: Causality in Discourse
The first study that we present investigates the role of conversational implicatures in
narrative descriptions of causal chains, and how usage patterns differ across different
causal chain types, across speakers of the same language, and across different
languages. We focus on the distribution of semantic underspecification of event
information: what types of event information do speakers (of particular languages)

make explicit versus leave underspecified, and how is this affected by the type of
causal chain they are describing.
Within a language, there are typically many different ways that a speaker could
describe an event. Consider the descriptions in (3) (adapted from Bohnemeyer et al.
(2010)): these could plausibly all describe the same event, although they differ with
respect to the information they entail versus leave underspecified.
(3) a. Floyd opened the door.
b. Floyd pushed the door open.
c. Floyd pushed the door and it opened.
d. Floyd pushed the door and opened it.
The underspecification of three different types of event information is illustrated
in (3): subevent relation (3c, 3d), subevent kind (3a, 3d), and shared subevent
identity (3d). (3b) specifies the kind of subevent for both the causing and resulting
subevents, the relationship between the two subevents, and does not describe the
same subevent twice. We assume that the semantics of (3) are something like those
in (4).
(4) a. ∃e1 .∃e2 . ACT(e1 ,Floyd’) ∧ UGR(e2 ,Door’)∧ Open(e2 ) ∧ CAUSE(e1 ,e2 )9
b. ∃e1 .∃e2 . ACT(e1 ,Floyd’) ∧ UGR(e1 ,Door’) ∧ Push(e1 ) ∧
UGR (e2 ,Door’)∧ Open(e2 ) ∧ CAUSE (e1 ,e2 )
c. ∃e1 .∃e2 . ACT(e1 ,Floyd’) ∧ UGR(e1 ,Door’) ∧ Push(e1 ) ∧
UGR (e2 ,Door’)∧ Open(e2 )
d. ∃e1 .∃e2 .∃e3 . ACT(e1 ,Floyd’) ∧ Push(e1 ) ∧ ACT(e2 ,Floyd’)
∧ UGR(e3 ,Door’)∧ Open(e3 ) ∧ CAUSE(e2 ,e3 )
More detailed explanations of each type of underspecification are given below.
Subevent relation: In narrative description, speakers do not necessarily make all

causal relations explicit, relying instead on stereotype implicatures10 (Levinson
2000, p. 114) to convey a causal relation.11 In descriptions like (3c), the
relationship between the events described in the two clauses is underspecified.
The most natural reading of (3c) is that pushing on the door caused it to open,
although it is still possible to force a reading that the two events are not causally
connected (as in (5)).
9 ACT and UGR stand for ‘actor’ and ‘undergoer’, respectively.

10 We assume that conversational implicatures are defeasible default interpretations, and unlike
presuppositions are polarity dependent. Entailments, on the other hand, are non-defeasible but also
polarity dependant.
11 An alternative to the Gricean account relies instead on coherence relations to motivate the
inference of a causal relation between two event descriptions (see Kehler & Cohen 2018 and
references therein).
(5) Floyd pushed the door and it opened when Sophie stepped in front of the
sensor.
A semantic representation for (3c) is shown in (4c): the causal relation between
e1 and e2 (CAUSE(e1 ,e2 )) is implicated. A pattern of causal underspecification
in narrative descriptions of causal chains was observed by Bohnemeyer et al.
(2010). Speakers of Dutch, Ewe, Japanese, Lao and Yucatec were asked to
describe what happened in video clips depicting short causal chains (similar to
those used in the present study), and would frequently describe the causal chains
over multiple clauses without specifying causal relations.
Subevent kind: Semantic information about the nature of a subevent is left
underspecified. While (3b) provides a semantic characterization of the type of
event which caused the door to become open (pushing), (3a) does not: Floyd
could have pushed the door, or pressed a button, or stood in front of a sensor.
(3a) still entails that Floyd was the Actor in some causing subevent (and that
this mystery causing subevent occurred), but the precise nature of the causing
subevent is underspecified. A semantic representation of (3a) is shown in (4a):
the nature of e1 (Push’(e1 )) is implicated (assuming a stereotypical door that
swings horizontally on hinges, as opposed to a sliding door or trapdoor).
Causativized lexical items and the causative senses of polysmous causative-
inchoative-alternating verbs (e.g., The door opened vs. Sally opened the door)
typically encode a semantically underspecified subevent (Sally opened the door
does not specify what Sally did to open the door – she might have twisted the
doorknob and pushed the door open, or she might have dynamited the door;
cf. the principle of ‘morpholexical transparency’ (Bohnemeyer 2007); ‘man-
ner/result complementarity’ Levin & Rappaport-Hovav (1995)). The same holds
for light verbs in periphrastic causative constructions (e.g., Sally made Floyd
reconsider his position again does not specify what it was that Sally did that
caused Floyd to reconsider: it might have been a suggestion, a threat, or Sally’s
own example). As with subevent relation underspecification, the underspecified
information can typically be recovered via a stereotype implicature: we infer that,
in the absence of a marked description, the nature of the causing event matches
that which is a stereotypical cause of the resulting event it is paired with (or
at least we assume it to be whatever we calculate as the most likely given the
context).
Shared subevent identity: Sometimes a causal representation includes two
subevent descriptions such that the intended interpretation of the representation
requires the inference that these two actually refer to the same subevent. This
shared identity may be an entailment or an implicature. In the latter case, we
may say that the shared identity of the two subevents is underspecified. An
example is (3d): the default reading is that the pushing caused the opening, and
not that Floyd pushed the door, and then opened it by pressing a button. The
description is still truth-conditionally compatible with the latter situation, and
the description underspecifies whether the pushing event, and the underspecified
causing event denoted by the transitive causative verb open are the same event
or not. A semantic representation of (3d) is shown in (4d): the shared identity of

e1 and e2 is implicated (e1 = e2 ).
The next section lays out the methodology for exploring the distribution of these
three kinds of underspecification in narrative descriptions of causal chains. Are
there certain types of causal chains in which one or more causal relations are more
likely to be left underspecified? Does the position in the causal chain affect the
likelihood of underspecification? Do speakers within a language behave uniformly?
Do speakers across languages behave uniformly?
3.3.1.1 Methodology
We collected descriptions of the CAL Clips from 10–20 speakers of English (Ger-
manic), Japanese (Japonic), Korean (Isolate), Russian (Slavic), Sidaama (Cushitic)
and Yucatec (Mayan).12 Each participant watched a clip, and was then asked
to respond to the question ‘What happened?’.13 In order to clarify the level of
informativity that they should provide, participants were instructed to respond as
though they were describing what happened in the clips to a person who had not
seen it. Specific examples of translations of ‘what happened?’ that were provided to
participants included ‘What would you say to your friend if she walked in soaking
wet?’ and ‘How would you ask about the contents of a novel or a TV episode?’
The open-ended nature of the task meant that participants were free to use any
strategy they liked for describing the clip. We designed an annotation system to
allow us to compare descriptions across clips and speakers in terms of: (1) which
of the events in the causal chain depicted in the clip were represented in the
description; (2) whether those events were semantically specified or underspecified;
and (3) whether the causal relation between each event in the causal chain was
entailed by the description or merely implicated. In order to compare descriptions
of the same clip across speakers, it was necessary to identify a maximal set of
(relevant) subevents for the causal chain depicted in each clip. For example, in clip
HO5_cuptower, in which a man slaps a tower built from paper cups, causing the
tower to collapse, the possible subevents that a speaker might mention are given
in (6):
(6) Event 1: man hits tower of cups

Event 2: tower of cups collapses/falls
12 The CAL Clips comprise 43 core clips and 15 supplementary clips. Descriptions of solely the
core clips were collected with Russian speakers. At the time of writing, data has also been collected
(but not yet analyzed) from Basque, Datooga (Nilotic, Tanzania), Ewe (Gbe, Ghana and Togo),
Mandarin, Nahuatl (Uto-Aztecan, Mexico), Spanish, Urdu (Indo-Aryan, Pakistan and India), and
Zarma (Songhay, Niger). Analysis is ongoing.
13 With the Japanese participants, an indirect question construction was used, since the direct form
was considered too brusque.

(7) a. Mužčina sloma-l piramidk-u iz staka-nov.

man(NOM.SG) break.down-PST tower-ACC.SG from cup-GEN.PL
‘(The/a) man broke down the tower of cups.’ [Russian, RUS5]14
b. Someone hit a stack of cups and then the stack fell on the floor. [English,
S5]
The descriptions in (7) were provided by speakers of English and Russian in

response to clip HO5_cuptower. By identifying the maximal set of subevents (6),
it is then possible to identify for each description which of these subevents are
encoded, whether each subevent is underspecified, and whether the relationship
between the two subevents is underspecified. (7a) exemplifies subevent kind under-
specification: it includes a transitive causative verb slomat’ ‘crack’, ‘break down’,
which encodes both a causing and a resulting subevent with a causal relation entailed
between them but leaves the causing subevent semantically underspecified (the
description does not specify the man’s action). (7b) exemplifies subevent relation
underspecification: it also encodes two subevents, providing semantically specific
information about each, but leaving the relationship between the two subevents
underspecified (it does not entail that the hitting event caused the falling event).
Because we aim to compare the mapping between description and subevents not
only for the same clip across speakers and language, but also for different clips,
we required a way to relate the subevents in one clip to the subevents in other
clips. This enables us to study more precisely the distribution of underspecification
strategies across causal chains, and answer questions like: is the causal link between
subevent X and subevent Y more likely to be underspecified for some causal
chain types/languages? Or: where in the causal chain are speakers more likely to
underspecify subevent kind? To achieve this, we included in the coding schema
generalized subevent categories according to the position of events in the causal
chain relative to each causal chain participant (causer/intermediator/affectee), and,
for each clip’s maximal set of subevents, determined which of these generalized
subevent categories they fell under. The maximal set of generalized categories is
shown in (8), and examples of the application of these labels to the maximum set of
subevents in some sample scenarios is shown in (9).
(8) CAUSER ACT , INTERMEDIATOR RESULT , INTERMEDIATOR ACT ,
AFFECTEE RESULT , AFFECTEE ACT
(9) a. HO5_cuptower:
CAUSER ACT : man hits tower of cups
AFFECTEE RESULT : tower of cups collapses
14 Key to morpheme glosses: 3 – 3rd person; A – Cross-reference ‘Set A’ (ergative/possessor);

ACC – Accusative; B – Cross-reference ‘Set B’ (absolutive/stative); CMP – Completive status (per-
fective aspect and declarative/realis mood); D2 – Anaphoric/distal particle; DEF – Definiteness; F –
Feminine; GEN – Genitive; INC – Incompletive status (imperfective aspect and neutral/unmarked
mood); NOM – Nominative; PL – Plural; PRV – Perfective aspect; PST – Past tense; SG – Singular.
b. HMO4_cups:
CAUSER ACT : woman pushes man
INTERMEDIATOR RESULT : man falls into tower of cups
AFFECTEE RESULT : tower of cups collapses
c. UC1_sing:
CAUSER ACT : woman 1 sings loudly/badly
AFFECTEE RESULT : woman 2 is annoyed
AFFECTEE ACT : woman 2 leaves room
Each description was annotated in terms of which of the subevents were encoded
in the description (and how many times each subevent was encoded), whether each
subevent encoding was semantically specified or not, and whether the causal relation
between each subevent encoding was entailed or not.
3.3.1.2 Results and Discussion
This annotation scheme produces a large quantity of data reflecting the distribution
of different kinds of underspecification in the narrative descriptions. A large number
of different questions could potentially be asked of this data, and analysis is still
ongoing.
We found all three types of underspecification (subevent relation, subevent kind,
and shared subevent identity) across all six languages. Yucatec and Sidaama speak-
ers in particular produced at least one type of underspecification in almost every
description. Subevent relation underspecification was most frequent in Sidaama
and Korean. Subevent kind underspecification was most frequent in Yucatec and
Japanese. Subevent identity underspecification was most frequent in Japanese and
Yucatec. English and Russian had the two highest percentages of descriptions which
did not contain any of the three kinds of underspecification.
Languages vary in the lexical and morphosyntactic resources they have available
for the representation of causal chains. This variation may be partially responsible
for the differences we found in underspecification strategies. For example, we
might expect a higher rate of subevent kind underspecification in languages with a
richer inventory of transitive causative verbs, causative morphology, or periphrastic
causatives (all of which encode complex events and typically include an underspec-
ified causing event) and we might expect less subevent relation underspecification
in languages with productive resultative or serial verb constructions (or complex
predicate types which semantically specify multiple subevents).
Aside from the properties of the languages involved, another working hypothesis
that may partially account for variation in underspecification rates is that speech
communities with high literacy rates among speakers and a strong written tradition
are more likely to prefer more explicit linguistic forms with less underspecification
particularly of causal relations. Written registers are typically more explicit than
spoken registers: they are not subject to the same working memory limitations, and
can thus use more words to express the same concept. At the same time, since most
writing happens outside the situation context that is being written about, the need for
explicitness is greater. If a high proportion of speakers are frequently using written
language, then a preference for greater explicitness may transfer from written to
spoken registers. The sample of languages in our study is currently too small for
any serious empirical test of this hypothesis, but it is a potential line of inquiry for
future work.
Another hypothetical cultural factor driving underspecification is politeness.
It has been suggested that attribution of responsibility may be habitually more
circumspect in cultures in which responsibility implies a high potential for face loss
(e.g., Keenan 1989; cf. also Brown & Levinson 1987). This nexus too remains to be
explored.
Lexical and morphosyntactic factors, literacy, and the community’s politeness
ethos would all potentially affect causal attributions independently from one
another, and their effects would thus counteract one another (and potentially cancel
one another out).
3.3.2 Case Study 2: The Semantic Typology of Causality
The second case study to be presented here aims at a ‘semantic typology’ of causal
language. Semantic typology is the crosslinguistic study of semantic categorization.
It compares languages in terms of the lexical and morphosyntactic resources
their speakers use for communications that involve concepts of a given domain
– in this case, the domain of causality. Included in the scope of investigation
are the morphosyntactic, semantic, and pragmatic properties of these devices and
the speech community’s pertinent practices of language use. Cf. Evans (2010),
Koptjevskaja-Tamm (2015), and Moore et al. (2015) for general introductions to
semantic typology.
With the exception of a small pilot study presented in Bohnemeyer et al. (2010),
which was a direct precursor of the present study, the research discussed in this
subsection is the first of its kind – the first semantic typology of causality ever
undertaken to our knowledge. The most basic property that sets this research apart
from previous typological studies on causative coding devices is its perspective
(cf. Comrie (1981), Dixon (2000), Escamilla (2012), Kemmer & Verhagen (1994),
Shibatani (1976), Shibatani & Pardeshi (2002), and Song (1996); inter alia). These
previous studies do not look systematically at how different kinds of causal chains
are expressed across languages, but rather single out a few constructions per
language that the researchers identify as causative on largely implicit criteria and
then compare their meanings and use to one another. In contrast, our study proceeds
by observing systematically how speakers of different languages communicate
about a range of related concepts. Regarding the problem of ensuring an ‘etically’
valid definition of the notion of ‘causality’ without imposing it on the ‘emic’
semantic analysis of language-specific constructions, we refer the reader to the
discussion in the beginning of Sect. 3.2 above. As stated there, we assume an
etic definition of ‘causality’ consistent with a ‘causal pluralism’ approach. This

assumption is built into the design of the video stimuli described in Sect. 3.2.2
by restricting them to scenes that instantiate all the properties that have been
suggested by previous research as being potentially involved in the cluster concept
of ‘causality’.
Due to its inherent perspective of mapping concepts to expressions, production
data plays a privileged role in most approaches to semantic typology. The study
presented here goes beyond this by combining production- and comprehension-
based designs. The production phase involves the collection of descriptions of the
CAL Clips introduced in Sect. 3.2. During the comprehension phase, these serve as
the basis for verbal stimuli whose goodness of fit with respect to the CAL Clips
is assessed via acceptability ratings. In preparation for the comprehension phase,
the participating researchers, who are experts on their field languages, extract the
major causative coding devices from the production data. An inventory of response
types is compiled, and for each clip, a set of descriptions is created that instantiate
all major response types in the inventory. Descriptions of each scene instantiating
the full range of major response types are created with the help of first-language
speakers. These stimulus descriptions are then rated for their acceptability by a
minimum of 12 speakers per language. The advantages of this multiphasic design
are the following:
• It provides insights into the use of causative coding devices in both production
and comprehension.
• It produces both positive and negative evidence – that is evidence regarding both
preferred and dispreferred uses.
• With an implementation such as the one we chose, it permits a distinction
between descriptions considered to be false and descriptions considered to be
truth-conditionally adequate but pragmatically infelicitous.
• It permits data collection from a potentially large number of speakers per
language while keeping transcription demands manageable.
The specific research question that has motivated the study presented here
concerns the role of iconicity in causative descriptions across languages. It has
long been argued that across languages, morphosyntactically simpler causative
devices are preferred for conceptually and semantically simpler, more direct causal
chains, while morphosyntactically more complex descriptions are preferred for
more complex, indirect chains. Haiman (1983) calls this the Iconicity Principle
(cf. also Comrie 1981; Dixon 2000; Kemmer & Verhagen 1994; Rappaport Hovav
& Levin 2010; McCawley 1976, 1978; Shibatani 1976; Shibatani & Pardeshi 2002;
Talmy 2000; and Verhagen & Kemmer 1997, inter alia). For a simple illustration,
consider the following examples from Yucatec Maya:
(10) Le=máak=o’ t-u=nik-ah le=bàaso-s-o’b=o’.

DEF=person=D2 PRV-A3=scatter-CMP(B3SG) DEF=cup-PL-PL=D2
‘The man, he scattered the cups’
(11) a. #Le=x-ch’úupal=o’ t-u=nik-ah

DEF=female:child=D2 PRV-A3=scatter-CMP(B3SG)
le=bàaso-s-o’b=o’.
DEF=cup-PL-PL=D2
‘The girl, she scattered the cups’
b. Le=x-ch’úupal=o’ t-u=mèet-ah
DEF=F-female:child=D2 PRV-A3=make-CMP(B3SG)
u=nik-ik le=bàaso-o’b le=máak=o’.
A3=scatter-INC(B3SG) DEF=cup-PL DEF=person=D2
‘The girl, she made the man scatter the cups’
Example (10) was produced as a description of CAL Clip HO5_cuptower.

It shows a man collapsing a cup tower by slapping it with his hand (cf. Fig. 3.1).
The description features a base-transitive causative verb. The same coding device is
rejected in (11a) as pragmatically misleading in response to HUO2_cups, in which
a woman or girl is shown sneaking up behind a man who is building a cup tower.
She purposely startles him and he collapses the cup tower (cf. Fig. 3.2). In this case,
a simple transitive causative verb would be appropriate with the actor role assigned
to the male character, but not to the female one. When the female is to be construed
as the causer, the periphrastic causative construction in (11b) is preferred.
While this contrast seems straightforward enough, a recent statistical examina-
tion of published data from a typologically and areally broadly varied sample of 50
languages by Escamilla (2012) failed to find a significant correlation between direct-
ness of causation and morphosyntactic complexity. Escamilla classified causative
coding devices in the languages of his sample based on the information provided
Fig. 3.1 HO5_cuptower

Fig. 3.2 HUO2_cuptower
in published resources. He notes that he often relied on examples provided by

his sources (op. cit. 82), raising the question to what extent his investigation was
influenced by translations.
Escamilla applied the set of semantic and lexical predictor variables proposed by
Dixon (2000). Dixon does not define ‘directness’. The examples he gives include
what we call ‘mediation’ (causal chains mediated by a intermediator are less direct
than unmediated causer-on-affectee chains), but also distinctions of force dynamics
(letting something happen is less direct than causing it) and domain of causation
(physical impact is more direct than psychological impact). This comes close to
the abstract view of directness espoused in Bohnemeyer et al. (2010) and the
present study, which treats directness not as one semantic predictor variable among
others, but rather as a superordinate or “meta-”variable that summarizes the effects
of all individual semantic predictors on morphosyntactic complexity. In contrast,
despite using ‘directness’ in a more abstract sense, Dixon treats it as one predictor
of morphosyntactic complexity among others, such as intentionality and control.
This exacerbates the absence of a definition: apparently, directness is understood
as a more specific notion than simply the aggregate of all semantic properties that
predict morphosyntactic complexity, yet no set of criteria is laid down by which it
could be decided what counts as direct and what does not. Escamilla adopts Dixon’s
classification, making it difficult to know how he coded the constructions he found
in the descriptions of the sample languages he worked with.
Escamilla’s results are difficult to interpret. He did not find a significant
correlation between ‘compactness’ (i.e., morphosyntactic complexity) and any of
Dixon’s semantic predictors. As he readily acknowledges, this is easily explained
by the lack of valid data that would have allowed him to score a given construction
for a given predictor. Nevertheless, Escamilla singles out the absence of a correlation
between compactness and directness as particularly noteworthy:
In other words, this data set failed to produce empirical support for the Iconicity Principle:
low compactness is claimed, crosslinguistically, to correlate with less direct causative action
(as in the now-famous I killed him vs. I let him die (. . . )). This claim has been found to
hold for other sets of languages, and I do not suggest that it is not a valid generalization;
however, I also have no good explanation for the fact of the near random patterning we see
here. (Escamilla 2012: 89)
The study presented in this subsection permits a validation of Escamilla’s

findings against a sample of so far just four unrelated languages from three
continents: Datooga (Nilotic, Tanzania; data collected and coded by A. Mitchell),
Japanese (Japonic, Japan; data collected and coded by K. Kawachi); Sidaama
(Cushitic, Ethiopia; data collected and coded by K. Kawachi), and Yucatec (Mayan,
Mexico and Belize; data collected and coded by J. Bohnemeyer). The investigation
is ongoing; the four data sets analyzed here represent just a snapshot. In contrast to
Escamilla’s approach, our research is based on the actual observation of the behavior
of at least 12 speakers per language vis-à-vis a large set of verbal and nonverbal
stimuli following a rigid protocol.
3.3.2.1 Methods
Stimuli
In a first step, descriptions of the CAL Clips were either specifically collected for
this study from a few speakers of each language or, where available, were taken from
the data collected for the subproject on the verbalization of causal chains in narra-
tives discussed in Sect. 3.3.1. The researchers, who are experts on the grammars and
lexicons of the target languages, then created inventories of major response types,
where a response type was understood as comprising a single causative coding
device or a combination of causative coding devices. Example (11b) illustrates such
a combination: a base-transitive causative verb embedded in the complement of
a periphrastic causative construction. Our working definition of ‘causative coding
devices’ included any lexical expressions or morphosyntactic constructions that
encode two or more events and in suitable contexts entail both the realization of the
events and a causal relation holding between them in the ‘etic’ sense of ‘causality’
discussed in the beginning of Sect. 3.2. Where researchers were in doubt as to
whether a certain construction really could be considered causative, they verified
with the help of native speaker consultants using entailment tests.
Once an inventory of response types had been established, a set of descriptions of
each CAL Clip was created with the help of first-language speaker consultants. For
each clip, this set of descriptions instantiated every response type. Where no suitable
lexical material was available – e.g., no transitive causative verb that expresses the
relevant kind of action, or no transitivized verb featuring causative morphology –
a form was made up by the researcher, expecting of course its rejection during the
acceptability rating phase.
A number of control sentences were added to the descriptions of a random

subset of the clips. These control sentences fell into three categories: (i) blatantly
ungrammatical; (ii) morphosyntactically wellformed but glaringly false of the scene
at issue; (iii) presenting information about the scene that was accurate, but irrelevant
to the task of communicating what is happening in the scene. The motivation behind
the inclusion of these control items was, first, to encourage the participants to make
use of the entire rating scale, and secondly, to have a baseline for the interpretation
of each participant’s ratings.
Training
Participants unfamiliar with the idea of rating scales were tutored on the concept by
discussing examples that, it was hoped, would serve to bridge it, such as grading
in school. All participants were then trained on the use of the 8-point rating scale
with the help of two training videos, one in which a woman is shown placing a
pencil on a table and one in which she is shown placing it in a cup on the table.
Using nontechnical language, the participants were instructed to distinguish among
ungrammatical descriptions (lowest ratings), incorrect descriptions (second-lowest
rating interval), correct but misleading or unhelpful descriptions (second-highest
rating interval), and descriptions that would be specifically useful for the purpose
of explaining the contents of the videos to somebody who has not seen them, but
for some reason needs to know what is ‘happening’ in the scenes. An example of
a correct but misleading description of the training scene with the woman putting
the pencil in the cup is ‘The woman put the pencil on the table’: this is not entirely
false, since the cup is on the table, but it is misleading. The procedure was continued
until the participants produced the expected ratings on more than two consecutive
descriptions. The training was conducted in the target languages.
Test Phase
Participants were assigned to four lists. Each list was shown the CAL Clips in a
different, pseudo-randomized order. The clips were shown in a PowerPoint presen-
tation. The order of presentation of the descriptions of each clip was randomized
with the help of an Excel spreadsheet. The same spreadsheet was used to record the
participants’ ratings. Participants watched each video at least once (and additional
times if they asked to). The researcher then read each description out aloud and
asked the participant to rate it before moving on to the next description. Participants
were encouraged to take as much time as they liked and were urged to rate each
description by itself rather than in comparison to the other descriptions of the
same video. They were reminded at regular intervals that they could assign any
rating as often as they saw fit to descriptions of the same scene. They were given
the opportunity to produce additional descriptions, including improved versions of
existing ones. The researchers would repeatedly encourage the participants to make
use of the entire scale and remind them of the distinction among ungrammatical,
incorrect, infelicitous, and felicitous descriptions. It would take participants between
under 30 and close to 90 min to complete the task. All participants completed the
task in a single sitting. The task was entirely conducted in the target languages.
Coding
The stimulus descriptions’ response types were coded by the participating

researchers for their morphosyntactic complexity level. The most morphosyn-
tactically compact descriptions involve only a single predicate, which encodes both
causing and resulting events. To categorize the morphosyntactic complexity of
descriptions which encode the causing and resulting events in separate predicates,
the Layered Structure of the Clause (LSC) model of Role and Reference
Grammar (van Valin 2005) was used. In this model, morphosyntactic complexity
is assessed in terms of two independent dimensions: the complexity level of the
constituents that combine to constitute a given expression and the morphosyntactic
relation between the constituents. These dimensions are called juncture and nexus,
respectively. The model assumes four juncture levels or ‘layers’: nucleus, core,
clause, and sentence (where the nucleus is an argument-taking head and constitutes
the core together with its syntactic arguments). The nucleus of an event description
is the lexical event descriptor; the core dominates the nucleus and its syntactic
arguments, and the clause dominates one or more core(s) plus additional material, in
particular operators related to finiteness and information perspective. Combinations
of these structural units, called ‘junctures’, occur at each of these structural levels.
Nuclear junctures are exemplified (non-exhaustively) by complex predicates, core
junctures by non-finite complementation constructions, and clause-layer junctures
by adverbial clause constructions. Junctures can be symmetrical or asymmetrical.
Asymmetrical junctures involve embedding of one unit (typically a core or clause)
in another. This embedding relation is called ‘subordinate nexus’ in this model.
The LSC model includes three nexus relations: coordination (defined in terms of
symmetry and independence in operators and modifiers), subordination (defined
in terms of asymmetry), and cosubordination (defined in terms of symmetry and
sharing of operators and modifiers). Coordination is assumed to be the loosest
and cosubordination the tightest form of integration of the constituents. Due to
the sharing of operators and modifiers, the constituents enjoy less autonomy in
cosubordination than in subordination, where such sharing is absent. Crossing the
three juncture types with the three nexus types results in nine logically possible
juncture-nexus types, although two of these, nuclear subordination and nuclear
coordination, are only marginally attested typologically. Juncture and nexus are
treated as projecting into a single hierarchy, with simplex nuclei representing
the tightest possible integration of subevent representations, followed by nuclear
cosubordination, and sentential coordination representing the loosest form of
integration. This single complexity hierarchy is one of the two properties that
motivated the adoption of the LSC model for present purposes, the other being its
broad (arguably universal) applicability regardless of language type.
3.3.2.2 Results
The participants’ ratings have been analyzed in terms of the factors that predict the
morphosyntactic compactness or juncture-nexus type (JNT) of the descriptions
that scored the highest rating for a given clip (the ‘ceiling rating’). Three predictive
variables have been considered: language, mediation (mediated vs. unmediated),
and domain (specifically, whether or not the causer makes physical contact with the
next participant in the chain). The heatmaps in Fig. 3.3 summarize the result for
each of the four languages.
As expected, and in line with the Iconicity Principle, more compact descriptions
(‘Simplex nucleus’ and ‘Nuclear cosubord’, representing base-transitive causative
verbs and complex predicates) were rated as acceptable for unmediated causal
chains than for mediated causal chains. Within each mediation level, physical
causation chains also were considered more compatible with compact descriptions
Fig. 3.3 Percentage of each juncture-nexus type for the most compact ceiling-rated description
for each clip + participant by language, domain and mediation
than non-physical ones. However, surprisingly, an ordinal mixed-effects logistic

regression model with most compact ceiling-rated JNT as dependent variable,
domain, language, and mediation as fixed factors; and clip, order, and participant
as random factors produced evidence of solely domain and language main effects,
whereas mediation mattered only in interactions with those factors (cf. Bellingham
et al. 2017 for details). However, see comments in Sect. 3.3.2.3 regarding limitations
of this type of analysis that result from imbalances in the current stimulus set.
3.3.2.3 Discussion
To understand the interplay between domain and mediation in our data, it is

important to know that the two correlate strongly in the design of the CAL
Clips: most scenes that feature three-participant (i.e., mediated) chains involve
psychological or speech act causation, and conversely, most scenes that involve
psychological or speech act causation also display mediation by a intermediator.
We believe that this correlation is not merely an artifact, but actually reflects biases
in the kinds of causal chains humans think and talk about most commonly. This
assumption remains to be tested against corpus data.
The observation that domain may be a stronger predictor of the morphosyntactic
complexity of causative descriptions than mediation does provide a potential clue
for the explanation of the failure of Escamilla (2012) to find a significant correlation
between directness and morphosyntactic complexity: both mediation and domain
appear to be tied up in the understanding of the directness variable in Dixon (2000),
and it is unclear how Escamilla’s coding policies dealt with these two factors.
At the same time, this very preliminary analysis of data from just four languages
did turn up evidence supporting the Iconicity Principle, provided one assumes
psychological and speech act causation is conceptually more complex (or less
direct) than physical causation. Data from additional populations is currently being
integrated into the analysis.
3.3.3 Case Study 3: Reasoning About Causality
The research described in this section was motivated by the need to see to
what extent cultural specificity in causal cognition is represented in or possibly
influenced by language. While we are not yet able to relate cognitive variation to
linguistic variation, the experiments discussed here serve as a launching point to this
investigation, and additional research into this question is currently underway. Much
of the work in linguistics that focuses on the mapping of form to meaning implicitly
treats causality and agency as universal notions – even in crosslinguistic research
(e.g., Comrie 1981; Dixon 2000; and Shibatani & Pardeshi 2002). Meanwhile, a
growing body of work in the field of social psychology calls the universality of
these notions very much into question (cf. references below). If these concepts are
subject to cultural variation, it is important to understand whether this variation

also affects concepts such as agentivity that typology and theories of the syntax-
semantics interface rely on. As a test case, we chose to examine whether the
contrast between intentional and unintentional actions has a different impact on
responsibility attribution in different populations.
A series of studies in social psychology have suggested cultural variation in atten-
tion to dispositional properties, with Chinese participants exhibiting less attention
to actor disposition – including intentions – compared to Americans (e.g., Morris
& Peng 1994; Chiu et al. 2000; Choi & Nisbett 1998; Choi et al. 1999; Maddux
& Yuki 2006; Menon et al. 1999; and Peng & Knowles 2003, inter alia). Although
we based our experiment design on this literature, we also recruited participants
from populations whose position on the sociocentrism-egocentrism spectrum is less
clear, since we are ultimately not primarily interested in the hypothetical nexus
between patterns of social organization and attention to dispositions, but more
broadly in any kind of culture-specificity in causal attributions. We plan to follow
up with all participants with a survey presented in Singelis (1994) that targets
the participants’ ‘self-construal’, specifically, the extent to which it involves social
interdependence vs. independence from others. The rest of this section discusses
the methodology employed in the responsibility assignment task and presents some
initial data showing the trends we found in causal attribution.
3.3.3.1 Method
Participants watched videos of two actors involved in a chain of events that

culminates in a resulting event. In each case, the chain is initiated by one actor,
dubbed the ‘causer’ (CR). The second actor is affected by the CR’s action and may
or may not in turn affect a third, inanimate, entity. This second actor is labeled CE.
After watching each video, participants divided 10 tokens into piles indicating their
assignment of responsibility for the resulting event. Piles represented CR, CE and
‘Neither’.
Materials
The experiment comprised a training phase involving 10 video clips and a test
phase with 24 video clips. The test items are described in Table 3.1 in terms of the
action/event involving the second actor (CE). These actions/events can all in one
way or another be understood as caused by the CR – in some cases via a physical
impact on CE; in others via a reflexive/uncontrolled or deliberate psychological
response to the CR’s behavior or as a response to a gestural command by the CR.
Three intentionality variables are represented as well: whether the CR intended their
action (I ⇒ A), whether the CR intended the outcomes of the chain (I ⇒ O), and
Table 3.1 Test phase video description

CE action CR I ⇒ A CR I ⇒ O CE intentional
CE breaks a plate Yes Yes Yes
CE breaks eggs Yes Yes Yes
CE collapses a cup tower Yes No No
CE collapses a cup tower Yes Yes No
CE collapses a cup tower Yes Yes No
CE falls Yes Yes No
CE falls No No No
CE falls Yes Yes No
CE is scared/falls over Yes Yes No
CE is startled No No No
CE is thrown a distance Yes Yes No
CE laughs Yes Yes No
CE leaves Yes No Yes
CE leaves Yes Yes Yes
CE sits down Yes Yes Yes
CE swings a swing Yes Yes Yes
CE tears a piece of paper Yes Yes Yes
CE tears a piece of paper Yes Yes No
CE tears a piece of paper No No No
CE tears a piece of paper Yes Yes No
CE tosses a ball into a box Yes Yes Yes
CE wakes Yes No No
CE yawns No No No
whether CE acted intentionally/volitionally.15 We adopted these variables from the

‘Culpable Control Model’ presented in Alicke (2000) on account of the model’s
positive reception in the social psychology literature.
Four of the training items featured scenes that fit the same parameters as the
test items. The remaining six items featured actions on which the two actors
collaborated, events that seemingly occurred without the involvement of either actor,
and events in which one actor destroyed an object while the other looked on.
15 Items that are represented in terms of the same description and configuration of variables in
Table 3.1 differed from one another in terms of (1) the use of an instrument by the CE, (2) for
unintentional CEs, the medium of interaction between the CR and the CE (physical (e.g., pushing)
vs non-physical (e.g., yelling loudly to startle) manipulation). The impact of these further variables
has not yet been analyzed.
Participants
For the initial study, 12 speakers of Yucatec Maya, 16 Mandarin speakers, and
20 Spanish speakers were recruited from and tested at sites in Barcelona and
Murcia, Spain, at Beihang University in Beijing, China, and in the village of Yaxley,
Quintana Roo, Mexico. In the follow-up study, we recruited 25 Basque speakers, 20
Japanese speakers in Tokyo, 12 Kupsapiny speakers from Kapchora in the Sebei
sub-region of Eastern Uganda, and 22 Sidaama speakers from Hawassa and Wondo
Genet in the Sidaama Zone of Ethiopia.
Training
The purpose of the training phase was to allow the participants to gradually
familiarize themselves with the ratings procedures and the concept of rating scales.
For this reason, we began with scenes in which the assignment of responsibility
seemed straightforward (either evidently neither actor was responsible, or only one
of them, or both to equal extent) and included four items similar in structure to
the test items at the end, where responsibility assignment seems less predictable as
responsibility may be shared asymmetrically between the characters. The training
phase commenced with the six clips that featured collaborative action, no involve-
ment of either actor, or one actor involved while the other was not. The experimenter
would play the first three of these, each time following up by apportioning the tokens
in the appropriate way and explaining why they did so. After this, the experimenter
would invite the participant to use the tokens to rate responsibility in the remaining
seven scenes. The experimenter would play a clip, establish which circle on the
paper represented each actor in the video, then replay the video and eventually
ask the participant to distribute the tokens. The experimenter would correct any
confusion about allocating the tokens and verify that the participant understood the
task.
Procedure
Participants were given 10 identical tokens (small glass stones or other objects
of similar size). To prevent confusion about the purpose of the task, no tokens
resembling currency were used. These tokens represented total responsibility for
end results in video clips observed during the task, such that each token represented
10% of total responsibility. Participants were also given a sheet of paper with three
circles drawn on it. The leftmost circle represented the character who ended in the
left-most position or final frame of the video clip, the center circle represented the
other character, and the right-most circle represented a portion of the responsibility
that could not be attributed to either character. Circles were arranged in a horizontal
row, or in two rows where the two circles representing actors were next to one
another in the top row and the ‘neither’ circle was drawn below them. The test
items were presented in one of four pseudo-randomized orders. Participants were

randomly and evenly distributed over these four orders.
During the test phase, participants watched the 24 test clips. After each clip, the
experimenter indicated which circle would represent each actor in the video and
then played the video a second time. The participant was then asked to distribute
responsibility for the final outcome of the clip between the actors. Responses were
recorded in a spreadsheet. After watching the 24 clips, the participant viewed each
clip again and provided a verbal description of the action in the video.
3.3.3.2 Results
Predictions
Suppose that members of sociocentric societies are relatively less likely to pay
attention to internal dispositions of the causer and more to situational factors in
their causal attributions, and suppose further that the mainstream cultural ethos of
China is relatively more sociecentric than that of many Western societies, with the
latter emphasizing individualism more strongly, as suggested by Morris & Peng
(1994). If this is the case, the intentionality of both actors should play a less
predictive role in the ratings of the Chinese participants than in those of the Spanish
participants. On the other hand, the findings in Le Guen et al. (2015) suggest that
causer intentionality may play an even greater role in the Yucatecans’ responsibility
assignments than in those of either of the other two groups.16 No predictions
were made for other populations due to lack of reported data on sociocentric and
egocentric values.
Analysis
An analysis of a subset of the data (Mandarin-, Spanish- and Yucatec-speaking

populations) suggests that I ⇒ A (causer intention to initiate an event) was
a significant factor in responsibility assignment while I ⇒ O (causer intention
for a particular outcome to occur) was not (see Evers et al. 2017 for details).
Figure 3.4 shows the mean CR responsibility ratings by population, suggesting
16 Le Guen et al. (2015) stand on a tradition of research into the role of so-called magical thinking
in causal attribution in traditional societies dating back to Evans-Pritchard (1937), and have
interpreted this tradition to entail that members of such cultures are more ready to accept intention
alone as the cause of an event even in the absence of observable actions. In a series of experiments,
they tested Yucatec attribution of causality where an actor intended an outcome they had no way of
affecting and found that intention to act impacted attribution of responsibility. One could interpret
the findings to say that Yucatecans weight intentionality to a greater degree than other cultures in
responsibility attribution.
Fig. 3.4 Average responsibility ratings for all (intentional and unintentional) causers by popula-
tion. Error bars represent 95% confidence interval
Fig. 3.5 Average responsibility ratings for causers by intentionality and population. Error bars
represent 95% confidence interval
small but significant differences in Spanish, Basque, and Mandarin responsibility

rankings. Figure 3.5 presents a breakdown by CR intentionality, suggesting all
populations but Basque and Mandarin speakers assigned more responsibility to
intentional than to unintentional CRs, as predicted. Figure 3.6 shows mean CR
responsibility ratings by population, comparing ratings when CEs are intentional
Fig. 3.6 Average responsibility ratings for CRs by CE intentionality and population. Error bars
represent 95% confidence interval
and unintentional. The results of this analysis show significant differences between
CR responsibility ratings depending on CE intentionality, where for all populations
except for Kupsapiny speakers, CRs are awarded significantly higher levels of
responsibility in the presence of an unintentional rather than intentional CE.
3.3.3.3 Discussion
In this study, we investigated the extent to which the contrast between intentional
and unintentional actions impacts responsibility attribution in different populations.
The presence of an unintentional (nonvolitional) second actor (as opposed to a
second actor who acted intentionally) significantly boosted attribution of respon-
sibility to the causer across populations. Overall CR responsibility ratings for
all populations were significantly lower than those of the Chinese participants
except for Sidaama speakers, although the differences for all were quite small.
Japanese, Kupsapiny, Sidaama, and Yucatec speakers were all fairly uniform in
overall responsibility attribution, while Spanish and Basque populations were
significantly lower than other groups. Ratings for unintentional and intentional
CRs were significantly different for Spanish, Yucatec, and Japanese populations
only, suggesting that sensitivity to intention when assigning responsibility may vary
by culture. Because differences in social organization between populations such
as speakers of Spanish and Basque are unclear, we are interested in evaluating
other possible social factors in the variation of responsibility attribution, including
language. Given that the representation of causality also has a significant impact on
the grammar and lexicon of natural languages, it is possible that differences in causal
cognition affect responsibility ratings awarded to causers, and that language may
actually be involved in shaping the transmission system of culture-specific cognitive
practices.
This study investigates the participant autonomy variable in the etic grid and
how it impacts responsibility assessment in a mediated causal chain. For this
study, we did not evaluate differences in mediation (CEs acting with or without
an instrument). We also did not distinguish between full and partial CE control
for psychological causation, but instead treated CE behavior as a binary between
intentionally participating in the causal chain (volitionally or under psychological
coercion), and unintentional participation in the causal chain through physical
impact.
3.4 Discussion
The three studies presented here apply the same etic grid of variables and variable
levels in three distinct research designs that target data gathering on speech produc-
tion (Sects. 3.3.1 and 3.3.2), speech comprehension (via acceptability judgments;
Sect. 3.3.2), and nonverbal cognition (Sect. 3.3.3). All three studies are ongoing:
data from additional populations is being collected, coded, and incorporated into
the analyses. Yet, all three studies have already produced interpretable results that
suggest tentative answers to the research questions they were designed to answer.
The study on causality in narratives found that the same underspecification strategies
are used across the languages included in the analysis so far, but that there are
differences in the extent to which the populations rely on the individual strategies.
The study on the semantic typology of causative coding devices has uncovered
preliminary evidence that domain, in the sense of the distinction between physical
and nonphysical causation, may be a more powerful predictor of morphosyntac-
tic complexity than mediation, in the sense of the number of participants and
subevents involved in the chain. The investigation of responsibility assignment
by members of different cultural communities has uncovered findings that so far
align with predictions arising from the social psychology paradigm that posits a
nexus between broad-scale patterns of social organization and the importance of
internal dispositions in judgments of responsibility. However, the investigation has
also found significant behavioral differences between populations that appear to
be broadly similar in social organization (Mayan vs. Sebei, Kupsapiny-speaking),
suggesting that factors beyond social organization may be at play or perhaps that
the sociocentrism-egocentrism variable is not sufficient to capture the relevant
differences in social organization. In addition, it remains to be seen to what
extent culture-specific patterns of responsibility assignment correlate with language-

specific patterns in the verbalization of causal relations.17,18
This is of course not to say that the grid and the CAL Clips are optimal tools for
this type of research, or even for the studies we have been carrying out. It is in fact
difficult to assess how close these tools come to being optimal. But, at least, we can
point out some shortcomings that have emerged.
One important deficiency of the CAL Clips is that they do not instantiate all
cells of the etic grid with the same frequency. Consequently, a data set collected
with the clips will comprise many more observations in some cells than in others.
When the number of observations is below a certain threshold, statistical analyses
such as the mixed-effects regression model mentioned in Sect. 3.3.2 may yield
spurious, unreliable results. We are currently planning to overcome this problem
by creating additional stimulus videos. We are also considering a redesign of
the studies that would allow us to target smaller sets of variables in separate
experimental conditions. This may make it possible to focus the analysis such that
each combination of variables is instantiated in enough clips.
There were also problems with particular videos. Several cases of ambiguity in
causal relations emerged. In the clip UM1_asleep, a woman is shown apparently
asleep in a chair, and a man walks across the room and apparently accidentally
trips over her foot, waking her up. We had intended the man to be the causer and
the resulting event to be the woman’s waking up, but across study populations, it
was perceived by some participants in this intended manner and by others with
the woman as the causer and the man’s tripping as the resulting event. In the scene
HM1_fall, a woman is shown sweeping, when another walks up in front of her and
stops there, apparently looking for something while unaware that she is impeding
the first woman’s action. The first woman then pushes the second, and she falls
to the floor. The clip was supposed to represent physical causation of motion with
a human causer and affectee. However, some participants viewed the woman who
winds up being pushed as the initiator of the causal chain and the caused motion
event as being itself the result of a caused psychological change (aggravation) in the
pusher.
Another kind of ambiguity problem influenced the identification of the characters
acting in the videos in some cases. There were several scenes where participants
17 That it was possible to reach these findings on the basis of the set of variables and levels we
started out with and the video clips we created to represent the possible combinations of these
variables and levels can be considered a proof of concept for the etic grid and stimulus set. An
additional study further strengthening the case for these tools is Hafeez (2018), which applied
them to the investigation of intricate agentivity-sensitive patterns of case alternations and light
verb selection in Urdu, following broadly the methodology of our semantic typology study (while
deviating from it in some details). Hafeez’s work in particular contributed to our understanding of
the interaction of these variables in the design of the CAL etic grid.
18 We think that intentionality and control are crucial for the verbal representation of causality in all
languages. Illustration of the importance of volitionality, intentionality, and control in the grammar
of causality comes from Indo-Aryan languages, some of which have been shown to have case
alternations and complex predicate constructions that are sensitive to these variables.
were misled in the attribution of gender due to clothing items and possibly
unfamiliarity with gender-specific facial traits in members of other ethnic groups
(exacerbated of course by the limitations of the videos in size and quality). Our
advice for future studies of this kind would be to make sure that actors appearing in
the same scene dress in distinct and easily identifiable colors.
A potential problem of particular interest for our purposes is culture-specific folk
theories of what kinds of events can cause what other kinds of events. It is important
to note that we did not observe this problem occurring with any level of generality,
with one exception: in the video UU1_yawn, a woman yawns, and a man yawns
in response. The idea of infectious yawning proved to be unfamiliar to many of our
non-Western participants.
Overall, the studies presented here suggest that crosslinguistic and cross-cultural
investigations of representations of causality that rely on an etic grid of potential
predictor variables and a set of nonverbal stimuli encoding the combinations of the
levels of these variables are feasible, and that their realization is not too daunting
within the context of a collaborative project with the relatively modest support
the CAL project has received. We believe, then, that this collection of studies can
serve as a model, not only for the exploration of other subdomains within causality
(e.g., biological and social causation), but also for the exploration of other domains
beyond causality.
3.5 Conclusions
We presented a set of variables and levels for the cross-population exploration of

verbal and cognitive representations of causality. We encoded the possible vari-
able/level combinations in a set of 58 video clips and applied these in three studies
to the collection of verbal production and comprehension data and of cognitive
categorization data. These studies’ preliminary findings can be summarized as
follows:
• In connected speech, speakers across languages appear to rely on the same basic
strategies for underspecifying information about subevent properties, subevent
identity, and causal relations.
• However, there was variation in the extent to which speakers of different
languages rely on each type of strategy. We hypothesize that such differences
may be driven both by the grammar and lexicon of the languages and by cultural
and demographic factors such as literacy.
• The preferred level of morphosyntactic complexity of a causative description
does indeed appear to reflect iconically the conceptual complexity of causal chain
that is represented.
• However, the distinction between physical and non-physical causation seems to
be a stronger predictor of morphosyntactic complexity than mediation, in the
sense of the number of potentially controlling participants involved in the chain.
The models discussed here do include some collinearity between mediation and
domain of causation, meaning that future research will be necessary in order to
assess the full significance of causal domain.
• There appear to be significant differences across populations in the extent to
which perceived causer intentionality drives responsibility assignments.
• These differences seem at least to partially align with suggested differences in
how members of different cultural communities conceptualize social organiza-
tion.
• However, it is not clear that all observed cross-population differences in respon-
sibility assignment can be attributed to differences in social cognition.
All three studies are ongoing at the time of writing and all results should be
considered as preliminary. It is our hope to have contributed an instrument that we
are both happy to share with other researchers in cognitive anthropology, linguistics,
and social psychology and that may inspire other cross-population studies in the
domain of causality and beyond.
Appendix 1: Causal Chain Properties of Core Stimuli
Each video clip in the core set of stimuli is listed below, along with a short
description of the causal chain depicted in the clip and the values intended for each
causal chain variable. See Sect. 3.2 for a description of the causal chain variables.
Causal chain participants:
CR = causer, CE = intermediator, AF = affectee, INS = instrument
HO6_paper A woman tears a piece of paper in half.

Mediation: No CE or INS. Participant type + degree of autonomy:
CR : HUMAN + INTENTIONAL ; AF : INANIMATE
Resulting event type: CHANGE OF STATE. Force dynamics: CAUSATION
HC1_leave A woman tells a man to leave the room, and he leaves.
CR : HUMAN + INTENTIONAL ; AF : HUMAN + INTENTIONAL
Resulting event type: CHANGE OF LOCATION. Force dynamics: CAUSATION
HOIproc1_swing A man pushes a swing with a tennis racquet and it moves
back and forth.
Mediation: INS but no CE. Participant type + degree of autonomy:
Resulting event type: PROCESS. Force dynamics: CAUSATION
HUO3_paper A woman sneaks up behind another woman and yells loudly,
which startles the other woman and makes her tear the piece of paper she is
holding.
Mediation: CE but no INS. Participant type + degree of autonomy:
CR : HUMAN + INTENTIONAL ; CE : HUMAN + REFLEXIVE ( NOISE ); AFFECTEE :
INANIMATE
HO2_egg A woman cracks an egg into a bowl.
NM2_reporter A reporter is blown away in strong wind.
CR : NATURAL FORCE ; AF : HUMAN + PHYSICAL IMPACT
HOI4_ball A man hits a ball off a wooden bench with a tennis racquet.
HO5_cuptower A man knocks over a cup tower
UO1_egg A woman trips while carrying eggs, and accidentally smashes them
into a bowl.
CR : HUMAN + UNINTENTIONAL ; AF : INANIMATE
UM3_faint A man faints onto another man and knocks him over.
CR : HUMAN + UNINTENTIONAL ; AF : HUMAN + PHYSICAL IMPACT
HMO4_cups A woman pushes another man into a stack of cups, and he knocks
it over.
CR : HUMAN + INTENTIONAL ; CE : HUMAN + PHYSICAL IMPACT; AF : INANIMATE
HU2_scare A girl jumps out of a box and shrieks, startling a boy, and he falls
over.
CR : HUMAN + INTENTIONAL ; AF : HUMAN + REFLEXIVE ( NOISE )
UO2_paper A woman is flipping through a book and accidentally tears a page.
HCO3_egg_new A man tells a woman to crack an egg into a bowl, so she does.
CR : HUMAN + INTENTIONAL ; CE : INTENTIONAL ; AF : INANIMATE
NC1_tsunami A man sees a giant wave heading towards him on a beach, so he

runs away.
CR : NATURAL FORCE ; AF : HUMAN + INTENTIONAL
HOI3_plate A woman shatters a plate with a broom handle.
UC1_sing A woman is singing poorly, so another woman covers her ears and
leaves the room.
CR : HUMAN + UNINTENTIONAL ; AF : HUMAN + INTENTIONAL
HCOI2_paper A woman tells another woman to cut up a piece of paper with
scissors, so she does.
Mediation: CE and INS. Participant type + degree of autonomy:
HO4_ball A man throws a ball into a box.
HM1_fall A woman pushes another woman to the floor.
CR : HUMAN + INTENTIONAL ; AF : HUMAN + PHYSICAL IMPACT
UMO2_cups A woman enters a room backwards, dragging a table. She bumps
into a man standing in front of a stack of cups, and he bumps the cups and they
fall to the floor.
CR : HUMAN + UNINTENTIONAL ; CE : PHYSICAL IMPACT; AF : INANIMATE
NM4_umbrella An umbrella blows away in the wind.
CR : NATURAL FORCE ; AF : HUMAN + PHYSICAL IMPACT
HOI1_paper A woman cuts a piece of paper into pieces with scissors.
HUO2_cups A woman sneaks up behind a man and yells loudly, which startles
the other man and makes him bump the stack of cups he is standing next to, then
the cups all fall to the floor.
CR : HUMAN + INTENTIONAL ; CE : REFLEXIVE ( NOISE ); AF : INANIMATE

UM1_asleep A woman is sleeping in a chair, and a man walks across the room
and accidentally trips over her foot, waking her up.
NU1_thunder A loud thunder clap startles a woman.
CR : NATURAL FORCE ; AF : HUMAN + REFLEXIVE ( NOISE )
HMO3_paper A woman pushes a woman who is holding a piece of paper, and
the paper tears.
CR : HUMAN + INTENTIONAL ; CE : PHYSICAL IMPACT; AF : INANIMATE
HOproc1_swing A man pushes a swing and it moves back and forth.
HCO2_paper A woman tells a woman to tear a piece of paper into pieces, and
so she does.
UM2_overboard A reporter standing on a boat steps backwards and bumps into
another man who is kneeling at the edge of the boat, knocking him (the kneeling
man) into the water.
UOproc1_swing A man accidentally bumps into a swing, causing it to move
back and forth.
HC2_sit A man tells a woman to sit, and so she does.
CR : HUMAN + INTENTIONAL ; AF : HUMAN + INTENTIONAL
HCOproc1_swing A woman tells a man to push a swing, and so he does.
UOI1_cuptower A man is sweeping next to his stack of cups, he turns and

accidentally knocks the cups over with the broom handle.
UU2_sneeze A woman sneezes behind another woman, startling her/making
her jump.
CR : HUMAN + UNINTENTIONAL ; AF : HUMAN + REFLEXIVE ( NOISE )
HU1_laugh_new A man pulls a funny face and makes a woman laugh.
CR : HUMAN + INTENTIONAL ; AF : HUMAN + REFLEXIVE ( URGE )
NCO1_umbrella It is raining, and so a man opens an umbrella.
CR : NATURAL FORCE ; CE : INTENTIONAL ; AF : INANIMATE
HCOI3_plate A man tells a woman to shatter a plate with a broom handle, and
so she does.
Mediation: CE and INS. Participant type + degree of autonomy:
UO3_ball A woman accidentally kicks a ball over her head and out of the room.
UUO2_paper A woman sneezes behind a man who is reading the newspaper.
He is startled, and tears the newspaper.
CR : HUMAN + UNINTENTIONAL ; CE : REFLEXIVE ( NOISE ); AF : INANIMATE
HCO4_ball A woman tells a man to throw a ball into a box, and so he does.
HM2_strongman A man picks up another man and throws him across the room.
CR : HUMAN + INTENTIONAL ; AF : HUMAN + PHYSICAL IMPACT
UU1_yawn A woman yawns, another man sees her yawning and so he yawns.
CR : HUMAN + UNINTENTIONAL ; AF : HUMAN + REFLEXIVE ( URGE )
Appendix 2: Causal Chain Properties of Supplementary

Stimuli
HClet_door A man blocking a woman from exiting a room sees her and moves
to let her pass.
CR : HUMAN + INTENTIONAL ; AF : INTENTIONAL
Resulting event type: CHANGE OF LOCATION. Force dynamics: LETTING
HO1_cup A woman throws a cup at the floor and it smashes.
Resulting event type: PROJECTILE BREAKING. Force dynamics: CAUSATION
UUO1_egg A man accidentally slams the door, which startles another man in the
room who is holding an egg, which makes him drop the egg and it smashes.
CR : HUMAN + UNINTENTIONAL ; CE : HUMAN + REFLEXIVE ( NOISE ); AF : INANI -
MATE
Resulting event type: PROJECTILE BREAKING. Force dynamics: LETTING
HO_let_ball A woman releases the ball she is holding, allowing it to fall.
HCO1_cup A man tells another man to throw a cup at the floor, so he does, and
the cup smashes.
CR : HUMAN + INTENTIONAL ; CE : HUMAN + INTENTIONAL ; AF : INANIMATE
HUO1_plate A woman sneaks up behind a man and yells loudly, which startles
the man and makes him drop the plate he is holding. It smashes on the floor.
CR : HUMAN + UNINTENTIONAL ; CE : HUMAN + REFLEXIVE ( NOISE ); AF : INANI -
MATE
UC_let1_doorway A woman tries to exit the room, but a man is blocking the
doorway (facing away from her). He doesn’t see her, but moves away from the
door and she passes through.
CR : HUMAN + UNINTENTIONAL ; AF : INTENTIONAL
HMO_let1_ball A woman pulls the arm of another woman who is holding a
ball, making her drop the ball.
UMO1_cup A woman enters a room holding a large bin which is blocking her
vision. She bumps into a man who is holding a cup, he drops the cup and it
smashes on the floor.
CR : HUMAN + UNINTENTIONAL ; CE : HUMAN + PHYSICAL IMPACT; AF : INANI -
MATE
UO4_cup A man is sitting at a desk, he moves his arm as he turns a page and
bumps a cup off the desk, and it smashes on the floor.
HMO1_plate A woman pushes another woman who drops the plate she was
holding. It smashes on the floor.
UCO1_ball A man faints near a woman who is holding a ball, she lets the ball
go to catch him and the ball falls to the floor.
CR : HUMAN + UNINTENTIONAL ; CE : HUMAN + INTENTIONAL ; AF : INANIMATE
NUO1_thunderclap A man is standing holding a plate, there is a loud
thunderclap which startles him and he drops the plate, which smashes on the
floor.
CR : NATURAL FORCE ; CE : HUMAN + REFLEXIVE ( NOISE ); AF : INANIMATE
UUO3_cup A man gestures for a woman sitting at a desk to hand him a jacket
hanging behind her. She reaches for the jacket, and knocks a cup off the table.
The cup smashes on the floor.
CR : HUMAN + UNINTENTIONAL ; CE : HUMAN + UNINTENTIONAL ; AF : INANI -
MATE
MClet_doorway A man blocking a woman from exiting a room does not move,
so she pushes him aside and exits.
CR : HUMAN + PHYSICAL IMPACT; AF : HUMAN + INTENTIONAL
Acknowledgements The materials presented here are based upon work supported by the National
Science Foundation under Grant No. BCS153846 and BCS-1644657, ‘Causality Across Lan-
guages’; PI J. Bohnemeyer. In addition, Kawachi’s research was supported by the Japan Society
for the Promotion of Science under grant KAKENHI Project ID 19K00565. We are grateful
to three anonymous reviewers and to the editors of the volume, Nora Boneh and Elitzur Bar-
Asher Siegal, for their constructive criticism. We would like to thank the members of the
University at Buffalo Semantics Typology Lab for assistance with the creation of the stimuli
(Katherine Donelson, Alexandra Lawson, Randi Moore and Karl Sarvestani) and piloting of the
Responsibility Assignment study design (José Antonio Jódar Sánchez) and the members of the
Beihang Research Group for Event Representation and Cognition for their assistance in testing
the Chinese participants (specifically, Enirile, Hongxia Jia, Fuyin Li, Jinmei Li, Sai Ma, Chenxi
Niu, and Mengmin Xu). We also gratefully acknowledge helpful advice from Dare Baldwin, the
late Sieghard Beller, Andrea Bender, and Bertram Malle, none of whom necessarily agree with the
views expressed in our chapter. The responsibility for any mistatements or omissions is naturally
ours alone.
References
Alicke, M. D. (2000). Culpable control and the psychology of blame. Psychological Bulletin,
126(4), 556–574.
Anscombe, G. E. M. (1971). Causality and determination. Reprinted in E. Sosa & M. Tooley (Eds.),
Causation (pp. 88–104). Oxford: Oxford University Press, 1993.
Bellingham, E., Evers, S., Kawachi, K., Mitchell, A., & Bohnemeyer, J. (2017). An experimental
approach to the semantic typology of causative constructions. Poster, 12th Association for
Linguistics Typology Conference (ALT 2017).
Bohnemeyer, J. (2007). Morpholexical transparency and the argument structure of verbs of cutting
and breaking. Cognitive Linguistics, 18(2), 153–177.
Bohnemeyer, J., Enfield, N. J., Essegbey, J., & Kita, S. (2010). The macro-event property: The
segmentation of causal chains. In J. Bohnemeyer & E. Pederson (Eds.), Event representation
in language: Encoding events at the language-cognition interface (pp. 43–67). Cambridge:
Cambridge University Press.
Brown, P., & Levinson, S. C. (1987). Politeness: Some universals in language use. Cambridge:
Cambridge University Press.
Chiu, C.-Y., Hong, Y.-Y., Morris, M. W., & Menon, T. (2000). Motivated cultural cognition: The
impact of implicit cultural theories on dispositional attribution varies as a function of need for
closure. Journal of Personality and Social Psychology, 78(2), 247–259.
Choi, I., & Nisbett, R. E. (1998). Situational salience and cultural differences in the correspondence
bias and in the actor-observer bias. Personality and Social Psychology Bulletin, 24, 949–960.
Choi, I., Nisbett, R. E., & Norenzayan, A. (1999). Causal attribution across cultures: Variation and
universality. Psychology Bulletin, 125(1), 47–63.
Comrie, B. (1981). Language universals and linguistic typology: Syntax and morphology. Chicago:
University of Chicago Press.
Croft, W. (1998). Event structure in argument linking. In M. Butt & W. Geuder (Eds.), The projec-
tion of arguments (pp. 1–43). Stanford: Center for the Study of Language and Information.
Davidson, D. (1969). The individuation of events. In N. Rescher (Ed.), Essays in honor of Carl G.
Hempel (pp. 295–309). Dordrecht: D. Reidel.
Dixon, R. M. W. (2000). A typology of causatives: Form, syntax and meaning. In R. M. W.
Dixon & A. Y. Aikhenvald (Eds.), Changing valency: Case studies in transitivity (pp. 30–83).
Escamilla, R. M. Jr. (2012). An updated typology of causative constructions: Form-function
mappings in Hupa (Californian Athabaskan), Chungli Ao (Tibeto-Burman) and Beyond. Ph.D.
Dissertation, University of California, Berkeley.
Evans, N. (2010). Semantic typology. In J. J. Song (Ed.), The Oxford handbook of linguistic
typology (pp. 504–533). Oxford: Oxford University Press.
Evans-Pritchard, E. E. (1937). Witchcraft, oracles and magic among the Azande. Oxford: Oxford
University Press.
Evers, S., Bellingham, E., Donelson, K., Du, J., Jódar Sánchez, J. A., Li, F., & Bohnemeyer, J.
(2017). The role of intentionality in causal attribution is culturally mediated: Evidence from
Chinese, Mayan, and Spanish populations. In Proceedings of the 39th Annual Meeting of the
Cognitive Science Society (pp. 343–348).
Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press.
Hafeez, S. (2018). Causality and agentivity in Urdu: Sensitivity of case clitics and light verbs to
volitionality, intentionality and control in Urdu. Qualifying Paper, University at Buffalo.
Haiman, J. (1983). Iconic and economic motivation. Language, 59(4), 781–819.
Harris, M. (2001). Cultural materialism: The struggle for a science of culture. Walnut Creek:
AltaMira Press.
Heider, F., & Simmel, M. (1944). An experimental study of apparent behavior. American Journal
of Psychology, 57, 243–259.
Keenan, E. (1989). Norm-makers, norm-breakers: Uses of speech by men and women in a
Malagasy community. In R. Bauman & J. Sherzer Explorations in the ethnography of speaking.
(2nd ed., pp. 125–143). Cambridge: Cambridge University Press.
Kehler, A., & Cohen, J. (2018). On convention and coherence. In G. Preyer (Ed.), Beyond semantics
and pragmatics (pp. 261–283). Oxford: Oxford University Press.
Kemmer, S., & Verhagen, A. (1994). The grammar of causatives and the conceptual structure of
events. Cognitive Linguistics, 5(2), 115–156.
Koptjevskaja-Tamm, M. (2015). Semantic typology. In E. Dabrowska & D. Divjak (Eds.),
Handbook of cognitive linguistics (pp. 453–472). Berlin: Mouton de Gruyter.
Le Guen, O., Samland, J., Friedrich, T., Hanus, D., & Brown, P. (2015). Making sense of (excep-
tional) causal relations. A cross-cultural and cross-linguistic study. Frontiers in Psychology, 6,
1–16.
Levin, B., & Rappaport-Hovav, M. (1995). Unaccusativity: At the syntax-semantics interface.
Cambridge, MA: MIT press.
Levinson, S. (2000). Presumptive meanings. Cambridge, MA: MIT Press.
Maddux, W. W., & Yuki, M. (2006). The “ripple effect”: Cultural differences in perceptions of the
consequences of events. Personality and Social Psychology Bulletin, 32, 669–683.
McCawley, J. (1976). Remarks on what can cause what. In M. Shibatani (Ed.), Syntax and
semantics VI: The grammar of causative constructions (pp. 117–129). New York: Academic.
McCawley, J. (1978). Conversational implicature and the lexicon. In P. Cole (Ed.), Syntax and
semantics IX: Pragmatics (pp. 245–258). New York: Academic.
Menon, T., Morris, M. W., Chiu, C.-Y., & Hong, Y.-Y. (1999). Culture and the construal of
agency: Attribution to individual versus group dispositions. Journal of Personality and Social
Psychology, 76(5), 701–717.
Moore, R. E., Donelson, K. T., Eggleston, A., & Bohnemeyer, J. (2015). Semantic typology: New
approaches to crosslinguistic variation in language and cognition. Linguistics Vanguard, 1(1),
189–200.
Morris, M. W., & Peng, K. (1994). Culture and cause: American and Chinese attributions for social
and physical events. Journal of Personality and Social Psychology, 67(6), 949–971.
Mourelatos, A. P. (1978). Events, processes, and states. Linguistics and Philosophy, 2, 415–434.
Parsons, T. (1990). Events in the semantics of English: A study in subatomic semantics. Cambridge,
MA: MIT Press.
Peng, K., & Knowles, E. D. (2003). Culture, education, and the attribution of physical causality.
Personality and Social Psychology Bulletin, 29, 1272–1284.
Pike, K. L. (1967). Language in relation to a unified theory of the structure of human behavior.
The Hague: Mouton.
Rappaport Hovav, M., & Levin, B. (2010). Reflections on manner/result complementarity. In M.
Rappaport Hovav, E. Doron, & I. Sichel (Eds.), Syntax, lexical semantics, and event structure
Shibatani, M. (Ed.). (1976). The grammar of causative constructions (Syntax and Semantics,
Vol. 6). New York: Academic.
Shibatani, M., & Pardeshi, P. (2002). The causative continuum. In M. Shibatani (Ed.), The grammar
of causation and interpersonal manipulation (pp. 85–126). Amsterdam: Benjamins.
Singelis, T. M. (1994). The measurement of independent and interdependent self-construals.
Personality and Social Psychology Bulletin, 20(5), 580–591.
Smith, C. S. (1979). The syntax and interpretation of temporal expressions in English. Linguistics
and Philosophy, 2(1), 43–99.
Song, J. J. (1996). Causatives and causation: A universal-typological perspective. London:
Longman.
Talmy, L. (1988). Force dynamics in language and cognition. Cognitive Science, 12(1), 49–100.
Talmy, L. (2000). Toward a cognitive semantics. Cambridge, MA: MIT Press.
van Valin, R. D. Jr. (2005). Exploring the syntax-semantics interface. Cambridge: Cambridge
University Press.
Vendler, Z. (1957). Verbs and times. The Philosophical Review, LXVI, 143–160.
Verhagen, A., & Kemmer, S. (1997). Interaction and causation: Causative constructions in modern
standard Dutch. Journal of Pragmatics, 27, 61–82.
von Wright, G. H. (1963). Norm and action. London: Routledge/Kegan Paul.
Cognition, 88(1), 1–48.
Wolff, P., Jeon, G.-H., & Li, Y. (2009). Causers in English, Korean, and Chinese and the
individuation of events. Language and Cognition, 1(2), 67–196.
Chapter 4
Asking Questions to Provide a Causal
Explanation – Do People Search for
the Information Required by Cognitive
Psychological Theories?
York Hagmayer and Neele Engelmann
Abstract In this paper, we give a brief overview of current, cognitive-psychological

theories, which provide an account for how people explain facts: causal model
theories (the predominant type of dependence theory) and mechanistic theories.
These theories differ in (i) what they assume people to explain and (ii) how they
assume people to provide an explanation. In consequence, they require different
types of knowledge in order to explain. We work out predictions from the theoretical
accounts for the questions people may ask to fill in gaps in knowledge. Two
empirical studies are presented looking at the questions people ask in order to get
or give an explanation. The first observational study explored the causal questions
people ask on the internet, including questions asking for an explanation. We also
analyzed the facts that people want to have explained and found that people inquire
about tokens and types of events as well as tokens and types of causal relations. The
second experimental study directly investigated which information people ask for
in order to provide an explanation. Several scenarios describing tokens and types of
events were presented to participants. As a second factor, we manipulated whether
the facts were familiar to participants or not. Questions were analyzed and coded
with respect to the information inquired about. We found that both factors affected
the types of questions participants asked. Surprisingly, participants asked only few
questions about actual causation or about information, which would have allowed
them to infer actual causation, when a token event had to be explained. Overall
the findings neither fully supported causal model nor mechanistic theories. Hence,
they are in contrast to many other studies, in which participants were provided
with relevant information upfront and just asked for an explanation or judgment.
We conclude that more empirical and theoretical work is needed to reconcile the
findings from these two lines of research into causal explanations.
Y. Hagmayer () · N. Engelmann

Department of Cognitive and Decision Sciences, Institute of Psychology, University of
Göettingen, Göttingen, Germany
e-mail: york.hagmayer@bio.uni-goettingen.de; neele.engelmann@uni-goettingen.de

https://doi.org/10.1007/978-3-030-34308-8_4
122 Y. Hagmayer and N. Engelmann
Keywords Causation · Causal reasoning · Explanation · Questions · Causal

models · Counterfactual reasoning · Mechanistic reasoning · Cognitive
psychology
4.1 Aims and Overview
In this paper, our first aim is to give a brief overview of current, cognitive-
psychological theories, which provide an account for how people explain facts. In
the literature, two major theoretical frameworks are often distinguished: dependence
theories (e.g., causal model theories) and production theories (i.e., mechanistic
theories). The theoretical frameworks differ in (i) what they assume people to
explain and (ii) how they assume people to provide an explanation. In consequence,
the theoretical frameworks require people to have different types of knowledge in
order to explain. The respective knowledge could be retrieved from memory, it could
be inferred from observed data, or it could be acquired by asking other people. We
are concentrating on the last of these three possibilities and work out predictions for
the questions people may ask to fill in gaps in knowledge.
The second aim of this paper is to present two empirical studies looking at
the questions people ask in order to get or give an explanation. Recent research
indicates that people tend to rely on the knowledge of other people, because their
own causal knowledge is usually only rudimentary (Sloman & Fernbach 2017).
Asking questions is presumably the primary way to tap into the knowledge of
others. The first study explored the causal questions people ask, including questions
asking for explanations. We analyzed the facts that people want to have explained.
It was an observational study analyzing questions people asked on the internet.
The second study directly investigated which information people inquire about in
order to provide an explanation. It was an experimental study. We presented several
scenarios describing facts to be explained and manipulated systematically whether
participants were familiar with the explanandum. Participants were allowed to ask
any question they would find helpful for explaining the respective fact. Questions
were analyzed and coded with respect to the information inquired about.
The aim of the final section is to discuss how the findings from the two studies
may inform cognitive-psychological theories of causal cognition. We will show that
investigating the questions people ask may lead to novel insights about how people
provide explanations.
4.2 Background
Causal reasoning may serve many different functions; both epistemic and pragmatic
(cf. Danks 2014, 2016). The first pragmatic function is to make predictions about
yet unobserved events. For example, knowledge about the causes of lung cancer
4 Asking Causal Questions to Provide a Causal Explanation 123
enables us to predict the risk of a specific person (or a group of people) to get
lung cancer. When we make a prediction based on causal knowledge, we infer
potential outcomes that may result from a set of causes. We may also infer how
likely different outcomes are. Note that these predictions are not based on statistical
correlations, which may also hold between events not causally related to each other.
This distinction is important for the second pragmatic function of causal reasoning:
to make decisions. When deciding on an action, we want to choose the option that
has the highest causal expected utility (Nozick 1993), that is, the option that is most
likely to generate the desired consequences. An option may be strongly correlated
with a desired outcome, but may not be causally related to it. For example, buying
sports shoes is statistically related to physical fitness, but it does not cause fitness.
Therefore buying sports shoes has no positive utility for achieving fitness. Doing
sports, by contrast, does. Causal reasoning allows the decision maker to make this
distinction and determine the utility of available options (see Hagmayer & Fernbach
2017; Sloman & Hagmayer 2006). Causal reasoning can also be used to assign
responsibility, guilt, and blame (Lagnado et al. 2013). In the legal domain it is
important not only to infer the cause of an event but also to establish responsibility.
For example, assume that a person’s heavy smoking caused her lung cancer. Still the
person may argue that the tobacco industry is responsible, as the industry knowingly
sold a highly addictive product that has a rather high likelihood of causing lung
cancer. Lawyers, jurors, and judges can resort to causal reasoning to evaluate this
argument and make a judgment. Finally, a pragmatic function of causal reasoning is
to regulate emotion and motivation (cf. Weiner 1985). For example, by attributing
failure to external factors that are not under their control, people avoid self-blame
and feeling ashamed. By attributing success to internal factors, people feel proud
and competent. Smokers who did not get sick may use these strategies and attribute
the positive outcome to their good genes and their otherwise healthy lifestyle to feel
good and safe despite smoking.
There are two epistemic functions of causal reasoning: to support the acquisition
of new causal knowledge and to derive explanations. To acquire new knowledge
about generic cause-effect relations, we have to have some causal background
knowledge and we have to make an inference from an often very limited set of data
(Tenenbaum et al. 2011). For example, to infer whether smoking causes lung cancer,
researchers had to consider epidemiological data on the correlation of smoking and
lung cancer and the results of experimental studies with animals (Proctor 2012).
In this paper, we focus on explanations. There are many things, which we may
want to explain. These include single instances and generic types of events. For
example, an oncologist may be asked to explain why a specific person Peter got
lung cancer. To do so, the oncologist may point out that smoking heavily for a
prolonged period of time, as Peter did, results in a rather high probability of lung
cancer, compared to non-smokers. A cancer researcher, by contrast, may be asked
why smoking causes lung cancer. To explain this fact, she may refer to the bio-
chemical mechanism by which toxins in the smoke lead to genetic defects that result
in an uncontrolled proliferation of cells. Causal reasoning also helps us to explain
statistical relations, which may not be causal at all. For example, we may want an
explanation why the number of deaths from lung cancer is rising more in Eastern
than in Western Europe. Research found that trends in deaths from lung cancer are
related to geography because the number of people smoking in Eastern Europe is
still rising but declining in the West (Didkowska et al. 2016).
Cognitive psychological theories try to give an account of how people reason
causally. Deriving explanations is only one form of causal reasoning they try to
describe and explain. In the next section we will provide a short overview of
two of the most important theoretical frameworks. The frameworks are based on
theoretical work on causation in philosophy (cf. Beebee et al. 2009; see Keim
Campbell et al. 2007, for philosophical analyses of the relation between causation
and explanation). In philosophy, an important distinction is between dependence
theories and productions theories of causation (cf. Hall 2004). While dependence
theories presume causes to be events probabilistically or counterfactually related
to their effects, production theories presume that there is an intrinsic relation
between cause and effect. This relation has been characterized as a transfer of a
conserved physical quantity (Dowe 2000) or any production process relating specific
entities (Machamer et al. 2000). Cognitive psychological theories build upon the
philosophical theories. They provide normative as well as descriptive accounts of
causal reasoning (see Waldmann 2017, for a recent and comprehensive overview).
Building upon the dependence framework, causal model theories (Gopnik et al.
2004; Griffiths & Tenenbaum 2005; Sloman 2005; Waldmann 1996) are one family
of currently dominant theories in cognitive psychology. Mechanistic theories (Ahn
et al. 1995; Koslowski 1996), which build upon production theories in philosophy,
are a second major account of causal reasoning. In this paper, we focus on these two
groups of theories and show how they account for causal explanation.1
4.3 Theoretical Accounts of Explanation in Cognitive

Psychology
At present, there is no agreed-upon definition of explanation in the cognitive-

psychological literature (cf. Keil 2006; Lombrozo & Vasilyeva 2017). There is
an agreement, however, that providing a causal explanation involves determining
the causes of the thing to be explained (i.e., the explanandum). Thus, current
theories in cognitive psychology concur with the old Aristotelian view of causes
as explanations (see Falcon 2019, for an introduction on the Aristotelian conception
of causes).
1 It
is important to mention that there are also dispositional theories of causal cognition (e.g., force
dynamics, Wolff 2007). Due to length considerations we do not discuss them here.
4.3.1 Causal Model Theories
Causal model theories assume that people construct a mental causal model of
the world, which represents the directed relations among causes and their effects.
Causes and effects are considered to be types of events. A direct relation in the
model is assumed to represent the existence of a causal mechanism that relates
a cause to its effect. No further specifics of the mechanism are assumed to be
represented (see Waldmann 1996; Sloman 2005; Waldmann et al. 2008). In addition,
many causal model theories assume that people represent the strength of the causes
(i.e., their causal power to generate the effect) and that people usually presume
that the influences of different causes are additive (see Waldmann 2000; Griffiths
& Tenenbaum 2005, for evidence that people probably have several conceptions of
how influences add).
Figure 4.1 depicts a graphical causal model of the causal relation between
smoking and lung cancer in general. Nodes represent types of events, arrows the
direct causal relations among them. The model also represents that lung cancer can
be generated by inhaling asbestos fibers. The causal model on the left hand side of
Fig. 4.1 represents only the structure of the causal relations. It shows which causal
relations exist (indicated by the arrows) and which do not (there is no arrow between
smoking and asbestos). The causal model on the right hand side also represents the
strength of the causal relations (i.e., it is a parameterized model). The power of
smoking to cause lung cancer (psmoking ) and the power of inhaling asbestos to cause
lung cancer (pasbestos ) represent the likelihood that the respective cause will generate
the effect when present. Powers are usually assumed to be learnt from observed
relative frequencies (see Cheng 1997, for a formal model). The probabilities of
smoking P(S) and asbestos P(A) capture their respective base-rates. According to
the model the probability of lung cancer (LC) is dependent on smoking and inhaling
asbestos, P(LC|S,A). Causal model theories assume that the dependencies between
effects and different combinations of causes are inferred from the causal powers and
the base-rates.
Smoking Smoking psmoking

P(S)
Lung Cancer Lung Cancer
P(LC|S,A)
Asbestos Asbestos pasbestos
P(A)
Fig. 4.1 Left hand side: Causal model of lung cancer. Nodes represent types of events, i.e. smok-
ing = actions of smoking, lung cancer = acquisition of lung cancer. Arrows are placeholders for
causal mechanisms by which the events are related to each other. Right hand side: Parameterized
causal model also representing the strength of the causal relations between smoking and lung
cancer and between inhaling asbestos and lung cancer
Causal Bayes nets (Pearl 2000; Spirtes et al. 2000) provide a formal account
for modeling causal relations in the world by combining directed acyclic graphs
of causal structures (i.e., graphs not containing feedback loops) with conditional
probability distributions representing the dependencies between the variables. They
also allow to compute changes in conditional probabilities when new information
arrives. Causal Bayes nets have been used widely in cognitive psychology and the
cognitive sciences to describe mental causal models as representations of causal
structures in the world and causal reasoning as inferences based on causal models
(e.g., Gopnik et al. 2004; Griffiths & Tenenbaum 2005).
There has been a lot of experimental research testing predictions derived from
causal model theories. A review of the findings is provided by Rottman & Hastie
(2014). It shows that people’s inferences and judgments are affected by assumptions
about causal structure. However, people’s inferences about the strength of causal
and non-causal relations often deviate substantially from predictions derived from
causal Bayes nets. Research on diagnostic reasoning (i.e., inferring causes from
effects) and causal attribution (i.e., inferring the causes of a particular instance of an
event) showed that people are sensitive to causal structure and causal strength (see
Meder et al. 2014).
4.3.1.1 Explanation
According to causal model theories, explananda can be types and tokens of

events. A type of an event is explained by the causal model, which represents the
causes directly influencing the event. For example, the causal model depicted in
Fig. 4.1 would be an explanation for the general occurrence of lung cancer. The
parameterized model would also explain the risk associated with smoking and why
the risk increases substantially when smoking and inhaling asbestos are both given.
To explain a particular instance of an event (e.g., Peter getting lung cancer), the
generic model has to be instantiated for the particular case. To instantiate the model,
the presence or absence of the causal factors within the model has to be established
(or inferred from other observations). Let’s assume that Peter smoked before getting
lung cancer but he was not in contact with asbestos. To provide an explanation, it
also has to be established whether the present causes actually generated the effect.
Even if a certain cause and the effect were present, the effect might have been caused
by another unknown factor. Thus, Peter’s lung cancer might be actually caused by
something else and not by smoking. Based on the given information, we can only
rule out asbestos. Cheng & Novick (2005) provide a formal account for inferring
actual causation based on PowerPC Theory, which is a special subtype of Causal
Bayes Nets (see also Meder et al. (2014) and Stephan & Waldmann (2018), for
further developments). Cheng and Novick’s model entails that a strong cause of an
observed event, which is present, is likely to have actually generated the event. It
also entails that a present cause, which has a low power to generate the effect, still
explains an observed effect if there are no other causes that may have generated the
event. Therefore, Peter’s smoking explains his lung cancer.
A causal model also provides the basis for an analysis of counterfactual
dependence between a particular instance of a cause and an effect. According to
counterfactual theories of causation in philosophy, a particular event is a cause of
another particular event if the latter would not have happened if the first did not (cf.
Lewis 1973; Menzies 2017). If a type of cause is necessary for a certain type of
event to occur, then counterfactual dependence will be given in every instance. But
even when a cause is not necessary for a type of event, counterfactual dependence
may hold for a particular instance. If there is counterfactual dependence between a
cause and an effect, then the cause can be considered an explanation.
Counterfactual dependence can be inferred from a causal model and the observa-
tions made in a particular case (see Pearl 2000; Halpern & Pearl 2005a, b; Halpern
2015, for a formal analysis). Thus counterfactual dependence can be established,
even when counterfactual cases (i.e., cases in which the cause in contrast to the
present situation was not present) were never observed before. Consider again the
causal model of lung cancer depicted in Fig. 4.1. Recall that we have observed
that Peter smoked, did not work with asbestos, and got lung cancer. Using these
observations, we can instantiate the causal model for Peter’s case. To infer what
would have happened if Peter did not smoke, we mentally undo smoking in the
instantiated causal model and infer the resulting probability of lung cancer. This
probability will be very low, because we know that Peter did not work with asbestos.
Hence, all causes that have a high power to generate lung cancer would be absent.
Note that such a Bayesian account of counterfactuals does not presume that the
probability of the effect given the counterfactual absence of the target cause is zero,
but it should be much lower than in the cause’s presence, and low in absolute terms.
There is a growing body of research investigating whether people consider
counterfactual dependence when asked to infer whether a cause actually led to a
particular effect (see Danks 2016, for a recent overview). The findings in general
support the counterfactual framework. But when participants in an experiment
are able to observe the mechanism generating the effect, they usually base their
judgment on the mechanism rather than the counterfactual dependence (see for
example Walsh & Sloman 2011, and Sect. 4.3.2).
4.3.1.2 Knowledge and Information Needed for an Explanation
In order to fully explain a type of event, the person needs to construct a generic
causal model representing all causes that make a difference for the event as well
as their strength and how they combine in causing the effect. If the type of event
is familiar, the person is likely to be able to recall all relevant causal factors from
memory. She will probably also know how strongly each causal factor affects the
event and whether and how they interact. If the event is unfamiliar, the person could
infer potential causes from observations. There are many indicators of causation
that the person may consider, including correlation, contiguity, and similarity of
cause and effect (see Lagnado et al. 2007, for a summary of respective experimental
research). As an alternative, the person may ask other people about the causes that
may affect the event to be explained, or check her own hypotheses about causal
factors with them.
In order to provide an explanation for a particular instance (i.e., a token event),
the person has to know about the generic causal model for the type of event to
be explained and the presence or absence of the potential causes. Based on this
knowledge, she has two options: she can infer how likely each present cause
actually generated the event. As an alternative, she can make an inference about
counterfactual dependence for each of the causes. Note that even when a person
is familiar with the type of event (i.e., knows about the generic causal model), the
presence of causes has to be found out. Whether a cause is present could be observed
or inquired about. According to the causal model approach, actual causation or
counterfactual dependence usually cannot be observed, but have to be inferred as
outlined above. Nevertheless, experts may be asked, because they may have better
means to assess actual causation or be better able to infer it.
4.3.2 Mechanistic Theories
While causal model theories assume that people focus on events and their causal
dependencies, mechanistic theories assume that people focus on the different types
of mechanisms through which the events are related. Thus, people are assumed
to care about the nature of the mechanisms. Knowledge about different types of
mechanisms in turn allows people to make inferences and provide explanations.
There is much less experimental research investigating mechanistic theories
than studies testing predictions derived from causal model theories. Nevertheless,
research on causal learning indicates that adults and children take into account
whether there is a plausible causal mechanism through which a potential cause and
an effect could be related to each other (e.g., Bullock et al. 1982; Koslowski et al.
1989). If there is none, the same probabilistic dependence is considered much less
causal. Research on causal attribution (i.e., on how people determine the cause of
a particular event) showed that people base their judgment on causal mechanisms
at least when they are observable (e.g., Walsh & Sloman 2011). For example, a
number of empirical studies investigated how people handle cases of preemption in
which two causes were present, each of which was sufficient, but not necessary for
an effect to occur. Consider again the exemplary case put forward by Hall (2004),
in which Billy and Suzy throw stones at the same bottle. Both stones are perfectly
on target, but Suzy’s stone reaches the bottle first and breaks it. In this case there
is no counterfactual dependence, but Suzy’s stone is identified as the actual cause
because there is a continuous process (a mechanism) by which her stone breaks the
bottle (see Walsh & Sloman 2011, for more experimental evidence using various
scenarios).
4.3.2.1 Explanation
The research inspired by mechanistic theories has focused on particular instances

of events as explananda. The framework could, however, be extended to account
for types of events by assuming that there are generic types of mechanisms.
For example, to explain lung cancer, the bio-chemical mechanism by which the
metabolites of the toxic substances in the smoke attach to the person’s DNA,
thereby alter the DNA’s replication and cause mutations that result in uncontrolled
replication (i.e., cancer), could be pointed out (see Hecht 2012).
To explain a particular instance, the mechanism that generated this event has to
be determined. This could be done by observing the mechanism connecting cause
and effect (e.g., billiard balls hitting each other, Michotte 1946) or by establishing
the presence of indicators of the mechanism. For example, to explain why Peter
got lung cancer no direct observation of the mechanism is possible. But it could be
found out whether toxins in the smoke caused the replication of DNA to be faulty,
which is a known mechanism in the causation of lungs cancer. It could also be found
out whether there are asbestos fibers in the lung cells and chromosomal aberrations
caused by them (see Barrett 1994, for details on the mechanisms).
4.3.2.2 Knowledge and Information Needed for an Explanation
To explain a type of event, the person needs to know which types of mechanisms
may lead to the type of event. If the event is familiar, the person may know about
potential mechanisms. There is, however, research indicating that people often have
an illusion of explanatory depth (Rozenblit & Keil 2002). People often assume
they know how things work, that is, what the underlying mechanisms are, but
upon scrutiny, they know very little. Professionals, by contrast, will know about
the mechanisms. Peter’s oncologist will know about the different mechanisms. If an
event is unfamiliar, probably none of the potential mechanisms is known. It is also
very unlikely that a person will be able to learn about mechanisms in everyday life
(apart from observable mechanical processes). Hence the person will have to ask
others for respective information.
In order to provide an explanation for a particular instance, the person has
to know about possible types of mechanisms and how their presence can be
established. Once the person knows about the mechanisms, their presence can be
assessed through observation or by asking experts that are more likely to be able to
access them.
4.3.3 Summary
Causal model theories and mechanistic theories account for people’s causal expla-
nations of events. They do so for types of events and specific tokens of events.
Both accounts assume that people should be able to provide an explanation when
presented with a familiar event, because they would have the required causal
knowledge. The theories, however, differ in what they assume people would
put forward as an explanation: For types of events, generic causal models or
types of mechanisms are suggested. For specific instances (i.e., tokens of events),
explanations should refer to present causes with high power and/or present causes
upon which the effect is counterfactually dependent (causal model theories), or
present mechanisms (mechanistic theories). If people have to provide an explanation
for an unfamiliar event, though, both accounts predict that people will have to
acquire new knowledge first. The accounts differ in what they suppose the required
knowledge is. For neither account, the required knowledge is easy to learn from
observation. Hence, we suspect that people will rather ask others for the knowledge
they consider relevant for providing an explanation. In fact, recent research showed
that people often resort to the community of knowledge when making judgments
and decisions (Sloman & Fernbach 2017). That is, they make use of the causal
knowledge other people have. To do so, asking questions seems to be the premier
option. If this hypothesis is correct, then the questions people ask will give us an
insight into what information people consider necessary for giving an explanation.
The questions will also show us, whether they search for information required by
causal model or mechanistic theories. In the next part of the paper, we will present
two empirical studies investigating the questions people ask.
4.4 Empirical Studies on the Questions People Ask in Order

to Explain
Investigating the questions people ask is a rather uncommon research strategy

in cognitive psychological research on causal attribution and explanation, that is,
research on how people infer the cause(s) of a particular instance of an event and
account for its occurrence.2 Usually, predictions derived from the theoretical models
(including the theories outlined above) are tested by presenting participants with an
artificial scenario containing relevant information (like in the Billy and Suzy case
presented above). Based on the scenario participants are asked for a judgment. The
information given to participants in these studies is manipulated systematically to
investigate how the given information affects people’s inferences and judgments.
The great advantage of this research strategy is the complete control over the
information participants can use in their causal reasoning. The downside is that this
research only shows how people consider information that is given to them. These
studies cannot show whether people will consider the same information when they
2 By contrast, there is quite a bit of research on information search in causal learning and hypothesis
testing (see Crupi et al. 2018, for an overview). There is also some research on information search
in decision making and problem solving (e.g., Huber et al. 1997).
would have to search for it. It might be that people take into account information
when it is presented to them, but not when they would have to look it up. Loosely
speaking, they may recognize the importance of a piece of information when they
see it, but they may not recall its importance when it is not right in front of them.
There are very few studies investigating the questions people ask and the
information they search for in causal attribution. Notable exceptions are studies
by Ahn et al. (1995) and, more recently and concerned with children’s causal
questions, Ruggeri & Lombrozo (2015). Ahn et al. investigated causal attribution.
Prior research (e.g., Kelley 1973; Cheng & Novick 1990) showed that people
consider the covariation of the type of event to be explained with potential causal
factors. People attribute the token event to the causal factor or combination of
factors that correlates with the type of event in general. That these findings support
causal model theories as covariation is an indicator for the existence of a causal
relation and its strength. Ahn et al. (1995) suggested that participants in these
prior studies based their judgments on covariations because they received only
information on covariation. Based on a mechanistic theory, they hypothesized that
in order to explain every-day events, people would prefer to look for the causal
mechanism that generated the event. Given the choice, people would not search for
covariation. To decide between the competing hypotheses of the two theories, Ahn
and colleagues ran studies in which they presented participants with descriptions of
specific token events (e.g., “Peter liked a certain piece of classical music today”) and
asked them to write down any question they would like to have answered in order
to identify the causes of the event. Participants’ questions were coded as inquiring
about covariation (e.g., “Do other people like the piece of music as well?”), testing
hypotheses about causes and/or causal mechanisms (e.g., “Was Peter in a good
mood today”), requesting general information (e.g., “Is the piece by Mozart?”),
inquiring about details of the effect (e.g. “How much did he like it?”), or other.
It turned out that on average more than 60% of questions tested hypotheses, while
only 10% asked for information about covariation. These findings and the findings
of subsequent studies clearly showed that people do not tend to inquire about
covariation when asked to identify causes of a particular instance. Unfortunately,
the findings do not clearly decide between causal model and mechanistic theories.
As outlined above, searching for the presence of potential causes is predicted by
causal model theories. It is also predicted by mechanistic theories, because the
potential cause is a part of the generating mechanism. Hence, the high number of
hypothesis testing questions does not provide clear evidence for either approach. A
more detailed analysis of the questions is needed to differentiate the two accounts,
which we will attempt in Study 2.
4.4.1 Study 1
The aim of the first study was to find out what people want to have explained
in everyday life. In other words, we wanted to find out what kind of explananda
people are interested in. Therefore, we analyzed questions posted on the internet on
answers.yahoo.com. People can submit any kind of question to the website. Other
users answer many but not all of the posted questions.
4.4.1.1 Sample
We sampled 1000 questions from answers.yahoo.com, 250 from each of the

four domains: natural sciences, health, psychology, and beauty. For each domain,
we picked a random question and then coded the subsequent 250 questions. If
people asked two or more distinct questions, we only coded the first question. No
demographic data on the people asking the question was available as users remain
anonymous on the website.
4.4.1.2 Coding of Questions
In a first step, questions were coded as causal or non-causal by a research assistant.

Causal questions were defined as any question which refers to causes, consequences,
effects, explanations, functions, interventions, relations, mechanisms, predictions,
diagnoses, interventions, or counterfactuals. When in doubt, questions were coded
as not causal.
In a second step, each causal question was classified into one the following
categories:
1. Explanation questions = questions asking for the cause(s) or explanation of
something (e.g., Why X? What is the cause of X? How does X work? Why does
X lead to Y?)
2. Causation questions = questions inquiring whether a certain causal relation holds
(e.g., Does X cause Y? Did X cause Y on this occasion?). In general, these
questions can be answered by Yes or No.
3. Prediction questions = questions asking for the consequences or effects of
something (e.g., What will result from X? What are the consequences of X?)
4. Utility questions = questions inquiring about the utility that results from X and
its consequences (e.g., Is doing X dangerous? Is eating X healthy?)
5. Intervention questions = questions about which action to perform in order to
obtain a desired outcome (e.g., What shall I do to get rid of X? Does doing X
will give me Y?). Questions not stating the relevant outcome were not coded as
intervention questions (e.g., Shall I do A or B?)
6. Other: questions that refer to causes, effects or a causal process but cannot be
assigned to any other category (e.g., questions about the time course of a process
or the strength of a cause)3
3 Note that these questions can provide important information for causal attribution. Information
about the time course of events can rule out certain causes as actual causes like in the Billy and Suzy
case, and information about causal power or strength can also help to establish actual causation.
Two raters coded causal questions into the categories. The agreement was 85%
overall, which can be regarded as good (Hartmann et al. 2004). Disagreements were
resolved through discussion.
In a third step, explanation questions were further coded into what they inquired
about:
(A) A single token instance (e.g., Why did X happen on this occasion) or a type
(e.g., Why does X happen in general?) and
(B) an explanation for an event/state/action/feature (e.g., Why do people do X?) or
an explanation of a relation/mechanism/process (e.g., Why does X cause Y?
How come that X leads to Y?)
4.4.1.3 Results
Table 4.1 depicts the results of the three coding steps. It turned out that roughly
half of the questions (49.5%) could be classified as causal questions. Note that the
percentage of causal questions probably depends on the domain of inquiry. People
may have more or less causal questions in other domains. More interesting are the
results of the second coding step. Of the causal questions, 29% were classified
as asking for an explanation and another 16% as asking about causation. These
questions presumably serve an epistemic function. More questions seem to serve a
pragmatic function: 29% of questions inquired about an intervention, another 16%
about the utility of something, and 8% asked about consequences to expect (i.e.,
predictions).
Of the explanation questions, most inquired about a type of event. For example,
one person asked “Why do people like watching people play video games over the
Table 4.1 Results of Study 1: Classification of 1000 questions from answers.yahoo.com from the
domains: natural sciences, health, psychology, and beauty
Step 1: Classification of all questions into causal and non-causal questions
Causal Non-causal
495 505
49.5% 50.5%
Step 2: Classification of causal questions
Explanation Causation Prediction Utility Intervention Other
145 79 38 77 142 14
29.3% 16.0% 7.7% 15.6% 28.7% 2.8%
Step 3: Classification of explanation questions
Token event Token relation Type of event Type of relation
27 6 79 33
18.6% 4.1% 54.5% 22.8%
Note. Absolute numbers and percentages are shown. Percentages add up to 100% for each step.
See text for definitions of categories
internet?” Also, a substantial number of questions asked about causal relations on

a type level, for example “Why is pizza fattening?” and “How can body mists cool
you down in summer?” Less questions inquired about specific token instances, for
example “Why does my 7 year old hoard litter, scraps, random things? ( . . . )“ and
“Why am I not losing weight/body fat? I’ve been exercising ( . . . ) every day for
almost a year ( . . . ) but I have not reached my goal weight yet.” Note that the first
question asks for an explanation of an action (hoarding), while the second asks about
the (missing) relation among exercise and weight loss for a specific person.
There were some additional, interesting findings. Although we analyzed 1000
questions, we did not find certain types of questions. We did not find any question
about covariation or the strength of a causal relation. This is in line with other
research on information search (cf. Huber et al. 1997; Ahn et al. 1995). We also did
not find any question whether a cause was necessary or sufficient for a particular
effect or for a type of event in general. Finally, we did not find questions involving
counterfactuals.
Of course, these findings might be due to the method we used. We analyzed
questions posted on the internet. Inquirers may assume that potential responders
might not be able to answer a counterfactual question, because they lack the
necessary knowledge about the specifics of the case. This hunch is supported by
the finding that quite a number of people described a specific token instance, but
then asked a question on a type level. For example, an apparently young woman
described being maltreated by her boyfriend after refusing oral sex. But her question
was “Why do you [other men] expect a girl to go down on you but you don’t go down
on her?”
4.4.1.4 Discussion
The primary aim of this first study was to find out what people want to have
explained in everyday life. It turned out that they asked for explanations of specific
instances and types of events. Most often, they asked for an explanation of an
event, action, or state. Less often, they asked for an explanation of a relation or
mechanism. This finding is interesting, because research on causal attribution and
explanation so far focused on explanations of particular instances. To the best of
our knowledge, there is no research on how people explain causal relations or
mechanisms. A mechanism is usually considered to be an explanans (something
that explains something else) rather than an explanandum.
The finding that there were no questions inquiring about quantitative aspects
(e.g., the strength of a cause) seems to be at odds with causal model theories.
Moreover, the lack of questions concerning information, which is necessary to
explain a particular instance of an event, does not seem to fit well with either
account. However, inquirers often directly asked for an explanation. Thus, they
might have been reluctant to ask more specific questions, which would have allowed
them to come up with an explanation themselves. Therefore, the findings provide no
counterevidence for causal model and mechanistic theories. In order to obtain direct
evidence, we must ensure that the inquirer will have to come up with an explanation
herself. In this case, the person will have to acquire all knowledge and information
she considers necessary for giving an explanation.
4.4.2 Study 2
The aim of the second, experimental study was to investigate directly which
questions people ask in order to provide an explanation. Participants were presented
with a scenario describing a fact, which was either a type of event or a specific
token instance of that event. For example, they were told that “Basil plants die
within two days” or that “Katrin’s basil plant died within two days”. We also
manipulated whether participants were familiar with the type of event. Basil plants
and their tendency to die within a short time period, for example, were assumed
to be familiar. All unfamiliar events were made up for this study, which precludes
any familiarity (see Table 4.2). As a third factor we also manipulated whether the
description highlighted that the event to be explained was abnormal (e.g., “basil
plants sometimes die within two days although they usually last for weeks”). As
abnormality was not the focus of this paper, we did not analyze the results with
respect to this factor yet and we will not include them in the results section.
Participants in all conditions were told that they would later have to provide an
explanation for the presented fact. Before doing so, they could ask any question
they considered helpful.
Table 4.2 Facts to be explained by participants in Study 2

Unfamiliar token event Familiar token event
Benni’s polinalyte cell culture turned purple today. Karin’s basil plant died within 2 days.
Tina is suffering from the disease Krokuritasis today. Erika suffers from stomach ache today.
Egon’s Lomodon, a newly discovered species, blew Susi’s dog peed in her apartment
steam from its nostrils today. today.
Sandra, who is member of a cult, performed a Petra failed an oral exam at university
unicorn-summoning-spell today. today.
Anna’s Fantasix, a new technical device, displayed a Anna’s computer took 30 s to boot
blue light today. today.
Unfamiliar type of event Familiar type of event
Polinalyten cell cultures turn purple. Basil plants die within 2 days.
People suffer from the disease krokuritasis. People suffer from stomach ache.
Lomodones, a newly discovered species, blow steam Dogs pee in the apartment.
from their nostrils. Students fail oral exams at university.
Cult members perform unicorn-summoning spells. Computers take 30 s to boot.
Fantasixs, a new technical device, display a blue light.
Note. All facts were presented in German, the participants’ primary language
4.4.2.1 Participants and Design
Forty-eight first-year students of psychology (mean age: 20.8 years, 88.1% female)
at the University of Göttingen, Germany, completed the study for course credit. A
short explanation of the task was provided and consent to participate was obtained.
Ethics was not required according to regulations at Goettingen University as we did
not misinform participants about the purpose of the study and the study involved no
risk for participants. Participants were randomly assigned to one of four between-
subjects conditions created by combining the factors type vs. token and abnormality
highlighted vs. not highlighted. The factor familiarity was manipulated within
subjects (i.e., each participant saw five familiar events and five unfamiliar events).
4.4.2.2 Procedure
The experiment was conducted online and in German. Participants in all conditions
were instructed that their task would be to explain a number of hypothetical events
(which would be briefly described to them in one or two sentences). However, since
participants would be lacking sufficient information to do so, they would first be
allowed to write down as many questions about the event as they wished. To increase
motivation for submitting sensible questions, we instructed participants that there
would be a second part of the study 2 weeks later, in which they would receive
answers to their questions and then would have to provide their best explanation for
each event. There was in fact no second part to the experiment, which was revealed
to participants at the end of the study. Each participant was presented with ten short
descriptions of events, in randomised order. In the token condition, all events were
described as single instances (see Table 4.2 for all items). In the type condition, the
same events were presented as general phenomena.
Familiarity of events was manipulated within subjects, with five scenarios
referring to familiar events (like the Basil plant example), and five other events
involving made-up entities about which participants could not have any specific
background knowledge (e.g., “Benni’s polinalyte-cell culture turned purple today”).
Events were taken from four domains (biological, technical, medical, and social),
and the same basic events were used in all four between-subject conditions.
After reading each event description, participants wrote down their questions into
five free text fields. They were told that they could enter more than one question per
field, if they wished to ask more.
4.4.2.3 Coding of Questions
Two independent, trained coders categorised participants’ questions according to

the requested information, using a classification scheme with five main categories
(see Table 4.3 for an overview). The classification scheme was designed to code
Table 4.3 Classification scheme for participants‘questions

Main categories Subcategories Examples for questions assigned to the category
1. Identification (a) Direct question What causes this disease?
of causes Is the disease caused by bacteria?
(b) Contiguous What happened before?
factors Did anything special happen before?
(c) Covariation Do other dogs also do this?
Does the machine show blue light at other times as
well?
(d) Generic How come that dogs cannot wait to pee?
mechanisma What the mechanism behind krokuritasis?
2. Sufficiency, NA Does a lack of education always lead to peeing
necessity, or problems?
causal strength of Is being threatened necessary for lomodones to blow
a factora steam from their nostrils?
How often does no watering kill basil plants within
2 days?
3. Presence of NA Did she travel recently?
causal factors What was the temperature on that day?
Is the animal sick?
4. Actual (a) Direct question Did the dog pee in the apartment because no one took
causation him for a walk?
(b) Mechanism Does she have stomach ache because there is an
important event coming up that makes her nervous?
(c) Counterfactual Would the plant have lived longer under different
circumstances?
5. Event or NA What kind of cell culture is this?
affected entity What is this machine used for?
How does this disease progress?
6. Others/unclear What are the consequences?
How was this species discovered?
a We assessed whether participants asked for generic mechanisms, the necessity or sufficiency of a
type of cause, or the causal strength/power of a type of cause. Participants did not do so. Therefore,
the examples shown are made up. Examples for other categories are questions participants actually
generated
the requested information into categories that allowed us to see whether participants
searched for the knowledge required by causal model and mechanistic theories.
The first category included all questions that inquired about potential causes of
the event to be explained. A question was classified as a direct question when it
inquired about the general causes of an event (“Why do people get stomach pain?”)
or whether a factor was causally related to the target event in general (“Is the disease
caused by an infection?”). Note that all examples given here were questions actually
asked by participants. Participants may also inquire about indicators of causation.
One indicator is contiguity. Questions were classified as asking for Contiguous
Factors when they requested information about what happened close to the event in
time or space (“Did anything special happen before?”, “What else happened in the
situation?”). A second important indicator is covariation. Causes correlate with their

effects. Apart from directly asking about a correlation, the participant may inquire
about consensus (“Do other dogs also show this kind of behaviour?”), consistency
(“Does she have stomach pain at other times as well?”), and distinctiveness (“Do
they [lomodones] blow steam from their noses only in certain situations?”). Answers
to these questions would allow the inquirer to infer whether the event is due to the
person/dog, the type of stimulus, or something particular in this situation (see Kelley
1973). We also subsumed questions inquiring about mechanisms that may generate
the event to be explained into the first category. Questions were assigned to this
category if they asked for a mechanism or process or asked how the event to be
explained comes about or is generated in general.
The second category were questions inquiring about the necessity, sufficiency,
and/or causal strength/causal power of a potential cause. All questions that referred
to quantitative aspects of a causal relation were coded here. As there were no
questions in this category, no examples can be provided.
The third category comprised all questions that refered to the presence of a
specific causal factor. An exemplary question was “Did she eat something bad?”
(in response to an item about a person suffering from stomach ache). In contrast to
Ahn et al. (1995), we differentiated between questions asking for the presence of a
causal factor and questions asking for the presence of a mechanism. We assigned
the latter to the next category (subcategory mechanism).
The next category of questions collated questions that would allow participants to
determine whether a cause actually generated a particular token event. Participants
may directly ask whether a potential cause was present and actually caused the effect
(“Was X the cause for Y?”, “Did X cause Y?”). Respective questions (e.g., “Did the
dog pee in the apartment because no one took him for a walk?”) were assigned to the
first subcategory. Participants may also inquire about the mechanism which actually
caused the event to be explained (“Did X lead to Y which resulted in Z?”, “What’s
the mechanism behind X?”). To be coded as a mechanism question, participants
either had to ask for a mechanism or process, ask “how (come)”, or describe a
sequence of events which resulted in the event to be explained. The last subcategory
were counterfactual questions (“Would X not have happened if Y had not happened
before?”). We categorised a question as counterfactual when it made any reference
to a hypothetical states of affairs.
The fifth category were questions that requested more information about the
event to be explained or about the entity (e.g., person, plant, technical device) that
was affected by it. We added this category after data collection because we realised
that a substantial number of questions fell into this category. All remaining questions
were assigned to an Other-category.
4.4.2.4 Hypotheses
As outlined above, both theoretical accounts expect a substantial difference between

familiar and unfamiliar events. Let us consider a type of event first. When a type
of event is familiar, causal model and mechanistic theories expect participants to

have causal knowledge, which would allow them to explain the type of event (by
either a causal model or a type of mechanism). This is not the case when the type
of event is not familiar. Thus, less questions would be expected for familiar types
of events. If the type of event to be explained is unfamiliar, then causal model
theories predict that participants will inquire about potential causes and their causal
strength. Mechanistic theories, by contrast, predict that participants will inquire
about potential mechanisms.
When a particular token event has to be explained, both theories predict that
participants will ask for additional information, even when they are familiar with
the type of event. If the type of event is familiar, causal model theories predict
that participants will inquire about the presence of potential causes in order to
instantiate the model for the particular case. Based on the instantiated causal
model, counterfactual dependence and actual causation can be inferred. Note that
the latter requires the person to know about the causal power of each cause.
Mechanistic theories predict that participants will inquire about the presence of
potential mechanisms or at least indicators of these mechanisms. If the particular
token event is unfamiliar, both theories predict that participants will first inquire
about causes or mechanisms for the type of event and then about the actually present
causes or mechanisms.
4.4.2.5 Results
Participants generated 1772 questions, which were assigned to the categories

described above by two independent coders. Agreement among coders was 81%.
Disagreements were resolved through discussion and every question was assigned
to only one of the categories. For each participant we calculated the number of
questions he or she asked for familiar and unfamiliar, token or type of events.
Relative to the number of questions, the percentage of each category of question
across scenarios was computed for each participant. It turned out that there were
two categories of questions for which no questions could be found: Category 1d:
generic mechanisms and Category 2: Sufficiency, necessity, or causal strength.
On average, participants asked 32 questions (SD = 11.8) for the 10 facts to be
explained. A multi-level linear model with a random intercept and type vs. token and
familiarity as fixed factors, and the number of questions as the dependent variable
revealed no differences between the number of questions asked, all ps > = .19.
Surprisingly, the number of questions was slightly lower for unfamiliar than familiar
scenarios, Mtoken, familiar = 16.3 (SD = 6.2), Mtype,familiar = 17.5 (SD = 4.7),
Mtoken, unfamiliar = 15.5 (SD = 4.6), Mtype,unfamiliar = 16.5 (SD = 6.6).
Figure 4.2 shows the mean relative frequencies of the four main categories
of questions across participants separated out for type vs. token and familiar vs.
unfamiliar events. Results for Category 2 are not shown as there were no questions.
To ease comparisons, we fixed the Y-axis to a range between 0% and 70%.
As the four graphs show, there were substantial differences between conditions.
Questions to identify cause(s) Questions about the presence of causes

70% 70%
60% 60%
Mean percentage
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%
High Low High Low High Low High Low
familiarity familiarity familiarity familiarity familiarity familiarity familiarity familiarity
Token Type Token Type
Questions to establish actual causation Questions about event or affected entity

70% 70%
60% 60%
Mean percentage
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%
High Low High Low High Low High Low
familiarity familiarity familiarity familiarity familiarity familiarity familiarity familiarity
Token Type Token Type
Fig. 4.2 Mean percentages of questions coded into categories 1–5 (+/−95% confidence interval)
in the four conditions. Category 2 is not shown as we did not find any question in this category
People asked more questions to identify a potential cause when asked to explain
a type of event compared to a token, and more questions to establish actual
causation when searching for an explanation of a particular token event compared
to a type. Interestingly, few questions about actual causation were asked, even
when participants were told that they would have to explain a particular instance.
Participants asked more questions about the presence of potential causes when they
were familiar with an event compared to when they were unfamiliar. Finally, they
asked more questions about the event to be explained or the affected entity when
they were unfamiliar with the explanandum compared to when they were familiar.
In order to test whether these observable differences between conditions hold
statistically, we ran a multi-level analysis for each category of questions with a
random intercept and with familiarity and type vs. token as well as their interaction
as fixed effects. Note that inter-individual differences of participants are captured
by the random intercept. The statistical analyses corroborated the findings from
the visual inspection. For questions inquiring about potential causes, there was a
difference between type and token events, t(46) = 3.18, p = .003. There was no
effect of familiarity and no interaction (p > .08). For questions to establish actual
causation, there was a clear effect for type vs. token, t(46) = −3.0, p = .004, no
35%
30%
25%
Mean percentage
20%
15%
10%
5%
0%
Familiar
Familiar
Familiar
Familiar
Familiar
Familiar
Unfamiliar
Unfamiliar
Unfamiliar
Unfamiliar
Unfamiliar
Unfamiliar
Token Type Token Type Token Type
direct question contiguity covariation
Fig. 4.3 Detailed analysis of questions asked to identify causes of an event. Graph shows mean
percentages of questions relative to all questions assigned to categories 1–5 (+/− 95% confidence
intervals). (Note. Percentages in each condition add up to the percentages shown in the graph in
the upper left corner of Fig. 4.2)
effect of familiarity, t(434 ) = −1.76, p = .085, and no interaction (p = .601). For

questions inquiring about the presence of causes, there was an effect of familiarity
t(43) = −7.95, p = .000, no effect of type vs. token, t(46) = −1.85, p = .071,
and no interaction (p = .120). Finally, for questions inquiring about the event
to be explained or the affected entity, a significant effect of familiarity resulted,
t(43) = 12.6, p = .000. All other effects were not significant (p > .80).
Unfortunately, it turned out that participants asked very few questions about
actual causation. Even in the token conditions they asked on average between 1 and
2 questions only. Therefore a more detailed statistical analysis did not make sense.
Nevertheless, we can report that participants sometimes asked for mechanisms (10
of 113 total actual causation questions) and for counterfactual dependence (14 of
113 total actual causation questions). Most often, however, they asked directly
whether a particular causal factor caused the event in the present case.
We were able, though, to conduct a more detailed analysis of the questions
people asked to identify causes. Figure 4.3 shows the results. It turned out that
they asked more direct questions about potential causes of an event when a type
of event had to be explained rather than a token, t(46) = 3.18, p = .001. They also
asked more for covariation given types, t(46) = 4.03, p = .000. No other effects
or interactions turned out significant in these analyses. By contrast, participants
4 Threeparticipants which were assigned to the token condition failed to respond to the unfamiliar
token events. Therefore the degrees of freedom were smaller for this comparison.
asked more for contiguous factors when they had to explain a token rather than
a type, t(46) = −2.13, p = .039,5 and they did so more when the event was
familiar compared to unfamiliar, t(43) = −3.60, p = .001. The interaction was not
statistically significant.
4.4.2.6 Discussion
The results of the second, experimental, study clearly showed that the two investi-
gated factors influenced the questions people asked assuming that they would later
have to explain the given event. When people were familiar with the presented event,
they asked more for the presence of particular causes. This makes sense, as they
probably knew about potential causes already. Causal model theories predict this
finding. This finding, however, is also in line with mechanistic theories, assuming
that participants asked about present causes, because these are the starting points
of causal mechanisms. When the event was not familiar, participants asked many
questions about the event and the affected entity, which they did not for familiar
events. This finding was surprising to us and was not predicted by any of the
theories.
Why would people ask questions about the effect or the affected entity in order to
explain an event? We can only speculate at this point. Knowing more about the event
and the affected entity could help participants to categorize the event and thereby
access higher-order domain-specific knowledge about causal factors and causal
relations. For example, finding out that krokuritasis is a type of gastro-intestinal
disease would enable participants to activate knowledge about gastro-intestinal
problems in general and their likely causes. Knowing more about the affected
entity can give pointers to potential causes as well. For example, knowing that the
person (or persons) affected by krokuritasis have food allergies points towards these
allergies as a potential cause. Future research will have to explore these speculations.
The second factor we investigated was whether participants had to explain a type
of event or a specific token instance of an event. In Study 1 we showed that people
ask for explanations of both. When they had to explain a specific token event rather
than a type, they asked more questions about actual causation, which we expected.
However, they still asked very few questions about actual causation overall, and
very rarely about mechanisms and counterfactuals. They also asked more about
contiguous factors for token events compared to types of events. When presented
with a type of event to explain, they asked more questions to identify causes, more
specifically they asked more questions about covariation and less about contiguity.
This makes sense, because covariation is a good indicator for causation on the type
level.
5 Note that this difference would not be statistically significant when controlling for the number of
analyses conducted (controlling for the number of analyses avoids an inflation of the risk for an
alpha error in statistical analyses). All other statistically significant results would still be significant.
4.4.2.7 Limitations
The first limitation of the second study is the number of participants in this study
(N = 48), which were all psychology undergraduates. One may argue that this
sample is rather small and not representative. We agree. However, participants took
the study seriously and generated more than 1700 questions, of which almost all
were very thoughtful and would have provided relevant information. Recall that
participants believed that they would get answers to these questions and then would
have to provide an explanation. A second limitation is the number of scenarios.
We presented participants with 10 scenarios from five different domains. We do
not know how the findings will generalize to scenarios in other domains. A third
limitation is that we did not provide participants with answers, allowed them to ask
additional questions, and then analyse their answers. Proponents of mechanistic and
causal model theories may argue that participants would have inquired about the
information they consider relevant if participants would have had the opportunity
to ask more questions after receiving answers to their first set of questions. It is
plausible that the first stage of information search in explanation tasks like ours
consists in constructing or instantiating a causal model for the event to be explained.
Once relevant causal factors and their presence or absence are established, people
might move on to figure out which of the present causes actually caused the event in
this case. Very recently a respective study was conducted in which participants were
given answers to their first set of questions and were then given the opportunity to
ask further questions before they had to give an explanation for a type or a token of a
familiar or unfamiliar event (Bertz 2018). It turned out that participants asked more
actual causation questions in round two than in round one, but the absolute number
of these questions was still very low. These findings corroborate the findings we
reported here.
4.5 General Discussion
4.5.1 Implications of Findings for Cognitive Psychological

Theories
In brief, neither causal model theories nor mechanistic theories were fully supported
by the results of the experimental Study 2 in which participants were requested
to ask questions in order to explain. Counter to assumptions made by mechanistic
theories, participants did not inquire about types of mechanisms to account for a
type of event. Even when asked to explain a specific instance of an event, they
rarely asked about a mechanism. Instead, they asked about the presence of causal
factors when the event was familiar and about the event or affected entity when
it was unfamiliar. Ahn et al. (1995) argued that questions about the presence of
causal factors are questions about mechanisms. Although we cannot rule out this
possibility, as present causal factors are the starting points of mechanisms, we saw
no positive evidence for this interpretation.
For causal model theories, the evidence seems to be mixed. On the positive
side, participants asked many questions, which allowed them to construct a causal
model for unfamiliar events. The questions participants asked about the event to
be explained may have allowed them to categorize the type of event and access
more abstract, higher order knowledge of the domain. This knowledge (e.g., about
types of diseases) could be used to construct a causal model. The model in turn
could be tested by subsequent questions about individual causal factors. Another
predicted finding – and therefore positive evidence – was that participants asked
many questions about the presence of known causal factors. Knowing about the
presence of causal factors is a prerequisite for explaining specific token events. On
the negative side, we did not observe questions inquiring about the strength of a
causal relation or the causal power of a causal factor, even when the event to be
explained was unfamiliar. There were similar findings in previous studies (Ahn et
al. 1995; Huber et al. 2011). Mental causal models, which do not represent the power
of causes, however, do not allow to infer actual causation (i.e., the probability that
an observed effect was caused by an observed cause) or counterfactual dependence
(cf. Pearl 2000). Hence, it is unclear how specific token events could be explained
based on the acquired knowledge.
Overall, the findings seem to indicate that people ask for information that allows
them to construct a causal model just representing the structure of causal relations,
that is, a model which only represents the causal factors affecting the type of event
to be explained. In addition, they ask which of these factors are present. They might
do so to reduce the number of causal factors within the model and to determine the
causes that could have impacted the event in a specific instance. Maybe participants
already consider present causal factors as explanations, because they are known to
affect the event to be explained in general. Maybe only a few participants considered
actual causation or counterfactual dependence as relevant for giving an explanation
and asked respective questions.
4.5.2 Conclusion
In this paper, we aimed to give a brief overview of two theoretical frameworks

in cognitive psychology that may account for explanation: causal model theories
and mechanistic theories. We showed that these theories make different predictions
about how people explain an event, which could be a type of event (e.g., getting
lung cancer) or a particular instance of an event (e.g., Peter getting lung cancer). To
give an explanation, these theories make assumptions about the knowledge people
have and the inferences they make based on this knowledge. If people do not
have the required knowledge, they would have to acquire it. In principle, people
would be able to get the knowledge through learning. But learning takes time and
needs valid, observable data. Such data may not be available in everyday life. This
is especially true for learning about mechanisms, like the mechanisms by which
smoking causes lung cancer. Instead, people could ask other, more knowledgeable
people. Therefore, we started to look at the questions people ask in order to give an
explanation. Study 2 was such a study.
Our second aim was to investigate whether people ask for the information
they would need to give an explanation according to causal model or mechanistic
theories. Study 1 showed that people quite often ask for explanations and that they
ask about explanations for particular instances of events, types of events, and causal
relations. Study 2 analysed the questions they asked in order to explain familiar
and unfamiliar, type and token events. The results did not clearly support any of
the theories. It is important to note that there is a lot of evidence for the role of
mechanisms and counterfactuals in explaining instances of events coming from
experimental studies using other research methods (see Danks 2017; Keil 2006;
Lombrozo & Vasilyeva 2017, for overviews). The crucial difference between these
studies and Study 2 is that they presented participants with information and then
looked whether and how participants used the given information. The results showed
that people consider mechanisms, causal power, and counterfactual dependence
when explaining, making a causal attribution, or judging actual causation. By
contrast, we did not provide any information to our participants, but analysed what
they asked for. We found that they rarely asked for causal power, mechanisms, or
counterfactual dependence. Hence, there is a gap between the findings of studies
using these two research methods. The task for the future is to find out how to
bridge the gap. We need theoretical work to create models that are able to account
for both sets of findings. We also need additional empirical work, more studies
on explanation that investigate information search and questions asked to validate
and extend the existing findings, and new research paradigms that investigate
information search and information processing while deriving an explanation.
References
Ahn, W. K., Kalish, C. W., Medin, D. L., & Gelman, S. A. (1995). The role of covariation versus
mechanism information in causal attribution. Cognition, 54, 299–352.
Barrett, J. C. (1994). Cellular and molecular mechanisms of asbestos carcinogenicity: Implications
for biopersistence. Environmental Health Perspectives, 102(Suppl 5), 19–23.
Beebee, H., Hitchcock, C., & Menzies, P. (2009). The Oxford handbook of causation. New York:
Bertz, L. (2018). Asking questions to provide a causal explanation – The role of familiarity.
Unpublished Bachelor’s thesis, Georg-August-University Göttingen, Göttingen, Germany.
Bullock, M., Gelman, R., & Baillargeon, R. (1982). The development of causal reasoning. The
Developmental Psychology of Time, 209–254.
Cheng, P. W. (1997). From covariation to causation: A causal power theory. Psychological Review,
104(2), 367–405.
Cheng, P. W., & Novick, L. R. (1990). A probabilistic contrast model of causal induction. Journal
of Personality and Social Psychology, 58(4), 545.
Cheng, P. W., & Novick, L. R. (2005). Constraints and nonconstraints in causal learning: Reply to
White (2005) and to Luhmann & Ahn (2005). Psychological Review, 112(3), 694–706.
Crupi, V., Nelson, J. D., Meder, B., Cevolani, G., & Tentori, K. (2018). Generalized information
theory meets human cognition: Introducing a unified framework to model uncertainty and
information search. Cognitive Science, 42, 1410–1456.
Danks, D. (2014). Unifying the mind: Cognitive representations as graphical models. Cambridge:
MIT Press.
Danks, D. (2016). Causal search, causal modeling, and the folk. In J. Sytsma & J. W. Buckwalter
(Eds.), A companion to experimental philosophy (pp. 463–471). Oxford: Wiley Blackwell.
Danks, D. (2017). Singular causation. In M. R. Waldmann (Ed.), Oxford handbook of causal
reasoning (pp. 201–215). Oxford: Oxford University Press.
Didkowska, J., Wojciechowska, U., Mańczuk, M., & Łobaszewski, J. (2016). Lung cancer epidemi-
ology: Contemporary and future challenges worldwide. Annals of Translational Medicine, 4(8),
150.
Dowe, P. (2000). Physical causation. Cambridge: Cambridge University Press.
Falcon, A. (2019). Aristotle on causality. In The Stanford encyclopedia of philosophy (Spring 2019
Edition). Retrieved from https://plato.stanford.edu/archives/spr2019/entries/aristotle-causality/
Gopnik, A., Glymour, C., Sobel, D. M., Schulz, L. E., Kushnir, T., & Danks, D. (2004). A theory of
causal learning in children: Causal maps and Bayes nets. Psychological Review, 111(1), 3–32.
Griffiths, T. L., & Tenenbaum, J. B. (2005). Structure and strength in causal induction. Cognitive
Psychology, 51(4), 334–384.
Hagmayer, Y., & Fernbach, P. (2017). Causality in decision-making. In M. Waldmann (Ed.), The
Oxford handbook of causal reasoning (pp. 495–512). New York: Oxford University Press.
Hall, N. (2004). Two concepts of causation. In J. Collins, E. Hall, & L. Paul (Eds.), Causation and
counterfactuals (pp. 225–276). Cambridge, Ma: MIT Press.
Halpern, J. Y. (2015). A modification of the Halpern-Pearl definition of causality. In Proceedings
of the 24th international joint conference on artificial intelligence (IJCAI) (pp. 3022–3033).
Halpern, J. Y., & Pearl, J. (2005a). Causes and explanations: A structural-model approach. Part I:
Causes. The British Journal for the Philosophy of Science, 56(4), 843–887.
Halpern, J. Y., & Pearl, J. (2005b). Causes and explanations: A structural-model approach. Part II:
Explanations. The British Journal for the Philosophy of Science, 56(4), 889–911.
Hartmann, D. P., Barrios, B. A., & Wood, D. D. (2004). Principles of behavioral observation. In S.
N. Haynes & E. M. Hieby (Eds.), Comprehensive handbook of psychological assessment. Vol.
3: Behavioral assessment (pp. 108–127). New York: Wiley.
Hecht, S. S. (2012). Lung carcinogenesis by tobacco smoke. International Journal of Cancer,
131(12), 2724–2732.
Huber, O., Wider, R., & Huber, O. W. (1997). Active information search and complete information
presentation in naturalistic decision tasks. Acta Psychologica, 95, 15–29.
Huber, O., Huber, O. W., & Bär, A. S. (2011). Information search and mental representation in
risky decision making: The advantages first principle. Journal of Behavioral Decision Making,
24, 223–248.
Keil, F. C. (2006). Explanation and understanding. Annual Review of Psychology, 57, 227–254.
Keim Campbell, J., O’Rouke, M., & Silverstein, H. (2007). Causation and explanation. Cam-
bridge, MA: MIT Press.
Kelley, H. H. (1973). The processes of causal attribution. American Psychologist, 28(2), 107.
Koslowski, B. (1996). Theory and evidence: The development of scientific reasoning. Cambridge,
MA: MIT Press.
Koslowski, B., Okagaki, L., Lorenz, C., & Umbach, D. (1989). When covariation is not enough:
The role of causal mechanism, sampling method, and sample size in causal reasoning. Child
Development, 60(6), 1316–1327.
Lagnado, D. A., Waldmann, M. R., Hagmayer, Y., & Sloman, S. A. (2007). Beyond covariation. In
L. Schulz & A. Gopnik (Eds.), Causal learning: Psychology, philosophy, and computation (pp.
154–172). Oxford/New York: Oxford University Press.
Lagnado, D. A., Gerstenberg, T., & Zultan, R. I. (2013). Causal responsibility and counterfactuals.
Cognitive Science, 37(6), 1036–1073.
Lewis, D. (1973). Counterfactuals. Malden: Blackwell.
Lombrozo, T., & Vasilyeva, N. (2017). Causal explanation. In M. Waldmann (Ed.), Oxford
handbook of causal reasoning (pp. 415–432). New York: Oxford University Press.
Machamer, P., Darden, L., & Craver, C. F. (2000). Thinking about mechanisms. Philosophy of
Science, 67, 1–25.
Meder, B., Mayrhofer, R., & Waldmann, M. R. (2014). Structure induction in diagnostic causal
reasoning. Psychological Review, 121(3), 277.
Menzies, P. (2017). Counterfactual theories of causation. In The Stanford encyclopedia of
philosophy (Winter 2017 Edition). Retrieved from https://plato.stanford.edu/archives/win2017/
entries/causation-counterfactual/
Michotte, A. E. (1946). The perception of causality. New York: Basic Books.
Nozick, R. (1993). The nature of rationality. Princeton: Princeton University Press.
Pearl, J. (2000). Causality. Cambridge, MA: Cambridge University Press.
Proctor, R. N. (2012). The history of the discovery of the cigarette–lung cancer link: Evidentiary
traditions, corporate denial, global toll. Tobacco Control, 21(2), 87–91.
Rottman, B. M., & Hastie, R. (2014). Reasoning about causal relationships: Inferences on causal
networks. Psychological Bulletin, 140(1), 109–139.
Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: An illusion of
explanatory depth. Cognitive Science, 26(5), 521–562.
Ruggeri, A., & Lombrozo, T. (2015). Children adapt their questions to achieve efficient search.
Cognition, 143, 203–216.
Sloman, S. (2005). Causal models: How people think about the world and its alternatives. New
York: Oxford University Press.
Sloman, S., & Fernbach, P. (2017). The knowledge illusion: The myth of individual knowledge and
the power of collective wisdom. New York: Penguin Random House.
Sloman, S. A., & Hagmayer, Y. (2006). The causal psycho-logic of choice. Trends in Cognitive
Sciences, 10(9), 407–412.
Spirtes, P., Glymour, C. N., Scheines, R., Heckerman, D., Meek, C., Cooper, G., & Richardson, T.
(2000). Causation, prediction, and search. Cambridge, MA: MIT Press.
Stephan, S., & Waldmann, M. R. (2018). Preemption in singular causation judgments: A
computational model. Topics in Cognitive Science, 10, 242–257.
Tenenbaum, J. B., Kemp, C., Griffiths, T. L., & Goodman, N. D. (2011). How to grow a mind:
Statistics, structure, and abstraction. Science, 331(6022), 1279–1285.
Waldmann, M. R. (1996). Knowledge-based causal induction. Psychology of Learning and
Motivation, 34, 47–88.
Waldmann, M. R. (2000). Competition among causes but not effects in predictive and diagnostic
learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 53–76.
Waldmann, M. R. (2017). The Oxford handbook of causal reasoning. New York: Oxford University
Press.
Waldmann, M. R., Cheng, P. W., Hagmayer, Y., & Blaisdell, A. P. (2008). Causal learning in rats
and humans: A minimal rational model. In N. Chater & M. Oaksford (Eds.), The probabilistic
mind. Prospects for Bayesian cognitive science (pp. 453–484). Oxford: Oxford University
Press.
Walsh, C. R., & Sloman, S. A. (2011). The meaning of cause and prevent: The role of causal
mechanism. Mind & Language, 26(1), 21–52.
Weiner, B. (1985). An attributional theory of achievement motivation and emotion. Psychological
Review, 92(4), 548–573.
Wolff, P. (2007). Representing causation. Journal of Experimental Psychology: General, 136, 82–
111.
Part III
Meaning Components of Causation
Chapter 5
Event Causation and Force Dynamics in
Argument Structure Constructions
William Croft and Meagan Vigus
Abstract This paper extends Croft’s theory of argument realization to include

nominals that denote events, instead of participants. Events are represented as
force dynamic interactions between participants and their subevents. That is, each
participant is associated with its own subevent; participants are related to each other
through force dynamic interactions as part of a causal chain (Croft 2012). There
is extensive cross-linguistic evidence that a participant’s place in the causal chain
determines its argument realization. In this paper, we tested three hypotheses about
the argument realization of event nominals against English sentences from VerbNet.
First, we define an event nominal as any nominal that refers to an event, regardless
of whether it is morpho-syntactically derived from a verb. Our three hypotheses are
(i) event nominals correspond to participant subevents, (ii) event nominals follow
the same argument realization rules as their associated participant, and (iii) when
both a participant and its subevent are realized as arguments, the subevent (i.e., the
event nominal) is construed as subsequent to its participant in the causal chain. We
find support for these hypotheses in the VerbNet data. In addition, we discuss more
complex cases of argument realization with event nominals.
Keywords Argument realization · Events · Event nominals · Force dynamics ·

Subevents · Participants
W. Croft () · M. Vigus

University of New Mexico, Albuquerque, NM, USA
e-mail: wcroft@unm.edu; mvigus@unm.edu

https://doi.org/10.1007/978-3-030-34308-8_5
152 W. Croft and M. Vigus
5.1 Introduction
There is a growing interest in the semantics of causation to account for mor-

phosyntactic patterns in argument realization (Levin & Rappaport Hovav 2005)
or argument structure constructions (Croft 1991, 2012). Much of this work has its
origins in Talmy’s model of force dynamics, which represents causation in terms of
force transmission from one participant in an event to another participant (Talmy
1976, 1988). Talmy generalizes the notion of transmission of force beyond physical
causation; the first author has further generalized the notion of transmission of force
beyond Talmy’s proposals. This extended model of force dynamics is presented
briefly in Sect. 5.2 of this paper. In Sect. 5.3, we briefly survey cross-linguistic,
diachronic and developmental data that support the hypothesis that a force dynamic
model of causation underlies event structure representation for linguistic expression.
However, causation is typically represented in philosophy and formal semantics
in terms of event causation: not as one participant transmitting force to another
participant, but as one event causing another event. These two ways of representing
causation seem difficult to reconcile, yet both are relevant for accounting for
language structure. Force dynamics appears to motivate cross-linguistic patterns of
argument realization, as in the realization of the car and the statue as subject and
object respectively in (1). Event causation, on the other hand, models causal connec-
tives, and causal interpretations of coordination, in complex sentence constructions,
as in (2).
(1) The car [SBJ] knocked over the statue [OBJ].
(2) a. The car hit the statue and it fell over.
b. The statue fell over because the car ran into it.
The first author developed an event structure representation that reconciles these
two ways of representing causation, in the course of the integrating force dynamic
and aspectual semantic structure of events (Croft 2012). The basic idea is that each
participant has its own subevent. The participant-specific subevents describe the
changes undergone by the participant over the time course of the event; this includes
the aspectual structure of the subevent. Force dynamic relations (e.g., transmission
of force) between participants are therefore equivalent to event causation between
the participants’ subevents. This model of event structure representation is briefly
described in Sect. 5.4.
Croft (2012) applies this model mostly to physical events, although there is also
some analysis of mental events and transfer of possession. In current work we
are extending the analysis more generally to mental and social events, basing the
survey of mental and social event types on VerbNet (Kipper-Schuler 2005; Palmer
et al. 2017), an online resource of verbal semantic classes that starts from Levin’s
(1993) classification of (mostly) physical events but extends it to a broader range
of verbal semantic classes including many mental and social events. Levin (1993)
is restricted to the analysis of argument structure alternations among verbs that
do not take infinitival or finite complement clauses. For the most part, the verbal
5 Event Causation and Force Dynamics in Argument Structure Constructions 153
semantic classes in Levin (1993) do not take event nominals (action nominals) as
arguments. The extension of VerbNet beyond Levin (1993) introduces many verbal
semantic classes that have arguments that themselves denote events: event nominals
(including gerunds) and infinitival and other complement types. We hypothesize that
event nominals and complement types that denote events represent the participant-
specific subevents posited in the event structure representation proposed in Croft
(2012). In Sect. 5.5, we test this hypothesis against the example sentences in
VerbNet that contain event nominals or complements. Although the hypothesis
largely holds, we suggest some refinements and qualifications to the hypothesis that
are necessary to account for the realization of subevent arguments in English.
5.2 The Force Dynamic Analysis and Its Extensions
The force dynamic analysis of argument realization has been influential in cognitive
semantics, beginning with Talmy (1976, 1988) and followed by DeLancey (1985)
and Langacker (1987) as well as by the first author (Croft 1991, 1993, 1994,
1998a,b, 2012). The causal approach is characterized by conceptualizing events as
a CAUSAL CHAIN linking participants in the event in terms of the transmission of
force from one participant to another. In addition to providing a model of event
lexicalization—predicates lexicalize segments of the causal chain—it also provides
a model of argument realization, articulated in greatest detail in Croft (1991, 2012).
The event structure that is proposed in the causal theory of argument realization
is a linear causal chain defined by the transmission of force from one participant to
another. Example (3) gives the causal event structure for the event expressed by the
sentence (the argument realizations below the causal chain will be explained below).
(3) Sue broke the coconut for Greg with a hammer.
The event structure is the causal or force dynamic chain. Causation is defined
in broad terms, to include a variety of causal relations. These involve not only
physical causation, but also an intentional being either initiating an action—what
Talmy (1976) calls volitional causation—or having one’s mental state altered as
the result of an action—what Talmy (1976) calls affective causation. Talmy (1976)
also introduces inducive causation, where one volitional agent interacts with another
volitional agent to induce the latter into engaging in an activity. Talmy (1976)
thus extends physical transmission of force relations to include the volitional
involvement of human agents as well as the physical interactions of physical objects.
Talmy’s (1988) force dynamic model further extends the analysis of one partici-
pant acting on another participant to include not only the prototypical “billiard-ball”
causation, where one participant applies force to a second participant which then
undergoes a change, as in I kicked the ball, but also “letting” (enabling) causation,
as in I dropped the ball, and forcefully maintaining a static situation (as in I was
holding the ball). Talmy also extends these generalized force dynamic relations to
social (interpersonal) events where one person allows or prevents another person’s
action.
In Croft (1991, 2012), the first author uses Talmy’s generalized force dynamic
model to account for argument realization patterns, using a model that avoids appeal
to participant roles. Participant roles defined in absolute terms appear to play little
role in argument realization, as many previous authors have noted (e.g., Dowty
1991). Instead, the ranking of participant roles is far more significant. In many
theories, the ranking of participant roles is independent of event structure: properties
of event structure are not used to define the ranking, and participant roles from
different kinds of events are lumped together in a single thematic role hierarchy. In
the causal theory, the ranking is defined solely within an event, and is defined as
the relative causal ordering of the participants in the event, that is, the causal chain.
In particular, the subject referent is antecedent to the object referent in the causal
chain.
The relative position of participants in the causal chain accounts for the high
degree of regularity in the mapping of the participants in transitive events to
subject and object roles across languages. Where there is variation across languages
(and within languages) in the choice of subject vs. object, it can be attributed to
indeterminacy in the ordering of participants in a causal chain. A clause construes an
event as a single linear, asymmetric causal chain, but not all events are of this type.
For example, in predicates involving mental states such as see, know and like, there
is substantial cross-linguistic variation in whether the experiencer or the stimulus
is coded as subject or object (Croft 1993, 2012: 233–36). This is illustrated by the
well-known English argument realization patterns in (4)–(6):
(4) I like Beethoven’s Seventh Symphony.

(5) I am enjoying Beethoven’s Seventh Symphony.
(6) Beethoven’s Seventh Symphony pleases me.
In mental events such as those expressed in (4)–(6), the force dynamics is

bidirectional: the experiencer attends to the stimulus, and the stimulus causes a
certain mental state in the experiencer. Hence the variability.
In fact, the variability is limited to examples like (4). Causative predicates that
focus on the change of mental state always have the stimulus as subject, as in (6),
because they describe the transmission of force from stimulus to experiencer. In
English, the only evidence of the causative construal is the argument realization,
however there are languages, such as Lakhota and Classical Nahuatl, that use
causative morphology on the verb when the stimulus is realized as subject (Croft
1993, 56). Activity verbs that describe how the experiencer is attending to the
stimulus always have an experiencer subject as in (5), because they describe
the transmission of force (in Talmy’s broad sense of “force”) from experiencer
to stimulus. The compatibility of the progressive aspect in examples like (5)
demonstrates that these don’t construe the mental event as a state, but as a process.
Although based in a rather different theory, Pesetsky (1995) comes to a similar
conclusion: when the experiencer is realized as object, the stimulus corresponds
to a Causer; when the experiencer is realized as subject, the stimulus corresponds to
the Target or Subject Matter of Emotion.
As mentioned above, it is examples like (4) that are expressed variably across
languages (Croft 1993); this is because these examples denote a state, and hence
there is no transmission of force. Examples (7)–(9) below demonstrate this cross-
linguistic variability with stative mental events.
(7) Kannada (Sridhar 1976, 583)

nanage ı̄ vicara gottu
I:DAT this fact know
‘I [SBJ] know this fact.’ (cf. ‘This fact is known to me.’)
(8) Japanese (Croft 1993, 67–8, from Kuno 1973, 79–95)
Dare ga eiga ga suki desu ka
who NOM movie NOM fond.of is INT
‘Who likes movies?’
(9) Eastern Pomo (Croft 1993, 67, from McLendon 1978, 3)
bé:kal wí ph i:lémka
3PL.PAT 1SG.PAT miss
‘I miss them.’
In English, as in (4), the experiencer is realized as subject and the stimulus as object.
Stative mental events in Kannada, shown in (7), realize the stimulus as subject;
the experiencer receives Dative marking. In Japanese, stative mental events have
both the experiencer and the stimulus as subject, marked with the Nominative ga
(when not replaced by Topic wa); this is illustrated in (8). Finally, Eastern Pomo
realizes both the experiencer and stimulus in stative mental events as objects, shown
in example (9).
Role designation is not stipulated in the causal model, but is part of the semantic
structure of the event. The solid arrows in example (3) represent the segment of the
causal chain that is denoted, or profiled, by the predicate in the clause (in this case,
break). We use the term ‘profile’ here basically as it is used in Cognitive Grammar
(Langacker 1987): it represents the concept denoted by a word against its semantic
frame (Fillmore 1982, 1985), in this case the entire causal chain in example (3).
Differences in verbal profile result in differences in argument realization. For

example, in the classic Locative alternation, different segments of the causal chain
are profiled, as in (10a) and (10b):
(10) a. Jack loaded the furniture on the truck.
b. Jack loaded the truck with furniture.
Examples (10a) and (10b) illustrate two further properties of the force dynamic
model of event structure representation. The first property is the construal of
noncausal relations as a causal chain. The caused location event represented in
examples (10a) and (10b) involves a noncausal relation: while the agent causes the
change in spatial configuration, the spatial relation of figure (the furniture) to ground
(the truck) is not causal. (Hence the link between figure and ground in the causal
chain lacks the arrowhead indicating directionality of causation.) In other words,
the force dynamic model is extended even further, to noncausal relations between
events.
Technically, noncausal relations are force-dynamically neutral: either participant
may be realized as antecedent to the other, as we observed cross-linguistically for
mental states above. However, certain (though not all) noncausal relations that recur
frequently in event structure are consistently conceptualized or construed with one
particular participant antecedent to the other. For example, it turns out that cross-
linguistically, the figure is construed as antecedent to the ground in a “causal” chain.
This noncausal yet directed relation is represented by a line without an arrowhead
in examples (10a) and (10b). Although the locative alternation differs in realization
of the figure and ground participants, the ordering of participants in the causal chain
remains the same, as can be seen in (10a) and (10b). That is, the figure (the furniture)
is construed as antecedent to the ground (the truck), regardless of which participant
is realized as object.
The second property is the differentiation of oblique case marking (adposition
or case affix) into two types, antecedent (labeled A.OBL in 10b) and subsequent
(labeled S.OBL in 10a). An antecedent oblique encodes a participant antecedent to
the participant realized as object in the causal chain; a subsequent oblique encodes
a participant subsequent to the participant realized as object in the causal chain.
Whether a participant is antecedent or subsequent depends of course on which
participant in the causal chain is encoded as object, which in turn depends on
which segments of the causal chain are profiled. This is how different argument
realizations, as in (10a) and (10b), may encode the same causal ordering of
participants: when the figure is realized as object, the ground is expressed with a
subsequent oblique, but when the ground is realized as object, the figure is expressed
with an antecedent oblique.
The division of obliques into antecedent and subsequent is a consequence of
the causal theory: it is simply a fact that participants realized as obliques occur
at different positions in the causal chain, relative to the participant realized as the
object. This consequence makes a prediction, namely, that there is a relatively sharp
linguistic division between antecedent and subsequent obliques. The next section
summarizes cross-linguistic, diachronic, and developmental evidence that supports
this prediction.
5.3 Evidence for the Antecedent-Subsequent Oblique

Distinction
The examples given in Sect. 5.2 to illustrate the force dynamic model of event
structure representation also illustrate the mapping rules that govern the encoding of
participants in syntactic roles. The mapping rules that accompany the causal event
representation are small in number and simple in formulation (Croft 1998b, 24,
2012, 207).
(i) Subject and object delimit the verbal profile
(ii) Subject is antecedent to object in the causal chain (SBJ → OBJ)
(iii) Antecedent oblique is antecedent to the object in the causal chain; subsequent
oblique is subsequent to the object in the causal chain (A.OBL → OBJ →
S.OBL)
The argument realization rules in (i)–(iii) are a more precise formulation of what
was called the Causal Order Hypothesis in Croft (1991, 186). The Causal Order
Hypothesis can be better described as a hypothesis about the structure of the causal
chain implicit in the realization rules. The hypothesis is presented in (11) (see also
Croft 1990, 53, 1991, 269, 1994, 91):
(11) Causal Order Hypothesis: a simple verb in an argument structure construc-

tion construes the relationships among participants in the event it denotes as
forming a directed, acyclic, and nonbranching causal chain.
The causal asymmetry between subject and object is overwhelmingly confirmed,

when causal relations are generalized to force dynamic relations, as described in
Sect. 5.2. This is what underlies most definitions of argument realization based on a
thematic role hierarchy or on proto-roles (Dowty 1991).
The realization rules in (i)–(iii) distinguish two classes of oblique syntactic
arguments, defined by their position relative to the position of the object in the
causal chain. Antecedent obliques are antecedent to the object in the causal chain;
antecedent obliques may or may not also be antecedent to the subject in the causal
chain. The instrumental phrase with a hammer in (3) is an example of an antecedent

oblique. Subsequent obliques are subsequent to the object in the causal chain. The
beneficiary phrase for Greg in (3) is an example of a subsequent oblique.
Evidence for the causal asymmetry between antecedent and subsequent obliques
is also quite robust, and is summarized here (see Croft 1991, ch.5, and 2012, ch.6
for a more detailed presentation of the evidence).
Traditional thematic roles can be translated into positions in the causal chain,
at least for canonical argument realization. English antecedent oblique expressions
with some examples of typical semantic roles are illustrated in example (12), and
English subsequent oblique expressions are illustrated in example (13):
(12) with (instrument, comitative, etc.):

a. Sue broke the coconut with a hammer.
b. I went to the park with Carol.
by (means, passive agent, etc.):
a. I went downtown by bus.
b. The cat food was eaten by raccoons.
of, metaphorical from/out of (cause):
a. The rabbit died from/of thirst.
b. He did it out of spite.
(13) for (beneficiary):

a. Sue broke the coconut for Greg.
metaphorical to, into (recipient, result):
a. Sally gave a copy of her slides to Jalon.
b. They smashed the statue to pieces.
c. The boy carved the stick into a knife.
Although one cannot predict which participant roles a specific oblique case
marking will express–case markers are usually quite polysemous–one can predict
that a specific oblique case marking will express only antecedent roles or only
subsequent roles. That is, one can generally categorize oblique morphosyntactic
markers as either antecedent or subsequent, as in (12) and (13). A cross-linguistic
study of oblique adpositions and case markers in a 40 language sample broadly
confirms this hypothesis (Croft 1991, 187–88); see Table 5.1. The examples of no
directionality in Table 5.1 are languages in which there is one highly general oblique
adposition or case marker that does not differentiate antecedent and subsequent
semantic roles, or it appears that a more highly differentiated oblique system is
breaking down and will end up as an undifferentiated system.
There is some variation in the argument realization of certain participants across
languages; however, they conform to the Causal Order Hypothesis. For example,
contact events vary as to whether the locus of contact or the “instrument” of contact
is realized as object. If the locus is realized as object, as in English, then the
Table 5.1 Syncretisms Syncretisms among antecedent thematic roles 39

among semantic roles in
Syncretisms among subsequent thematic roles 39
oblique adpositions and case
markers No directionality in the case marking system 5
Syncretisms across subsequent and antecedent roles 2
“instrument” is realized as an antecedent oblique, with with; see example (14). If

the “instrument” is realized as object, as in Chechen-Ingush (example from Nichols
1984, 188; see Croft 1991, 190), then the locus of contact is realized as a subsequent
oblique (dative); see example (15).
(14) a. Father beats his son with a stick.
b. *Father beats a stick on/to his son.
(15) da:s woPa: Gam j-iett

father:ERG son:DAT stick beats
‘(The) father beats (his) son with a stick.’
In examples (12) and (13), we noted that metaphorical uses of spatial directional
path markers, such as ablative (source) from and out of and allative (goal) to and
into, function as subsequent and antecedent obliques respectively. This is the result
of a general metaphorical mapping of spatial directional path meaning into the
direction of transmission of force (Croft 1991, 192–98):
(16) Space → Causation metaphor:

Causation: antecedent role object subsequent role
↑ ↑ ↑
Space: ablative/source locative allative/goal
The metaphorical mapping between spatial direction and direction of transmis-

sion of force is well attested in the same 40 language sample (Croft 1991, 196); see
Table 5.2.1
In examples (10a) and (10b) in Sect. 5.2, we observed that when two participants
in an event are in a spatial figure-ground relation, cross-linguistically the figure is
1 The exceptions all involve the use of the allative for manner. This is likely due to the fact
that different types of stative secondary predicates, including manner and resultative, share
constructions (Verkerk 2009a,b); manner is antecedent since it is a property of the event, and
resultative is subsequent since it expresses a resulting event, and typically takes subsequent case
marking such as the case marking for allative.
Table 5.2 Syncretisms Syncretisms between ablative and antecedent roles 13

among spatial and causal
Syncretisms between locative and object marking 1
semantic roles in oblique
adpositions and case markers Syncretisms between allative and subsequent roles 15
Exceptional syncretisms 3
almost always construed as antecedent to the ground. Croft (1991, 200–1) gives
examples from Modern Irish, German, Russian and Hungarian. If the causal relation
is opposite to the figure-ground relation, as in These beams support the roof or The
bowl contains fruit, then the causal relation determines the argument realization, not
surprisingly.
A similar construal of possessum as antecedent to possessor is also crosslinguis-
tically widespread (Croft 1991, 207):
(17) a. The dean presented an award to the valedictorian.
b. The dean presented the valedictorian with an award.
Again, when the possessor (recipient) is realized as an oblique, a subsequent

oblique is used (in English, to); when the possessum is realized as an oblique, an
antecedent oblique is used (in English, with), in conformity with the construal of
the possessum as antecedent to the possessor. The possessum-first construal applies
when possession is lost as well as gained:
(18) a. The mayor stole the land from the peasants.
b. The mayor robbed the peasants of their land.
Transfer of possession also uses a double-object (ditransitive) construction, in

which case neither possessum nor possessor is construed as antecedent to the other.
The antecedent-subsequent semantic role distinction manifests itself in other
grammatical domains in which event structure plays a role. In adnominal modifica-
tion, the figure-first and possessum-first construals also dictate choice of preposition
for the modifying noun. In the lid for the jar, the spatial ground dependent (the
jar) is subsequent to the spatial figure head (the lid) and so takes the subsequent
for; but in the jar with a lid, the figure dependent (the lid) is antecedent to the
ground head noun and so takes the antecedent with. Likewise, in the food for the
cats, the possessor dependent (the cats) is subsequent to the possessum head (the
food) and so takes the subsequent for; but in the man with a knife, the possessum
dependent (the knife) is antecedent to the possessor head (the man) and so takes the
antecedent with (Croft 1991, 228–231). In nominalization, there is a widely attested
syncretism of agent and instrument nominalizations (e.g., English writ-er [agent]
and stapl-er [instrument]), as well as syncretism of agent, instrument, and location
nominalizations, based on the metaphorical mapping from locative to the verbal
profile referred to above (Croft 1991, 231).
In addition to the cross-linguistic evidence supporting the antecedent vs.
subsequent force dynamic distinction, there is developmental evidence indicating
the psychological reality of the antecedent-subsequent distinction, at least in English
(Croft 1998b, 40). Children use an inappropriate antecedent preposition for another
antecedent function but not for a subsequent function, and vice versa (Bowerman
1983, 463–65; Bowerman 1989; Clark & Carpenter 1989, 19, Table 10). Even more
striking, when overgeneralizing argument structure alternations, children choose an
appropriate antecedent or subsequent preposition (Bowerman 1982, 338–39):
(19) ‘I don’t want it because I spilled it [toast] of orange juice.’ [E 4;11]
In (19), the child makes the ground the object of spill, which is unacceptable in
adult English. Needing to realize the figure argument as an oblique, she chooses an
antecedent preposition of (as in strip the trees of bark), again in conformity with
the figure-first construal. This indicates that the child has figured out the universal
antecedent oblique-subsequent oblique distinction, but has not yet figured out
which English-specific antecedent oblique preposition goes with which antecedent
participant role—an at least partly idiosyncratic fact of English—nor has the child
yet figured out which predicates can realize only the figure as object or only the
ground as object.
Studies of the grammaticalization of adpositions and case markings expressing
oblique participant roles (e.g., Lehmann 1982, 1995, 2002) generally find that they
respect the distinction between antecedent and subsequent roles. That is, a form
encoding an antecedent role grammaticalizes into expressing another antecedent
role and the same goes for forms encoding a subsequent role. At the last stage
of grammaticalization, subsequent oblique forms may come to be used for overtly
coded objects (i.e., accusative) and antecedent oblique forms may come to be used
for overtly coded subjects (i.e., ergative) (Lehmann 1982, 1995, 2002, 99).
(20) a. Subsequent grammaticalization paths:
benefactive, directional > dative > accusative
b. Antecedent grammaticalization paths:
comitative > instrumental > ergative
locative > instrumental, ergative
ablative > genitive > ergative
The fact that the antecedent-subsequent semantic distinction is maintained in the

grammaticalization of case markers should not be surprising, since the synchronic
polysemy (syncretisms) presented above is the result of diachronic processes.2
Finally, the recent Leipzig Valency Project documents the grammatical encoding
of verb-specific participant roles (called microroles) for around eighty basic verbs,
selected from across the verbal semantic classes in Levin (1993). The microroles
are all separately coded for their case marking (subject, object, different types of
oblique). Hartmann et al. (2014) plot the microroles in a spatial model generated by
multidimensional scaling (MDS), which positions microroles in the space such that
roles are positioned closer to each other depending essentially on how frequently
the microroles are expressed by the same linguistic form across the languages in the
sample. Their MDS spatial model with our interpretation is presented in Fig. 5.1.
The spatial model of semantic relations between the microroles clearly indicates
two dimensions of contrast. The dimension from lower left to upper right represents
the core argument (subject, object) – peripheral (oblique) argument contrast. The
dimension from upper left to lower right represents the antecedent – subsequent
contrast. For core argument microroles, the microroles typically realized as subject
are antecedent to the microroles typically realized as object in the causal chain.
Oblique microroles are scaled from antecedent to subsequent as described earlier in
this section. (There is a blurring of the core-peripheral contrast in the subsequent
region due to the variation in the expression of recipients, addressees and other
human subsequent microroles as objects or as datives.)
In sum, larger-scale cross-linguistic studies have confirmed the importance of
the antecedent-subsequent oblique distinction in case systems, thereby confirming
the Causal Order Hypothesis. In addition, these larger-scale studies have specified
the structure of the conceptual space of participant roles in finer-grained detail than
is implied simply by the antecedent-subsequent role distinction. Nevertheless, the
boundary between antecedent and subsequent oblique forms is generally adhered
to, although there are subtler patterns of syncretism among antecedent roles and
among subsequent roles; and although there are paths of semantic change outside
the causal domain (in the spatial and intentional domain) by which a case form may
acquire functions across the antecedent-subsequent divide.
2 The one exception to the sharp division between antecedent and subsequent roles is the dative
> ergative pattern. This is the result of the grammaticalization of a possessive construction with
dative possessor construal into a perfective ergative construction (Anderson 1977; Trask 1979;
Lehmann 1982, 1995, 2002, 98; Haig 2008).
Fig. 5.1 MDS spatial model of relations between microroles in the ValPaL database (Valency
Project; http://www.valpal.info, accessed 17 March 2017), with interpretation of the two dimen-
sions of the spatial model. (Cf. Hartmann et al. 2014:470, Fig. 3. Fig. 5.1 differs in that it is derived
from the raw ValPaL data and uses the Optimal Classification unfolding algorithm of Poole 2000)
5.4 The Three-Dimensional Model: Integrating Causal and

Aspectual Structure
In Sect. 5.1, we briefly described the reconciliation of the force dynamic represen-
tation of causal relations with event causation in Croft (2012). Each participant
is associated with its own subevent, and force dynamics represents the causal
relation between the initiating participant’s subevent and the endpoint participant’s
subevent. Here we briefly describe the nature of the subevents before describing
their integration into the representation of causation.
Each subevent represents how an event unfolds over time—that is, the aspectual
structure of the subevent. This definition implictly requires two dimensions. The
Fig. 5.2 The

two-dimensional
representation of aspect
Fig. 5.3 Two alternative

profiles for see
first, of course, is time. The second is what it means to say the subevent “unfolds”.
Unfolding characterizes the states and changes of state of the individual that take
place over the time interval in which the subevent occurs. These are its phases.
A number of linguists have proposed phasal models of how an event unfolds,
that is, a temporal decomposition of the event into discrete phases (e.g., Dowty
1979; Parsons 1990; Binnick 1991; Jackendoff 1991; Breu 1994; Bickel 1997;
Rappaport Hovav & Levin 1998; see also Croft 2009, 149–51, 2012, 45–52). The
model presented here is also a phasal model, but unlike most previous proposals,
it treats the qualitative states as points on a second dimension, and change as
transitions from one state to another on that dimension. Figure 5.2 illustrates the
model for the perceptual event of seeing.
The x axis is the time dimension (t), and the y axis is the qualitative state
dimension (q). The dotted contour in Fig. 5.2 is how the seeing event unfolds.
Seeing has two defined states on q: not seeing something and seeing something.
Seeing something is a transitory state, that is, one starts and stops seeing a particular
object over one’s lifetime. Seeing has at least three phases: not seeing something;
the transition from not seeing something to seeing it, which is construed as an
instantaneous jump from one state to the other and represented by a vertical line;
and seeing that thing. The sequence of phases describes the aspectual contour of the
event.
The English verb see in a particular usage profiles one phase of the event; a
solid line indicates the profiled phase. Hence the aspectual contour functions as the
semantic frame for the profiled phase or phases of the event. The two representations
in Fig. 5.3 describe two different aspectual construals of the predicate see in English.
The left-hand representation in Fig. 5.3 profiles the resulting state, as in I see the
watchtower. In the right-hand representation in Fig. 5.3, there is an alternative,
grammatically acceptable construal of English see where it profiles the transition
from not seeing something to seeing it—an achievement, in Vendler’s sense—as in
Suddenly I saw the mountain lion. Part of the challenge in analyzing aspect is the
great flexibility of predicates in English to occur in different aspectual construals,
without any morphological change in the verb form (Croft 2009, 2012).
The model of aspectual structure presented here is presented in greater detail
in Croft (2012). The two-dimensional t/q diagrams allow us to provide distinct
representations of all of the aspectual types (or construals) that have been discussed
in the aspectual literature, and it also allows us to make sense of the bewildering
variety of aspectual construals. These aspectual types go under different names as
several of them have been discovered independently in different analytical traditions
(generative, formal semantic and cognitive semantic). This is not the primary
concern of this paper; the interested reader is directed to Croft (2009), (2012),
chapters 2–4.
The causal model for argument realization described in Sect. 5.2 is simply to add
the causal chain as a third “dimension” to the two-dimensional aspectual represen-
tation (Croft 2009, 161–64, 2012, ch. 5–6); see Fig. 5.4.3 The third “dimension” in
event structure is the nonbranching, acyclic, directed causal chain (see Sect. 5.3),
though it is really a graph structure rather than a continuous geometric dimension.
The crucial feature of this representation, as noted above, is that each participant
has its own subevent in the causal chain.4 The subevent is the aspectual pro-
file/contour for that participant’s activity in their role in the larger event. Informally,
this can be thought of as what each individual participant does or undergoes during
the course of the event. Each participant’s subevent then stands in a causal relation
to the subevent of the next participant in the causal chain—or a noncausal relation,
e.g. a spatial relation as in the locative alternation described in Sect. 5.2.5 In some
sense, the participant and the subevent are the same thing in the event structure
representation, using the notion of an entity as a history of what that entity is or
does over its lifetime. The subevent is the participant history for that time interval
and that complex event.
Figure 5.5 gives the three-dimensional representation for example (3) in Sect. 5.2
(Croft 2009, 163; 2012, 214). Each participant has its own subevent: Sue applies
force to the hammer, the hammer makes impact with the coconut, the coconut
undergoes an irreversible change of state, and Greg comes to benefit from the
3 Three-dimensional representations are of course difficult to apprehend on a two-dimensional page
or screen; the representation in Fig. 5.4 more or less collapses the causal and qualitative state
dimensions onto the vertical dimension (Croft 2009, 161–62, 2012, 212–213). The advantage
of this way of reducing the three-dimensional representation onto two dimensions is that the
temporal alignment of the subevents is clearly indicated. The qualitative state scales for each
participant/subevent are kept separate, in order to remind the viewer that they actually belong
on a third dimension.
4 Although this is similar to Rappaport Hovav & Levin’s (1998) Argument Realization Condition
that requires each argument to be associated with a subevent in the event structure, Rappaport
Hovav & Levin’s requirement applies to the syntactic realization of arguments, whereas Croft
(2012) stipulates that each semantic participant, regardless of argument realization, has an
associated subevent.
5 This separation of aspectual structure from causal structure is reminiscent of Jackendoff’s (1990)
separation of a thematic tier from an action tier. While Jackendoff’s action tier corresponds
straightforwardly to causal relations, the thematic tier corresponds more to the qualitative state
dimension, as opposed to aspectual structure. Jackendoff (1990) also distinguishes participants on
both of these tiers in terms of Roles, in contrast to the three-dimensional representation discussed
here.
Fig. 5.4 Three-dimensional

representation modified for
display
Fig. 5.5 Causal-aspectual

structure of example (1)
outcome. All of the subevent profiles must be aligned temporally; the entire event
is punctual. There is no longer any problem with defining the endpoint of the verbal
profile: the coconut is involved in only one subevent.
Figure 5.6 illustrates how a durative event is represented, namely Jane read “War
and Peace”. The agent is engaged in an undirected activity, namely scanning the
text, while the text is undergoing an incremental change (as a representational source
theme, in Dowty’s terms). The transmission of force takes place for the profiled
temporal phase of the event, but for convenience it is only represented by the causal
arrows at the beginning and the end of the profiled phase.
5.5 Argument Realization and Event Nominals as Arguments
All of the examples in Sects. 5.2, 5.3 and 5.4 involve physical events. Many basic
verbs, in particular the verbs in the verb classes in Levin (1993), describe physical
Fig. 5.6 Representation of a

durative event
events: the participants expressed as arguments are physical objects or persons,

and the events represent physical (or agentive but ultimately physical) interactions
between the participants. When we move beyond physical events to mental events
(perception, cognition, emotion, and desire/intention) and social events (involving
interpersonal interactions), we find that events (states of affairs) and propositions
function as semantic arguments of the main predicate. Events/propositions as
arguments are realized in English and other languages as event nominals (also called
action nominals) and as complements, including infinitival complements.
Event arguments are not generally discussed in analyses of argument realization.6
However, the model of event structure presented in Sect. 5.4 offers a natural
extension of argument realization rules to event arguments. We propose three
hypotheses regarding the argument realization of event nominals:
(I) Event nominals express participant subevents.
(II) Event nominals follow the same argument realization rules as ordinary
nominals, in terms of realization as subject, object, antecedent oblique or
subsequent oblique.
(III) If both a participant and its subevent are realized as distinct arguments of a
predicate, then the subevent, expressed as an event nominal, is construed as
subsequent to its participant.
Our basic hypothesis (I) is that event arguments correspond to participant
subevents. The second hypothesis (II) is that event nominals and complements,
to the extent that the latter allow oblique adpositions or case marking, should
behave with respect to the argument realization rules just like the participants whose
6 Grimshaw (1990) does discuss event nominals in terms of argument realization, arguing for a
distinction between process nominals that have argument structure and result nominals that do not.
We have found, however, that the distinction between process and result nominals does not appear
to be relevant to the realization of participants and their subevents with antecedent or subsequent
obliques.
subevents they correspond to. The second hypothesis relies on the veracity of the
first: event nominals must correspond to participant subevents in order to be able
to ascertain whether they follow the same argument realization rules as partici-
pants. The converse, however, is not true: event nominals may express participant
subevents, but not follow the same argument realization rules. Our final hypothesis
(III) is that a subevent is construed as subsequent to its associated participant.
The first and second hypotheses fall out from the model discussed in Sect. 5.4:
if force dynamic relations exist not between participants, but between partici-
pant/subevent pairs, it follows that either a participant or its subevent may be
expressed in a particular context.7 Furthermore, the participant and its subevent
should follow the same argument realization rules. That is, whether the participant or
the subevent is directly expressed in a sentence, it refers to both the participant and
its subevent; therefore, the same argument realization rules are expected to apply.
In this section, we test these hypotheses against verbs in the verb classes in
VerbNet (Kipper-Schuler 2005; Palmer et al. 2017), an online resource of verbal
semantic classes and the argument structure constructions that realize them, that take
event arguments. We were able to analyze 192 example sentences from VerbNet that
include event arguments.
We focus here primarily on event nominals. Event nominals are defined in the
typological literature as forms derived from verbs that denote events and allow for
a broad, if not full, range of case marking (case inflections or adpositions) (Comrie
1976; Koptjevskaja-Tamm 1993). This definition of event nominals includes English
gerunds, which take a range of prepositions. In this paper, we take a morphologically
broader view of event nominals: any nominal referring to an event, regardless of
whether it is derived from a verb (e.g., incident). The identification of a nominal
as an event nominal is not based solely on the lexical item, but the context as
well. Example (21) below illustrates how the same lexical item may or may not
be interpreted as an event nominal.
(21) a. One student spilled coffee on their exam.

b. It took the students 3 hours to finish the exam.
In (21a), the context makes it clear that exam refers to a physical object; therefore
we would not consider this an event nominal. In (21b), exam is described by its
duration and therefore clearly refers to an event; we would consider this an event
nominal.
The following subsections explain and illustrate the three hypotheses, and later
subsections discuss more difficult cases.
7 The present paper does not put forth any generalizations about when, or under which circum-
stances, an event nominal corresponding to a participant’s subevent may be expressed instead of
a nominal referring to the participant. As mentioned above, it appears that event nominals tend to
occur more often as arguments of mental or social predicates, however an in-depth study would be
necessary in order to propose more solid generalizations.
5.5.1 Event Nominals as Participant Subevents
The first hypothesis is that event nominals express the subevent of a participant in
the clause. As described above, each participant is associated with a subevent that
represents the qualitative phases of that participant during the event in the main
clause. Examples (22) and (23) below illustrate this hypothesis. Both the event
nominal and its associated participant are in bold.
(22) a. The clown amused the children.

b. The clown’s antics amused the children. (VerbNet)
(23) a. The President shocked the Democrats.
b. The President’s tweets shocked the Democrats.
c. The tweets shocked the Democrats.
Examples (22) and (23) demonstrate how event nominals, like antics or tweets, are
used to refer to the subevent associated with a participant in the sentence. That is,
antics refers to the clown’s subevent and tweets refers to the President’s subevent, as
can be seen in the representation. In cases like (22) or (23), the construction allows
for either the participant or its subevent to be expressed as an argument, without
a drastic change in meaning. Whether the participant or the subevent is expressed,
essentially, they both refer to the combination of participant and subevent, i.e. the
bottom portion of the causal-aspectual representation.
There are examples in VerbNet in which event nominals are not the subevent of
an expressed participant in the clause.
(24) The enemy soldiers submitted to demands. (VerbNet)
In (24), the initiator of the event nominal demands is not expressed in the clause.
However, it is likely that the identity of the demanders would be present in the
discourse context. This would be an example of null anaphora, or Definite Null
Instantiation, following the theory of null instantiation in construction grammar
(Fillmore 1986; Lambrecht & Lemoine 2005; Lyngfelt 2012). The null-instantiated,
i.e. unexpressed, participant is definite, or known, in the context. The same is
probably true of example (25).
(25) I interrogated him about the incident. (VerbNet)

There are other examples in VerbNet where it is not clear that the participant
associated with the subevent would be known in the discourse context. These can
be seen below in examples (26)–(28).
(26) I learned about the drinking. (VerbNet)

(27) They tolerate smoking. (VerbNet)
(28) Success requires hard work. (VerbNet)
In example (26), the participant associated with the drinking subevent, the
drinker(s), may or may not be known in the discourse context. That is, (26) may be
uttered in either of the two contexts shown below:
(29) How is John doing? I learned about the(/his) drinking.

(30) I found someone’s empty bottles in the break room; that’s how I learned about
the drinking.
In (29), the drinker is mentioned in the discourse context and therefore the
drinking is easily replaced with a definite pronoun, his drinking. In (30), the
participant associated with drinking corresponds to the indefinite pronoun someone
and therefore is not known in the discourse context. This represents what is called
Free Null Instantiation (FNI) of the identity of the drinkers; the identity of the
null-instantiated participant may or may not be available in the discourse context.8
Finally, examples (27) and (28) are more general statements and they represent
examples of what Lyngfelt calls Generic Null Instantiation (GNI). That is, the null-
instantiated participant corresponds to a generic “people”.
There are certain types of events which tend to have event nominals as argu-
ments with null-instantiated participants, such as communication events, shown in
examples (31)–(33).
(31) John discussed his own presentation.

(32) John discussed Bill’s presentation.
(33) John discussed the presentation.
In communication events, the topic or subject matter is often expressed by an event

nominal, such as presentation. In some cases, the participant whose subevent is
expressed by the topic event nominal may also be the speaker in the communication
event, as in (31). However, this is not necessarily the case: in (32), the event nominal
presentation corresponds to Bill’s subevent; Bill may or may not be a participant in
the communication event. Often, the participant whose subevent is expressed by the
event nominal is null instantiated, as in (33); this appears to be a case of FNI.
8 This example also shows Indefinite Null Instantiation (INI) of what is drunk, conventionally
interpreted as an alcoholic drink.
5.5.2 Event Nominals and the Argument Realization Rules
The second hypothesis predicts that subevents follow the same argument realization
rules as their participants. That is, subevents are ordered in the causal chain along
with their participants. Other participants (and subevents) in the causal chain are
ordered with respect to subevents in the same way as they are with participants.
This can be seen in examples (34) and (35) below.
(34) John confronted it with emergency measures. (VerbNet)
(35) Russia subjugated Mongolia with overwhelming force. (VerbNet)
In (34) and (35), the initiators of the causal chains, John and Russia, are realized as
subject and the endpoints of the causal chain, it and Mongolia, as object. As can be
seen in the representation in (35), the event nominal represents a subevent associated
with the initiator of the causal chain. Therefore, the event nominals, emergency
measures and overwhelming force, are expressed by the antecedent oblique (with).
Since the event nominal represents the subevent of the initiator, it follows the
argument realization rules in that it is realized as antecedent to the object in the
causal chain.
In examples (36) and (37) below, the event nominal expresses the subevent
associated with the endpoint of the causal chain (as opposed to the initiator in 34
and 35).
(36) I needed his cooking. (VerbNet)
(37) I saw their laughing and joking. (VerbNet)
In examples (36) and (37), the initiators of the causal chains, both I, are realized
as subject and the endpoints, their and his, as possessors of the event argument. The
event arguments, laughing and joking and cooking, are realized as object. Since the
event argument is associated with the participant at the endpoint of the causal chain,
it is realized as object, subsequent to the initiator of the causal chain realized as
subject.
In 192 VerbNet examples, there is no exception to the generalization that event
arguments follow the same argument realization rules as ordinary nominals. Event
arguments correspond to participant subevents, and are construed to occur at the
same position in the causal chain as the participant whose subevent they express,
relative to other participants/subevents. That is, subevents associated with the
initiator of the causal chain will be construed as antecedent to the endpoint of
the causal chain (and its subevents). Subevents associated with the endpoint of the
causal chain will be construed as subsequent to the subject (and its subevents). If
both a participant and its subevent are expressed in a single clause, then the subevent
will be construed as subsequent to the participant.
This generalization does not entail that arguments denoting participants and
arguments denoting their subevents are interchangeable, however. In fact, our
impression is that this is generally not the case. Understanding the circumstances
under which an argument denoting a subevent or an argument denoting a participant
are grammatically appropriate is a challenging task that is unfortunately beyond the
scope of this paper (see fn. 7).
5.5.3 Participants and their Subevents Both Expressed as

Arguments
In many VerbNet examples, both the participant and the participant’s subevent are
realized as arguments of the main predicate, as in (38)–(40) below.
(38) He managed the climb. (VerbNet)
(39) I tried exercising. (VerbNet)
(40) I forced him into coming. (VerbNet)
Since both the participant and the participant’s subevent are arguments, one can
ask if there is a regular construal of the two arguments with respect to argument
realization. The third hypothesis predicts that event arguments are realized as
subsequent to the participant whose subevent they express. That is, a participant’s
subevent is construed as subsequent to the participant itself.
In examples (38) and (39) above, the participants, he and I, are realized as
subject and their subevents, climb and exercising, as object. In example (40), he
is realized as the direct object and the event he is engaged in, coming, is expressed
as a subsequent oblique, into. In (41) and (42) below, the subevents, task and waking
up, are realized with a subsequent oblique, on and to, and the participants, they are
he, are realized as subject. Lastly, in (43), the subevent, hard work, is realized as
the object, with the participant, us, realized with the antecedent oblique from. These
examples show the different ways that the construal of participant as antecedent to
subevent (or, subevent as subsequent to participant) can be grammatically realized.
(41) They worked on the task. (VerbNet)

(42) He adapted to waking up early. (VerbNet)
(43) Success requires hard work from us. (VerbNet)
It is also possible that, grammatically, the relative position of the subevent and its
participant is indeterminate. This occurs when the participant is realized as subject,
with the event nominal/subevent as an antecedent oblique, as in (44) below.
(44) He managed with dealing the cards. (VerbNet)
Antecedent obliques are only ordered with respect to the object, and not the subject,
and therefore these types of examples are also compatible with the construal of
subevent as subsequent to its participant. This is similiar to the comitative use of
with as in Johan wrote the paper with Carla, in which Carla is co-located in the
causal chain with Johan.
We describe the relation between a participant and its subevent, when both are
expressed as arguments, as an Engage relation. It is not really a force dynamic
relation, which exists only between subevents. It expresses a different kind of
semantic relation, but it is integrated into the pattern by which participants/subevents
are realized grammatically in argument structure.
Of the 192 VerbNet examples analyzed in this paper, 13 appear problematic
for the third hypothesis. That is, it is not clear that the participant is realized as
antecedent to its subevent. These problematic cases fall into two types.
The first type involves event nominals realized with obliques that may be
antecedent in other types of constructions, as in (45) and (46).
(45) I suspected him of lying. (VerbNet)

(46) I helped him with homework. (VerbNet)
In both of these examples, the participant associated with the event nominal, him
in both examples, is realized as object. The event nominals are realized with of
and with, respectively. If of and with are considered antecedent obliques, then these
constructions would construe the subevent as antecedent to its participant. Although
there are constructions in which of or with may be considered an antecedent oblique,
there are also constructions in which they are not. Both of and with may be used
to co-locate an argument in the causal chain. This can be seen with of in partitive
constructions (bowl of soup) and with with in associative constructions (She ordered
spaghetti with mushrooms). Thus, of and with may be analyzed as co-locating
the participant with its subevent, and therefore not a direct violation of the third
hypothesis.
The other type of problem case concerns a particular type of causation. This can
be seen below in examples (47)–(49).
(47) The rules forbid us from smoking. (VerbNet)

(48) They excluded us from going to the party. (VerbNet)
(49) He withdrew from the trip. (VerbNet)
In both (47) and (48), the participant is realized as object and its subevent with
the preposition from. Thus, the subevent/event nominal appears to be construed
as antecedent to the participant. This construal is based on the event expressed by
the main predicate, namely that it is preventing the participant from taking part in
the subevent expressed by the event nominal. Example (49) also expresses that the
participant will not be taking part in the subevent expressed by the event nominal,
the trip. We propose that there is a distinct relationship expressed in examples (47)–
(49): the participant is NOT engaged in the expressed subevent. We call this
relationship Refrain. It is a different type of metaphorical extension of spatial
relations, where the allative spatial relation is used for the positive relationship
between participant and subevent (Engage), and the ablative spatial relation is used
for the negative relationship between participant and subevent (Refrain).
5.5.4 Multivalent Event Nominals
For monovalent event nominals, the generalizations and examples presented above
work fairly straightforwardly. However, many event nominals describe bivalent, or
more generally multivalent, events. For bivalent events, the event nominal may
either refer to the subevent of the initiator of the causal chain or the endpoint of
the causal chain, as can be seen in (50) and (51) below.
(50) The doctor performed the surgery.

(51) The patient underwent surgery.
Examples (50) and (51) both contain the event nominal surgery, which is a
bivalent event. We therefore analyze surgery as involving two subevents, the doctor’s
subevent or the patient’s subevent. Examples (50) and (51) demonstrate that the
bivalent event nominal surgery may refer to either subevent. In (50), surgery is
construed as the doctor’s actions during the surgery. In (51), surgery is construed
as the change(s) that the patient undergoes during the surgery. These illustrate that
event nominals, even if they are not monovalent, still refer to a single participant’s
subevent.
For multivalent or bivalent event nominals, we tentatively suggest the following

rule to account for how an event nominal of a multivalent event is associated with a
single participant’s subevent.
(i) Associate nominals of multivalent events with the one expressed participant.
(ii) If two participants are expressed as dependents of the main clause, associate
the nominal of the multivalent event with the initiator (unless a patient-oriented
predicate such as undergo is present).
(iii) If two participants are expressed as dependents of the main clause, one as a
core argument and the other as an oblique phrase, associate the nominal of the
multivalent event with the core participant.
That is, multivalent event nominals will be associated with whichever participant
is expressed. If more than one of the event nominal’s participants is expressed, then
the event nominal is usually associated with the initiator. The first rule is illustrated
in (50) and (51) above. Since only one participant is expressed in each example, the
event nominal refers to the respective participant’s subevent.
The other participant can be expressed by an oblique in the same clause, as in (52)
and (53) below.
(52) The doctor performed the surgery on the patient.
(53) The patient underwent surgery by the doctor.
In both cases, the event nominal does NOT express the subevent associated with
the participant realized in the oblique phrase. More generally, our tentative proposal
associates the subevent realized by the event nominal with the highest expressed par-
ticipant in the grammatical relations hierarchy (subject, object, oblique), including
participants expressed in the main clause.
Both patient in (52) and doctor in (53) follow the realization rules of the Causal
Order Hypothesis. In example (52), the patient is realized with the subsequent
oblique on. In example (53), the doctor is realized with the antecedent oblique by.
5.5.5 Participants Expressed as Dependents of Event Nominals
The situation becomes even more complex when multivalent event nominals have
participants expressed as dependents of the event nominal. This can be seen in
examples (54)–(57) below.
(54) I accepted their writing novels. (VerbNet)
(55) I saw her bake the cake. (VerbNet)
(56) I succeeded in climbing the mountain. (VerbNet)
(57) I used the cupboard to store food. (VerbNet)
There are three basic grammatical realizations of event argument dependents
found in the VerbNet examples, possessives as in (58), objects as in (59), and
dependent obliques as in (60).
(58) I saw their laughing and joking. (VerbNet)
(59) We promoted writing novels. (VerbNet)
(60) They excluded us from going to the party. (VerbNet)
Event arguments and their dependents may be analyzed as subordinate clauses
that create subchains which are embedded within the main clause causal chain.
Possessives may be used for both initiator and endpoint participants of the event
expressed by an event nominal, but objects (of gerunds) and dependent obliques are
only used with endpoints of the subevent expressed by the event nominal. Although
(some of) the participants are realized as dependents of the event nominal, they still
follow the second rule above: the event nominal is associated with the initiator of the
subevent it expresses. The endpoints of the subevent expressed by the event nominal
are expressed as objects of the event nominal. These objects are the endpoints of the
subchain created by the event nominal and its dependents.
Examples (61) and (62) below illustrate how these subchains will be analyzed
and embedded in the causal chain expressed by the main clause and the main
clause argument phrases. In the notation below, participants are represented as
nodes in roman face and subevent arguments as nodes in italics. The argument
roles are displayed underneath the nodes. Argument roles in all capitals represent
the argument phrases in the main clause; argument roles in lower case represent
argument phrases in the subordinate clause. The = symbol represents an Engage (or
Refrain) relation between a participant and its subevent. The labels on the arrows,
lines, and Engage relation indicate the predicate or adposition that expresses the
relation.
The notation is illustrated for example (61) below:

(61) I spent the resources on buying books. (VerbNet)
The full causal chain expresses a relationship between a buyer, money, and the
goods of the value that the money can buy. The main clause expresses the buyer as
subject, the money as object, and an event nominal, buying, as a subsequent oblique.
The event nominal realizes the subevent of the resources (cf. Fifty dollars will buy
you three books). The resources are in an Engage relation with the buying subevent.
The realization of the resources as object and the buying subevent as the subsequent
oblique (with on) conforms to the hypothesis that a participant is construed as
antecedent to its subevent. The phrase books is realized as an object dependent of
the gerund buying, that is, it forms a subchain of the full causal chain. This subchain
is then embedded in the main causal chain, formed by the main predicate spent, by
means of the Engage relation between resources and buying.
Example (62) shows an alternative construal of the event in (61).
(62) I frittered away all my savings by buying books.
In this example, the construal is that my actions caused the loss of my savings,
whereas in (61) the main goal is buying books and spending resources was the
means to achieve the goal. Therefore, in (62), there is an Engage relation between
the initiator of the main causal chain, I, and the subevent expressed by buying (cf. I
bought the books), expressed by the antecedent oblique by. The Engage relation here
specifies that the main clause initiator’s subevent is overtly expressed by the means
clause predicate, buying. As in (61), books is a dependent of an event argument and
is therefore part of a subchain embedded within the main causal chain.
When participants are dependents of the event argument and not the main
verb, their realization need not conform to the main causal chain, but only
to their (subordinate) clause’s subchain. Each of those sets of participants and
their subevents (i.e., each causal subchain) has to conform to the Causal Order
Hypothesis, but the Causal Order Hypothesis does not apply to all participants
realized in different clauses/event nominal phrases in a single sentence at once.
While the examples shown so far are fairly straightforward, we will now show how
this subchain analysis is necessary to analyze more complex examples with more
participants and subevents.
Croft & Vigus (2017) presents an analysis of the RISK frame (Fillmore & Atkins
1992) as involving both participants and their subevents. The main elements of the
RISK frame are shown below (Croft & Vigus 2017, 150):
Fig. 5.7 RISK frame

participants and subevents
(63) Actor: entity that performs the Deed

Deed: action that brings about the potential of Harm to the Valued Object
Valued Object: entity that may be hurt, lost, or otherwise damaged if the
Harm occurs
Harm: potential negative outcome of the Deed
Purpose: potential positive outcome of the Deed
These elements are combined into the causal-aspectual representation shown

below in Fig. 5.7 (Croft & Vigus 2017, 151, Figure 4). Each participant is paired
with a subevent: the Actor with the Deed, the Valued Object with the Harm, and the
Beneficiary with the Purpose. The Actor’s Deed causes the possibility of Harm to the
Valued Object, and also the possibility of the Purpose subevent for the Beneficiary.
Both the Harm and Purpose subevents are unrealized (or, at least not necessarily
realized). Semantically, the risk event may be more of a branching causal chain
because it is possible for the Purpose subevent to be realized without the Harm
subevent. However, it is consistently realized grammatically as a non-branching
chain with the Valued Object/Harm subevent antecedent to the Beneficiary/Purpose
subevent. Therefore, Fig. 5.7 below represents the linguistic construal of the RISK
frame.
Both participants and subevents can be realized as arguments. In example (64)
below, only participants are realized as arguments.
(64) Why did he risk his life for a man he did not know? (Fillmore & Atkins 1992,
88)
The Actor, he, is realized as subject, the Valued Object his life as object and the
Beneficiary a man with the subsequent oblique for.
However, sentences can also express a mixture of participants and subevents, as
in (65) below. (The labels of the links in the causal chain have been suppressed to
save space.)
(65) He had risked two of his submarines by sending them to the edge of the
American beaches. (Fillmore & Atkins 1992, 90)
The Actor, he, is realized as subject of the main clause. The Actor is in an Engage
relation with the event nominal sending, for which it functions as the unexpressed
subject. The means subordinate clause corresponds with the Deed subevent. The
Engage relation is between the Actor and the Deed, as it should be following
Fig. 5.7. The event nominal sending is expressed by the antecedent oblique by, since
the Deed subevent is antecedent to the Valued Object’s subevent, which is realized
as the object of the main clause.
The Harm—the Valued Object’s subevent—is left implicit. However, the sub-
marines are also involved in the Deed subevent, expressed by them in the means
subordinate clause. Therefore, submarines also appear in the subchain as the object
of the event argument sending. That is, the same participant (the submarines) is
involved in two subevents (the Deed and the Harm), and there is a causal relationship
between the two subevents: sending the submarine causes its loss, if the Harm is
realized. This is a cycle in the causal chain; but the causal chains expressed by
individual clauses in example (65) are individually acyclic.
There is also another participant involved in the Deed subevent, beaches, that
is realized with the subsequent oblique to and is a part of the means subordinate
clause’s subchain. This demonstrates how these subchains work: beaches is realized
with a subsequent oblique because it is subsequent to the object of the subordinate
clauses’s subchain. It is not, however, subsequent to the object of the main clause
chain; in fact it is antecedent because it is part of the Deed subevent.
Example (66) below is even more complex, with multiple subordinate clauses.
(66) Mrs. Gore even risked the wrath of the record industry by campaigning to
have warning labels put on particularly offensive records. (FrameNet)
The Actor, Mrs. Gore, is realized as the subject of the main clause. The Harm,
wrath, is realized as the object of the main clause. The participant record industry is
in an Engage relation with the wrath subevent and is realized as a possessive of the
event nominal (we use italicized capitals to distinguish this phrase, dependent on the
event nominal wrath, from dependents of other event nominals in the sentence). As
with the submarines in the causal chain in example (65), Mrs. Gore appears twice
in the causal chain, but in different causal subchains. In the wrath subchain, Mrs.
Gore is the unexpressed stimulus of the emotional state experienced by the record
companies.
In the means clause subchain, on the other hand, Mrs. Gore is in an Engage
relation with the event argument campaigning, expressed by the antecedent oblique
by. The subordinate clause introduced by the antecedent oblique by expresses
the Deed. The Deed itself involves both participants (Mrs. Gore, warning labels,
records) and subevents (campaigning, have, put on). The subchain represented by
this subordinate clause realizes the have subevent with a subsequent oblique, as it
is subsequent to the campaigning subevent. The participants in this subchain also
follow the realization rules: warning labels is realized as object of the purposive
infinitive subevent realized by to have, hence subsequent to the unexpressed agents
whom Mrs. Gore wants to act. The warning labels are in turn antecedent to the
records in the subchain. The warning labels are the implicit subject of the passive
participle complement put which is a subevent dependent on have. The records are
realized as a subsequent oblique phrase of put.
By allowing for subordinate clauses to represent subchains of the main causal
chain, even more complex examples, like (65) and (66), can be represented with a
non-branching causal chain. However, these causal chains allow cycles, represented
by re-entrant nodes in the causal chains in examples (65) and (66). Instead, the
subordinate clauses introduce subchains, with their own ordering of participants and
subevents, that then themselves fit into the main causal chain of the main clause.
5.6 Conclusion
The causal chain model of event structure representation accounts for broad
cross-linguistic patterns of argument realization, in particular the realization of
participants in two different types of oblique phrases, antecedent obliques and
subsequent obliques. The “three-dimensional” model of event structure allows for
the integration of event causation represented as transmission of force with event
causation as causation between subevents, by associating each participant in an
event expressed by a single clause with its own subevent. The three-dimensional
model of event structure can be extended to account for the argument realization
of event nominals, albeit with a number of additional realization rules and rules of
associating the subevents denoted by event nominals with their participants.
Most of the examples in this paper include participant subevents expressed as
either event nominals or gerunds. Gerunds, like event nominals, take case marking
and therefore fit straightforwardly into the general argument realization rules. Partic-
ipant subevents can also be expressed by other types of complements, specifically
infinitives (bare infinitives and to-infinitives), bare adjectives and participles, and
finite complements. Finite complements are more clearly caseless, as they can
be objects and (rarely) subjects. To-infinitives on the other hand, could either be
treated as caseless or as subsequent obliques using to. The fact that to-infinitives can
occur as subjects, including when there is also an object realized, argues for them
being treated as caseless, and hence analyzed as being realized as either subject or
object. The extension of this model of event structure to complements awaits further
research.
Funding The research described in this paper was supported by grant number HDTRA1-15-1-
0063 from the Defense Threat Reduction Agency to the first author.
References
Anderson, S. R. (1977). On mechanisms by which languages become ergative. In C. Li (Ed.),

Mechanisms of syntactic change (pp. 317–364). Austin: University of Texas Press.
Bickel, B. (1997). Aspectual scope and the difference between logical and semantic representation.
Lingua, 102, 115–131.
Binnick, R. I. (1991). Time and the verb. Oxford: Oxford University Press.
Bowerman, M. (1982). Reorganizational processes in lexical and syntactic development. In E.
Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of art (pp. 319–45).
Bowerman, M. (1983). Hidden meanings: The role of covert conceptual structures in children’s
development of language. In D. R. Rogers & J. A. Sloboda (Eds.), The acquisition of symbolic
skills (pp. 445–470). New York: Plenum.
Bowerman, M. (1989). When a patient is the subject: Sorting out passives, anticausatives, and
middles in the acquisition of English, paper presented at the Workshop on Voice, University of
California, Santa Barbara.
Breu, W. (1994). Interactions between lexical, temporal and aspectual meaning. Studies in
Language, 18, 23–44.
Clark, E. V., & Carpenter, K. L. (1989). The notion of source in language acquisition. Language,
65, 1–30.
Comrie, B. (1976). The syntax of action nominals: A cross-language study. Lingua, 40, 177–201.
Croft, W. (1990). Possible verbs and event structure. In S. Tsohatzidis (Ed.), Meanings and
prototypes: Studies on linguistic categorization. London: Routledge.
Croft, W. (1993). Case marking and the semantics of mental verbs. In J. Pustejovsky (Ed.),
Semantics and the Lexicon (pp. 55–72). Dordrecht: Kluwer Academic.
Croft, W. (1994). Voice: Beyond control and affectedness. In: P. Hopper & B. A. Fox (Eds.), Voice:
Form and function (pp. 89–117). Amsterdam: John Benjamins.
Croft, W. (1998a). Event structure in argument linking. In M. Butt & W. Geuder (Eds.), The
projection of arguments: Lexical and compositional factors (pp. 21–63). Stanford: Center for
the Study of Language and Information.
Croft, W. (1998b). The structure of events and the structure of languages. In M. Tomasello (Ed.),
The new psychology of language: Cognitive and functional approaches to language structure
(pp. 67–92). Mahwah: Lawrence Erlbaum Associates.
Croft, W. (2009). Aspectual and causal structure in event representations. In V. Gathercole (Ed.),
Routes to language development: In honor of Melissa Bowerman (pp. 139–166). Mahwah:
Lawrence Erlbaum Associates.
Croft, W. (2012). Verbs: Aspect and causal structure. Oxford: Oxford University Press.
Croft, W., & Vigus, M. (2017). Constructions, frames, and event structure. In The AAAI 2017
Spring Symposium on Computational Construction Grammar and Natural Language Under-
standing (pp. 147–153). AAAI Press. http://www.aaai.org/Library/Symposia/Spring/ss17-02.
php, AAAI Technical Report SS-17-02.
DeLancey, S. (1985). Agentivity and syntax. In W. H. Eilfort, P. D. Kroeber, & K. L. Peterson
(Eds.), Papers from the Parasession on Causatives and Agentivity, Twenty-first Regional
Meeting, Chicago Linguistic Society (pp. 1–12). Chicago: Chicago Linguistic Society.
Dowty, D. (1979). Word meaning and montague grammar. Dordrecht: Reidel.
Dowty, D. (1991). Thematic proto-roles and argument selection. Language, 67, 547–619.
Fillmore, C. J. (1982). Frame semantics. In The Linguistics Society of Korea (Ed.), Linguistics in
the morning calm (pp. 111–137). Seoul: Hanshin.
Fillmore, C. J. (1985). Frames and the semantics of understanding. Quaderni di semantica, 6,
622–654.
Fillmore, C. J. (1986). Pragmatically-controlled zero anaphora. In C. J. Fillmore, V. Nikiforidou,
M. V. Clay, M. Niepokuj, & D. Feder (Eds.), Proceedings of the Twelfth Annual Meeting of the
Berkeley Linguistics Society (pp. 95–107). Berkeley: Berkeley Linguistics Society.
Fillmore, C. J., & Atkins, B. T. S. (1992). Towards a frame-based lexicon: The semantics of RISK
and is neighbors. In A. Lehrer & E. F. Kittay (Eds.), Frames, fields and contrasts: New essays
in semantics and lexical organization (pp. 75–102). Hillsdale: Lawrence Erlbaum Associates.
Grimshaw, J. B. (1990). Argument structure. Cambridge: MIT Press.
Haig, G. L. J. (2008). Alignment change in Iranian languages: A construction grammar approach.
Berlin: Mouton de Gruyter.
Hartmann, I., Haspelmath, M., & Cysouw, M. (2014). Identifying semantic role clusters and
alignment types via microrole coexpression tendencies. Studies in Language, 38, 463–484.
Jackendoff, R. (1990). Semantic structures. Cambridge: MIT Press.
Jackendoff, R. (1991). Parts and boundaries. Cognition, 41, 9–45.
Kipper-Schuler, K. (2005). VerbNet: A broad-coverage, comprehensive verb lexicon. Ph.D. thesis,
University of Pennsylvania.
Koptjevskaja-Tamm, M. (1993). Nominalizations. London: Routledge.
Kuno, S. (1973). The structure of the Japanese language. Cambridge: MIT Press.
Lambrecht, K., & Lemoine, K. (2005). Definite null subjects in (spoken) French: A Construction-
Grammar account. In M. Fried & H. C. Boas (Eds.), Grammatical Constructions: Back to the
roots (pp. 13–55). Amsterdam: John Benjamins Publishing.
Langacker, R. W. (1987). Foundations of cognitive grammar (Theoretical prerequisites, Vol. I).
Stanford: Stanford University Press.
Lehmann, C. (1982/1995/2002). Thoughts on grammaticalization: A programmatic sketch (Vol.
I). Arbeiten des Kölner Universalien-Projekts, 48, Institut für Sprachwissenschaft, Köln,
revised edition published by LINCOM Europa, München, 1995. Revised edition reprinted as
Arbeitspapiere des Seminars für Sprachwissenschaft der Universität Erfurt, 9. Erfurt: Seminar
für Sprachwissenschaft der Universität.
Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago:
Levin, B., & Rappaport Hovav, M. (2005). Argument realization. Cambridge: Cambridge Univer-
sity Press.
Lyngfelt, B. (2012). Re-thinking FNI: On null instantiation and control in construction grammar.
Constructions and Frames, 4, 1–23.
McLendon, S. (1978). Ergativity, case, and transitivity in Eastern Pomo. International Journal of
American Linguistics, 44, 1–9.
Nichols, J. (1984). Direct and oblique objects in Chechen-Ingush and Russian. In F. Plank (Ed.),
Objects (pp. 183–209). New York: Academic.
Palmer, M., Bonial, C., & Hwang, J. D. (2017). VerbNet: Capturing English verb behavior, meaning
and usage. In S Chipman (Ed.), The Oxford handbook of cognitive science (pp. 315–336).
Parsons, T. (1990). Events in the semantics of English: A study in subatomic semantics. Cambridge:
MIT Press.
Pesetsky, D. M. (1995). Zero syntax: Experiencers and cascades. Cambridge: MIT Press.
Poole, K. T. (2000). Non-parametric unfolding of binary choice data. Political Analysis, 8(3),
211–237.
Rappaport Hovav, M., & Levin, B. (1998). Building verb meanings. In M. Butt & W. Geuder
(Eds.), The projection of arguments: Lexical and compositional factors (pp. 97–134). Stanford:
Center for the Study of Language and Information.
Sridhar, S. N. (1976). Dative subject. In S. Mufwene, C. A. Walker, & S. B. Steever (Eds.),

Papers from the Twelfth Regional Meeting, Chicago Linguistic Society (pp. 582–593). Chicago:
Chicago Linguistic Society.
Talmy, L. (1976). Semantic causative types. In M. Shibatani (Ed.), The grammar of causative
constructions (Vol. 6, pp. 43–116). New York: Academic.
Talmy, L. (1988). Force dynamics in language and cognition. Cognitive Science, 2, 49–100.
Trask, R. L. (1979). On the origins of ergativity. In F. Plank (Ed.), Ergativity (pp. 385–406). New
York: Academic.
Verkerk, A. (2009a). Secondary predication in a typological context: The encoding of resultatives,
depictives and manner predications and their placement within a conceptual space. Master’s
thesis, Radboud University Nijmegen.
Verkerk, A. (2009b). A semantic map of secondary predication. Linguistics in the Netherlands, 26,
115–126.
Chapter 6
Resultatives and Constraints on
Concealed Causatives
Beth Levin
Abstract A well-formed transitive resultative construction must show a relation

of direct causation between its causing and caused subevents; that is, resultatives
conform to the same well-formedness condition as lexical causatives. Yet the best
formulation of this condition is the subject of continued discussion. This paper
revisits this question in the context of transitive resultatives. They are ideal for this
investigation as their verbs provide explicit information about the causing subevent,
while lexical causatives are silent about this subevent. This paper investigates the
relation between the causing and caused subevents through case studies of resul-
tatives with the result phrases dry and awake. The case studies probe the complex
interplay between the subject, the verb, the postverbal NP, and the result phrase
using naturally occurring examples. The last case study investigates why resultatives
with certain verb–AP combinations disallow a particular interpretation. Together
these case studies support the prototypical understanding of direct causation in the
literature.
Keywords Causal chains · Concealed causatives · Direct causation · Emission

verbs · Intervening causers · Manner verbs · Resultative constructions
6.1 Introduction
Any discussion of the semantics of the English resultative construction, illustrated

in (1) and (2), seems necessarily to be enmeshed with the notion of cause.
(1) The waitress comes back, wiping the silverware dry with a cloth napkin
before laying it out. (Jaffe, Michael Grant. 1996. Dance real slow, 24. New
York: Farrar Straus Giroux)
B. Levin ()
Department of Linguistics, Stanford University, Stanford, CA, USA
e-mail: beth.levin@stanford.edu

https://doi.org/10.1007/978-3-030-34308-8_6
186 B. Levin
(2) Last night, the dog poked me awake every hour to go outside. (Dunford, Gary.
1994. Charity’s for the birds. The Toronto Sun, November 27, 6)
This paper aims to better characterize the type of causation implicated in this
construction and in so doing to illuminate the linguistic representation of causation
more generally. The link between resultative constructions and causation arises
because transitive resultative constructions—those with an NP following the verb,
as in (1) or (2)—are easily given a causative paraphrase.1 For example, (1) is
paraphrasable as ‘The waitress wiped the silverware causing it to become dry’.
The availability of a paraphrase in terms of a causative relation between two
events2 suggests that resultatives should receive an analysis which involves explicit
reference to causation even though the construction itself does not show any overt
causative element. As Bittner (1999: 1) puts it, ‘the causal relation appears to come
from nowhere’.
This paper aims to clarify the nature of this causative relation and does not
try to explain its source. In (1) the causing event—the wiping—and the caused
event—the drying—share a participant—the silverware. It is this participant that
the result state—i.e. dry in (1)—is predicated of. That is, a causer manipulates the
shared participant, whose state then changes. One hypothesis is that such a shared
participant is a necessary component of the causative relation between subevents.
However, a consideration of a wider range of transitive resultatives suggests that
this may not be the best way to characterize this relation. Certain resultatives,
although equally amenable to a causative paraphase, are not characterized by such
an obvious link between the causing and caused events. Consider (3), which could
be paraphrased as ‘The roosters’ crowing caused me to awake’.
1 For discussion of the subtypes of resultative constructions see Levin & Rappaport Hovav
(1995: 34–41); see also Rappaport Hovav & Levin (2001: 793–794) for a list. Specifically,
transitive resultative constructions contrast with intransitive resultative constructions, which lack a
postverbal NP; their result phrase is predicated directly of the (surface) subject, as in The cookies
burned black. They are ignored here as they are not usually taken to be causative (e.g., Goldberg
& Jackendoff 2004; Levin & Rappaport Hovav 1999: 204–207; Rappaport Hovav & Levin 2001:
784). This assumption receives support from the unavailability of causative paraphrases for such
constructions. For example, ‘the lake’s freezing caused it to become solid’ does not truly capture
the sense of The lake froze solid; here solidness is the endpoint of the freezing process. However,
causative paraphrases are not always odd: The clothes steamed dry on the radiator might be
paraphrased as ‘the clothes’ steaming on the radiator caused them to become dry’. Further,
intransitive resultatives bear a syntactic resemblance to anticausative constructions, and some
researchers analyze anticausative constructions as causative in nature (e.g., Alexiadou et al. 2015;
Chierchia 2004; Koontz-Garboden 2009; Levin & Rappaport Hovav 1995; Reinhart 2002: 241–
242, but see Rappaport Hovav 2014; Rappaport Hovav & Levin 2012). Thus, the causative status
of intransitive resultatives may merit further scrutiny.
2 I assume a ‘biclausal’ or ‘bieventive’ analysis of causation as argued for by Dowty (1979: 91–
96), Parsons (1990), and Shibatani (1976a,b), among others. This contrasts with work that models
causation as a relation between an individual and a proposition (McCawley 1968, 1971) or as
a relation between individuals (Croft 1991: 162–163), an approach adopted by much work in
cognitive linguistics (e.g., Talmy 1976). See also Sect. 6.2.
6 Resultatives and Constraints on Concealed Causatives 187
(3) When the roosters that scratch in the yard of Brastagi’s best hotel crowed me
awake that dawn a few months ago, I knew it was destined to be a memorable
day . . . (Robbins, Tom. 1986. True adventure: crowned king of the cannibals.
The New York Times, March 16, Sect. 6, Part 2, 8)
Here the causing event—the crowing—describes an action which does not involve
physical manipulation of any entity, not even the entity that attains the result state.
In fact, this action need not even be directed at a particular entity: the roosters could
simply be crowing for their own reasons. This interpretive property has a syntactic
reflex. Constructions such as (3) are referred to as nonselected NP resultatives as the
NP following the verb—the apparent ‘object’ of the verb—is not selected (or, more
formally, ‘subcategorized’) by it. For instance, in (3) the rooster does not ‘crow me’;
that is, *The rooster crowed me. Thus, (3) contrasts with (1) and (2), where the NP
is selected by the verb (e.g., The waitress wiped the silverware; The dog poked me).
The selected vs. nonselected NP resultative subtypes are recognized here because
they prove useful for organizing the data as it bears on the goals of this paper.3 Many
formal accounts analyze all resultatives as having nonselected postverbal NPs; see
den Dikken & Hoekstra (1994), Hoekstra (1988, 1992a,b,c), and McIntyre (2004:
542–547) for syntactic arguments and Grône (2014) and Kratzer (2005) for semantic
arguments. I use the label ‘postverbal NP’ rather than ‘object’ for the NP following
the verb to remain agnostic about the best syntactic analysis of the construction,
as the choice of analysis determines whether the postverbal NP is indeed ever an
object.
As this suggests, nonselected NP resultatives have received considerable atten-
tion with respect to the best syntactic analysis of the construction; in contrast, they
have not received sustained attention in the context of a theory of causation, with
Kratzer (2005) being an exception. Yet precisely because they have a causative
interpretation despite the nonselected NP, they seem particularly worthy of interest.
As this paper shows, an examination of the well-formedness conditions on both
selected and nonselected NP resultatives, especially in the context of other con-
structions which express causation, has much to tell us about the types of causative
relations between events that are linguistically privileged. The major question to be
3I use the selected vs. nonselected NP resultative distinction since it makes useful cuts in the data;
however, a reviewer asks whether the line between resultatives of these two types is always clear.
This question is motivated by parallels the reviewer notes between nonselected NP resultatives
with the verb–AP combination pour–full and the locative alternation. In its basic use pour takes
what will be the contents of a container as its object (e.g., pour tea into the mug), but it may take
the container as postverbal NP in a resultative (e.g., pour the mug full); see Sect. 6.4. Although not
usually so described, the object of a locative alternation verb can be characterized as contents (e.g.,
pack the clothes) or container (e.g., pack the suitcase) (Waltereit 1999: 239–240). Given this, when
pour takes a container postverbal NP in a resultative, perhaps the NP is not truly nonselected. The
reviewer’s suggestion, however, seems to build on the assumption that the objects in both locative
alternation variants are selected, an assumption denied in many analyses (Beavers 2017; Mateu
2010).
188 B. Levin
considered in the remainder of the paper is: What is the best way to characterize the
causative relation between their causing and caused events?
Through several case studies I show that across both selected and nonselected
NP resultatives, the action described by the verb—that is, the causing event—is an
(often usual) way of bringing about the state described by the result phrase—the
caused event. Thus, I focus on the relation between these events, and consider the
contribution of the postverbal NP in this context. In a selected NP resultative, the
entity denoted by the postverbal NP qualifies as a ‘basic’ participant in the causing
event. For example, the silverware is a clear participant in the causing event—the
wiping—in (1). In a nonselected NP resultative, the causing event still impinges
on the entity denoted by the postverbal NP, although as I discuss it can be difficult
to decide whether it truly qualifies as a participant in the causing event. Its exact
role, if any, in this event depends on the nature of the action that brings about
the relevant result state. For instance, the writer in (3) is not a participant in the
causing event—the crowing—in the strong, manipulated sense that the silverware
is in the selected NP resultative (1); nevertheless, he is impinged on by this event
and perhaps could even be considered a participant in a rather weak sense. Much of
this paper is devoted to identifying the nature of this impingement, as it is critical
for understanding the type of causation characteristic of resultatives. This goal,
however, can be met without fully resolving whether the referent of the nonselected
NP is a participant in the causing event; I leave this issue for future work.
This paper probes these issues using naturally occurring data that illuminate
the complex interplay among the verb, the result phrase, and the postverbal NP—
whether selected or not—in controlling the well-formedness of a resultative. These
data are drawn from a collection of just under 1250 naturally occurring transitive
resultatives with adjective-headed result phrases, as in examples (1)–(3).4 The data
are predominantly taken from newspapers and fiction written since the mid-1980s;
some recent examples from web searches have been added to explore particular
verb–AP combinations further. There is a limitation to the data. The examples were
primarily collected opportunistically and are not drawn from a ‘balanced’ corpus
designed to be representative of current English. Thus, they bear on claims about
possible options—claims which are important in their own right. However, I refrain
from giving counts, and any quantitative assessments in this paper should at best be
taken to be suggestive of patterns that may exist.
Section 6.2 sets the stage by introducing notions of causation, particularly
direct causation, as they pertain to resultatives. Section 6.3 begins to examine the
constraints on the causative relation between the causing and caused events in a
resultative—what I refer to as the tightness condition on resultatives. It introduces
4 Examples are drawn from a larger collection which includes intransitive resultatives with result
APs and both transitive and intransitive resultatives with result PPs (e.g., Robinson roared him into
silence), as well as examples of the way construction (e.g., She swam her way to good health).
These are all ignored here. See Grône (2014) for interesting recent discussion of the semantics
of transitive resultatives with result PPs; such resultatives too can be classified into selected and
nonselected NP types.
two key observations: the choice of result phrase constrains the choice of verb
describing the causing event and in most resultatives this verb is a manner verb.
Relatedly, it points out that the causing event is often a usual means of bringing
about the result state. Section 6.4 further probes the nature of the causing event–
caused event relation in nonselected NP resultatives. Sections 6.5 and 6.6 present
case studies of resultatives whose result APs are headed by the adjectives dry and
awake, respectively, to illuminate the tightness condition. They explore whether
properties of the types of actions required to bring about each of these result states
play out in the form of the resultative (i.e. selected vs. nonselected NP). Section 6.7
further clarifies the tightness condition by examining why some resultatives cannot
receive certain, apparently plausible nonselected NP interpretations. Section 6.8
concludes with a discussion of the well-formedness of resultatives in light of the
preceding sections and its implications for the tightness condition.
6.2 A First Look at the Relation Between the Subevents
Both selected and nonselected NP resultatives are deserving of interest in the study
of causation because their constituent subevents are taken to show the kind of tight
semantic integration characteristic of the subevents of so-called ‘lexical causatives’
(Rappaport Hovav & Levin 2001: 783). Lexical causatives are transitive sentences
such as The waitress dried the silverware or The dog woke me that allow a causative
paraphrase (e.g., ‘The waitress caused the silverware to be dry’, ‘The dog caused me
to awake’). Yet, like resultatives, they lack an overt causative element; they too are
what Bittner (1999) calls ‘concealed causatives’. In contrast to resultatives, lexical
causatives leave the causing event implicit and only specify the caused event. That
is, in Tracy warmed the soup, we do not know whether Tracy heated the soup in a
pot on the stove, in a bowl in a microwave, or perhaps even by another, more far-
fetched method. What makes resultatives particularly relevant to illuminating the
form of causation that falls under ‘concealed causation’ is that resultatives, unlike
lexical causatives, include explicit information about the causing event (via the
verb), as well as about the caused event (via the result phrase). Thus, in The waitress
wiped the silverware dry, we know that the waitress brought about the silverware’s
dry state by wiping it (and not, say, airdrying it). Similarly, in The dog poked me
awake, we know that the dog woke me by nudging me (and not, say, barking). Thus,
resultatives provide more information about what causing event–caused event pairs
qualify for a tight linguistic expression.
The relation between the subevents in both lexical causatives and resultatives
is taken to instantiate a special type of causation, whose hallmark is a fairly
tight relation between the two subevents. This tightness is sometimes described as
enabling the two subevents to form a single, but complex event. This event structure
is posited in part because of a syntactic reflex: syntactically, lexical causatives
show evidence of a monoclausal structure (Shibatani 1976a,b). This proposed tight
connection is often supported by contrasting lexical causatives with periphrastic
190 B. Levin
causatives, which contain an overt causative verb—cause in the examples in (4);

however, like lexical causatives and unlike resultatives, they too are silent about the
causing event.
(4) a. The waitress caused the silverware to be dry.
b. The dog caused me to awake.
The literature repeatedly notes that the link between the events in periphrastic
causatives is looser than in lexical causatives. For instance, not all situations that can
be described using periphrastic causatives can be described by the corresponding
lexical causative (e.g., Dowty 1979: 96–97; Hall 1965: 28; Shibatani 1976b: 28–
31), as in (5).5
(5) a. The low air pressure caused the water to boil.
b. ∗ The low air pressure boiled the water. (Hall 1965: 28, (2–33), (2–34))
Similarly, certain situations that can be described by periphrastic causatives cannot
be described by the corresponding resultatives. For example, the intended sense of
The dog caused me to awake by scratching at the bedroom door is not captured by
The dog scratched me awake; rather, the resultative is only appropriate if the dog
actually scratches the sleeper.
The difference between lexical and periphrastic causatives is often described in
terms of whether the relation between the two events necessarily involves what is
commonly referred to as ‘direct causation’—as lexical causatives are said to do—or
may also allow for ‘indirect causation’—as in periphrastic causatives (e.g., Bittner
1999; Fodor 1970; McCawley 1978; Pinker 1989: 66; Shibatani 1976b: 31–39;
Smith 1970; and Wolff 2003). Resultatives again are said to pattern with lexical
causatives (e.g., Bittner 1999; Carrier & Randall 1993: 124–125; Dowty 1979: 220;
Goldberg 1995: 194–195; Kratzer 2005: 196–197; Levin & Rappaport Hovav 1999:
211–212; Pustejovsky 1991: 64–65; and Rappaport Hovav & Levin 1998, 2001:
783).6 Bittner (1999: 2), in fact, introduces the hypothesis in (6) in her work on
concealed causatives:
5 Counterfactuality is often said to be a necessary ingredient of causation (Dowty 1979: 99–110;

Lewis 1973). In fact, Levin & Rappaport Hovav state such a condition on resultatives: ‘The result
subevent would not have happened if the causing subevent had not happened and all else had
remained the same’ (1999: 212, (29c)). However, counterfactuality holds periphrastic and lexical
causatives as well as resultatives, so there must be another condition that differentiates lexical
causatives and resultatives from periphrastic causatives. Moreover, even rather loosely related
events can be counterfactually related, so this observation too suggests that it is necessary to look
beyond counterfactuality.
6 Neeleman & van de Koot (2012) call into question direct causation as a condition on lexical
causatives, citing examples such as A slip of the lip can sink a ship (2012: 28, (8d)), and replace
it with the notion ‘contributing causal factor’. However, Martin (2018) and Wolff (2003: 33–34)
offer ways of accommodating such examples under the umbrella of direct causation. Whatever
the implications of such examples, there is nevertheless agreement that some kind of a tightness
condition must be met by two subevents to be expressible by a lexical causative—or resultative—
and it is this condition that I examine in this paper.
(6) Concealed Causative Semantics: If a causal relation is syntactically concealed

(only its arguments are overtly expressed), then it is semantically direct (no
intermediate causes). (Bittner 1999: 2, (C))
Shibatani (1976b: 31) characterizes the type of causation found in lexical causatives
narrowly as ‘manipulative’ causation, reflecting the intuition that in many such
causatives the causer physically manipulates the causee. Indeed, the prototypical
form of causation associated with concealed causatives is manipulative causation.
Other researchers propose a range of overlapping characterizations of the type of
causation characteristic of concealed causatives; see Wolff (2003: 4, Table 1) for
a compendium. Although these characterizations are less specific than Shibatani’s
manipulative causation, they subsume it. In (6), Bittner defines direct causation in
terms of a lack of intermediate causes, a definition which is elaborated by Kratzer
(2005: 196–198) within an event-cause-event model of causation.7 Rappaport
Hovav & Levin (2001), building on Levin & Rappaport Hovav (1999), make a
related proposal in setting out a set of well-formedness conditions on the relation
between the two subevents of resultatives.8
(7) There is no intervening event between the causing subevent and the result
subevent; that is, causation is direct. (Rappaport Hovav & Levin 2001: 783,
(45d))
Rappaport Hovav & Levin are working in an event-cause-event model of causation,
so the assumption implicit in (7) is that an intervening event introduces an
intermediate cause.
Definitions of direct causation within an approach that models causation as a
relation between individuals explicitly mention intervening causers, as in the much
cited definition of direct causation in (8) from Wolff (2003).
(8) Direct causation is present between the causer and the final causee in a causal
chain (1) if there are no intermediate entities at the same level of granularity9
as either the initial causer or final causee, or (2) if any intermediate entities that
are present can be construed as an enabling condition rather than an intervening
causer. (Wolff 2003: 4–5)
7 Despite rejecting a notion of direct causation, Neeleman & van de Koot introduce a notion of
‘accountability’ to deal with intentional entities in a causal chain (2012: 31). Interestingly, they
then introduce a condition that once a causal chain has an accountable individual in it, there cannot
be another intervening accountable individual (2012: 33, (16)).
8 The other well-formedness conditions on resultatives proposed by Rappaport Hovav & Levin
(2001: 783, (45)) are: (i) the subevents need not be temporally dependent, (ii) the result subevent
cannot begin before the causing subevent, and (iii) only the result subevent can bound the event as
a whole.
9 See Wolff (2003: 33–34) for discussion of granularity in the context of direct causation and Croft
(1991: 162–165) for more general discussion of the granularity of event descriptions.
192 B. Levin
This definition assumes an individual-act-on-individual model of causation. On this

model (Croft 1991: 167–174), events are represented in terms of a ‘causal chain’
composed of a set of segments, each of which relates two event participants. The
segments in the causal chain and, thus, the participants are ordered in the direction
of ‘transmission of force’ from one participant to the next. Kratzer (2005: 197)
also uses the term ‘causal chain’, but defines it in terms of a series of events,
with each non-initial event counterfactually related to the preceding event in the
chain. Since events have participants, even when adopting the event-cause-event
model of causation, it is possible to refer to event participants. Concomitantly,
an individual-act-on-individual representation could be extracted from an event-
cause-event representation. Therefore, for convenience I may sometimes refer to
the participants in the events rather than the events themselves in this paper.
Although there is disagreement concerning whether ‘direct’ causation in any
form is the right characterization of the tightness condition on lexical causatives
and resultatives (see fn. 6), there is some kind of tight relation between their
subevents, and it is the tightness of this relation that has led to proposals that
they are construable as a single event—a property that sets them apart from
periphrastic causatives. This relation must capture the intuitions behind the term
‘direct causation’; these include the lack of an intervening cause and in prototypical
instances the presence of physical manipulation of the causee by the causer.
However, to remain open about the best characterization of the relation and to avoid
any presuppositions that might accompany the term ‘direct causation’, I refer to a
tightness condition, particularly as this paper’s central aim is to investigate its nature
further. As I show, in most instances the tightness condition as demonstrated by the
resultatives in the case studies conforms to direct causation in the sense of a causer
directly affecting a causee, often via physical manipulation.
Existing characterizations of direct causation, such as those cited here, make
reference to notions such as ‘intermediate’ or ‘intervening event’, ‘intervening
causer’, or ‘intermediate entity’. All these notions involve the part of the causal
chain where the causing event and caused event come together, which I refer to
as the ‘middle’ of the causal chain. This portion of the causal chain is critical to
understanding the tightness condition; however, since lexical causatives leave the
causing event implicit, they do not provide an optimal domain for exploring it.
In contrast, transitive resultatives explicitly express the causing event; thus, they
provide a laboratory for exploring the properties that the causing event must have
so that the relation between it and the caused event meets the tightness condition.
The goal is to characterize this relation in a way that encompasses both selected
and nonselected NP resultatives, while shedding light on why the postverbal NP is
understood as selected or not.
6.3 The Result State’s Implications for the Verb in the

Resultative
Resultative constructions are concealed causatives. Although they lack an explicit

marker of causation, the result state, represented by the result phrase, is understood
to be brought about by the causing event, represented by the verb. Thus, the causing
event must be one that can lead to the result state, while conforming to the tightness
condition. As a given result state constrains the set of causing events that can bring
it about, looking in the corpus at the set of verbs representing the causing event for
a given result phrase should provide information about the nature of the tightness
condition.
A survey of the corpus shows that a preponderance of the attested result states
are physically instantiated, as in (9), although there are exceptions, including the
mental states listed in (10).
(9) awake, bare, barkless, black, blank, bloody, clean, clear, closed, coarse, dark,
dry, empty, flat, full, free, hoarse, . . .
(10) alert, calm, clueless, crazy, helpless, loopy, speechless, witless, . . .
Unattested in the corpus are adjectives naming individual-level states, supporting
observations in the literature (Hoekstra 1992a: 162; Levin & Rappaport Hovav
1995: 55).10
When a resultative has a result phrase that names a physically instantiated state,
the causing event must describe an action that can bring about such a state. Such
actions must necessarily involve physical manipulation of the entity that changes
state; such states cannot be achieved by action at a distance (magic aside!). Thus,
force exertion, such as pushing or pulling, is often used to open or close a door
or window and surface contact, such as sweeping or mopping, is often used to
clean floors. Indeed, many of the verbs attested in such resultatives in the corpus
lexicalize actions that involve physical manipulation such as surface contact, impact,
or force exertion. These actions are usually, although not always, performed by
a volitional, animate agent. Since resultatives with a physically instantiated result
phrase typically have an agent, they instantiate what Croft (1991: 168), who as noted
in Sect. 6.2 adopts the individual-act-on-individual model of causation, considers
the most unmarked type of causation: a volitional agent physically manipulates
an entity to bring about a physically instantiated state in it. Such scenarios also
instantiate what are considered prototypical instances of direct causation.
The verbs in resultatives whose result phrases describe a mental state of an
animate entity such as (11) are more diverse; they may describe physical actions,
10 Previous work notes several other constraints on the adjectives in result phrases. Most important,
Wechsler (2005, 2012) shows that they must be maximum endpoint closed scale adjectives (e.g.,
clean, empty); see also Goldberg (1995: 195–197). Open scale adjectives (e.g., cool, long, wide)
and minimal endpoint closed scale adjectives (e.g., dirty, wet) are not attested (except in rare
instances where they are coerced to have a maximum endpoint interpretation).
194 B. Levin
which can provoke these states, as in (11a), as well as acts of communication, which
can too, as in (11b).
(11) a. He cracked the window and let the cool, dry desert air slap him alert.
(Dunlap, Susan. 1998. No immunity, 214. New York: Delacorte)
b. Well, the conclusion was that my mistress grumbled herself calm.
(Brontë, Emily. 1965 [1847]. Wuthering Heights, 78. London: Penguin)
I focus on resultatives whose result phrases describe physically instantiated states,
but the paper’s conclusions can be extended to result phrases that describe mental
states.
The verbs found in resultatives with physically instantiated result states are
almost exclusively instances of what are called ‘manner’ verbs, a set which contrasts
with ‘result’ verbs (Levin & Rappaport Hovav 1991, 2013, 2014; Rappaport Hovav
& Levin 1998, 2010). Result verbs lexically specify a change in a scalar valued
property of an entity (Hay et al. 1999; Kennedy & Levin 2008; Rappaport Hovav
2008); that is, they describe the attainment of a result state of an action (e.g., remove,
cover, empty, clean), including the types of result states that are expressed in the
result phrases of resultatives. Thus, many of them are the verbs found in lexical
causatives. Manner verbs, in contrast, are not verbs of scalar change and have been
called ‘manner’ because many specify a way of carrying out an action, such as a gait
(e.g., walk, amble, prance) or a way of making contact with a surface (e.g., pound,
sweep, wipe). Some of these actions are regularly intentionally performed to bring
about one or more result states. In fact, some involve tools designed precisely to
bring about a particular state (e.g., crank, mop, towel). Thus, wiping is used to clean
surfaces such as tables or counters, while mopping is used to clean floors. A manner
verb does not entail its intended result, if there is one (Talmy 2000: 265–267); in
contrast, the result cannot be denied with a result verb. Compare (12) and (13).
(12) I just wiped the counter, but it’s still dirty/sticky/covered in crumbs.
(13) # I just cleaned the counter, but it’s still dirty/sticky/covered in crumbs.
A small number of resultatives have a result verb. Such resultatives have result
phrases that further specify the result lexicalized in the verb, as in the verb–result AP
combination fill–full, consistent with a constraint against having an action leading
to two result states (e.g., Goldberg 1991: 368, Tenny 1987: 183–184, 1994: 68).
In the preponderance of resultatives, the result phrase is typically found with a
manner verb, and specifically a manner verb which describes an action used to
bring about the relevant result state.11 This observation holds equally of selected
and nonselected NP resultatives. Just as wiping is performed to dry silverware, as
in the selected NP resultative (1), repeated as (14), so is pouring performed to fill a
container, as in the nonselected NP resultative (15).
11 Furthersupport for the link between certain manners and results comes from an observation in
Alexiadou et al. (2017). They point out that manner verbs in French show actuality entailments
involving usually associated results when they take inanimate cause subjects.
(14) The waitress comes back, wiping the silverware dry with a cloth napkin
before laying it out. (Jaffe, Michael Grant. 1996. Dance real slow, 24. New
York: Farrar Straus Giroux)
(15) Audrey flipped a mug into the air, caught it by its handle, and poured it full.
(Greenlaw, Linda. 2008. Fisherman’s bend, 219. New York: Hyperion)
Although in (14) and (15) the causing event is performed intentionally, what matters
is that the causing event is a way of bringing about the result state. If this is the case,
the causing event need not be performed intentionally. Consider the nonselected NP
crow–awake example (3), repeated as (16), as well as a second example with the
same result phrase, (17). In (16) the subject—the sound emitter—is animate, but is
not making noise in order to wake someone; in (17) the emitter is inanimate and,
thus, lacks intention.
(16) When the roosters that scratch in the yard of Brastagi’s best hotel crowed me
awake that dawn a few months ago, I knew it was destined to be a memorable
day . . . (Robbins, Tom. 1986. True adventure: crowned king of the cannibals.
The New York Times, March 16, Sect. 6, Part 2, p. 8)
(17) At exactly midnight, the phone beside Helma’s bed jangled her awake.
(Dereske, Jo. 2001. Miss Zukas shelves the evidence, 47. New York: Avon)
These examples exploit our knowledge that a loud noise often wakes a sleeper. What
all the examples have in common—(14) and (15) as well as (16) and (17)—is that
the causing event is a typical way of bringing about the result state.12
The observed relation between a manner verb and a result phrase—that is, that the
action lexicalized by the verb is a typical way to bring about the relevant result seems
to be the reason for taking the relation between the causing and caused subevents in
a resultative to qualify as tight in a way that allows its expression via a concealed
causative.13 The general idea appears in the literature on resultatives. Wechsler
(1997: 310, (10)) writes that in selected NP resultatives the result must denote “a
‘canonical’ or ‘normal’ result state of an action of the type denoted by the verb.”
Kaufmann & Wunderlich (1998: 15) make a comparable point about nonselected
NP resultatives. Iwata (2014: 253–259) examines certain unacceptable resultatives
and points out that the action described by their verb could not bring about the
relevant result, couching the discussion in the force dynamic terms often employed
in an individual-act-on-individual model of causation. Finally, Kratzer (2005: 198–
199) proposes that in order for a resultative to be well-formed, instances of the event
where the relevant result is achieved must be in the extension of the resultative’s
12 Even though in some instances the verb–result phrase association may be ‘one off’, it must be
understood to qualify as ‘tight’ given the discourse context in the sense elaborated in the remainder
of this paper. Such examples might qualify as instances of what Martin (2018) calls ‘ex posto facto’
causation.
13 As Malka Rappaport Hovav (p.c.) points out, taking the causing event to usually bring about the
caused event is not the same as taking the two events to be ‘close’ in a causal chain; however, these
notions might often fall together.
196 B. Levin
verb. Most likely, this move works because the result state is usually expected to
come about when the causing event, the event represented by the verb, occurs.
This section has provided a general overview of how the choice of result phrase
constrains the choice of verb in a resultative: the verb must describe an action that
can bring the relevant result about. This perspective turns out to be particularly
useful for understanding how nonselected NP as well as selected NP resultatives can
meet the tightness condition. It suggests that case studies of particular result phrases
found in resultives of both types can be fruitfully used to probe what properties
of the causing event lead to the different status of the postverbal NP. Sections 6.5
and 6.6 present such case studies.
6.4 More on Nonselected NP Resultatives
This section further sets the stage for the case studies by examining how the
association between the causing event and the result state plays out in nonselected
NP resultatives, that is, those resultatives whose postverbal NP is not understood
as the object of the verb describing the causing event. A priori this property might
suggest such resultatives involve a weaker link between the causing and caused
events since the postverbal NP, by virtue of being nonselected, is not a participant
in the causing event in the strong sense that a selected postverbal NP is. Thus, this
section zooms in on the relation between the causing and caused events—the middle
of the causal chain—in nonselected NP resultatives with respect to the participants
in these events. The major focus is on understood, but unexpressed participants in
both the causing and caused events, including whether the entity denoted by the
nonselected NP may be involved in the causing event as such a participant. These
considerations are also relevant to understanding how the causing and caused events
are able to meet the tightness condition when the postverbal NP is not selected.
The criterion used to determine whether the postverbal NP in a transitive
resultative is selected or not is whether it can be understood as the object of the
verb when the verb is used outside of the resultative. In a selected NP resultative,
as in the (a) sentences in (18) and (19), it can, as shown in the (b) sentences. Such
constructions are necessarily found with transitive verbs.
(18) a. Last night, the dog poked me awake every hour to go outside. (Dunford,
Gary. 1994. Charity’s for the birds. The Toronto Sun, November 27, p. 6)
b. The dog poked me.
(19) a. She snipped off the end of the cotton and patted the mend flat. (Curzon,
Clare. 1990. Three-core lead, 69. New York: Doubleday)
b. She patted the mend.
In nonselected NP resultatives, as in the (a) sentences in (20)–(22), the postverbal
NP is not understood to be the object of the verb, as shown in the (b) sentences.
(20) a. He had set an alarm, which rang at five thirty the following morning,
shrilling them both awake. (Pilcher, Rosamunde. 1984. Voices in
summer, 116. New York: St. Martin’s)
b. ∗ The alarm shrilled them.
(21) a. ‘Before you go, crank me flat.’ (Roberts, Lillian M. 1998. Almost human,
17. New York: Ballantine)
b. ∗ You cranked me.
c. You cranked the hospital bed.
(22) a. Maxey stood up to get a glass and pour it full of milk. (Cail, Carol. 1995.
Unsafe keeping, 146. New York: St. Martin’s.)
b. ∗ Maxey poured the glass.
c. Maxey poured the milk.
The verb may be an intransitive verb, as in (20), as well as the earlier example (3);
that is, a verb that does not typically select an object.14 Alternatively, it may be
a transitive verb as in (21) and (22); see the (c) sentences. In such resultatives,
the verb’s object though unexpressed, is still understood, as many have noted
(Carrier & Randall 1993: 125–126; Kaufmann & Wunderlich 1998: 5; Levin &
Rappaport Hovav 1995: 37–39). In (21a), the addressee is cranking something; here,
its identity—a hospital bed—is inferrable from context. In (22a), we assume that
the liquid being poured is milk since milk is mentioned as the complement of the
adjective full.
Resultatives with understood objects always involve manner verbs. As discussed
in Levin (1999) and Rappaport Hovav & Levin (1998, 2010), two-argument manner
verbs need not express their non-subject (perhaps, more accurately, non-agent or
more broadly non-effector (van Valin & Wilkins 1996)) argument. This property
is just the prerequisite needed to give rise to a nonselected NP resultative, and
Rappaport Hovav & Levin (1998, 2010) tie this distributional fact to argument
realization differences inherent to manner and result verbs. Such analyses build
14 InEnglish, it can be tricky to attribute intransitivity to a verb because many verbs that are
considered intransitive can take certain special types of objects, most notably cognate objects,
as well as reaction objects (Levin 1993: 95–96, 97–98). Thus, smile is typically considered
intransitive, but it can take a cognate object as in She smiled the smile of a Mona Lisa or a
reaction object as in She smiled her approval. Not all intransitive verbs allow such objects so
readily. Consider shrill in (20): The alarm clock shrilled a shrill seems quite odd, although perhaps
The alarm clock shrilled a warning is better. I assume that a verb is intransitive if it can be found
without a postverbal NP without the existence of such an NP having to be inferred. This caveat
is necessary because transitive verbs like eat can be found with or without an object, but in the
absence of an object, one is still understood: Casey ate means that ‘Casey ate something’.
198 B. Levin
on earlier observations that nonselected NP resultatives are only found with

unergative verbs, which lack objects altogether, or with those transitive verbs that
independently allow unexpressed objects (Carrier & Randall 1993; Goldberg &
Jackendoff 2004: 548; Levin & Rappaport Hovav 1995). However, the restatement
in terms of manner verbs is superior because some manner verbs that seem not
to allow unexpressed objects in isolation have such objects in nonselected NP
resultatives.15 An example is the verb crank, which sounds odd in an unexpressed
object use (*Pat cranked), yet is attested in the nonselected NP resultative (21a).
It appears that the recoverability conditions on unexpressed objects with transitive
manner verbs are more easily met in resultatives.
Although unexpressed objects are ‘intermediate entities’ in the causal chain, the
acceptability of resultatives such as (21a) and (22a) means that such objects do not
count as the ‘intervening causers’ or ‘intermediate causes’ mentioned by Bittner
(1999) and Wolff (2003). In this respect, they are like instruments, which also do
not count as ‘intervening causers’ according to Wolff (2003), who takes them to
qualify as enabling conditions, rather than true causers which could ‘disrupt’ the
causal chain, i.e. lead to the ill-formedness of a concealed causative description.16
A key question is what types of intermediate entity can be present, while allowing
the causing and caused events to maintain a tight relation.
Some researchers also note an implicit relation between the postverbal NP and
the causing event, perhaps mediated by the participant denoted by the unexpressed
object, if one is available (Kaufmann & Wunderlich 1998: 32). They suggest that
although the postverbal NP is not selected by the verb in the causing event, the
entity it denotes is nevertheless sometimes a participant in this event. For instance,
Jackendoff (1990: 226–227) proposes that the postverbal NP bears an ‘oblique’
relation to the verb describing the causing event; along similar lines, Sato (1987: 83)
proposes that it bears a location or goal relation to the verb. Thus, (24) shows that
the container that coffee—the unexpressed participant in (15), repeated as (23)—is
poured into may be expressed in a locative PP complement of pour.
(23) Audrey flipped a mug into the air, caught it by its handle, and poured it full.
(Greenlaw, Linda. 2008. Fisherman’s bend, 219. New York: Hyperion)
(24) Audrey poured coffee into the mug.
However, the postverbal NP does not always bear such a clear relation to the
verb in the causing event. Consider (20a), repeated in (25a); here it does not seem
15 Unexpressed objects must be pragmatically recoverable in context. See Brisson (1994) for some
discussion of the recoverability condition. The wider availability of unexpressed objects in some
constructions than in others is noted in Levin (1999: 244, fn. 10), but remains to be explained.
16 The literature distinguishes between those instruments that have their own energy source
and can perform an action independently and those that cannot. The former, sometimes called
‘intermediary’ instruments, can occur as subjects, while the latter, sometimes called ‘facilitating’
or ‘enabling’ instruments, cannot (Marantz 1984: 247; McKercher 2001: 52–54; Ono 1992;
Wojcik 1976: 165). Facilitating instruments do not qualify as intervening causers, but intermediary
instruments do, as discussed further in Sect. 6.7. See Wolff et al. (2010) for discussion of the place
of these two types of instruments in a causal chain.
possible to accommodate the nonselected NP in a sentence with the verb, as shown

in (25b).
(25) a. He had set an alarm, which rang at five thirty the following morning,
shrilling them both awake. (Pilcher, Rosamunde. 1984. Voices in
summer, 116. New York: St. Martin’s)
b. ?? The alarm shrills at/to them.
In other instances this option might seem to exist, but on closer scrutiny it may not
capture the precise sense of the resultative. Consider (3), repeated in (26a): as noted
in Sect. 6.1, roosters could crow someone awake at dawn without explicitly crowing
at them.17
(26) a. When the roosters that scratch in the yard of Brastagi’s best hotel crowed
me awake that dawn a few months ago, I knew it was destined to be a
memorable day . . . (Robbins, Tom. 1986. True adventure: crowned king
of the cannibals. The New York Times, March 16, Sect. 6, Part 2, p. 8)
b. The roosters crowed at me that dawn a few months ago.
Finally, in (21a), repeated as (27), the cranking event involves the addressee, as
agent, and the bed, as the manipulated entity; it would seem odd to say that the
speaker is a participant in this event. Yet the cranking happens precisely because the
speaker is in the bed and will be affected by an action on it.
(27) ‘Before you go, crank me flat.’ (Roberts, Lillian M. 1998. Almost human,
Thus, elaborating on what was noted in Sect. 6.1, although in nonselected
NP resultatives the causing event clearly impinges on the entity denoted by the
postverbal NP, in certain instances it is at best a participant in the causing event
in a weak sense, while in others it is not strictly speaking a participant at all. These
observations underscore the importance of further investigating the conditions that
govern the well-formedness of nonselected NP resultatives. The case studies in the
following two sections are intended to do this, and they return to the issues and
examples discussed in this section. As discussed in Sect. 6.3, since the result state
influences the possible causing events, they are organized around particular result
phrases—adjective phrases headed by dry and awake—and not particular verbs.
6.5 A Case Study: Result APs Headed by the Adjective Dry
Resultatives whose result AP is headed by the adjective dry—henceforth, the result

AP dry—provide a good domain for examining the factors governing the well-
formedness of both selected and nonselected NP resultatives and particularly for
17 Jackendoff (1990: 227) suggests that crow at does capture the sense of resultatives such as (26a),
but I do not find that to be the case.
200 B. Levin
investigating how the link between the causing and caused events qualifies as tight
in nonselected NP, as well as selected NP, uses. The reason is that in the corpus this
result AP, unlike many others, is prevalent in resultatives of both types, allowing the
conditions on each to be compared.
An examination of the data shows that the type of resultative overwhelmingly
correlates with the nature of the entity that the adjective dry is predicated of, and
specifically with whether it is a surface or a container. By container, I mean an
entity designed to contain something, such as a bottle or bowl; thus, it must be 3-
dimensional and have an interior. By a surface, I mean an entity that is conceived of
as 2-dimensional, such as a table or plate; however, 2-dimensionality is sometimes
a matter of construal. Thus, a tub is prototypically a container: it is designed to be
filled with water or other liquid, say for bathing; however, sometimes a tub may be
conceived of as a surface; for instance, when being wiped with a sponge or scrubbed
with a brush.
Section 6.5.1 examines resultatives where the result AP dry is predicated of a
surface. Section 6.5.2 turns to those where this result AP is predicated of a container.
The data reveal that dry has slightly different senses when predicated of surfaces
and containers. I show that relatedly the manners of causing the type of dryness
holding of surfaces vs. containers give rise to differential preferences for selected
vs. nonselected NP resultatives. Section 6.5.3 extends the observations to the result
APs empty and full. Section 6.5.4 considers the larger implications of the case study
in the context of the tightness condition.
6.5.1 The Result AP Is Predicated of a Surface
When predicated of a surface, dry indicates that the surface has no liquid on it, as in
a dry floor/counter. This sense of dry is found in resultatives where it is predicated
of a surface.
(28) a. The waitress comes back, wiping the silverware dry with a cloth napkin
before laying it out. (Jaffe, Michael Grant. 1996. Dance real slow, 24.
New York: Farrar Straus Giroux)
b. He took the towel from her hands and patted her face dry. (Meyers,
Annette. 1997. The groaning board, 266. New York: Doubleday)
This state is brought about by removing any liquid from the surface. This in turn
is usually accomplished through contact with and motion over the surface, that is,
by an action directed at the surface. The precise action depends on the nature of the
surface and often involves an instrument designed to absorb liquid such as a sponge,
dishcloth, or towel. Thus, dishes are usually dried with a dishcloth, while a face is
usually dried with a towel. The verbs attested in resultatives with this sense of dry
lexicalize such actions, as in (29). (Some of these actions can also be carried out on
a dry surface (e.g., pat, rub, wipe).)
(29) blot, brush, dab, lick, rub, spin, wipe, . . .
Unsurprisingly, as these actions are directed at the surface whose state is being
changed, they are lexicalized by verbs which take the surface as object. Thus, the
postverbal NP is understood as both a participant in the action denoted by the verb
and the holder of the result state, giving rise to a selected NP resultative. In such
resultatives, then, the causing and caused events share a participant. Further, they
involve direct causation in the strong sense. The entity that changes state is directly
manipulated in the causing event, and the only intermediary entity is the instrument,
if any, used in the causing event to facilitate bringing about the result state, such as
the napkin in (28a) or the towel in (28), but such entities do not qualify as causers;
see fn. 16.
6.5.2 The Result AP Is Predicated of a Container
When predicated of a container, dry indicates that the container is empty of liquid,
as in a dry well/tank or even dry throat/lungs, where body parts are being taken
to be containers.18 This sense is found in resultatives where the result AP dry is
predicated of a container which is fulfilling its function, such as a teapot or kettle.
(30) a. Having . . . drunk the teapot dry . . . (Dark, Eleanor. 1986 [1959].
Lantana Lane, 94. London: Virago)
b. One of them [=tea kettles] must’ve whistled itself dry . . . (Conant, Susan
J. 1995. Ruffly speaking, 76. New York: Doubleday)
This state is usually brought about by removing liquid from the container. Such
actions are often directed at the liquid in the container—the container’s contents—
rather than at the container itself. Actions of two types can bring about this state,
and the relevant type depends on the nature of the container.
With a prototypical container or something construed as such, the actions
are designed to (re)move the liquid, perhaps through the use of an appropriate
18 McIntyre (2004: 546) notes that dry has the ‘empty’ sense in nonselected NP resultatives, but
does not mention that this sense is predicated of containers, nor discuss how this property plays
into the licensing of nonselected NP resultatives. He writes that dry in this sense cannot appear
after a copula, but such uses are actually found, as in After three years of drought, the well is dry.
Even The cellar/boiler is dry, which he finds unacceptable, are well-attested on the web.
Nevertheless, the verb dry is almost exclusively used in descriptions of changes involving
the dryness that holds of surfaces. For instance, Pat dried the tub is unambiguous and can
only mean that the tub’s surface was dried and not that the tub no longer has water in it. Web
searches reveal only very few examples with the intended meaning (e.g., along the lines of A
supernatural occurrence had dried the well). Given this, it is unsurprising that attested nonselected
NP uses of dry are not paraphrasable with the verb dry. For example, although ‘She dried her
face’ can paraphrase the selected NP resultative (28), ‘We dried the teapot’ does not paraphrase
the nonselected NP resultative (30a). However, this is not a general property of nonselected NP
resultatives. Both the pour–full example (23) and the crow–awake example (26a) can be given
such paraphrases, as in ‘Audrey filled the mug’ and ‘The rooster woke me’. This issue deserves
further investigation.
202 B. Levin
instrument (e.g., a pump); thus, they are lexicalized by verbs that take the liquid
as their object, as in (31).
(31) boil, drain, drink, pump, slurp, suck, . . .
In such examples the liquid (e.g., tea in (30a)) is the understood object of the verb
lexicalizing the action denoted in the causing event; the NP denoting the container
(e.g., the teapot in (30a)) is not the object of this verb and, thus, qualifies as a
nonselected NP. However, the container and the liquid being removed from it are
spatially contiguous so that removing the liquid—the manner—brings about the
state of dryness in the container—the result. Thus, there is a strong association
between the causing and caused event. It is by virtue of this relation that these
resultatives, despite their nonselected NP, meet the tightness condition. As in some
examples discussed in Sect. 6.4, the causing event brings about the result state
although the container is not a clear participant in the causing event: drinking, for
instance, involves a drinker and a liquid.
Instances of the second type involve a body part such as the lungs or vocal tract,
or even the body itself, which can also be construed as a container.
(32) a. Davina and I erupted from the knife-sharp grass, shrieking our lungs dry
. . . (Meyers, Margaret. 1995. Swimming in the Congo, 29. Minneapolis:
Milkweed)
b. After the funeral yesterday she thought she’d cried herself dry. (Sefton,
Maggie. 2005. Knit one, kill two, 7. New York: Berkley)
The actions involved in ‘drying’ the body part involve the secretion (usually by a
human) of a substance or the emission of a substance or sound—actions which may
result in a body (part) becoming dry (i.e. empty).
(33) boil, cry, shriek, sweat, talk, whistle, . . .
The secretion/sound is usually unexpressed in resultatives, but outside of such
constructions, it may be the object of the verb lexicalizing the action (e.g., shriek
an ear-shattering shriek, cry a mournful cry), even though the verb is often taken
to be intransitive.19 In (32a) the nonselected NP is a body part, but in other
instances, including (32b), this NP may be a reflexive pronoun which stands in
for the whole body; see Sect. 6.6.2 for more discussion of reflexive pronouns as
the nonselected NP. In these examples the result is often an incidental, although
necessary consequence of the action denoted in the causing event. The causer in the
causing event may also be an inanimate entity, as in (30b), repeated as (34): here an
inanimate entity which is viewed as having an internal energy source is portrayed as
drying out its insides by emitting steam—a direct result of the substance emission
which accompanies the whistling.
19 Dan Lassiter (p.c.) suggests that perhaps (32a) is actually a surface construal because the
interpretation involves a ‘subjective feeling of dryness around the sides of the lungs’.
(34) One of them [=tea kettles] must’ve whistled itself dry . . . (Conant, Susan J.
1995. Ruffly speaking, 76. New York: Doubleday)
In each instance, the action, typically when done excessively, leads to the relevant
result. And, again, there is a relation of spatial contiguity between the container and
the understood substance whose removal from the container leads to its dryness.
6.5.3 Results APs Headed by the Adjectives Empty and Full
Since states of containers can be altered by affecting their contents, we predict that
result phrases headed by adjectives that are near-synonyms or antonyms of container
dry like empty and full should pattern like it in resultatives; that is, they should be
found in nonselected NP resultatives with the container as the postverbal NP.20 They
are indeed found in resultatives comparable to those with container dry, as in (22),
repeated as (35), and (36).
(35) Maxey stood up to get a glass and pour it full of milk. (Cail, Carol. 1995.
Unsafe keeping, 146. New York: St. Martin’s)
(36) Tom waggled the bottle at me, and swigged it empty when I declined.
(Boneham, Sheila Webster. 2013. The money bird, 11. Woodbury, MN:
Midnight Ink)
The adjectives full and empty, unlike dry, may hold of solids as well as liquids.
Thus, a wider set of actions can be performed on containers to achieve these states.
Some directly affect the container, giving rise to selected NP resultatives with full
and empty. For instance, the corpus includes several examples of ‘shake NP empty’,
among them (37). In this example, the container is moved back and forth, and
its motion causes its contents to fall out, leaving it empty. Thus, the container
participates in both events.
(37) She knelt before him and taking one of his hands in hers, shook the bag
empty. (Patterson, Pat. 2002. Spirit path, 94. Lincoln, NE: iUniverse)
Also among the selected NP resultatives with full are several instances of ‘fill NP
full’. Although this result AP seems to reiterate the final state associated with a
change of state verb, in several of the examples full is modified, so the result AP
provides additional information relevant to resolving any vagueness or imprecision
in the interpretation of full (Kennedy 2007; Lasersohn 1999).
(38) ‘Don’t fill the bags too full.’ (MacPherson, Rett. 1997. Family skeletons, 64.
New York: St. Martin’s)
20 Kaufmann & Wunderlich (1998: 32) note the importance of the container–contents relation in
discussing a German nonselected NP resultative whose result AP is the German counterpart of full.
204 B. Levin
(39) I filled a serving bowl with cat food, set it on the floor, along with the biggest
pot I owned, filled full of water. (Thornton, Betsy. 2001. High lonesome
road, 166. New York: Thomas Dunne)
6.5.4 The Importance of Contiguity to the Well-Formedness of

Resultatives
A resultative is used to convey a change in the result state of the postverbal NP.
The dry case study shows that the nature of this NP affects the type of action
needed to effect the relevant result state. Certain states in an entity, such as dryness
can be brought about by acting directly on that entity or by acting on an entity
that is contiguous to that entity. This difference, in turn, is behind why selected
vs. nonselected NP resultatives are found with dry: nonselected NP resultatives
emerge since states of containers can be altered by affecting their contents.21 Thus,
due to the different ways that dryness is brought about in surfaces and containers,
NPs denoting surfaces and containers are found in different types of resultatives.22
The sensitivity of resultatives to whether dry is predicated of a container or a
surface is not surprising. The notion of ‘container’ is privileged conceptually, if not
linguistically. First, consider the phrase a cup of milk; although taken literally it
refers to ‘a cup filled with milk’, it may also be used to refer to ‘a quantity of milk
equal to a cup’, whether or not the milk itself is in a cup. In this second instance,
the container is really referring to its contents (Apresjan 1973: 7; Ostler & Atkins
1992: 90). Further, spatial relational terms comparable to the English preposition
in, which are used to refer to a figure contained in a ground, are present in even
small inventories of such terms (Levinson et al. 2003). Further, this spatial relation
is defined functionally rather than purely geometrically. Thus, English in and some
of its counterparts in other languages can be used to describe both partial and full
inclusion of the figure by the ground as long as containment is present. That is, they
apply equally to flowers in a vase, which are unlikely to be fully contained by the
21 A question that arises is why these manners of drying a container are rarely lexicalized by a
verb with the container as object. I am aware of only one such verb, drain, and it allows either
the contents (e.g., drain the water) or the container (e.g., drain the tub) as object; see also fn. 3.
In contrast, although the state of a surface is often affected by affecting what is on it (typically
unwanted material or detritus), the verbs lexicalizing the relevant activities such as sweep or wipe
tend to take the surface as their object (Levin & Rappaport Hovav 1991; Rappaport Hovav & Levin
1998). This question relates to possible verb meanings, and I leave it aside.
22 Having isolated the conditions which determine whether a resultative with the adjective dry
predicated of a container will have a selected vs. nonselected NP, we can now predict the conditions
which would give rise to a nonselected NP resultative with the dry predicated of surfaces. If the
dryness of a surface could be altered by an action directed at some entity contiguous to the surface,
then we might expect a nonselected NP resultative. I have not found such resultatives.
vase, and an apple in a bowl, where full containment is most often the case (Feist
2004; Levinson et al. 2003).
The importance of contiguity between contents and container to ensuring that the
tightness condition between the causing and caused subevents is satisfied in certain
nonselected NP resultatives is brought out by an observation made by Kratzer. She
(2005: 196) points to an unavailable nonselected NP interpretation of the German
counterpart of They drank the teapot empty. As in English, this sentence can describe
drinking the contents of the teapot, thus causing it to be empty. However, Kratzer
notes that this sentence cannot be used to describe the following scenario: drinking
from a well to such an extent that there is no water left in the well to fill the teapot,
so it remains empty.23 Kratzer uses this contrast to refine her formal definition of
direct causation, including the definition of a causal chain. What she does not notice
is that in the unavailable interpretation the unexpressed NP—the water in the well—
does not denote the contents of the teapot, so there is no contiguity relation. The
generality of Kratzer’s observation can be demonstrated using (35), repeated here.
(40) Maxey stood up to get a glass and pour it full of milk. (Cail, Carol. 1995.
Unsafe keeping, 146. New York: St. Martin’s)
In this sentence, we understand the liquid that is poured to be the liquid that fills
the glass. This sentence cannot receive an interpretation where Maxey pours some
liquid onto the floor, while the glass somehow gets full of milk. Such a scenario too
would lack contiguity.
If contiguity can contribute to meeting the tightness condition, we might ask
whether the contiguity relation could play out in the other direction in resultatives;
that is, could manipulating a container affect its contents? This possibility would
manifest itself in a nonselected NP resultative whose verb describes an action on
a container while its result state holds of the contents of this container, denoted by
the postverbal NP. In fact, the crank–flat example (27), repeated as (41), is precisely
such an example.
Here, the action is directed at the bed—the container—but it is the person in the
bed—the contents—whose state change is relevant.
The contiguity relation, then, is essential to the well-formedness of certain
nonselected NP resultatives; that is, it allows the tightness condition to be met.
The reason most likely is that due to contiguity, manipulating the contents of a
container is in some sense manipulating the container or, alternately, manipulating
the container is, in some sense, manipulating its contents. Thus, there is a tight
23 As Dan Lassiter (p.c.) points out, the unavailable interpretation is not a typical resultative
interpretation because the teapot remains empty rather than becomes empty. If this is a worry,
then consider the following scenario, which also cannot be the described by the resultative: people
drink all the water in the well, so no water is left causing people to drink all the water in the teapot.
206 B. Levin
relation between the causing and caused subevents, so that ‘direct causation’ as it is
prototypically understood is instantiated in such examples.
6.6 A Second Case Study: Result APs Headed by the

Adjective Awake
This section explores resultatives with the result AP awake, another result phrase
prevalent in both selected and nonselected NP resultatives. Although a sleeper may
awake naturally, this state may also be brought about via actions that impinge on the
sleeper. In many instances the causer—the doer of the relevant action—is someone
other than the sleeper, but in some instances a sleeper might wake him/herself.
There are several ways in which the state of wakefulness ensues. First, a causer
can cause a sleeper to awake through physical manipulation. Second, a causer can
cause a sleeper to awake by making a sound—usually a loud or high-pitched sound;
relatedly, sometimes someone can cause a sleeper to awake by simply looking at
the sleeper. Third, a sleeper might wake him/herself through an involuntary bodily
process or through a deliberate activity intended to restore wakefulness; here the
sleeper is the causer. Considering these scenarios in the context of the case study of
dry leads to predictions about whether each one would be expected to be expressed
using a selected or nonselected NP resultative. These predictions are introduced and
verified in the following subsections.
6.6.1 Selected NP Resultatives
The first prediction is that when the causing event involves a causer directly
manipulating the sleeper, who then awakes, the scenario should be expressed by a
selected NP resultative as the causing and caused events share a participant. Attested
selected NP resultatives with the result AP awake indeed conform to the prediction;
they involve causers waking the sleeper through some sort of physical contact or
manipulation, as in (42). Thus, their causing events usually involve impact or force
exertion verbs, as in (43), as they lexicalize such actions.
(42) a. Last night, the dog poked me awake every hour to go outside. (Dunford,
Gary. 1994. Charity’s for the birds. The Toronto Sun, November 27, p. 6)
b. . . . the moment he was deeply asleep Vinck was tugging him awake . . .
(Clavell, James. 1980. Shogun, 652. New York: Atheneum)
(43) bump, hug, jerk, kiss, poke, slap, tickle, tug, . . .
These are selected NP resultatives since these verbs take the sleeper—the event
participant that the result AP is predicated of—as their object.
6.6.2 Nonselected NP Resultatives
I turn now to a second prediction: when a causer emits a sound or speech which
impinges on the sleeper, the scenario should be expressed by a nonselected NP
resultative since only the emitter or speaker is necessarily involved in the sound
emission or speaking event. Indeed, many attested nonselected NP resultatives with
the result AP awake involve a causer, either animate or inanimate, emitting a sound
or speech.
(44) a. When the roosters that scratch in the yard of Brastagi’s best hotel crowed
me awake that dawn a few months ago, I knew it was destined to be a
memorable day . . . (Robbins, Tom. 1986. True adventure: crowned king
of the cannibals. The New York Times, March 16, Sect. 6, Part 2, p. 8)
b. Half an hour later I had finished my day’s work, shouted Howard awake,
and headed to my truck . . . (Andrews, Sarah. 1994. Tensleep, 143. New
York: Otto Penzler)
Such resultatives involve sound emission verbs (e.g., crow, jangle) and manner of
speaking verbs (e.g., scream, shout)—verbs of the types that lexicalize these actions,
as in (45).
(45) bark, buzz, clang, crow, jangle, scream, shout, shrill, . . .
Furthermore, the attested verbs tend to be those members of the relevant classes
that involve loud or high-pitched sounds, precisely the types of sound most likely to
wake someone.
Such verbs are typically intransitive, with the emitter or speaker as subject, and
to the extent they allow an object, it denotes a sound or speech (e.g., shout an
answer, scream a fierce animal scream). As the verb’s object is not the sleeper,
in such resultatives, then, the sleeper is expressed as a nonselected postverbal NP.
Although the sound/speech is left unexpressed, it does impinge on the sleeper: the
sound waves or the speech move across space and make ‘contact’ with the sleeper in
an abstract sense. In (44b) the understood entity, the speaker’s words, make ‘contact’
with the entity denoted by the postverbal NP, the sleeper. However, it may seem odd
to call the sleeper a ‘recipient’ of the sound: in some instances the causing event may
not have an intended recipient, either because the emitter is inanimate or because the
emitter, although animate, may simply emit the sound for its own sake, as in (44a).
Concomitantly, some of the attested verbs describe actions that may be directed at
someone or something (e.g., shout at Howard), but that may, as in (44b), or may not,
as in (44a), be the case in a given resultative. The sleeper, then, seems to qualify as
a participant in the causing event at best only in a weak sense, a point raised in
Sect. 6.4.
The second prediction can be generalized from events of sound emission and
speaking—the domain of sound—to events of light emission and looking—the
208 B. Levin
visual domain.24 The corpus includes a few resultatives with the result AP awake
with the looking verb stare.
(46) Even Charlotte had been unable to stare her awake as she usually did.
(McGown, Jill. 2004. Unlucky for some, 203. New York: Ballantine)
In such examples a causer directs his or her gaze at a sleeper, who senses the gaze
and awakes. Further, web searches reveal a few resultatives with the result AP awake
with light emission verbs (e.g., flash, shine), as in (47).
(47) My Bonamassa warning light has just flashed me awake . . .
(http://theafterword.co.uk/eyes-wide-open-aynsley-lister/; accessed April
10, 2019)
They describe scenarios in which emitted light impinges on and wakes a sleeper,
as in (47). Such examples are much less frequent than comparable resultatives
with sound emission verbs probably because sound is more effective than light
as a means of waking a sleeper. These scenarios are analogous to those discussed
involving sound emission and speaking events, and it is not surprising that they too
are described with nonselected NP resultatives.
A third prediction involves a scenario where the sleeper is both the causer and
the one that awakes. In such a scenario a sleeper wakes him/herself through a
bodily process or other activity, which although not explicitly directed at the sleeper,
nevertheless disrupts sleep. In fact, the relevant actions mostly involve a single event
participant. Thus, although the causer and the sleeper refer to the same person,
they constitute distinct event participants. Such scenarios, too, would be expected
to be expressed via nonselected NP resultatives. Indeed, a small number of such
nonselected NP resultatives are attested.
(48) a. Estelle hugged him, but . . . he squirmed down, standing by her knees as
he blinked himself awake. (Havill, Steven F. 2008. The fourth time is
murder, 221. New York: St. Martin’s)
b. Yarborough was ‘a biblio-holic’ and history buff who ‘read himself
awake each morning’. (Gonzalez, John. 1996. Hundreds mourn Yarbor-
ough. The Fort Worth Star-Telegram, January 31, Texas Section, p. 17)
The hallmark of these resultatives is a reflexive pronoun as the postverbal NP, which
indicates the coreference relation between the causer and sleeper.25
24 Ithank Cleo Condoravdi (p.c.) for asking about this extension.

25 Nonselected NP resultatives with reflexive pronouns as the postverbal NP have received special
attention because some early researchers took the pronouns simply to have a syntactic function:
they allow a result phrase to be predicated of the subject, allowing the Direct Object Restriction—
the generalization that result phrases must be predicated of (underlying) objects (Simpson 1983:
145; Levin & Rappaport Hovav 1995)—to be satisfied. However, on most accounts, including
the one outlined here, the reflexive pronoun and the subject are taken to instantiate distinct event
participants, even though their referents are the same.
In some of the attested examples sleepers wake themselves through a bodily

process, as in (48a); often these are involuntary processes which are not under the
sleeper’s control. In others the causing event involves a deliberate activity on the
sleeper’s part, as in (48b). Concomitantly, the examples involve verbs describing
bodily processes whose occurrence may cause a sleeper to wake up, as in (49a),
as well as other activities, including screaming or shouting, that someone might
involuntarily engage in when sleeping—say, due to a nightmare—that bring back a
state of wakefulness, as in (49b).
(49) a. blink, cough, puff, snort, stretch, . . .
b. read, scream, shout, . . .
As with the second prediction, almost all the relevant verbs are intransitive, and to
the extent they allow an object, it is a cognate object (e.g., blink a little blink, cough
a loud cough) or denotes a sound or speech (e.g., the content of the communication,
shout an answer). Again, the sleeper is not the object of these verbs; rather, there
is often an understood entity—the sound or cognate object—which impinges on the
sleeper, just as it does in examples falling under the second prediction.
To conclude this section, both selected and nonselected NP resultatives with the
result AP awake maintain tight relations between the causing and caused events,
either via direct physical manipulation of the holder of the result state or some other
more abstract type of impingement on this entity.
6.7 The Significance of Unavailable Interpretations of

Resultative Constructions
In perusing the verb–AP combinations in the corpus, it becomes evident that there
are certain scenarios that they are not used to describe. Specifically, certain verb–
AP combinations are found with a selected NP resultative interpretation, but are
not attested with certain potential nonselected NP interpretations. I argue that the
relevant verb–AP combinations simply cannot have the relevant interpretations.
A closer look at these scenarios shows that they involve causal chains with true
intervening causers. Thus, they provide support for a formulation of the tightness
condition that does not allow for intervening causers, as in the definitions of direct
causation cited in Sect. 6.1.
Before considering specific examples, I point out that it is not a priori impossible
for a given verb–AP combination to be found in both selected and nonselected NP
resultatives, so the lacuna I present cannot be attributed to this possibility. Although
my corpus does suggest that certain verb–AP combinations are more likely to be
found in selected NP resultatives and others in nonselected NP resultatives, there
are nevertheless a few combinations which are found in both. One such combination
is rub–raw. This combination receives a selected NP interpretation in (50) and a
nonselected NP interpretation in (51).
210 B. Levin
(50) The salt [in the ocean water] rubbed their feet raw. (Alvarez, Lizette. 2014.
For Cubans in Miami, the gulf to their homeland narrows. The New York
Times, December 21, p. 21)
(51) . . . the author had rubbed her hands raw while scrubbing the hems of her
older sisters’ long dresses . . . (Hill, Marion Moore. 2008. Death books a
return, 238. Corona del Mar, CA: Pemberley Press)
In (50), the water directly rubs some people’s feet causing their rawness. In (51)
the author is scrubbing—and, thus, rubbing—the dresses with her hands, and this
activity causes her hands to be raw. She is not rubbing her hands directly.
Certain verb–AP combinations are attested in the resultative corpus only with
selected NP interpretations. Thus, attested resultatives of the form ‘kick NP
open/closed/shut’ all receive a selected NP interpretation.26 That is, (52) is under-
stood as ‘Sam makes contact with the door with his foot, causing it to open’. On
this interpretation a causer directly comes into contact with the entity denoted by
the postverbal NP using a body part.
(52) Sam kicked the door open.
Resultatives of the form ‘kick NP open/closed/shut’ are never attested with a
nonselected NP interpretation. For instance, (52) never means ‘Sam kicks a ball
which hits the door, causing it to open’, and my intuition is that this interpretation is
simply impossible. On this interpretation, a projectile set in motion by a causer
comes into contact with another entity, changing its state. That is, kick has an
understood, but unexpressed object, a ball, so that the door is now a nonselected
NP. Such a scenario seems plausible, and might even have seemed a candidate for
description by a resultative.27
I propose that the relevant interpretation of (52) is unavailable because it violates
the tightness condition. As I now argue, the ball that figures on this interpretation
qualifies as a cause; thus, it would be an intervening causer between Sam’s kicking
action and the change of state in the door. A launched ball is what Kearns (2000:
241), drawing on Cruse (1973: 19–20), terms a projectile: an entity that moves due to
the kinetic energy it receives from an imparted force; see also Wolff et al. (2010: 96).
Such an entity may itself impart this force to another entity through contact, just like
other causes—agents, natural forces, and instruments with their own energy source
(i.e. the ‘intermediary’ instruments of fn. 16)—may (Wolff et al. 2010: 96). Further,
projectiles pattern with other causes with respect to common diagnostics for causes
26 Iam assuming that there is a transitive use of kick, as in The horse kicked the groom, as well as
an intransitive use, as in The horse kicked (e.g., moved its leg in a particular fashion) and that the
examples relevant to this section involve transitive kick.
27 The resultative shoot NP dead may appear to be just such an example, but it is not. It does
not have an interpretation comparable to the missing interpretation of (52) since the meaning
lexicalized by shoot involves firing a gun and not a bullet; it is the latter which would be the
analogue of the ball in (52).
(Cruse 1973: 19–20). First, they pass the ‘what X did’ test, just as agents, natural
forces, and autonomous machines do.28
(53) a. What the ball did was break the window.
b. What Cameron/the wind/the crane did was break the window.
Second, they may be subjects of certain transitive verbs, again patterning like agents,
natural forces, and autonomous machines.
(54) a. The ball broke the window.
b. Cameron/the wind/the crane broke the window.
As a second example, consider (55), which involves the verb–AP combination
push–open. It too is attested in the corpus with a selected NP interpretation. For
instance, a selected NP interpretation is clearly available for (55): ‘Tracy pushed
(on) the door, causing it to open’. But, the sentence cannot describe a situation
where Tracy pushes on a red button that sets a mechanism in operation that opens
the door. This scenario would involve a nonselected NP interpretation.
(55) Tracy pushed the door open.
Once again, the missing interpretation involves an intervening causer. The button
here serves as a (proxy for a) mechanism with its own energy source, qualifying as
a cause. Again, it passes the diagnostics for a cause, as shown in (56).
(56) a. What the red button did was open the door.
b. The red button opened the door.
It is worth noting that the missing interpretations discussed in this section are
qualitatively different from the missing interpretation of Kratzer’s (2005: 196)
example They drank the teapot empty, discussed in Sect. 6.5.3. Kratzer’s example
lacks an interpretation not because the unexpressed argument is an intervening
causer, but because there is too large and too loose a ‘gap’ in the causal chain.
In concluding this section, I acknowledge the potential limitations of drawing
conclusions from non-occurring data and intuitions about unavailable interpreta-
tions, but I believe that these lacunae are real. Interestingly, the missing inter-
pretations involve an intervening causer. The absence of resultatives with such
interpretations, then, supports proposals that the tightness condition includes a no-
intervening-cause condition. Including such a condition still allows for the types of
link between subevents found in the resultatives examined in the dry and awake case
studies, which are even tighter. Once again, however, nonselected NP resultatives
provide a source of insight into the tightness condition.
28 DeLancey (1984: 203) uses the example The axe broke the window to show that a facilitating
instrument is odd as the subject of break. This sentence can describe an axe which falls off a shelf
onto a window, a scenario where it qualifies as a projectile, but it would not be used to describe a
scenario where someone uses an axe to hit a window, except perhaps in a contrastive context.
212 B. Levin
6.8 Conclusion: On the Well-Formedness of Transitive

Resultatives
In the introduction I asked: What is the best way to characterize the causative
relation between the causing and caused events described by a resultative? Returning
to this question, a goal of this paper was homing in on the relation between the
subevents in selected and nonselected NP resultatives in order to gain insight into the
tightness condition on concealed causatives, and, thus, into the nature of causation
as it matters to language. A related question under consideration was: How are the
causing and caused events able to be integrated into a resultative when the postverbal
NP is not selected by the verb representing the causing event? This question arises
because the relation between the causing and caused events might seem to be looser
precisely because of the lack of selection. I now review what the case studies
revealed before presenting some more general concluding remarks.
6.8.1 Selected NP Resultatives
In selected NP resultatives the result state in the entity denoted by the postverbal
NP is one that a causer, perhaps using an instrument, brings about by acting directly
on this entity via physical manipulation. Concomitantly, verbs denoting actions
involving contact with a surface or exertion of a force on an entity are prevalent
in selected NP resultatives. The choice among these semantic types depends on the
nature of the result state. There is no ‘intermediate entity’ (except perhaps for an
enabling—or facilitating—instrument) and, thus, no ‘intervening causer’ (let alone,
an ‘intervening event’ introduced by this causer): the causer directly affects the
entity denoted by the postverbal NP, bringing about the result state. Thus, in such
resultatives the tightness between the subevents falls under the most prototypical
understanding of direct causation as manipulation of a physical object by a causer
to cause a change of state in it.
6.8.2 Nonselected NP Resultatives
In nonselected NP resultatives the result state in the entity denoted by the postverbal
NP is again one that a causer, perhaps using an instrument, brings about by acting in
some way that causes a change of state in the entity denoted by the postverbal NP.
This may happen in several ways, as the case studies show, with different types of
understood entities mediating the relation between the causing and caused events. I
focus on two.
The action may be on an understood entity which is spatially contiguous with the
entity denoted by the postverbal NP, e.g., in a contents–container relation to it. In
(57) the understood entity, tea, is contained in the entity denoted by the postverbal
NP, while in (58) the understood entity, the bed, contains the entity denoted by the
postverbal NP.
(57) Having . . . drunk the teapot dry . . . (Dark, Eleanor. 1986 [1959]. Lantana
Lane, 94. London: Virago)
Such actions typically involve physical manipulation of this understood entity, and,
concomitantly, the verb denotes such an action. Due to their contiguity, when the
causer acts on the understood entity, the causer also acts on the entity denoted by
the postverbal NP.
In other instances, the action may involve the production of a sound, speech,
light, or a gaze, and, concomitantly, the verbs involved are verbs of substance, light,
or sound emission, manner of speaking, or looking. There is again an understood
entity: the emitted sound, speech, light, or gaze which impinges on the entity
denoted by the postverbal NP.
(59) He had set an alarm, which rang at five thirty the following morning, shrilling
them both awake. (Pilcher, Rosamunde. 1984. Voices in summer, 116. New
York: St. Martin’s)
More generally, there is abstract ‘contact’ between the understood entity and the
entity that changes state. However, despite this impingement it is not always clear
whether the entity denoted by the postverbal NP truly qualifies as a participant in
the causing event.
In instances of both types there is ‘contact’ with the entity denoted by the
postverbal NP. However, the understood entity does not constitute an ‘intervening
causer’ (or introduce an intervening event) since it does not have any internal
energy source of its own (Wolff et al. 2010). Thus, despite the nonselected NP
such resultatives satisfy the tightness condition, including meeting the previously
proposed direct causation conditions.
6.8.3 Back to Causation
This study has explored transitive resultatives in order to shed light on the nature of
the tightness condition on concealed causatives. It confirms that something like the
notions of direct causation found in the literature are indeed important to the well-
formedness of both selected and nonselected NP resultatives. Intervening causers
disrupt tightness. Physical manipulation of event participants falls under tightness,
as do some other contiguity relations between event participants; these are relevant
to certain resultatives with the result APs dry, empty, flat, and full. The case studies
further show that more abstract relations that seem to be generalizations of these
214 B. Levin
notions matter as well, such as the ‘impingement’ or abstract ‘contact’ relevant to

certain resultatives with the result AP awake.
However, these generalizations are largely based on case studies of transitive
resultatives whose result APs are headed by awake or dry. Thus, it is imperative to
carry out studies of a wider range of result phrases to confirm the generalizability
of these points. Preliminary case studies of another half-dozen common result
APs suggest that they are. However, the observed relations may in part reflect the
focus both here and in these additional studies on resultatives whose result phrases
describe physically instantiated states. Although the paper’s conclusions should
extend to result phrases that describe purely mental states, other more abstract
relations may be at play. Further, resultatives with PP result phrases were set aside
to limit the scope of the investigation. The insights that emerged from this study of
resultatives with AP result phrases should carry over to resultatives with PP result
phrases, and this expectation too awaits future confirmation.
Despite these limitations, this paper shows that an in-depth examination of
the make-up of transitive resultatives has much to tell us about the nature of the
causative relations between events that are linguistically privileged, and, for this
reason, most likely also relevant to non-linguistic forms of reasoning. Future fine-
grained studies that address this study’s limitations should illuminate the nature of
causation even further.
Acknowledgements I thank Dan Lassiter, Malka Rappaport Hovav, Tim Sundell, and three
reviewers for valuable comments on earlier versions of this paper. I have also presented this
material in several venues, and I am grateful to the audiences for their feedback.
References
alternations. Oxford: Oxford University Press.
Alexiadou, A., Martin, F., & Schäfer, F. (2017). Optionally causative manner verbs: When implied
results get entailed. Handout. Roots V. Queen Mary University of London/University College
London.
Apresjan, J. D. (1973). Regular polysemy. Linguistics, 142, 5–32.
Beavers, J. (2017). The spray/load alternation. In M. Everaert & H. C. van Riemsdjik (Eds.), The
Wiley Blackwell companion to syntax (2nd ed.). Oxford: Blackwell. https://doi.org/10.1002/
9781118358733.wbsyncom066.
Brisson, C. (1994). The licensing of unexpressed objects in English verbs. In Papers from the 30th
Regional Meeting of the Chicago Linguistic Society (CLS) (Part 1: The main session, pp. 90–
102). Chicago: Chicago Linguistic Society.
Carrier, J., & Randall, J. H. (1993). Lexical mapping. In E. Reuland & W. Abraham (Eds.),
Knowledge and language. Vol. II: Lexical and conceptual structure, pp. 119–142). Dordrecht:
Kluwer.
Chierchia, G. (2004). A semantics for unaccusatives and its syntactic consequences. In A.
Alexiadou, E. Anagnostopoulou, & M. Everaert (Eds.), The unaccusativity puzzle: Explorations
of the syntax-lexicon interface (pp. 22–59). Oxford: Oxford University Press.
Croft, W. A. (1991). Syntactic categories and grammatical relations. Chicago: University of

Chicago Press.
Cruse, D. A. (1973). Some thoughts on agentivity. Journal of Linguistics, 9, 11–23.
DeLancey, S. (1984). Notes on agentivity and causation. Studies in Language, 8, 181–213.
den Dikken, M., & Hoekstra, E. (1994). No cause for a small clause? (Non-)arguments for the
structure of resultatives. Groninger Arbeiter zur germanistischen Linguistik, 37, 89–105.
Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht: Reidel.
Feist, M. I. (2004). Talking about space: A cross-linguistic perspective. In K. D. Forbus,
D. Gentner, & T. Regier (Eds.), Proceedings of the Twenty-Sixth Annual Conference of the
Cognitive Science Society (pp. 375–380). Mahwah: Lawrence Earlbaum.
Fodor, J. A. (1970). Three reasons for not deriving kill from cause to die. Linguistic Inquiry, 1,
429–438.
Goldberg, A. E. (1991). It can’t go up the chimney down: Paths and the English resultative. In
Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society (BLS) (pp. 368–
378). Berkeley: Berkeley Linguistics Society.
Goldberg, A. E. (1995). Constructions: A construction grammar approach to argument structure.
Chicago: University of Chicago Press.
Goldberg, A. E., & Jackendoff, R. (2004). The English resultative as a family of constructions.
Language, 80, 532–568.
Grône, M. (2014). Les résultatives de l’anglais: une étude de leur syntaxe et de leur productivité à
l’aune de la sémantique lexicale et de la pragmatique. Ph.D. thesis, Université Paris Diderot—
Paris 7.
Hall [Partee], B. (1965). Subject and object in English. Ph.D. thesis, MIT, Cambridge, MA.
Hay, J., Kennedy, C., & Levin, B. (1999). Scalar structure underlies telicity in ‘degree achieve-
ments’. In Proceedings of Semantics and Linguistic Theory (SALT) (Vol. 9, pp. 127–144).
Ithaca: Cornell Linguistics Circle Publications.
Hoekstra, T. (1988). Small clause results. Lingua, 74, 101–139.
Hoekstra, T. (1992a). Aspect and theta theory. In I. M. Roca (Ed.), Thematic structure: Its role in
grammar (pp. 145–174). Berlin: Mouton de Gruyter.
Hoekstra, T. (1992b). Small clause theory. Belgian Journal of Linguistics, 7, 125–151.
Hoekstra, T. (1992c). Subjects inside out. Revue Québécoise de Linguistique, 22, 45–75.
Iwata, S. (2014). Aspect and force dynamics: Which is more essential to resultatives? English
Linguistics, 31, 234–263.
Jackendoff, R. S. (1990). Semantic structures. Cambridge, MA: MIT Press.
Kaufmann, I., & Wunderlich, D. (1998). Cross-linguistic patterns of resultatives. Ms. University
of Düsseldorf.
Kearns, K. (2000). Semantics. New York: St. Martin’s.
Kennedy, C. (2007). Vagueness and grammar: The semantics of relative and absolute gradable
predicates. Linguistics and Philosophy, 30, 1–45.
Kennedy, C., & Levin, B. (2008). Measure of change: The adjectival core of verbs of variable
telicity. In L. McNally & C. Kennedy (Eds.), Adjectives and adverbs in semantics and discourse
Koontz-Garboden, A. (2009). Anticausativization. Natural Language and Linguistic Theory, 27,
77–138.
Kratzer, A. (2005). Building resultatives. In C. Maienborn & A. Wöllstein (Eds.), Event arguments:
Foundations and applications (pp. 177–212). Tübingen: Niemeyer.
Lasersohn, P. (1999). Pragmatic halos. Language, 75, 522–551.
Levin, B. (1999). Objecthood: An event structure perspective. In Proceedings of the 35th Annual
Meeting of the Chicago Linguistic Society (CLS) (Part 1: The main session, pp. 223–247).
Chicago: Chicago Linguistic Society.
Levin, B., & Rappaport Hovav, M. (1991). Wiping the slate clean: A lexical semantic exploration.
Cognition, 41, 123–151.
216 B. Levin
Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-lexical semantics
interface. Cambridge, MA: MIT Press.
Levin, B., & Rappaport Hovav, M. (1999). Two structures for compositionally derived events. In
Proceedings of Semantics and Linguistic Theory (SALT) (Vol. 9, pp. 199–223). Ithaca: Cornell
Linguistics Circle Publications.
Levin, B., & Rappaport Hovav, M. (2013). Lexicalized meaning and manner/result complemen-
tarity. In B. Arsenijević, B. Gehrke, & R. Marín (Eds.), Studies in the composition and
decomposition of event predicates (pp. 49–70). Dordrecht: Springer.
Levin, B., & Rappaport Hovav, M. (2014). Manner and result: A view from clean. In R. Pensalfini,
M. Turpin, & D. Guillemin (Eds.), Language description informed by theory (pp. 337–357).
Amsterdam: John Benjamins.
Levinson, S., Meira, S., & The Language and Cognition Group (2003). ‘Natural concepts’ in the
spatial topological domain—adpositional meanings in crosslinguistic perspective: An exercise
in semantic typology. Language, 79, 485–516.
Lewis, D. (1973). Causation. The Journal of Philosophy, 70, 556–567.
Marantz, A. P. (1984). On the nature of grammatical relations. Cambridge, MA: MIT Press.
verbs. In U. Sauerland & S. Solt (Eds.), Proceedings of Sinn und Bedeutung (SuB) 22. (Vol. 2,
pp. 107–124). https://semanticsarchive.net/sub2018.
Mateu, J. (2010). On the l-syntax of manner and causation. In M. Duguine, S. Huidobro, &
N. Madariaga (Eds.), Argument structure and syntactic relations: A cross-linguistic perspective
(pp. 89–112). Amsterdam: John Benjamins.
McCawley, J. D. (1968). Lexical insertion in a transformational grammar without deep structure.
In Proceedings of the 4th Annual Meeting of the Chicago Linguistic Society (CLS) (pp. 71–80).
Chicago: Chicago Linguistic Society.
McCawley, J. D. (1971). Prelexical syntax. In R. J. O’Brien (Ed.), Report of the 22nd Annual
Roundtable Meeting on Linguistics and Language Studies (pp. 19–33). Washington, DC:
Georgetown University Press.
McCawley, J. D. (1978). Conversational implicature and the lexicon. In P. Cole (Ed.), Syntax and
semantics 9: Pragmatics (pp. 245–259). New York: Academic.
McIntyre, A. (2004). Event paths, conflation, argument structure, and VP shells. Linguistics, 42,
523–571.
McKercher, D. A. (2001). The polysemy of with in first language acquisition. Ph.D. thesis, Stanford
University, Stanford, CA.
Neeleman, A., & van de Koot, H. (2012). The linguistic expression of causation. In M. Everaert,
M. Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the interface (pp. 20–
51). Oxford: Oxford University Press.
Ono, N. (1992). Instruments: A case study of the interface between syntax and lexical semantics.
English Linguistics, 9, 196–222.
Ostler, N. D. M., & Atkins, B. T. S. (1992). Predictable meaning shift: Some linguistic properties
of lexical implication rules. In J. Pustejovsky & S. Bergler (Eds.), Lexical semantics and
knowledge representation (pp. 87–100). Berlin: Springer.
Parsons, T. (1990). Events in the semantics of English. Cambridge, MA: MIT Press.
Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge,
MA: MIT Press.
Pustejovsky, J. (1991). The syntax of event structure. Cognition, 41, 47–81.
Rappaport Hovav, M. (2008). Lexicalized meaning and the internal temporal structure of events.
In S. Rothstein (Ed.), Crosslinguistic and theoretical approaches to the semantics of aspect
Rappaport Hovav, M. (2014). Lexical content and context: The causative alternation in English
revisited. Lingua, 141, 8–29.
Rappaport Hovav, M., & Levin, B. (1998). Building verb meanings. In M. Butt & W. Geuder
(Eds.), The projection of arguments: Lexical and compositional factors (pp. 97–134). Stanford,
CA: CSLI Publications.
Rappaport Hovav, M., & Levin, B. (2001). An event structure account of English resultatives.
Language, 77, 766–797.
Rappaport Hovav, M., & Levin, B. (2012). Lexicon uniformity and verbal polysemy. In M.
Everaert, M. Marelj, & T. Siloni (Eds.), The Theta System: Argument structure at the interface
Reinhart, T. (2002). The Theta System—An overview. Theoretical Linguistics, 28, 229–290.
Sato, H. (1987). Resultative attributes and GB principles. English Linguistics, 4, 91–106.
Shibatani, M. (1976a). Causativization. In M. Shibatani (Ed.), Syntax and semantics 5: Japanese
generative grammar (pp. 239–292). New York: Academic.
Shibatani, M. (1976b). The grammar of causative constructions: A conspectus. In M. Shibatani
(Ed.), Syntax and semantics 6: The grammar of causative constructions (pp. 1–40). New York:
Academic.
Simpson, J. (1983). Resultatives. In L. Levin, M. Rappaport, & A. Zaenen (Eds.), Papers in
Lexical-Functional Grammar (pp. 143–157). Bloomington: Indiana University Linguistics
Club.
Smith, C. S. (1970). Jespersen’s ‘move and change’ class and causative verbs in English. In
M. A. Jazayery, E. C. Polomé, & W. Winter (Eds.), Linguistic and literary studies in honor
of Archibald A. Hill (Descriptive linguistics, Vol. 2, pp. 101–109). The Hague: Mouton.
Talmy, L. (1976). Semantic causative types. In M. Shibatani (Ed.), Syntax and semantics 6: The
grammar of causative constructions (pp. 43–116). New York: Academic.
Talmy, L. (2000). Towards a cognitive semantics. Vol. I, Concept structuring systems. Cambridge,
MA: MIT Press.
Tenny, C. L. (1987). Grammaticalizing aspect and affectedness. Ph.D. thesis, MIT, Cambridge,
MA.
Tenny, C. L. (1994). Aspectual roles and the syntax-semantics interface. Dordrecht: Kluwer.
van Valin, R. D., & Wilkins, D. P. (1996). The case for ‘effector’: Case roles, agents, and agency
revisited. In M. Shibatani & S. A. Thompson (Eds.), Grammatical constructions (pp. 289–322).
Oxford: Clarendon Press.
Waltereit, R. (1999). Grammatical constraints on metonymy: On the role of the direct object.
In K.-U. Panther & G. Radden (Eds.), Metonymy in language and thought (pp. 233–255).
Wechsler, S. (1997). Resultative predicates and control. In Texas Linguistic Forum 38: The syntax
and semantics of predication (pp. 307–321). Austin: Department of Linguistics, University of
Texas.
Wechsler, S. (2005). Resultatives under the ‘event-argument homomorphism’ model of telicity. In
N. Erteschik-Shir & T. Rapoport (Eds.), The syntax of aspect (pp. 255–273). Oxford: Oxford
University Press.
Wechsler, S. (2012). Resultatives and the problem of exceptions. In I.-H. Lee et al. (Eds.), Issues
in English linguistics (pp. 119–131). Seoul: Hankookmunhwasa.
Wojcik, R. (1976). Where do instrumental NPs come from? In M. Shibatani (Ed.), Syntax and
semantics 6: The grammar of causative constructions (pp. 165–180). New York: Academic.
Wolff, P., Jeon, G.-H., Klettke, B., & Li, Y. (2010). Force creation and possible causers across
languages. In B. Malt and P. Wolff (Eds.), Words and the mind: How words capture human
experience (pp. 93–111). Oxford: Oxford University Press.
Chapter 7
Deconstructing Internal Causation
Malka Rappaport Hovav
Abstract This paper argues that the set of verbs characterized by Levin &
Rappaport Hovav (1995) as internally caused do not form a grammatically coherent
class. In particular, the verbs which have been termed internally caused change
of state verbs show grammatical properties which distinguish them from all the
other internally caused verbs. It is argued that within the class of change of state
verbs, there is no grammatically relevant classification of verbs or roots in terms of
internal and external causation. The internally caused change of state verbs often
describe events which come about in the natural course of events, and it is shown
that the general principles governing lexical causatives, in particular, the (non)
appearance of cause arguments dictate that these verbs most often appear in the
intransitive variant and with external arguments which refer to ambient conditions.
This, however, is just a tendency, and follows from the nature of the events described
by the verbs, and not from any grammatical property of the verbs. The verbs bloom,
blossom and flower, often taken to represent the class of internally caused COS
verbs are shown to be better analyzed as a special class of substance emission verbs.
The property which is suggested to unify the non-change-of-state internally caused
verbs is that of selecting a force-creator.
Keywords Actual cause · Causal factor · Causative alternation · Change of state

verbs · Direct causation · External causation · Force creator · Internal
causation · Verbs of emission
M. Rappaport Hovav ()

Department of Linguistics; Language, Logic and Cognition Center, The Hebrew University of

https://doi.org/10.1007/978-3-030-34308-8_7
220 M. Rappaport Hovav
7.1 Introduction
Prominent among the many argument alternations discussed by linguists is the

causative alternation. The alternation in English is illustrated in the pairs of
sentences below:
(1) a. The cat broke the vase.
b. The vase broke.
(2) a. My son/the heat melted the chocolate.
b. The chocolate melted.
(3) a. I/The strong wind dried the clothes.
b. The clothes dried.
(4) a. The players bounced the ball.
b. The ball bounced.
In (1)–(4), the meaning of the transitive use of the verb in each pair can roughly
be paraphrased as “cause to V-intransitive”. For example, (1a) can be paraphrased as
“The cat caused the vase to break.” In English,1 the same morphological form of the
verb appears in both the transitive and the intransitive frame. I will call a sentence
with the transitive form of a causative alternation verb the causative variant, and a
sentence with the intransitive form the anticausative variant.2
Not all intransitive verbs have causative counterparts.3
(5) a. My kids played.

b. *The camp director played my kids.
(6) a. The trees bloomed.
b. *The farmer bloomed the trees.
(7) a. Trump trembled.
b. *Putin trembled Trump.
1 There is great variation both within and across languages with respect to the morphological
relationship between the variants of the causative alternation and there is much to say about the
patterns of morphological marking, but this is far beyond the scope of this paper. For discussion,
see, among many others, Haspelmath (1993, 2017), Haspelmath et al. (2014), Alexiadou et al.
(2015).
2 In the literature, the term ‘anticausative’ often refers to a morphologically marked intransitive
member of a causative alternation pair. In this paper, it will refer to the intransitive variant
regardless of the morphological form. Given that I am concentrating on English, the role of
morphology in the account is minimal.
3 I choose to scrutinize the relation between the variants from the perspective of intransitive verbs
because it is not always clear whether a particular transitive verb is causative. That is, it is more
difficult to clearly delineate a group of transitive verbs with causative semantics and look for
counterparts without the causative component, than it is to begin with intransitive verbs and look
for transitive verbs with a causative paraphrase containing the intransitive verb.
7 Deconstructing Internal Causation 221
(8) a. The soldiers’ wounds glowed in the dark.

b. *The bacteria glowed the soldiers’ wounds in the dark.
(9) a. The bone deteriorated.
b. *The doctors deteriorated the bone (with radiation).
A central component of a complete account of the causative alternation is the
delineation of the verbs which participate in the alternation. Levin & Rappaport
Hovav (1995; L&RH) introduced a distinction between verbs which they called
externally caused and those which they called internally caused4 (building on
insights from Smith 1970), in order to account for the distribution of verbs in the
causative alternation. Specifically, verbs participating in the alternation are charac-
terized by them as externally caused and those not participating are characterized
as internally caused. Examples of verbs from the two classes, as they appear in the
appendix in L&RH, are shown in (10) and (11).
(10) externally caused:

a. Change of state break, cool, dry, lengthen, loosen, open, narrow, sink,
smooth, whiten, widen . . .
b. Motion bounce, rock, roll, spin, swing, twirl, twist . . .
(11) internally caused:
a. Agentive dance, exercise, laugh, play, ponder, sing, smile, think . . .
b. Non-agentive Activity blush, hesitate, quiver, shudder, tremble . . .
c. Emission beam, flash flicker, gleam, glitter, glow, sparkle, twinkle . . .
buzz, gush, gurgle, jingle, jangle, wail, dribble, drip, drool, gush, ooze,
spurt, sweat . . .
d. Change of state blister, bloom, blossom, burn, corrode, decay,
deteriorate, erode, ferment, flower, germinate, grow, molder, molt,
ripen, rot, rust, sprout, stagnate, swell, tarnish, wilt, wither5 . . .
As seen above, the major classes of internally caused verbs on L&RH’s analysis
include: (i) agentive activity verbs; (ii) a variety of other (aspectually) activity verbs,
4 L&RH use the terms internally and externally caused verbs, though this is something of a
misnomer since the verbs themselves are not caused. But this is the term that has been used in
subsequent literature and for expository purposes here I won’t make an attempt to improve the
nomenclature.
5 grow does not appear in the L&RH’s appendix listing internally caused COS verbs, but it does
appear in Levin’s (1993) class of entity-specific COS verbs, which are claimed in L&RH to
be internally caused. This verb figures prominently in subsequent discussions of the causative
alternation in Distributed Morphology, where an explanation is offered for the lack of a causative
counterpart for the verb (see, for example, Marantz 1997 and Harley & Noyer 2000). Interestingly,
this verb shows somewhat atypical behavior even for a verb describing internally caused changes of
state (Rappaport Hovav 2014: 13). ripen appears in L&RH as externally caused, probably because
of its morphology: The appendix includes a sub-class of alternating verbs with -en. However, this
verb clearly shares the properties of other verbs which have been termed internally caused (COS)
verbs, especially as they are characterized in Sect. 7.3.
like tremble, shudder, slouch and hesitate which are typically predicated of humans
but are not agentive; (iii) verbs of emission (e.g., light, sound, smell and substance
emission); and (iv) a subset of change of state (COS) verbs. The two major classes
of externally caused verbs are (COS) verbs and non-agentive verbs of manner of
motion.6 Some kind of distinction regarding the nature of causation as internal
or external has subsequently been adopted, by many researchers: Marantz (1997),
Harley & Noyer (2000), Alexiadou et al. (2006, 2015), Alexiadou (2010), McKoon
& Macfarland (2000) and Wright (2002), among many others.
The main claim of this paper is that the verbs listed in (11) do not form a
unified class; there is no coherent semantic characterization which covers all of
these verbs, nor do all the verbs in (11) have any shared grammatical property.
Most strikingly, internally caused COS verbs are unaccusative cross-linguistically
(L&RH: 159–162), while the rest of the verbs in (11) are unergative. Despite the
fact that most of the attention in the literature devoted to internally caused verbs
has focused on verbs in (11d), distinguishing them from externally caused COS
verbs, I argue that there is no grammatically encoded distinction between internally
and externally COS verbs. For example, in principle, all COS verbs – including the
verbs in (11d) – participate in the causative alternation. The same set of principles
governs the causative alternation with all COS verbs, including principles which
determine semantic restrictions of the external argument and the (non)appearance of
the external argument. Differences between specific verbs with respect to patterns
of their participation in the alternation have less to do with their lexically encoded
properties and more to do with the nature of the events they describe. In contrast, the
verbs in classes (11a–c) do form a distinct grammatically significant group and have
argument realization properties which clearly distinguish them from the verbs in
(11d) (though each subgroup has other distinguishing grammatical properties, some
of which will be discussed later in this paper). In particular, they are unergative.
There is then no possibility of adding an external argument which can be considered
the cause of the event denoted by the base verb. This explains the fact that they do
not participate in the causative alternation.
This paper is structured as follows. In Sect. 7.2, I briefly review L&RH’s original
motivation and for presentation of the distinction between internal and external
causation in the classification of verbs. In Sect. 7.3, I take a closer look at how
the classes are delineated and point out that there are two partially overlapping but
essentially distinct characterizations of the class of internally caused verbs. In Sect.
7.4, I show that of the three subclasses of internally caused verbs, the COS verbs
display a pattern of argument realization behavior in general, and participation in the
6 Verbs such as roll, bounce, spin and rock can be agentive, of course, but need not be. They share
properties with verbs classified as externally caused when the theme of motion is non-agentive.
When they are used agentively, they show properties of agentive manner of motion verbs and as
such they display different properties with respect to the causative alternation (see fn. 11).
causative alternation in particular, which distinguish them from the other classes in
(11). Of the two characterizations of internally caused verbs, one is more appropriate
for COS verbs and the other for the rest of the verb classes in (11). In Sect. 7.5, I
show that at least for English, the distinct grammatical properties which have been
attributed to internally caused COS verbs do not in fact hold; this supports my claim
that there is no grammatically encoded distinction between internally and externally
caused COS verbs, though it is convenient to continue using the term internally
caused change of state verbs descriptively. In Sect. 7.6, I turn to the semantic and
pragmatic principles which determine the range of subjects one finds with causative
alternation verbs and the principles which determine the (non) appearance of the
cause argument. These principles allow us to derive the special semi-grammatical
signature found with internally caused COS verbs, without having to recognize a
grammatical distinction between verbs of internal versus verbs of external causation.
In Sect. 7.7, I show that the verbs bloom, blossom and flower have a set of argument
realization properties which sets them apart from the rest of the verbs in (11d) and
that they are best analyzed as a special class of verbs of substance emission, and not
COS verbs, as they are typically classified in the literature. In Sect. 7.8, I suggest
that if a grammatically relevant class of internally caused verbs is to be recognized,
it should encompass the verbs in (11a–c) to the exclusion of the verbs in (11d),
and that the property which they share, by virtue of which they all have underlying
external arguments, is the selection of a force creator. Section 7.9 concludes.
7.2 L&RH’s Motivation for the Distinction Between Internal

and External Causation
In order to assess the grammatical relevance of the internal/external causation

distinction, we need to ask what exactly it is that we are classifying: roots, verbs,
events in the world, event descriptions? Though L&RH was written before the
distinction between roots and verbs was clearly articulated, the authors are fairly
explicit that theirs is a classification of verbs and not events in the world or even
event descriptions.7 More specifically, they take the distinction to be relevant to the
7 This position can be contrasted, for example, with what is known about telicity which is not
associated strictly with verbs (many verbs have variable telicity), but also does not categorize
actual happenings in the world (the same happening can be described with a telic event description
or an atelic event description) (Krifka 1998; Rappaport Hovav & Levin 2002; Rothstein 2004).
While both telicity and internal/external causation are linguistic categories and not classifications
of happenings in the world, telicity is now known to be a property of event descriptions
compositionally built from the verb and its arguments (and modifiers). L&RH take internal/external
causation to be a classification of verbs. Verbs that show variable behavior with respect to this
distinction can be classified either way; no one has ever suggested that it is possible to derive one
kind of event description from another compositionally in the way possible with respect to telicity,
and I also do not think this is possible in general.
lexical representation of verbs as encoding construals of events (happenings in the

world). This is made clear in the following quote:
(12) “The distinction between internally and externally caused eventualities is a
distinction in the way events are conceptualized and does not necessarily
correspond to any real difference in the types of events found in the world. In
general, the relation between the linguistic description of events and the
events taking place in the real world is mediated by the human cognitive con-
strual of events, which is what we take our lexical semantic representations
to represent.” (L&RH 1995: 99)
In terms of the semantic explication, L&RH have two partially overlapping
descriptions of the semantic property the distinction is supposed to capture. The
first has to do with the relation between the argument undergoing change and the
cause of the change. They write:
(13) “With an intransitive verb describing an internally caused eventuality, some

property inherent to the argument of the verb is ‘responsible’ for bringing
about the eventuality. For agentive verbs such as play and speak, this property
is the will or volition of the agent who performs the activity. Thus the concept
of internal causation subsumes agency.” (L&RH 1995: 91)
(14) “This . . . reflects the nature of internal causation, which involves causation
initiated by, but also residing in, the single argument, and hence dependent
on its properties.” (L&RH 1995: 94)
The second characterization of the distinction has to do with the nature of the
event itself and L&RH apply this to internally caused COS verbs.
(15) “That is, the changes of state that they describe are inherent to the natural
course of development of the entities they are predicated of and do not need
to be brought about by an external cause (although occasionally they can be,
and in such instances causative uses of these verbs are found).” (L&RH
1995: 97)
L&RH assign different representations to the members of the classes of internally
and externally caused verbs. The representations are meant to be connected on the
one hand to the semantic explication and on the other to the grammatical behavior
the distinction is meant to capture. They argue that internally caused verbs are
monadic – they are lexically associated only with the argument whose change or
activity is described by the verb.
(16) “The adicity of a verb is then a direct reflection of a lexical semantic property
of the verb, namely, the number of open positions in the lexical semantic re-
presentation.” (L&RH 1995: 95)
In contrast, externally caused verbs are dyadic8 ; they lexically include a cause
argument. L&RH argue that all alternating verbs are externally caused9 ; the basic
lexical representation is then associated with the transitive (causative) variant which
reflects the categorization of the verb as externally caused. A subclass of externally
caused verbs – those that do not specify anything about the causing event in a
causative structure – can undergo a process of lexical binding of the external
argument (cf. Reinhart 2002; Horvath & Siloni 2011, 2013 for a similar analysis;
see also Koontz-Garboden 2009 for a different perspective on the derivation of
the intransitive variant from the transitive variant). The lack of specification of the
causing event is manifest in the absence of selectional restrictions on the subject. In
(17) and (18) we see the correlation between the lack of selectional restrictions on
the subject (which is expressed in the range of semantic types of subjects that can
appear with alternating verbs) and the ability of a verb to alternate.
(17) a. The vandals/The rocks/The strong winds broke the windows.

b. The windows broke.
(18) a. The insurgents/*The poison/*The flood assassinated the president.
b. *The president assassinated.
The characterization of the verbs which participate in the causative alternation
then is: externally caused verbs with no semantic restrictions on the external
argument.
The process of lexical binding was taken by L&RH to be an operation on the
lexical representation of the verbs. The anticausative variant, which is the output of
lexical binding, still contains a representation of a lexically bound causal element
(Fig. 7.1). The existence of this lexically bound causal element was argued to be
diagnosed by the ability of the anticausative to appear with the phrase by itself in
the sense of “without outside help” (Chierchia 2004; L&RH 1995).10
(19) The vase broke by itself.

(20) #Mary laughed by herself. (can only mean ‘unaccompanied’)
On this analysis, internally caused verbs do not participate in the alternation
because the alternation derives intransitive verbs from transitive verbs. Internally
8 Infact of matter, adicity is not an indication or even a reflection of the relevant kind of causation.
For example, there are dyadic unaccusative verbs of (like appear) which are not externally caused;
see Chapter 3 of L&RH. And internally caused verbs are certainly not all monadic. For example,
agentive verbs are internally caused, but there are dyadic agentive verbs (Levin 1999). Verbs of
emission are classed as internally caused and they are dyadic (see Sec. 7.6.).
9 But not all externally caused verbs undergo the alternation; verbs of destruction such as destroy,
wreck and ruin, and verbs of killing such as kill, murder and assassinate, are externally caused but
do not alternate in English, though they do in other languages. See (34)–(36) below.
10 The other interpretation of by itself is “alone, unaccompanied” (L&RH: 88–89). There is actually
much more to be said about the distribution of this phrase, but a detailed discussion of this matter
is beyond the scope of this paper. See Horvath & Siloni (2011, 2013), Schäfer (2007), Schäfer &
Vivanco (2016).
Intransitive break:
LSR [[x do-something] cause [y become BROKEN]]
¯
Lexical binding 0/
Linking rules ¯
AS <y>
Fig. 7.1 The input and output of lexical binding, L&RH: p. 108. (LSR = Lexical semantic
representation, a predicate decomposition representation of event structure; AS = argument
structure)
caused verbs are inherently intransitive, and there is by assumption no rule in

English which freely adds an argument, only a rule which lexically binds an external
argument. There is, then, no way to derive the causative variant of these internally
caused verbs.11 The distinction between internally and externally caused verbs has
been adopted by many researchers, as we will see.12
We have seen in this section that L&RH provide two distinct characterizations
of internally caused verbs. In the next section I will show that the two intuitive
characterizations roughly single out two subclasses of the original class of internally
caused verbs. I show that these two subclasses have distinct grammatical properties
and that the notion of internal causation having to do with the inability of the verb to
participate in the causative alternation is appropriate for only one of the subclasses.
7.3 Redefining the Classes
One of the most significant problems for L&RH’s analysis is that it offers no real
probe for the semantic distinction between internally and externally caused verbs
independent of the phenomena the distinction was meant to account for to begin
with. The different and intuitively defined explications given above are not based on
well-established semantic categories that can be independently identified.13 These
11 It is true that agentive manner of motion verbs have causative variants (e.g., The trainer ran the
horses around the field). However, as argued by L&RH and others (e.g., Folli & Harley 2006;
Alexiadou 2014) this alternation is distinguished from the causative alternation in many respects.
It is also narrowly restricted lexically.
12 Another piece of evidence that L&RH bring to support their analysis is the frequent pattern of
morphological marking whereby the anitcausative variant is morphologically marked (based on

data drawn from Nedjalkov & Slinitskiy 1973). They take this to indicate that the anticausative
variant is derived. However, the morphological marking of the variants in the causative alternation
is much more complicated; there is a certain amount of correlation between the semantics of the
verb and the morphological marking but this is evident only at the level of large typological studies
(e.g., Haspelmath 2017; Haspelmath et al. 2014).
13 We can contrast the intuitive characterizations of internal causation with a notion such as scalar
change (Rappaport Hovav 2008, 2014b; Rappaport Hovav & Levin 2010; Beavers 2008), which
characterizations then do not allow us to test our hypotheses. The following quote
drives this point home; see also the text in parentheses in (15) above.
(21) “B. Levin once heard her landlord say The pine needles are deteriorating the
roof. Although to our ears this sentence is unacceptable, probably because we
conceive of deterioration as always being internally caused, it appears that the
landlord’s conceptualization was different.” (L&RH 1995: 99)
The only evidence for different conceptualization of course is the participation of
the verb in the causative alternation. The circularity in the argumentation is evident:
verbs are classified in an intuitive way and then when the data go contrary to the
classification, verbs are suggested to be either wrongly classified or to allow more
than one classification.14
A criterion L&RH suggest for distinguishing between internally and externally
caused verbs is the degree to which the verbs exert selectional restrictions on the
argument of the caused event; the idea is that there are much stricter selectional
restrictions on arguments undergoing internally caused changes than those under-
going externally caused changes. This is related to the characterization in (13):
“some property inherent to the argument of the verb is ‘responsible’ for bringing
about the eventuality”. In fact, Levin (1993) uses the term “verbs of entity specific
change of state” for the class which has come to be called internally caused COS
verbs: “These verbs describe changes of state that are specific to particular entities.
That is, these verbs impose very narrow selectional restrictions on their arguments.
That is, silver and some other metals tarnish, flowers and plants wilt, and so on” (p.
246). However, externally caused COS verbs also exert semantic restrictions on their
objects (after all, as demonstrated by Fillmore 1970, only certain kinds of entities
break and they break because of properties internal to them). It seems, then, that this
diagnostic is also of limited value. Moreover, verbs of emission also impose tight
selectional restrictions on the subject (Levin 1993: 233); yet, as we will see, verbs
of emission show different argument realization properties than internally caused
COS verbs, including different causative alternation patterns. I conclude from this
that the imposition of tight selectional restrictions on an argument is not a semantic
property relevant to participation in the causative alternation.
A problem specific to the characterizations in (13) and (14) is that for many
verbs classed as internally caused, it clearly is possible to isolate causes, and they
is composed of specific semantic components. Justifying the claim that a verb or predication
encodes scalar change involves an exact specification of how the components of scalar change
are instantiated in the meaning of the verbs or predication.
14 L&RH were in fact aware of this. They write “What is important is that the nature of the
externally versus internally caused distinction leads to expectations about where fluctuations
with respect to verb classification both within and across languages may be found” (p. 100).
That is, while there is no independently available criterion for diagnosing the distinction, the
intuitive characterization, though not making exact predictions, still makes predictions about the
general zone where the borderline between the classes is expected to be found. We seek a better
characterization of the properties of verbs that determine the zone in which they fall.
certainly reside external to the entity undergoing the change. For example, when
metal rusts, there are surely causes for this event: most immediately, moisture
in the surroundings, but also circumstances which bring about the moisture. All
these would be considered causes, for example, under a counterfactual analysis of
causation (e.g., Lewis 1973; Dowty 1979).
(22) If the weather hadn’t been so humid, the gate would not have rusted.
On the other hand, metal does rust in the natural course of events, given these
basic circumstances. In fact, looking at the class of verbs in (11d) it appears that the
characterization in (15), rather than those in (13) and (14) fits; one can say that they
generally describe changes which come about in the natural course of events (in the
context of particular ambient conditions).
On the other hand, it is not appropriate to characterize the verbs in (11a–c) as
describing changes which take place in the natural course of events. People don’t
laugh, dance or think (verbs in 11a) in the natural course of events. In fact, the
volitional action of an agent is in some sense the epitome of what does NOT happen
in the natural course of events: people’s actions are fairly inherently unpredictable.
Though the verbs in (11b) describe events which are typically non-volitional, they
still do not occur in the natural course of events, given particular circumstances.
The verbs in (11c) do not really describe changes, and so the notion of change in
the natural course of events in not relevant to them either. In the next section, I show
that the verbs in (11a–c) also display distinct behavior from the verbs in (11d) with
respect to the causative alternation.
7.4 Patterns of Participation in the Causative Alternation
The problems mentioned in the last section have to do with ways of independently
characterizing the class without appealing to the participation of the verbs in the
causative alternation – which is what the characterization was meant to account for
to begin with. We saw that the intuitive characterization of internally caused verbs
isolated two distinct classes of verbs. In this section, I show that these two classes in
fact show distinct grammatical properties. The verbs in (11d) are COS verbs which
do not lexically select a cause. These verbs do indeed participate in the causative
alternation, in contrast to the verbs in (11a–c).
A variety of studies (most particularly McKoon & Macfarland 2000 (henceforth
M&M) and Wright 2002; see also Rappaport Hovav & Levin 2012, Alexiadou
2014, Alexiadou & Anagnostopoulou 2019) have shown that verbs which have been
categorized as internally caused COS verbs (the verbs in 11d) do in fact appear in
causative variants, although there are differences among verbs with regard to their
tendency to appear in causative or anticausative variants. This is seen in Table 7.1,
taken from M&M, which lists some of the verbs classed by L&RH as internally
caused and the probabilities of a transitive variant for each based on a large corpus
study.
Table 7.1 Probability of Verb Probability Verb Probability

transitive construction –
internally caused verbs Blister .22 Germinate .06
(McKoon & Macfarland Bloom .00 Rot .08
2000: 837) Blossom .00 Rust .14
Corrode .63 Sprout .26
Decay .00 Stagnate .02
Deteriorate .01 Swell .37
Erode .68 Tarnish .98
Ferment .54 Wilt .06
Flower .00 Wither .12
Note: There were two transitive sentences with bloom;
rounding makes the entry .00 (M&M provide one
example of the transitive variant of bloom from their
corpus, and it turns out that it is not causative, but rather
an example of an emitter subject; see Sect. 7.7 below)
Table 7.2 Probability of External causation verbs

transitive construction –
Low prob. Higher prob.
externally caused verbs
(McKoon & Macfarland Verb Trans. Verb Trans.
2000: 838) Abate .10 Dissipate .41
Atrophy .03 Fossilize .60
Awake .03 Fray .52
Crumble .05 Redden .24
Explode .07 Splinter .49
Fade .01 Thaw .61
Shrivel .11
Vibrate .03
Mean .06 Mean .48
But, a similar gradience of the probability of verbs to appear in the causative

variant occurs with verbs which have been classified as externally caused, as shown
in Table 7.2, also taken from M&M. Thus, despite what is often claimed (e.g.,
Alexiadou 2014), the gradience in the probability of verbs to appear in the causative
variant is not a property of the class of internally caused verbs.
Nonetheless, there is a very marked tendency for verbs classified as internally
caused to appear with subjects which are not agents in their causative variant.15
If we restrict our attention to the uses of these verbs with concrete objects, then
15 Infact, this is probably the reason researchers thought that these verbs do not participate in
the alternation. At a time when generative grammarians relied mainly on their judgments of
constructed sentences without context, in testing a verb for participation in the causative alternation
they typically formed transitive sentences with agent subjects. For example, the overwhelming
majority of the sample causative alternation sentences used to illustrate the grammatical properties
of semantic classes of verbs in Levin (1993) have agent subjects.
the propensity of these verbs to appear with subjects that are not agents might be
taken to be criterial for the classification of the verbs as internally caused. Among
the 71 transitive sentences selected from M&M’s corpus with concrete objects
and verbs classified by L&RH as internally caused COS verbs, 57 had natural
entities as subjects and only 3 had animates (Table 3 in M&M; p. 843). This is
in marked contrast with verbs which are classified by L&RH as externally caused,
where the subjects of causative variants with concrete objects are more evenly
distributed among the different semantic categories, according to M&M. In a list
of 111 transitive sentences with concrete objects and what have been classified as
externally caused COS verbs, 23 had natural entities as subjects and 37 had animates
(Table 4 in M&M; p. 843).
More specifically, the subjects we typically find with verbs classified as internally
caused when they take concrete objects can be characterized as “ambient condi-
tions” (Rappaport Hovav & Levin 2012). Further examples appear in the sentences
below, taken from Wright (2001).
(23) a. Salt air and other pollutants can decay prints. (LN 1982)
b. The onset of temperatures of 100 degrees or more, on top of the drought,
has withered crops. (NYT 1986) (Wright 2002: 341)
(24) a. *The photographer/*the new method can decay the prints.
b. *The farmer withered the crops.
(25) a. Light will damage anything made of organic material. It rots curtains, it
rots upholstery, and it bleaches wood furniture. (LN)
b. Salt air rusted the chain-link fences. (LN)
c. Bright sun wilted the roses. (LN) (Wright 2001: 112)
The idea that there is a rather restricted range of subjects for internally caused
COS verbs is strengthened by a survey task reported in Wright (2002). She asked
subjects to list three typical causers for a variety of verbs normally classed as
internally caused and also for a variety of verbs normally classed as externally
caused. The difference in results is striking: on average 8.5 distinct causers were
listed for internally caused COS verbs, whereas for externally caused COS verbs on
average 14.8 typical causers were listed (Wright 2002: 344–345). We return to these
data in Sect. 7.6 and suggest an explanation for them.
However, for most verbs which are generally included in the class of internally
caused COS verbs, even this property of having a restricted range of subjects in the
causative variant is just a marked tendency. Despite what is often said, these verbs
can appear with agent subjects (see also discussion in Rappaport Hovav & Levin
2012):
(26) I used red onion rather than white and sliced shiitake mushrooms, and I
wilted my kale just a bit. http://www.eatingwell.com/recipe/250328/hearty-
kale-salad/
(27) With that, I withered the second peach tree, which needed the pollination
of the first tree to grow. http://newworcesterspy.net/the-fatherly-guide-to-
success/
(28) I rusted the metal by using a water and vinegar solution. The acid in the
vinegar speeds up the rusting process. https://rufflesandrust.co.za/up-cycle-
project/
(29) I corroded the seat post fastener and threaded rod on my old proprietary
aero seat post. https://forum.slowtwitch.com/forum/Slowtwitch_Forums_C1/
Triathlon_Forum_F1/Sweat_Corrosion_P6834011/
(30) I blistered my hands sanding. https://www.instagram.com/malinda_chan/p/
Bt7jWHuAupN/
In fact, I have checked each and every verb in (11d) and found that without
exception they can appear with agent subjects (see Sect. 7.7 below on the excep-
tional properties of bloom, blossom and flower). I will discuss the properties of
these verbs in the causative alternation further in Sect. 7.6.
Furthermore, it emerges from M&M’s data that internally caused COS verbs with
abstract objects appear much more frequently in causative variants in comparison
to the same verbs with concrete objects, and with a wider range of semantic types
of subjects, as illustrated for the verb erode below. That is, when they have abstract
objects, verbs considered internally caused display behavior more similar to verbs
considered externally caused. This suggests that the properties isolated are not
properties of the verb per se. Rather it is the property of the entire predication (the
event description) which determines whether the verb can have a transitive variant
and what the range of subjects for the transitive variant can be.
(31) Markets eroded the morals of the people involved. https://www.
huffingtonpost.com/2013/05/13/markets-morals-study_n_3267995.html
(32) He eroded my self-confidence and my dignity. https://womenintheworld.
com/2017/12/19/dustin-hoffman-accusers-share-details-about-his-alleged-
abuse-and-its-impact/
(33) The financial interests of biotech and drug companies have eroded the
values of the medical profession . . . https://www.dailybreeze.com/2012/01/
04/helen-dennis-we-have-more-control-over-aging-than-we-think/
This situation holds for all verbs classed as internally caused COS verbs; with
concrete objects they overwhelmingly take ambient conditions as subject, and with
abstract objects they take a wider variety of subjects, including agents, abstract
entities, states and events.16 One might want to claim that these verbs have separate
lexical entries when taking abstract objects, and that these verbs can be classified as
internally caused with concrete objects and externally caused with abstract objects.
The idea would be that when COS verbs have abstract objects they are used
metaphorically, and their metaphorical uses have separate lexical entries. But this
would just be begging the question of why it is that changes of state with abstract
objects are to be considered externally caused.
Furthermore, it is worth noting that the transitivity options of a COS verb do not
necessarily change when a verb is used with an abstract argument, hence understood
metaphorically. For example, verbs of destruction have the property of not de-
transitivizing in English:
(34) a. The Romans destroyed/ruined/wrecked the city.

b. *The city destroyed/ruined/wrecked.
This feature holds consistently of all verbs of destruction even when they take
abstract objects:
(35) a. You destroyed my hopes.

b. *My hopes destroyed.
(36) a. The tension wrecked our relationship.
b. *Our relationship wrecked.
This seems to indicate that the strong transitivity of these verbs is indeed a
grammatically encoded property (Rappaport Hovav 2014a). We see, then, that when
we isolate a clear grammatical property associated with a particular verb, varying
the kind of object and the status of the verb as concrete or metaphorical does
not affect that grammatical property. In contrast, what controls the transitivity of
internally caused COS verbs (their participation in the causative alternation) appears
not to be a grammatical property.
The conclusion I draw from these data is that for COS verbs, it is not a property
of the verb per se that determines the range of subjects allowed, but rather the more
specific kind of change the verb describes on a particular use. The exact nature of
the change is to a large degree determined by the choice of the internal argument –
the theme of the change of state. I return to the analysis of COS verbs in Sect.
7.6. But first it is important to point out that these patterns of distribution in the
causative alternation are not displayed by the verbs in (11a–c). These verbs much
more consistently resist causativization, and varying the semantic type of subject
does not change the picture significantly.
16 Ihave not done a systematic corpus study to establish this fact, but targeted web searches make
this abundantly clear.
(37) a. *I cried/hesitated the interviewee.

b. I caused the interviewee to cry/hesitate.
c. *The tense circumstances/my attitude cried/hesitated the interviewee.
d. The tense circumstances/my attitude caused the interviewee to cry/hesitate.
This generalization is straightforwardly accurate for the verbs in (11a,b). The
picture with respect to verbs of emission is a bit different. It is in general true that
emission verbs also do not have causative variants (Levin 1993: 234–237; L&RH:
92)
(38) a. *The bacteria glowed the wounds.

b. *The wind glowed the embers. (Alexiadou 2014: 880).
c. *The compliment glowed her (face).
Certain classes of verbs of emission have occasional causative uses – most
strikingly verbs of sound emission, but also a small number of light emission verbs;
however, these verbs can causativize only when the emission is the result of direct
physical manipulation (Levin et al. 1997). The exact conditions controlling the
causative variants of emission verbs deserve further attention, but it is clear that
the causative variants are far more sporadic and are governed by a different set of
conditions which govern the causative variant of change of state verbs. For example,
even physical manipulation is not fully sufficient to license the causative variant:
(39) a. The stagehand flashed the lights.

b. The lights flashed
(40) a. *The jeweler sparkled the diamond (with a special cloth).
b. The diamond sparkled.
And it is certainly NOT the case that ambient conditions typically serve as the
subject of the causative with verbs of sound and light emission. Rather, when the
causative variant is licensed, it is with agents. In contrast, as we have seen, for
the class of internally caused COS verbs, it is usually ambient conditions and not
agents which are found in the causative variants. See Sect. 7.7 for discussion of
verbs of substance emission. While causativization of COS verbs is subject to a
direct causation restriction as are all causatives, COS verbs do not typically require
direct manipulation in order to satisfy the direct causation requirement.
The conclusion I come to from the discussion so far is that the verbs in (11d)
belong to a distinct class from the other verb classes in (11). If we do want a
grammatical distinction between internal and external causation, it will contrast the
verbs in (11a–c) with the intransitive verbs in (10), and the verbs in (11d) should be
classified along with the other verbs in (10). Before moving on to a fuller analysis
of the patterns of participation of the verbs in (11d) in the causative alternation,
I demonstrate in the next section that at least in English the class of internally
caused COS verbs have no grammatically isolatable properties, despite what has
been argued in the literature.
7.5 Against a Syntactic Representation of Internally Caused

Change of State Verbs
The distinction between different kinds of causation that are lexicalized in verbs was
adopted in the Distributed Morphology framework by Harley & Noyer (2000) and
Marantz (1997) and further developed by Alexiadou, Anagnostopoulou & Schäfer
(2006, 2015; henceforth AAS) and in other work. In this framework, roots, encoding
the idiosyncratic semantic core of a lexical item, are clearly distinguished from their
“first phase” syntactic environment (Ramchand 2008), which syntactically encodes
the event structure properties associated with the word built around the root.
In contrast to the derivational analyses discussed in Sect. 7.2, AAS (2006,
2015) present a non-derivational analysis of causative alternation verbs (see also
Piñón 2001; Doron 2003 for a similar account in Hebrew & Schäfer 2009 for a
general discussion of the non-derivational approach), where the two variants of the
alternation share a root and the difference between the two variants resides in the
amount of structure built on top of the root.
In the case of COS verbs, the causative, anticausative, passive and adjectival
passive forms are all built on a root encoding the lexicalized state with the addition
of functional layers. The anticausative involves v – which categorizes the root as a
verb and introduces event implications. The structure consisting of v followed by a
state root gives rise to a causal interpretation whereby the event is identified as the
cause of the result state. The causative variant involves the addition of Voice which
introduces the external argument and bears features relating to agentivity (Kratzer
1996; Pylkkänen 2008, AAS). Passive involves a different feature specification in
Voice than the active transitive, which does not allow the external theta role to be
assigned to [Spec, Voice]. The adjectival passive adds participial morphology which
may attach above vP or VoiceP; it stativizes its verbal complement. See, for example,
AAS for details.
AAS, following ideas of Harley & Noyer (2000),17 take the type of causation
to be a property of roots, not verbs. The distinction between internal and external
causation is taken to be part of the encyclopedic information associated with a root
which determines the syntactic contexts (amount of structure) minimally needed to
be associated with the root when it is realized as a verb. AAS (p. 54) classify COS
roots into four different sub-classes (adding one to the three subclasses in Harley &
Noyer 2000) which determine the range of syntactic structures built on the roots.
17 Interestingly,Marantz (1997) and AAS recognize, as I do here, that unergative verbs should be
classified differently from (internally caused) COS verbs, but they classify the roots of COS verbs
in terms of internal and external causation, unlike the position I take here.
(41) a. agentive: (like murder): The change in these verbs is always instigated by
an agent. Verbs based on these roots never alternate in any language – they
are always transitive; as a consequence, they must always be in a structure
with Voice.
b. internal causation: (like blossom and wilt): The action is always
dependent on the argument undergoing the change of state. Also
characterized as spontaneous.
c. external causation: (like kill and destroy): The action must be instigated
by an argument other than the one undergoing the action. As a consequence,
these verbs must appear with Voice.
d. cause-unspecified: (like break and open): The action may causally
originate either with the object of the action or with another argument. The
suggestion made by AAS (based on Harley & Noyer) is that when they are
transitive they express external causation and when they are intransitive they
express internal causation.18
The classes in (41a) and (41c) do indeed have grammatically identifiable
properties. For example, agentive COS verbs do not alternate cross-linguistically;
as far as I know, this generalization has never been challenged.19 With respect to
the verbs in (41c), at least in English this class is constituted by verbs of killing
and destruction. These verbs do not alternate in English (though they do in other
languages such as Greek and Hebrew, and when they alternate they must appear
with non-active Voice morphology). As mentioned above, this is a grammatical
property and is not affected by altering the choice of argument or with metaphorical
use. I suggest that this grammatical property does not follow from the nature of
the encyclopedic information associated with the state the roots encode but rather
because it is convenient for languages to have words which describe changes of state
having to do with death and destruction whose use entails that the change did not
come about in the natural course of events (Rappaport Hovav 2014a: 19) but with
some identifiable, if not identified, cause.
In Rappaport Hovav (2014a: 20–25) I provide arguments against characterizing
alternating verbs as being cause-unspecified, that is, as externally caused in the
transitive variant and internally caused in the intransitive variant. One argument
which brings the point home is the fact that the intransitive variant of a COS verb
can come immediately after a clause in which the cause of the change of state is
explicitly mentioned, as in (42).
18 Inthe end, AAS characterize this last class of roots as falling in the intermediate zone of a
spontaneity scale; the exact mode of participation in the causative alternation depends on their
account on the Voice system of the language.
19 This is true for what has been called lexical causativization. Languages often have a more
productive ‘syntactic’ causativization strategy (like the English cause X to V construction) which
allows the entire range of verbs. It should be pointed out that there are precious few COS verbs
which are necessarily agentive, murder and assassinate being the prime examples.
(42) a. I pounded on the piggy bank and it finally broke.

b. I leaned against the door and it opened. (Rappaport Hovav 2014a: 25)
It is difficult to argue that in these cases the root is understood as internally caused
when an external cause is explicitly mentioned in the previous clause. This is in
marked contrast to the verbs of killing and destruction in (43) and (44). With these
verbs, mentioning the cause in a previous clause does not license the intransitive
variant. That is, these verbs are strongly grammatically transitive.
(43) a. *This time I aimed carefully, fired accurately, and the victim finally
murdered. (Rappaport Hovav 2014a: 17)
b. *This time I aimed carefully, fired accurately, and the deer finally killed.
(44) a. *I spilled a cup of coffee on the precious document and it completely
ruined.
b. *The hurricane ripped through our area and the houses completely
wrecked.
I argue, however, that there is no grammatical distinction between classes (41b)
and (41d). This is in contrast to the position laid out in Alexiadou (2014) (see also
Alexiadou & Anagnostopoulou this volume), who suggests that what distinguishes
between internally caused and externally caused change of state verbs is not
participation in the alternation but rather the following three properties:
(45) (i) There is active morphology on the intransitive variant of the alternation
in languages which mark the distinction between active and non-active
morphology (i.e., this is a labile alternation);
(ii) The subject is restricted to be a cause (rather than an agent);
(iii) The transitive variant of the verb cannot passivize.
On Alexiadou’s account there is a structural distinction between the causative
variants of verbs built on internally and externally caused roots.20 The purported
restriction that the subjects of internally caused verbs be causes (excluding, for
example, agents) is attributed to the well-known direct causation restriction on
lexical causatives (see Sect. 7.6 below). Her proposal is that the subject of a verb
built on an externally caused root (and unergative verbs as well) is licensed in [Spec,
Voice] and the subjects of verbs built on internally caused roots are licensed in
[Spec, vP]. Since v introduces, as mentioned, a causal relation, Alexiadou assumes,
20 Alexiadou further suggests that the roots of the class of internally caused COS verbs themselves
are divided into two sub-classes: those which act like regular transitive verbs and those which have
the three properties above. The former class is represented by the verb ferment and the latter by
blossom. It is unclear what makes verbs like ferment different from externally caused COS verbs
(or cause-unspecified verbs under that characterization of alternating verbs). Therefore, I discuss
only her analysis of what she calls the blossom verbs, which are claimed to display the properties
(i) – (iii) above. Though see Sect. 7.7 for further discussion of the verb blossom.
following Solstad (2009),21 that the causer subjects of these verbs are a type of event
modifiers. She writes:
(46) “A defining property of causers is their inherent eventivity, as they are taken
to be responsible for the bringing about of an action or a result . . . Natural
forces are inherently eventive by definition. Causers name/explicate the event
that leads to the resultant state of the theme.” (p. 896)
Since by hypothesis cause subjects are generated in vP and appear in a structure
which lacks Voice, these verbs are expected not to passivize, since passive is a
voice alternation and passive morphology is the exponent of a head which appears
in Voice. We have seen, however, that at least in English, all internally caused
COS verbs can appear transitively and can have agentive subjects in their transitive
forms.22
We might then hypothesize that the subject is generated in [Spec vP] when it is a
cause and in [Spec, Voice] when it is an agent. This hypothesis comes with a clear
prediction – there should be a correlation between the nature of the subject and
passivizablity of the verb. We should find that the verb can be passivized with an
agent argument but not with a cause. However, this prediction is not borne out – it is
not at all difficult to find internally caused COS verbs in the passive, and, moreover,
many of these passives appear with causers in the by phrase. Note that the by phrases
in these examples refer to the causing event.23
(47) In the Eastern U.S., the dreadful summer of 1955 will be remembered for a
long time to come. Beginning in July, the region was withered by drought
and a heat wave, the worst on record, with temperatures in the 90s for a
large part of the month. http://content.time.com/time/magazine/article/0,
9171,823,875,00.html
21 According to AAS’s analysis, from-PPs as in The window cracked from the pressure are
realizations of the cause (Schäfer 2012), and are also modifiers of the causing event. Internally
caused COS verbs, like all other COS verbs, appear with these modifiers, as in The tents rotted
from the sun.
22 AAS report that these verbs do not appear with agent subjects in German. A more systematic
comparison of the behavior of internally caused COS verbs in different languages is clearly called
for.
23 Alexiadou (2014) also argues that internally caused COS verb display a labile morphological
pattern in the alternation (property (i) above in this section), though she does not claim that all
verbs showing a labile pattern are internally caused. Indeed, in Alexiadou (2010) there are verbs
which show a labile pattern of alternation, but are classified as cause-unspecified. Scrutiny of the
class of verbs she includes as internally caused COS verbs based on morphological criteria does
not convince the observer that these are all indeed internally caused COS verbs. For example, she
lists in her (34) verbs such as the Greek counterparts to thin and cool as internally caused, though
I see no semantic reason for this; the argumentation seems somewhat circular, reminiscent of the
circularity of argumentation in L&RH.
(48) Beach fill placed in April 2001 at Torrey Pines State Park, located on the
border between San Diego and Del Mar, CA about 6 km north of Scripps
Submarine Canyon (Fig. 7.1), was eroded by a storm in November 2001.
(49) This pole by the food area was rusted by the oxidation in the air. https://
sites.google.com/site/justinxie712/justin’sweatheringthings?tmpl=
%2Fsystem%2Fapp%2Ftemplates%2Fprint%2F&showPrintDialog=1
I conclude, then, that there is no defining feature of a set of roots which give rise to
an identifiable class of internally caused COS verbs.
If internally caused COS verbs appear in identical syntactic contexts as other
COS verbs, and there is no grammatically relevant distinction between internally
and externally caused verbs or roots, it remains to be explained why these verbs
appear less frequently in causative variants (as shown by M&M’s & Wright’s 2002)
data, and, furthermore, why these verbs tend to appear with ambient condition
subjects when they take concrete objects. The answer to these questions will emerge
from an account of the conditions which govern the (non)appearance of a cause in
COS event descriptions and the conditions which govern the choice of a cause, the
topic of the next section.
7.6 Semantic and Pragmatic Constraints on the External

Argument with COS Verbs
I assume that all COS verbs can freely add an external argument which will be
interpreted as a cause. For a complete account of the alternation, we need to specify,
among other things:
– the semantic constraints on the external argument, when expressed, and what
they follow from;
– the conditions under which an external cause argument can/must/may not be
expressed.
I discuss each in turn here. The necessary semantic condition on the external
argument is that it be construable as a direct cause. The idea that lexical causatives
must express direct causation goes back to Fodor (1970), McCawley (1978),
Shibatani (1978), and more recently Goldberg (1995), Bittner (1999), Piñón (2001),
Wolff (2003), Kratzer (2005) and Levin (this volume). There is a great deal of
discussion of the exact conditions which give rise to direct causation. Rappaport
Hovav & Levin (2012) and Rappaport Hovav (2014a) rely on the definition provided
in Wolff (2003):
(50) “Direct causation is present between the causer and the final causee in a
causal chain: (i) if there are no intermediate entities at the same level of
granularity as either the initial causer or final causee, or (ii) if any inter-
mediate entities that are present can be construed as an enabling condi-
tion rather than an intervening causer.” (Wolff 2003: 5)
Rappaport Hovav & Levin (2012) and Rappaport Hovav (2014a) suggest that this
accounts for the fact discussed in Sect. 7.4 that verbs typically classified as internally
caused have a marked tendency to appear with external arguments which describe
ambient conditions and not agents. Recall that this especially holds when the verbs
take concrete objects.
(51) a. ?The workers rusted the fence.
b. Heavy rains over the years must have rusted the fence.
bribieislandenvironmentprotection.org.au/wp-
content/uploads/.../Problem_Parks.pdf
In Rappaport Hovav (2014a) I provide the following explanation:
(52) “The most direct causes of such changes are natural forces and ambient condi-
tions which trigger or facilitate these changes. In order to introduce an agent
in an event of this sort, the agent would have to precede the natural force or
ambient condition in the chain of causation. For the agent to then qualify as a
direct cause in the causal chain, the natural force or ambient condition must be
considered an enabling condition (part (ii) of [50]), but this is not possible as
the agent does not have control over them.” [cf. page 237] (Rappaport Hovav
2014a: 22)
To be a bit more explicit, we can say that ambient conditions – like
water/humidity for rust or corrosion – are the most immediate causes because
they physically act upon the patient and bring about the COS – the acting of the
ambient conditions on the theme of the change of state is the proximate event in the
chain of causation. Given this, by (50ii) an agent can be a cause only if the ambient
conditions can be considered enabling. According to Wolff (2003), an intermediary
can be considered an enabler if it does something that is in concordance with the
tendency of the causer. Tendency is perhaps not the right property to attribute to an
agent, but we can say that ambient conditions can be considered enablers if they
effect a change in accordance to the volition of the agent. And it certainly appears
to be true that volitional control over ambient conditions allows an agent to appear
as the cause of an internally caused COS verb, as in (28), repeated here (53), or
(54a,b), taken from Wright (2001).24
24 The notion of direct causation is still a topic of intense investigation. For a recent discussion,
see Neeleman & van de Koot (2012). Agentivity can enhance the ability of a verb to appear in
the causative, but it is not the case that the subject of a causative, even for internally caused COS
verbs, has to be agentive. For example, I accidentally rusted my keyboard. https://www.reddit.com/
r/MechanicalKeyboards/comments/6af6pa/photos_i_accidentally_rusted_my_keyboard/
(53) I rusted the metal by using a water and vinegar solution. The acid in the
vinegar speeds up the rusting process. https://rufflesandrust.co.za/up-cycle-
project/
(54) a. The scientist germinated the seeds.
b. The winemaker fermented the grapes. (Wright 2001: 163)
With abstract objects, agents appear as subjects of internally caused COS verbs
more often than with concrete objects (Table 6, p. 844 in M&M). If we look at some
examples, we might suggest that the intervening events in such cases are more under
the control of the agent. For example, (32), repeated here as (55), appears to describe
a situation in which a manipulative person controls the factors which directly govern
the self-confidence of a victim. See also the discussion in Rappaport Hovav & Levin
(2012).
(55) He eroded my self-confidence and my dignity.

What we have said so far is a first step toward an explanation of the distribution
of the different kinds of external arguments which appear with verbs when they
describe changes which come about in the natural course of events. It does not,
however, explain why these verbs are overwhelmingly used in their intransitive
variants when they take concrete objects. For this, we need a theory which accounts
for the (non) appearance of the external argument.
In Rappaport Hovav (2014a), I provide a pragmatic account of the
(non)appearance of external arguments with change of state verbs. I begin with
the following assumption:
(56) “In the description of a change of state, the cause of the change of state is rel-
evant; therefore, since an utterance which specifies the cause of the change of
state is more informative than one which expresses just the change of state, it
is to be preferred, all things being equal.” (Rappaport Hovav 2014a: 23)
More recently, and in a similar vein, Schäfer & Vivanco (2016) have shown that
causatives and their non-causative counterparts form scalar pairs in the sense that
the causatives entail the non-causatives.25 This being the case, when faced with the
option of describing an event using the causative or the anticausative, the choice of
the anticausative will often generate an implicature that the stronger statement (the
causative) is not true. They show that when the causative follows the negation of
the intransitive variant, as in (57), the negation shows properties of metalinguistic
negation. See Schäfer & Vivanco (2016) for details.
25 Schäfer & Vivanco’s main goal is to argue against an analysis of the anticausative variant of
causative alternations verbs (see Beavers & Koontz-Garboden 2013a, b) which assigns the verbs in
this variant a reflexive representation. Under such an analysis the causative variant does not entail
the anticausative. I find Schäfer & Vivanco’s arguments fully convincing, and so assume without
further comment that the causative variant entails the anticausative.
(57) The vase did not break, (#but) you broke it. (Schäfer & Vivanco 2016: 48a)
However, given that changes of state in general have causes, it is not clear what
it means that the use of the intransitive variant generates the implicature that the
stronger statement is not true. That is, the use of a sentence such as The vase
broke, clearly does not generate the implicature that there was no cause for the
breaking. But there may be other reasons not to use the stronger statement other
than the assumption that the stronger statement is false. For example, if the cause is
recoverable in some way from context, then the sentence with the cause expressed
is no longer more informative than the corresponding sentence which expresses just
the change of state. In fact, it may be considered redundant. In that case, the latter
may be preferred from considerations like Grice’s (1989) Maxim of Manner. There
are a number of factors which may lead to the cause being recoverable. For example,
it may have been mentioned previously in the discourse as in (58):
(58) a. In a fit of rage he threw the plate on the floor. We all came to see what
happened and saw that it broke.
b. Using his bare hands and sharpened sticks, Lame Hawk began to tunnel
under Nimbock’s limp body. He worked tirelessly, ignoring his blistered and
bleeding hands and watching with satisfaction as the ditch deepened
(attested example). (Rappaport Hovav 2014a: 26)
Without the previous context, the intransitive use of deepen in (59b) would sound
infelicitous. That is, just given the causative and anticausative uses without any
context, the following judgments are gotten:
(59) a. They deepened the ditch.

b. *The ditch deepened.
Or it may be the case, that the speaker does not know the cause, in which case,
though the transitive would be more informative, the speaker cannot truthfully
specify the cause. Rappaport Hovav (2014a) illustrates this with the contrast
between following two sentences, taken from McCawley (1978):
(60) a. The door of Henry’s lunchroom opened and two men came in.
b. The door of Henry’s lunchroom opened and two men went in. (McCawley
1978: 246)
McCawley points out that in (60a), the reader infers that either the men or
someone else opens the door, while in (60b) the reader infers that the men did not
open the door themselves. The logic behind these inferences stems from the fact
that the verb come, as a deictic verb expressing motion toward the deictic center
(the speaker), puts the speaker inside the lunchroom, while the verb go, as a deictic
verb expressing motion away from the deictic center, places the speaker outside the
lunchroom. In the first case, then, the speaker inside the lunchroom is assumed not to
see who it is that opened the door. Therefore, though the transitive variant would be
more informative, the speaker will leave the agent unmentioned if she is ignorant of
who opened the door and so cannot truthfully utter the transitive variant. However,
in the second case, with the speaker placed outside of the lunchroom, chances are
that the speaker sees the men. In such a situation, if the speaker sees the men, (56)
would require her to use the transitive version. Since she does not, one can infer that
this is because someone else has opened the door and the speaker does not mention
whom, because she cannot see the agent of the opening.
When a change of state comes about in the natural course of events, the
cause may not be deemed relevant. Consider for example, a situation in which
an ice bucket is left on the kitchen table. The assertion The ice melted does not
generate the implicature that there is no cause to the ice melting; there certainly
is a cause for the melting. In this particular case, The room temperature melted
the ice (https://tinyhouseblog.com/tiny-house/winterizing-tiny-house/) may be true,
but specification of the cause seems superfluous and the causative is not more
informative than the anticausative. This is so, because speakers all know what the
cause is. In contrast, in the following example, the cause serves to situate the change
of state at a particular time of day. It then becomes informative.
(61) This railing is just outside my door. On a very cold winter morning before the
temperature melted the ice, I was able to capture this image. The iron work
is long and the ice is also long and cold. https://www.istockphoto.com/gb/
photo/wintery-icey-wonderland-gm1007850930-271907964
It appears, then, that (56) needs to be modified somewhat. It is not the case that
the cause of a change of state is always relevant. Some states are such that they have
a propensity to change in the natural course of events, and this affects the relevance
of the cause of the change of state. We might put it thus:
(62) For a given state and a given entity there is a default expectation of whether
the state (or the degree to which the state holds) will or will not change in the
natural course of events, i.e., whether the entity has the disposition to undergo
a change in state. The cause of a change of state is relevant only if for the
given state and the given entity, there is no default expectation of change.
For example, days lengthen in the natural course of events and not just under
specific circumstances. Though the lengthening of days is fully predictable, it may
be worthy of mention, particularly as a way of making reference to a specific time
of the year as in:
(63) The festival got its name from an ancient law regulating the working hours of
the guild members. Dependent on daylight, the craftsmen worked from dawn
to dusk in winter, but when the days lengthened in spring, they had to rely on
the six o’clock bells to know when to put down their tools. https://www.
swissinfo.ch/eng/exploding-snowman-welcomes-spring/2642254
Because days lengthen in the natural course of events (63) without the mention
of the cause (62) is not violated, and indeed the anticausative of lengthen when the
theme of change is days is by far the most common. It would indeed be odd to
replace the relevant sentence in (64) with
(64) But when the tilt of the Earth toward the sun lengthened the days in the
spring . . .
However, when “the days” is understood to mean not the number of hours of
sunlight, but rather a contract-specified number of work hours, this change does
not come about in the normal course of events, and the transitive version is more
felicitous than the intransitive version.
(65) But Board of Education members said they lengthened the days to ensure
students receive the equivalent of 180 days of instruction. http://news.google.
com/newspapers?nid=1957&dat=19960406&id=g4hGAAAAIBAJ&sjid=r-
kMAAAAIBAJ&pg=1078,1019911 (Rappaport Hovav 2014a: 24)
This particular example underscores an important point. The verb lengthen is
considered an ordinary causative alternation verb. It would be categorized by L&RH
as externally caused and by AAS as cause unspecified. But as these examples
illustrate, whether or not a verb describes a change which occurs in the natural
course of events depends not only on the state lexically encoded in the root, but on
the argument the state is predicated of.
To take another example, corpses decay in the natural course of events. And in
an informal web search it was not difficult to find examples such as ‘The corpse
decayed’ and variations thereof. The causative variant, with a specification of the
cause of decay is, in contrast, difficult to find. Nonetheless, in an article discussing
plants used for preservation of the dead in an Indian village we find the following:
(66) The smoke also gets rid of bacteria and organisms that will decay the corpse.
https://www.fs.usda.gov/Internet/FSE_DOCUMENTS/stelprdb5347125.pdf
Here the cause is still an expected cause, but it becomes relevant because the
sentence is not reporting a change of state but rather the property of the smoke
which gets rid of the cause of the change of state. If changes of state which have been
characterized as internally caused are those which come about in the natural course
of events, we can understand the following properties of verbs which express those
changes, discussed in the literature (M&M, Wright 2002 and Alexiadou 2014):
(67) (i) The intransitive variant is most frequent;

(ii) The verbs typically occur with a restricted range of subjects;
(iii) The cause subject often includes a modifier and modification improves
grammaticality judgments.
We have already discussed (67i) and (67ii). With respect to (iii), Wright (2002)
reports that the transitive versions of what are considered internally caused COS
verbs are consistently ranked lower in acceptability judgment tasks, and that
modification of the subject improves the acceptability rating. (68) and (69) are her
examples.
(68) a. ?Last July, sunlight wilted the begonias.
b. Last July, the intense sunlight wilted the begonias.
(69) a. ?The past summer, moisture rotted the tomatoes.
b. This past summer, extremely moist conditions rotted the tomatoes.
(Wright 2002: 345):
The explanation for this fact will emerge from a deeper understanding of how
scalar alternatives relevant for the description of a change of state are determined.
It is impossible to determine what the causative alternative is for a given change
of state without context. In the case of a sentence such as “I have three children,”
one doesn’t need context to determine that the alternatives are sentences in which
“three” is replaced by other numbers. One needs context only to determine which of
the scalar alternatives is appropriate for the utterance. But for a sentence describing
a change of state, there are any number of possible causes one can supply to form
the causative as the scalar alternative. In a particular situation, in order for a speaker
to decide whether an anticausative or a causative is pragmatically more appropriate
to a particular situation, she must supply the causative with an appropriate cause
subject. I suggest that in the case of the description of change of state which comes
about in the natural course of events, there may be difficulty in choosing a cause
subject.
The distinction between causal factors and actual causes found in discussions of
causation in the philosophical literature might give us some insight to this. Changes
which come about in the natural course of events typically have a variety of causal
factors. For example, a tree grows, a cliff erodes and teeth decay from a variety of
factors which are typically co-occurring, common and predictable. For convenience,
I use Dowty’s (1979) formulation of the distinction between causal factors and
actual causes, but see also Hitchcock & Knobe (2009) for a more recent discussion:
(70) [· CAUSE Ψ ] is true iff (i) · is a causal factor for Ψ , and (ii) for all other ·
such that · is also a causal factor for Ψ , some ¬·-world is as similar, or
more similar, to the actual world than any ¬· -world is.
Most of the causes for internally caused changes of state would fall under ·
since we assume them to be a part of the world as we know it, which is why the
changes they support take place in the natural course of events. For this reason,
these causal factors are not good candidates as actual causes (the particular cause
deemed to be “the” cause of the change of state). The less expected the causal factor,
the more appropriate it is as choice of actual cause in a causal statement (Hitchcock
& Knobe 2009).26 This, I suggest, helps explain why verbs which describe changes
of state which occur in the natural course of events are used intransitively most
frequently.27 Modification, however, helps pick out one of the causal factors as being
unusual or one aspect of the causal factor as being unpredictable or less expected,
and this explains the judgments in (68) and (69).
But when predicated of abstract entities, the change can no longer be considered
one which comes about in the natural course of events, and so the transitive variant
will be preferred. Not surprisingly, this is what we see in Table 7.3, taken from
Table 7.3 Objects of full transitive sentences with internally caused COS verbs (McKoon &
Macfarland p. 841)
Artifacts Nature Animate Body parts Abstract
Blister 3 8
Bloom 1
Corrode 19 1 2 4 22
Deteriorate 3
Erode 3 4 48
Ferment 4 1
Germinate 1
Rot 2 3 1 2
Rust 5
Sprout
Stagnate 1
Swell 1 1 3 12
Wilt 1 8
Wither 7 3
Total 37 18 2 16 101
26 This idea goes back to Mill (1872) who notes “If we do not . . . enumerate all the conditions, it is
only because some of them will in most cases be understood without being expressed, or because
for the purpose in view they may without detriment be overlooked. For example, when we say, the
cause of man’s death was that his foot slipped in climbing a ladder, we omit as a thing unnecessary
to be stated the circumstance of his weight, though quite as indispensable a condition of the effect
which took place.” Some conditions are not mentioned because they are taken to be known to the
listener; stating them explicitly would be superfluous.
27 The way in which actual causes are chosen should not in principle make any distinction between
lexical and periphrastic causatives. Yet in sentences such as The sunlight caused the begonias to
wilt there is no hint of infelicity, as opposed to the lexical causative. We know that the conditions for
lexical causatives are more stringent than the conditions of periphrastic causatives and violating the
conditions gives rise to a sense of infelicity, not just redundancy. See, for example, Martin (2018)
and Bar-Asher Siegal & Boneh (2019). Therefore, it is clear that the explanation given above needs
further elaboration for full coverage of the data.
M&M; in a randomly chosen part of their corpus the objects of transitive sentences
with verbs they classified as internally caused are overwhelmingly abstract.
The transitive use is made more felicitous if the cause subject describes some
unexpected circumstance. Roots or verbs which more easily alternate tend to specify
changes which are compatible with a wider range of entities; sometimes the changes
they specify then do come about in the natural course of events, but sometimes not.
That depends on the entity they are predicated of. There is a difference between
dimensional verbs such as widen or lengthen and verbs classified as internally
caused COS verbs such as rust or corrode. Many entities can change in degree of
width or length and whether they do so in the natural course of events depends on
the particular entity and how the change in value is understood (see, for example
the contrast between the two examples of days lengthening in (64) and (65) above).
However, a blanket statement that alternating verbs appear in the intransitive only
when they describe changes which come about in the natural course of events is
still inaccurate, since, as we have already seen, the anticausative may be licensed
because the cause is recoverable from some other reason, or because it is not known
to the speaker.
7.7 Blossom Verbs as Verbs of Emission
Bloom is taken by M&M to be the prototypical internally caused change of state

verb (p. 833) and Alexiadou takes blossom to represent the class. I suggest, however,
that these verbs, along with flower and sprout should not be considered COS verbs.
A glance at Table 7.1 above shows that bloom, blossom and flower are the only
verbs in M&M’s corpus, besides decay,28 which have a .00 probability of transitive
occurrence. They point out that there were two examples of transitive bloom in their
corpus, and provide one such example: it [a shrub] blooms white flowers in summer
(p. 838). But notice that in this example, the subject of the transitive sentence is
not a cause. This sentence does not have the paraphrase The shrub caused the white
flowers to bloom. A better paraphrase would be: The shrub produced while flowers
(by blooming).29 In fact, careful scrutiny of the grammatical behavior of the verbs
28 While decay also has a 0 probability of transitive occurrence in M&M’s data, this verb shows
properties of being a COS verb. For example, it is not difficult to find examples of “X decayed the
teeth” on the web, and the subject of the transitive can always be analyzed as a cause.
29 A piece of evidence that the subjects of these verbs are not causers is the fact that these verbs can
appear with the source/emitter subject without the second argument. This is shown for a standard
substance emission verb in (i) and for blossom in (ii). In contrast, internally caused COS verbs
cannot in general appear with the cause argument but not the theme of the COS (iii):
(i) The wound oozed (blood) for several days.
(ii) The rod blossomed (almond flowers).
(iii) The high tides eroded *(the coast).
bloom, blossom, and flower indicates that they do not generally show patterns of
argument realization typical of COS verbs. I argue that they constitute a special
class of verbs of substance emission; I will call them blossom verbs. I leave the
discussion of sprout to the end of this section.
Verbs of substance emission typically have two arguments: an emitter/source
and an emitted substance/entity (Levin 1993: 237), though often the non-subject
argument need not be explicitly expressed. These verbs show what Levin calls the
source/substance alternation. Either the emitter/source can be the subject, in which
case the emitted substance/entity can be a direct object (the (a) sentences below),
or the emitted substance/entity can be the subject, and the source expressed in a PP
with a source-marking preposition (the (b) sentences below).
(71) a. The well gushed (oil).

b. Oil gushed (from the well).
(72) a. The wound oozed (pus).
b. Pus oozed (from the wound).
(73) a. The faucet dripped (water).
b. Water dripped (from the faucet).
In the same way, the blossom verbs take two arguments, an emitter – typically a
plant or tree – and an emitted entity – a kind of flower or blossom.30 Since I take
them to be verbs of emission, it is not surprising that either the emitted entity or
the emitter can be subject; i.e., these verbs show a variety of the substance/source
alternation. For each verb, I give examples of emitter subject and emitted entity
subject. Note the source phrases in the emitted entity-subject variant (75, 77, 79). In
some of the examples, they take goal phrases, where the source is implicit (79b).
Bloom emitter subject
(74) a. The chinaberry trees in front of my house bloomed tiny white flowers,
which fell like snow into the puddles on the sidewalk below. (Dana Sachs,
The House on Dream Street, Algonquin Books, Chapel Hill, NC, 2000,
p. 61)
b. Near the abandoned Islip Speedway, he pointed out a rare alpine family
member known as pyxie. A one-inch-wide, low-growing perennial shrub,
it blooms white flowers in summer. (Anne C. Fullam, “Botanist-Sleuth
Searches Out Long-hidden Plants”, Section 11LI, New York Times,
November 8, 1987, p. 2)
Bloom emittee subject
(75) a. Tiny white flowers bloomed from the chinaberry tree.

b. The bud bloomed from the branch of a cactus. (The Girl from: Based
on a True Story, Vanessa Voth, Freisen Press, p. 148)
30 It
is perhaps not accidental that the blossom verbs are denominal – or are at least zero-related to
nouns, all referring to the emitted entity.
Blossom emitter subject
(76) a. They resemble a tomato plant and each branch has blossomed flowers
on each level of the leaves. https://questions.gardeningknowhow.com/
tag/plant-identification-2/page/4/
b. The plant will blossom flowers when light is provided in abundance.
https://www.theaquariumguide.com/articles/crinum-calamistratum
Blossom emittee subject
(77) a. The almonds which blossomed from the rod of Aaron for the tribe of
Levi . . . https://archive.org/stream/arcanacoelestiah06swed_0/
arcanacoelestiah06swed_0_djvu.txt
b. And he came . . . first his head, then his body . . . tall and untidy-haired
like Harry, the smoky, shadowy form of James Potter blossomed from
the end of Voldemort’s wand. Harry Potter and the Goblet of Fire . . . .
Flower emitter subject
(78) a. I am not ignorant that ‘the Ancients’ had frames, probably warmed
green-houses–since they flowered roses at mid-winter–and certainly
conservatories. https://www.gutenberg.org/files/32205/32205.txt
b. It flowered about 50 flowers each plant for the whole season and got a
very bad case of blackspot. https://davesgarden.com/guides/pf/go/948/#b
Flower emittee subject
(79) a. I am a happy content soul until about November, when the last flower
has flowered and the soil gets wet and cold. http://www.hoehoegrow.co.
uk/2014/01/
b. Lo & behold this spring it shot out new shoots from the base and a
beautiful bunch of roses flowered forth. https://www.pinterest.com/
suetodd1111/josephine-bonaparte
Emission verbs show unergative behavior typical of verbs of emission in the
variants with emitter subjects, and unaccusative behavior in the variants with emittee
subjects (Levin & Krejci 2019). We find that the blossom verbs display the same
behavior. First, the emitter subject variant can take a direct object (74, 76, 78); the
ability to assign accusative case is a hallmark of unergative verbs. Relatedly, it is
well-known that unergative verbs can appear with a variety of non-subcategorized
objects, whereas unaccusative verbs do not (L&RH 1995, among many others).
Blossom verbs with emitter subjects can appear, for example, in fake reflexive
resultative constructions, as in (80).
(80) a. . . . until the plants drooped as though they had bloomed themselves to
death. (Gwen Bristow, “The Handsome Road”) https://books.google.com/
books?isbn=1480485160
b. Having bloomed themselves silly several weeks ago, the daffodils are now
busy photosynthesizing. http://gardenersapprentice.com/gardeningtips/
growing-growing/
c. Asters and golden-rod and Spanish needles had blossomed themselves into
seedy exhaustion long ago. (Elinor Brooke, “Out of the Fire”, The American
Magazine 21, 1886, p. 529; https://books.google.com/books?id=
ingqAAAAMAAJ
d. However the original plant transferred to the greenhouse (as insurance?)
had flowered itself to death. (Trevor Wray, “Orostachys”, Northant’s News
23.1, Spring 2012, British Cactus and Succulent Society; http://northants.bcss.
org.uk/nl231/nl231oro.htm
On the other hand, the emitted entity subject variant is unaccusative, being a verb
of directed motion (Levin & Krejci 2019). They do not appear transitively; there are
no attested examples of fake reflexive resultatives for these verbs on their emittee
subject variant.
As already mentioned, the transitive uses presented so far, are not causative. The
transitive variants of blossom verbs do not passivize:
(81) *White flowers were blossomed by the tree.
The fact that they do not passivize is not surprising – transitive uses of verbs of
emission do not passivize either.
(82) a. The wound oozed pus.
b. *Pus was oozed by the wound.
(83) a. The well gushed oil.
b. *Oil was gushed by the well.
I do not provide an explanation for the lack of passivization for these verbs; the
important point is that the blossom verbs pattern like verbs of substance emission.
We might ask whether there are true causative variants of emission verbs, and if
there are, whether they are causatives of the emitter-subject variant or the emittee
subject variant. Wright (2002) brings an example of a causative of an emitter subject
use of blossom and in (85) I bring a passive version of an emitter-subject use of
blossom. But these are very, very rare.
(84) Early summer heat blossomed fruit trees across the valley.
(85) We learned about the magnificent Methuselah tree, a date palm that was
blossomed by Dr. Elaine Solowey, from the remains of a seed that was
found from 2000 years ago on Masada. Since that original germination, Dr.
Solowey has sprouted many other trees, and shortly one will bear fruit.
https://seniorisraelexperience.wordpress.com/2018/01/09/environment-of-
the-desert-and-kibbutz-life/
This is not surprising. While there are emission verbs which have causative
variants, they are typically from the sound emission class and to a lesser degree
the light emission class. See Levin et al. (1997) for a discussion of causative uses
of sound emission verbs. Before concluding this section, it is worth noting that the
verb sprout, also seems to be a blossom verb. It too, is denominal, and displays a
source/substance alternation:
Sprout emitter subject
(86) a. As they grew, they sprouted buds and then bloomed. (Music for Alice, by
Allen Say, Boston, Houghton Mifflin Company, 2004. p. 16)
b. In time they sprouted buds, then burst open into papery purple, pink,
yellow, or white flowers. https://kiwords.blogs.com/kiwords/2005/05/_this_
afternoon.html
Sprout emittee subject
(87) a. Many buds sprouted from the stumps in April. From “Dormancy and
spring development of lateral buds in mulberry”, Physiologia Plantarum 75.2
b. little green sprout in the spring sprouted from the earth. https://stock.adobe.
com/images/little-green-sprout-in-the-spring-sprouted-from-the-earth/
128372107
However, unlike other blossom verbs, sprout does seem to have causative uses
(88), and, concomitantly, have passive uses as well (89). The subjects in (88) can
range from agents to natural causes, typical of causative uses. As in the blossom
examples of (84) and (85), these are causatives of the emitter subject variant.
(88) a. They simply sprouted the beans they carried. http://europe.chinadaily.
com.cn/epaper/2018-05/04/content_36137499.htm
b. the company which sprouted the seeds is able to trace the batch supplied to
the supermarket . . . http://traceabilitytraining.food.gov.uk/module11/
overview_1.html#.XDto-FwzbIU
c. the warm, rainy weather sprouted the wheat before it could be gathered.
https://cdnc.ucr.edu/?a=d&d=SDU18721102.2.32
(89) a. The seeds were sprouted by five sprout producers and then sold. http://
www.outbreakdatabase.com/search/?vehicle=sprout
b. These beans were sprouted by Vinitha and they tasted really crunchy!
c. . . . kumquat blossoms and jasmine. In earlier times, shallot, onion and
madder plants were sprouted by the same method. https://www.
livinginseason.com/celebrations/chinese-new-year/
I do not have an explanation for the unusual behavior of sprout, but offer some
speculative remarks. The overriding condition on the addition of a cause argument
is directness of causation between the causing and caused events as we have seen.
In the case of sprouting, as opposed to blossoming and blooming, the event is one
of the first stages in a plant’s life cycle. This fact might facilitate the construal of the
causing event as a direct cause.31
Summarizing this section, it appears that it is best to remove the class of blossom
verbs from the list of COS verbs. These verbs in general can appear in both transitive
and intransitive variants, but the transitive variant is not an instance of a COS verb
with a cause subject, but rather an emission verb with an emitter subject and a
substance/emitted entity as object.
7.8 Non-COS Internally Caused Verbs
It is probably significant that of the verbs in (11), those classified by L&RH

as internally caused, only the COS verbs are unaccusative. All the others are
unergative. We might then ask what the verbs in (11a–c) have in common by virtue
of which they lexically select an external argument, thus precluding the formation
of lexical causatives. I suggest that if there is a class of verbs which deserves to
be considered internally caused, it is this class, and not the class of COS verbs
which have ironically received the most attention in the context of discussion of
internal vs. external causation. The alternative characterization of internal causation
in L&RH, in (13) and (14) above seems more appropriate for this class of verbs.
It seems to me that force-dynamic analyses of event structure (Talmy 1988; Croft
1990, 1991; Copley & Wolff 2014; Copley 2019; Wolff et al. 2010; see relatedly,
Folli & Harley 2007) can provide further insight. In particular, it appears to me
that the notion of internal causation can be translated into a notion such as ‘force
creator,’ developed in Wolff et al. (2010). Wolff et al. discuss three categories of
force-creation, two of which are relevant to us here. They discuss force creation
through energy conversion, mainly when an entity has some internal source of
energy which translates into some kind of action (kinetic energy). This seems to be
relevant to the classes in (11a), the agentive verbs and (11b), non-agentive activities.
Somewhat speculatively, we might say that verbs of emission, on their unergative
variants, appear with subjects that are described as emitters. An emitter can be
conceptualized as creating a force which causes the emitted entities to be emitted. If
internal causation can be equated with force-creation, we might then propose that in
general, force-creators are realized as external arguments, and this is the reason that
they do not have causative forms which always involves the addition of an external
argument. Ideally one would want to avoid the pitfalls of the use of terms such as
internal causation in a purely intuitive way, and find a syntactic/semantic way of
grounding the notion of force creator. This, however, is the topic for another paper.
31 Ithank Beth Levin for enlightening discussion of this point. I should point out that there are
other properties of these verbs which deserve more attention. For example, many of them appear
in existential there constructions and in locative inversion construction, suggesting that they may
act as verbs of appearance.
7.9 Conclusion
The verbs which L&RH classed together as internally caused do not form a
semantically unified or a grammatically relevant class. Of these verbs, the sub-class
of COS verbs was shown to have argument realization properties which distinguish
them from the other verbs. In particular, these verbs participate in the causative
alternation. I suggested that the most appropriate characterization of the class of
COS verbs which have been termed internally caused is the class of verbs which
typically (but not always) describe events which come about in the natural course
of events, sometimes in the context of specific ambient conditions and sometimes
in general. It is this property which determines the pattern of participation of
these verbs in the causative alternation. The way in which a verb participates in
the causative alternation is partly dependent on the entity the change of state is
predicated of, suggesting that the relevant notion of internal causation is a property
of predicates or event descriptions, and not a property of verbs or roots. The
same principles which govern the ways in which verbs participate in the causative
alternation relative to the theme of the change of state is uniform across all change
of state verbs. I have therefore argued against a grammatically relevant distinction
between COS verbs or roots as internally or externally caused. Internally caused
COS verbs are unaccusative, as all other anticausative COS verbs. In contrast, the
other sub-classes of internally caused verbs discussed in L&RH are all unergative.
I suggested that what these verbs have in common is that they have subjects which
can be characterized as force creators. This determines the unergative syntax and
also explains why these verbs do not participate in the causative alternation.
*Acknowledgments I thank Beth Levin for discussion of much of the material presented in this
paper and for helpful comments on drafts of the paper. Thanks also to three anonymous reviewers
for extremely helpful comments on an earlier version of the paper, which led to what I hope are
significant improvements.
References
Alexiadou, A. (2010). On the morpho-syntax of (anti-)causative verbs. In M. Rappaport Hovav, E.

Doron, & I. Sichel (Eds.), Syntax, lexical semantics and event structure (pp. 177–203). Oxford:
Alexiadou, A. (2014). The problem with internally caused change-of-state verbs. Linguistics 2014,
52(4), 879–909.
Alexiadou, A., & Anagnostopoulou, E. (2019). Novel experiencer-object verbs and clitic doubling.
Syntax. https://doi.org/10.1111/synt.12172.
Alexiadou, A., & Anagnostopoulou, E. (this volume). Experiencers and causation. In E. A. Bar-
Asher Siegal & N. Boneh (Eds.), Perspectives on causation. Cham: Springer.
Alexiadou, A., Anagnostopoulou, E., & Schäfer, F. (2006). The properties of anticausatives cross-
linguistically. In M. Frascarelli (Ed.), Phases of interpretation (pp. 187–212). Berlin: Mouton.
alternations: A layering approach. Oxford University Press.
Bar-Asher Siegal, E. A., & Boneh, N. (2019). Sufficient and necessary conditions for a non-unified
analysis of causation. In R. Stockwell, M. O’Leary, Z. Xu, & Z. L. Zhou (Eds.), Proceedings of
the 36th west coast conference on formal linguistics (pp. 55–60). http://www.lingref.com/cpp/
wccfl/36/index.html.
Beavers, J. (2008). Scalar complexity and the structure of events. In J. Dölling & T. Heyde-Zybatow
(Eds.), Event structures in linguistic form and interpretation. Berlin: Mouton de Gruyter.
Beavers, J., & Koontz-Garboden, A. (2013a). In defense of the reflexivization analysis of
anticausativization. Lingua, 131, 199–216.
Beavers, J., & Koontz-Garboden, A. (2013b). Complications in diagnosing lexical meaning: A
rejoinder to Horvath & Siloni (2013). Lingua, 134, 210–218.
Chierchia, G. (1989/2004). A semantics for unaccusatives and its syntactic consequences. In A.
Alexiadou, A. Elena & E. Martin (Eds.), The unaccusativity puzzle: explorations of the syntax–
lexicon interface (pp. 22–59). Oxford: Oxford University Press.
Copley, B. (2019). Force dynamics. In R. Truswell (Ed.), Oxford handbook of event structure.
versa. In B. Copley & F. Martin (Eds.), Causation in grammatical structures (pp. 11–57).
Croft, W. (1990). Possible verbs and the structure of events. In S. Tsohatzidis (Ed.), Meanings and
prototypes: Studies in linguistic categorization (pp. 48–73). London: Routledge.
Croft, W. (1991). Syntactic categories and grammatical relations. Chicago, IL: University of
Chicago Press.
Doron, E. (2003). Agency and voice: The semantics of the semitic templates. Natural Language
Dowty, D. (1979). Word meaning and Montague Grammar. Dordrecht: Kluwer.
Fillmore, C. (1970). The grammar of hitting and breaking. In R. A. Jacobs & P. S. Rosenbaum
(Eds.), Readings in English transformational grammar. Waltham: Ginn.
Fodor, J. (1970). Three reasons for not deriving ‘kill’ from ‘cause to die’. Linguistic Inquiry, 1,
429–438.
Folli, R., & Harley, H. (2006). On the licensing of causatives of directed motion: Waltzing Matilda
all over. Studia Linguistica, 60(2), 121–155.
Folli, R., & Harley, H. (2007). Teleology and animacy in external arguments. Lingua, 118, 190–
202.
Goldberg, A. (1995). Constructions: A construction grammar approach to argument structure.
Chicago: University of Chicago Press.
Grice, H. P. (1989). Studies in the way of words. Cambridge: Harvard University Press.
Harley, H., & Noyer, R. (2000). Licensing in the non-lexicalist lexicon. In B. Peeters (Ed.), The
lexicon/encyclopedia interface (pp. 349–374). Amsterdam: Elsevier.
Haspelmath, M. (1993). More on the typology of inchoative/causative verb alternations. In B.
Comrie & M. Polinsky (Eds.), Causatives and transitivity (pp. 87–120). Amsterdam: John
Benjamins.
Haspelmath, M. (2017). Universals of causative and anticausative verb formation and the spon-
taneity scale. Lingua Posnaniensis, 58(2), 33–63.
Haspelmath, M., Calude, A., Spagnol, M., Narrog, H., & Bamyaci, E. (2014). Coding causal–
noncausal verb alternations: A form–frequency correspondence explanation. Journal of
Linguistics. https://doi.org/10.1017/S0022226714000255.
Horvath, J., & Siloni, T. (2011). Anticausatives: Against reflexivization. Lingua, 121, 2176–2186.
Horvath, J., & Siloni, T. (2013). Anticausatives have no cause(r): A rejoinder to Beavers & Koontz-
Garboden. Lingua, 131, 217–230.
Koontz-Garboden, A. (2009). Anticausativization. Natural Language and Linguistic Theory, 27,
77–138.
Kratzer, A. (1996). Severing the external argument from its verb. In J. Rooryck & L. Zaring (Eds.),
Phrase structure and the lexicon (pp. 109–137). Dordrecht: Kluwer.
Kratzer, A. (2005). Building resultatives. In C. Maienborn & A. Wollstein (Eds.), Event arguments:
Foundations and applications (pp. 177–212). Tübingen: Niemeyer.
Krifka, M. (1998). The origins of telicty. In S. Rothstein (Ed.), Events and grammar. Dordrecht:
Kluwer.
Levin, B. (1993). English verb classes and alternations. Chicago: Chicago University Press.
Levin, B. (1999). Objecthood: An event structure perspective. Proceedings of CLS, 35, 223–247.
Levin, B. (this volume). Resulatives and causation. In E. A. Bar-Asher Siegal & N. Boneh (Eds.),
Perspectives on causation. Cham: Springer.
Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax–lexical semantics
interface. Cambridge, MA: MIT Press.
Levin, B., & Krejci, B. (2019). Talking about the weather: Two Construals of precipitation events
in English, Glossa. A Journal of General Linguistics, 4(1), 58.
Levin, B., Song, G., & Atkins, S. (1997). Making sense of corpus data: A case study of verbs of
sound. International Journal of Corpus Linguistics, 2, 23–64.
Lewis, D. (1973). Causation. Journal of Philosophy, 70(17), 556–567.
Marantz, A. (1997). No escape from syntax: Don’t try morphological analysis in the privacy of your
own lexicon. In A. Dimitriadis, L. Siegel, C. Surek-Clark, & A. Williams (Eds.), University
of Pennsylvania working papers in linguistics (pp. 201–225). Philadelphia: University of
Philadelphia.
verbs. Proceedings of Sinn und Bedeutung, 22(2), 107–124.
McCawley, J. (1978). Conversational implicature and the lexicon. In P. Cole (Ed.), Syntax and
semantics 9: Pragmatics (pp. 245–259). New York: Academic.
McKoon, G., & Macfarland, T. (2000). Externally and interally caused change of state verbs.
Language, 76, 833–858.
Mill, J. S. (1872). System of logic. In J. M. Robson (Ed.), Collected works of John Stuart Mill (Vol.
VII and VIII). Toronto: University of Toronto Press.
Nedjalkov, V., & Silnitsky, G. (1973). The typology of morphological and lexical causatives. In F.
Kiefer (Ed.), Trends in soviet theoretical linguistics (pp. 1–32). Dordrecht: Reidel.
Neeleman, A., & van der Koot, H. (2012). The linguistic expression of causation. In M. Everaert,
T. Siloni, & M. Marelj (Eds.), The theta system: Argument structure at the interface. Oxford:
Piñón, C. (2001). A finer look at the causative-inchoative alternation. In Proceedings of semantics
and linguistic theory 11. Ithaca: Cornell Linguistics Circle.
Pylkkänen, L. (2008). Introducing arguments. Cambridge, MA: MIT Press.
Ramchand, G. (2008). Verb meaning and the lexicon: A first phase syntax. Cambridge: Cambridge
University Press.
Rappaport Hovav, M. (2008). Lexicalized meaning and the internal structure of events. In S.
Rothstein (Ed.), Theoretical and crosslinguistic approaches to the semantics of aspect (pp.
13–42). Amsterdam: John Benjamins.
Rappaport Hovav, M. (2014a). Lexical content and context: The causative alternation in English
revisited. Lingua, 141.
Rappaport Hovav, M. (2014b). Building scalar changes. In A. Alexiadou, B. Borer, & F. Shcaefer
(Eds.), The syntax of roots and the roots of syntax (pp. 259–281). New York: Oxford University
Press.
Rappaport Hovav, M., & Levin, B. (2002). Change of state verbs: Implications for theories of
argument projection. In Proceedings of the 28th annual meeting of the Berkeley Linguistics
Society (pp. 269–280).
(pp. 21–38). Oxford, UK: Oxford University Press.
Rappaport Hovav, M., & Levin, B. (2012). Lexicon uniformity and the causative alternation. In M.
Everaert, M. Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the interface
(pp. 150–176). New York: Oxford University Press.
Reinhart, T. (2002). The theta system – an overview. Theoretical Linguistics, 28, 229–290.
Rothstein, S. (2004). Structuring events. Dordrecht: Kluwer.
Schäfer, F. (2007). By itself. Ms., Universität Sttutgart.
Schäfer, F. (2009). The causative alternation. Language and Linguistics Compass, 3(2), 641–681.
Schäfer, F. (2012). Two types of external argument licensing – the case of causers. Studia
Linguistica, 66(2), 128–180.
Schäfer, F., & Vivanco. (2016). Anticausatives are weak scalar expressions, not reflexive expres-
sions. Glossa: A Journal of General Linguistics, 1(1), 18. 1–36.
Shibatani. (1978). The grammar of causative constructions: A conspectus. In M. Shibatani (Ed.),
The grammar of causative constructions, syntax and semantics (p. 6). New York: Academic.
Smith, C. S. 1970. Jespersen’s ‘move and change’ class and causative verbs in English. Linguistic
and literary studies in honor of Archibald A. Hill. Vol. 2: Descriptive linguistics, ed. by
Mohammad A. Jazayery, Edgar C. Polomé & Werner Winter, 101–109. The Hague: Mouton
de Gruyter.
Solstad, T. (2009). On the implicitness of arguments in event passives. In A. Schardl, M. Walkow,
& M. Abdurrahman (Eds.), Proceedings of NELS 38 (Vol. 2, pp. 365–374). Amherst: GLSA.
Talmy, L. (1988). Force dynamics in language and cognition. Cognitive Science, 12, 49–100.
Wolff, P., Jeon, G.-H., Klettke, B., & Yu, L. (2010). Force creation and possible causers across
languages. In B. Malt & P. Wolff (Eds.), Words and the mind: How words capture human
experience (pp. 93–110). Oxford/New York: Oxford University Press.
Wright, S. K. (2001). Internally caused and externally caused change of state verbs. Doctoral
dissertation. Evanston, IL: Northwestern University.
Wright, S. (2002). Transitivity and change of state verbs. BLS, 28, 339–350.
Chapter 8
Aspectual Differences Between Agentive
and Non-agentive Uses of Causative
Predicates
Fabienne Martin
Abstract This paper aims to provide an account for why, across languages, the
zero-change (or failed-attempt) use of causative predicates is easier to obtain with
agent subjects than with causer subjects. The paper is structured as follows. Sec-
tion 8.2 reports experimental studies suggesting that the degree of acceptance of the
zero-change use at study varies across languages and across types of causative verbs,
focusing on Mandarin run-of-the-mill (extensional) monomorphemic causative
verbs on one hand, and French and English defeasible (modal) causative predicates
on the other. This difference is accounted for in Sect. 8.7. Section 8.3 identifies the
source of the zero-change use for these two sets of languages, namely outer aspect
for Mandarin extensional causatives, and sublexical modality for English or French
defeasible causatives. Section 8.4 spells out an issue raised by Martin’s (2015)
account of the link between agentivity and non-culmination. Section 8.5 shows how
the Voice head introducing agents vs. causers combines with causative VPs, and how
the semantic difference between these two voice heads influences the interpretation
of the VP-event, and, in particular, the way the causing event type denoted by the VP
is tokenized. More precisely, it is argued that in the agentive use, the causative event
type denoted by the VP is ‘fleshed out’ by complex events composed of an action
of x and a change-of-state (CoS) of the theme’s referent y, whereas in the non-
agentive use, the very same causing event type is fleshed out by CoS of the theme’s
referent only, themselves caused by the eventuality denoted by the subject. On this
view, if we abstract away from the external argument, a non-agentive causative VP
is interpreted the same way as its anticausative counterpart. It is then argued that the
semantic difference between the two voice heads ultimately explains why typically,
zero-change uses of causative VPs are acceptable with agents only, starting with
extensional causative verbs in Sect. 8.6.1, and then addressing defeasible (modal)
causative verbs in Sect. 8.6.2. Section 8.8 accounts for why the zero-change reading
is occasionally accepted by some speakers even with a causer subject.
F. Martin ()
Humboldt-Universität zu Berlin, Berlin, Germany
e-mail: fabienne.martin@hu-berlin.de

https://doi.org/10.1007/978-3-030-34308-8_8
258 F. Martin
Keywords Telicity · Perfectivity · Non-culminating accomplishments ·

Agentivity · Causation · Extensional causative verbs · Defeasible causative
verbs · Anticausative verbs · Sublexical modality · Agents · Causers · Voice ·
French · English · Mandarin
8.1 Introduction
A standard assumption in the literature on argument structure is that external

arguments are not arguments of their verbs (Kratzer 1996).1 A not uncommon
conclusion from this assumption is that the alternation between agent (animate) and
causer (inanimate) external arguments of causative predicates is irrelevant for the
aspectual properties of the VP, for they lie outside the event structure relevant for
the calculation of these properties.
This conclusion has been challenged in recent work establishing that the
aspectual properties of some lexical causative predicates very much vary with the
thematic properties of the subject. The generalization observed across languages is
that with a subset of causative verbs (whose extension and properties partly vary
across languages), a change of the theme’s referent is implied, but crucially not
entailed, by lexical causative statements, if the subject is associated with some
agentive properties. By contrast, a causative sentence built with a verb of the same
set tends to entail, or at least much more strongly implies, that at least a part of
a change developing towards a result state of the type encoded by the predicate
(henceforth P-state) occurs when the subject is a (non-instrumental) inanimate entity
or an accidental agent.2 This is what Demirdache & Martin (2015) call the Agent
Control Hypothesis, which says that so-called failed-attempt or zero-change uses
of causative predicates require the predicate’s external argument to be associated
with ‘agenthood’ properties. Many (genetically unrelated) languages confirm this
correlation; see Jacobs (2011) on Salish languages, Demirdache & Martin (2015),
Liu (2018) & van Hout et al. (2017) on Mandarin, Park (1993) and Lee (2015) and
Beavers & Lee (Forthcoming) on Korean, Kratochvíl & Delpada (2015, 230) on
Abui (Papuan), Sato (2019) on Indonesian.
Take for instance the Mandarin example (1). The first clause of this sentence
is true not only if Yuehan successfully burned the book, but also if he put the
book into the fire, and the book didn’t start burning at all before I took it away
(because it was too humid to catch fire, for instance). Similarly, the first clause of
(2) may also be true if Lulu tried to close the door, but the door didn’t start closing
because something blocked it. Park (1993) and Beavers & Lee (Forthcoming) make
1 Abbreviations used: ACC = accusative; CL = classifier; IMP = imperfective; INT = intransitivizer;

PFV = perfective; NEG = negation; NOM = nominative; PROG = progressive; REFL = reflexive;
TR = transitivizer.
2 Atypical
agents and instruments are beyond the scope of this paper. On the former, see Beavers
& Lee (Forthcoming) and Martin (2016) and references therein.
8 Agentive and Non-agentive Uses of Causative Predicates 259
similar observation for Korean; see also (3), and Sato (2019) reports that failed-
attempt/zero-change uses are also possible in Indonesian. The French or English
examples (5)–(6) are similar, but with an important difference: while the predicates
in (1)–(4) are run-of-the-mill causative predicates, those in (5)–(6), although also
causative, embed a sublexical modal operator, and are in that sense not run-of-the-
mill (extensional) causatives (see Sect. 8.3.2).3 I call causative predicates encoding
a modal operator at a sublexical level defeasible causative verbs.
(1) MANDARIN, Demirdache & Martin (2015)

Yuēhàn shāo le tā-de shū, dàn gēnběn méi shāo-zháo.
Yuehan burn PFV 3SG-DE book but at.all NEG.PFV burn-ignite
Literally: ‘Yuehan burned his book, but it didn’t get burnt at all.’
(2) MANDARIN, Martin et al. (2018a)

Lùlu guān-le nèi-shàn mén, dàn gēnběn méi guān-shàng.
Lulu close-PFV that-CL door but at-all NEG close-up
Literally: ‘Lulu closed that door, but it didn’t get closed at all.’
(3) KOREAN, Jiyoung Choi (p.c.)

Chelswu-nun mwul-ul el-li-ess-una, mwul-i an
Chelswu-TOP water-ACC freeze-CAU-PAST-but water-NOM NEG
el-ess-ta.
freeze-PFV-DEC
Literally: ‘Chelswu froze the water, but the water did not freeze.’
(4) INDONESIAN, Sato (2019)

Esti men-tutup pintu, tapi tidak ter-tutup.
Esti TR-close door but NEG INTR-close
Literally: ‘Esti closed the door, but it didn’t close.’
(5) FRENCH, Martin (2015)

Dr Li m’a soigné, mais je n’ai pas guéri du tout.
dr Li me=has treated but I NEG=has NEG cured at all
‘Dr. Li treated me, but I didn’t recover at all.’
3I take for granted that the predicates used in the examples (1)–(6) have a causative event
structure; for explicit arguments in favour of this assumption, see Koenig & Davis (2001) (English),
Martin & Schäfer (2013) (French, English and German), Martin et al. (2018a) (Mandarin), Park
(1993) and Beavers & Lee (Forthcoming) (Korean). Note that Beavers & Lee (Forthcoming)
analyse the Korean causative predicates licensing a zero-change reading as encoding a sublexical
modal operator, and thus as non extensional causatives, on a par with teach-verbs in English.
The alternative option briefly discussed in Sect. 8.9 is that the Korean predicates are extensional
causative predicates just like in Mandarin (or the English translations of the same verbs), and the
zero-change use is licensed by the simple past marker -ess-.
260 F. Martin
(6) ENGLISH, adapted from Oehrle (1976)

John taught Mary how to iron sheets (although in spite of the fact that she saw
him do it, she still doesn’t know how it is done).
By contrast, sentences built with the same predicates used non-agentively entail at
least a part of a change developing towards a P-state in the theme’s referent (the
zero-change reading is generally not felicitous); see (7)–(12).4
(7) MANDARIN, Demirdache & Martin (2015)

Huǒ shāo le tā-de shū, #dàn méi shāo-zháo.
fire burn-PFV 3SG-DE book but NEG.PFV burn-touch
Intended: ‘The fire burned her book, but it didn’t get burnt at all.’
(8) MANDARIN
Nà-zhen feng guān-le nèi-shàn mén, #dàn gēnběn méi
that-CL wind close-PFV that-CL door but at-all NEG . PFV
guān-shàng.
close-up
Intended:‘That gust of wind closed that door, but it didn’t get closed at all.’
(9) KOREAN, Jiyoung Choi (p.c)

Hanpa-ka gangmwul-ul el-li-ess-una #ganmwul-i an
cold.wave-NOM river-ACC freeze-PAST-BUT river-NOM NEG
el-ess-ta.
freeze-PAST-DEC
Intended: ‘A cold wave froze the river, but the river didn’t freeze.’

Angin men-tutup pintu, #tapi tidak ter-tutup.
wind TR-close door but NEG INTR-close
Intended: ‘The wind closed the door, but it didn’t close.’
(11) FRENCH, Martin (2015)

Ce séjour chez sa soeur l’a soignée, (#mais elle n’a pas
this stay at her sister she=has treated but she NEG=has NEG
guéri du tout).
cured at all
Intended: ‘This stay at her sister’s cured her, (but she didn’t recover at all).’
4 Note that in Mandarin, the use of causer subjects is quite restricted with monomorphemic change-
of-state verbs, see in particular Tham (2019) on this point.

(12) ENGLISH, adapted from Oehrle (1976)

John’s demonstrations taught Mary how to iron sheets (#although in spite
of the fact that she saw him do it, she still doesn’t know how it is done).
For Romance and Germanic, Martin & Schäfer (2012) gather around 50 defeasible
causative verbs. In Mandarin, extensional causative monomorphemic verbs allowing
the zero-change use are few (less than twenty), but of very frequent use. Martin et al.
(2018a) note that when these Mandarin verbs are used intransitively, the zero-change
reading is infelicitous; at least a partial change is entailed. This is also true for the
Korean, Indonesian, or French, see (13)–(17); cf. Lyutikova & Tatevosov (2010,
p.64) for a related observation in Karachay-Balkar.

Mén guān le, (#dàn gēnběn méi guān-shàng).
door close PFV but at-all NEG.PFV close-up
Intended: ‘The door closed (but it didn’t get closed at all).’

shū shāo-le, #dàn gēnběn méi shāo-zháo.
book burn-PFV but at all NEG.PFV burn-ignite
Intended: ‘The book burned, but it didn’t get burnt at all.’
(15) KOREAN, Jiyoung Choi (p.c.)

Kang-i el-ess-ciman #kang-i el-ci anh-ass-ta.
river-NOM freeze-PAST-but river-NOM freeze-NEG-PFV-DEC
Intended: ‘The river froze, but the river didn’t freeze.’

Pintu ter-tutup, #tapi tidak ter-tutup.
door INTR-close but NEG INTR-close
Intended: ‘The door closed, but it didn’t close.’
(17) FRENCH
Ma blessure s’est soignée (toute seule), #mais elle n’a pas
My wound REFL.is treated (by itself) but she NEG.has NEG
guéri du tout.
cured at all
Intended: ‘My wound cured (lit.: treated) by itself, but it didn’t cure at all.’
The main goal of this paper is to provide an account for the contrasts presented
above, elaborating on the analysis developed in Martin (2015) and solving some
of its shortcomings. The paper is structured as follows. Section 8.2 reports exper-
imental studies suggesting that the degree of acceptance of the zero-change use
at study varies across languages and across types of causative verbs, focusing on
262 F. Martin
Mandarin run-of-the-mill (extensional) monomorphemic causative verbs on one

hand, and French and English defeasible (modal) causative predicates on the other.
This difference is accounted for in Sect. 8.7. Section 8.3 identifies the source of the
zero-change use for these two sets of languages, namely outer aspect for Mandarin
extensional causatives, and sublexical modality for English or French defeasible
causatives. Section 8.4 spells out an issue raised by Martin’s (2015) account of the
link between agentivity and non-culmination. Section 8.5 shows how the Voice head
introducing agents vs. causers combines with causative VPs, and how the semantic
difference between these two voice heads influences the interpretation of the VP-
event, and, in particular, the way the causing event type denoted by the VP is
tokenized. More precisely, it is argued that in the agentive use, the causative event
type denoted by the VP is ‘fleshed out’ by complex events composed of an action
of x and a CoS of the theme’s referent y, whereas in the non-agentive use, the
very same causing event type is fleshed out by CoS, themselves caused by the
eventuality denoted by the subject. On this view, if we abstract away from the
external argument, a non-agentive causative VP is interpreted the same way as its
anticausative counterpart. It is then argued that the semantic difference between the
two heads ultimately explains why typically, zero-change uses of causative VPs are
acceptable with agents only, starting with extensional causative verbs in Sect. 8.6.1,
and then addressing defeasible (modal) causative verbs in Sect. 8.6.2. Section 8.8
accounts for why the zero-change reading is occasionally accepted by some speakers
even with a causer subject.
8.2 Degree of Acceptance of the Zero-Change Reading of

Causative Predicates
8.2.1 Mandarin
The availability of incomplete interpretations of causative predicates has been tested

in several experimental studies. However, only a few of them distinguish between
situations where no change towards a P-state is observable—Tatevosov’s (2008)
failed attempt situations, e.g., the door doesn’t even start closing while an agent
tries to close it—and situations such that a change towards a P-state obtains,
although not leading to a P-state—Tatevosov’s (2008) partial result situations, e.g.,
the door started closing, but did not end up closed. These two types of incomplete
situations are conflated in Chen’s (2017) truth-value judgement tasks on adult vs.
child Mandarin, and in Arunachalam & Kothari’s (2010) truth-value judgement
tasks on adult Hindi. Cross-linguistically, the partial-change reading of causative
predicates is much less constrained than the zero-change use; it is felicitous for a
much broader range of predicates and does not require an agentive subject as the
zero-change reading does (Demirdache & Martin 2015). For Mandarin, however,
van Hout et al. (2017) and Liu (2018) carefully distinguish them. Besides, they
test these predicates in both agentive and non-agentive uses, and thus enable one
to evaluate the impact of the thematic properties of the external argument on the
aspectual properties of the ensuing causative statement.
The results of the truth-value judgement tasks reported in van Hout et al. (2017)
and Liu (2018) are interesting in two respects. Firstly, they confirm that the change
inference triggered by some Mandarin causative simple (monomorphemic) verbs
seems indeed defeasible when they are used with an agent subject, while change
is perceived as entailed by most speakers when the same predicates are used with a
causer subject. These results also suggest that even with an agent subject, the change
inference is quite strong; for instance, among the 30 adult speakers she tested, Liu
(2018) observed a mean of 40% acceptance (across all verb types) for perfective
causative statements in a situation where no change towards a P-state obtains. This
is certainly higher than what is observed by van Hout et al. (2017) with extensional
causative predicates such as open in adult English, Dutch, Spanish or Basque (where
none of the adults tested accepted the zero-change use for these predicates in
perfective sentences), but it is nevertheless far from unconditional acceptance. Liu
(2018) also reports an important variation from predicate to predicate; for instance,
she observes that while 70% of the subjects accept the perfective form of zhé ‘cut’
in a zero-change situation, only 37% of them accept it for the perfective form of the
verb kāi ‘open’ (see Liu (2018) and Martin et al. (2018b) for an explanation of this
difference).
In summary, the available data suggest that the zero-change reading of causative
simple (monomorphemic) verbs in Mandarin, although possible, is nevertheless
quite restricted. This restriction has yet to be accounted for.
8.2.2 Romance and Germanic
Experimental data on the interpretation of defeasible causatives such as teach in

Romance or Germanic are scarce, but the available data suggest that in striking
contrast with Mandarin, these predicates used with an agentive subject are very
easily accepted by adult speakers in a context where the targeted change does not
obtain at all. For instance, Kazanina et al.’s (Forthcoming) Experiment 1 shows
that 90% of the 29 adult speakers tested accepted sentences such as Jane threw
the frisbee to Woolly as a description of an intended but failed transfer (e.g.,
situations where Woolly fails to catch the frisbee).5 A paper and pencil judgement
survey I conducted on French verbs led to similar results. The survey tested three
conditions (Agent, Instrument and Causer subjects) with two different predicates,
5 Kazanina et al. use the to-variant of these verbs, which has been claimed to license the failed
attempt use more easily than the double-object variant. But Oehrle (1976) shows that many verbs
of transfer of possession entail successful transfer on either variant, while with agentive subjects,
verbs such as offer fail to entail caused possession in either variant; see also the discussion in
Rappaport Hovav & Levin (2008, section 5).
264 F. Martin
Table 8.1 SURVEY 1: Mean score judgements on a [0–5] scale for the zero-result use of soigner
and enseigner (0 = totally unacceptable; 5 = totally acceptable)
N = 19 AGENT INSTRUMENT CAUSER
soigner ‘treat/cure’ 4,8 2,8 1,7
enseigner ‘teach’ 4,8 4,1 2,3
namely soigner ‘treat/cure’ and enseigner ‘teach’. Participants (N = 19) had to rate
causative statements followed by a denial of causal efficacy (sentences such as (5)
vs. (11) above), on a [0–5] scale (0 = totally unacceptable; 5 = totally acceptable). As
the results in Table 8.1 shows, causal efficacy is perceived as entailed with causer
subjects, but can be very easily denied with agents, and this at a much higher level
than with extensional causative monomorphemic verbs in Mandarin.
We therefore have one research question more on our agenda: why is the change
inference with agents stronger with Mandarin extensional causatives than with
French or English defeasible causatives? This question is addressed in Sect. 8.7.
In the next section, I turn to the question of the source of the zero-change use of
causative predicates in the different types of languages under study.
8.3 The Source of the Zero-Change Use Across Languages
8.3.1 Via Outer Aspect
For South and East Asian languages such as Thai and Hindi, Koenig & Muansuwan
(2000) and Altshuler (2014, 2017) suggested that the source of incomplete (non-
culminating) interpretations is the perfective marker in these languages. As these
authors argue, the definition of the perfective adopted for languages such as French
or English is not appropriate for languages such as Thai or Hindi. According to
this definition, the perfective operator PFV entails that the event e it existentially
quantified over falls under the respective predicate P , and since predicates denote
properties of complete events, e is complete with respect to P , see the definition in
(18), where C stands for ‘complete’ and P is a variable for an eventuality predicate
(the neo-Kleinian relation between the topic time and the event time, not relevant
for our purposes, is here ignored; see Altshuler (2016) for a comparison between
Kleinian and non-Kleinian approaches of (im-)perfectivity).
(18) PFVC = λP ∃e[P (e)] (perfective requiring event completion)
These authors propose to distinguish two types of perfective operators. The

perfective in Thai or Hindi entails event maximality, but not event completion, and
is in that sense a partitive operator (Altshuler 2014): the reported event has to cease,
but does not necessarily culminate.
Altshuler (2014) offers a modal definition of event maximality, according to
which MAX(e, P ) is satisfied if e is a complete P -event or ceases to develop

further towards a P -event in the actual world. Simplifying things, Altshuler’s (2014)
definition is as follows:6
(19) MAX(e, P ) :=
a. e is a part of a possible P -event and
b. e is not a proper part of any actual event that is part of a possible P -event.
Take for instance the first clause of (20a). Its semantic representation is given in
(20c), where the variables e range over events and the variables s over states, and
MAX(e, P ) indicates that e is maximal with regard to the predicate P as defined
in (19) (the full derivation is in (52)). The formula in (20c) is true if there is an
event e such that e is maximal with regard to the Lulu-cause-that-door-to-be-open
event property (i.e., if e is a complete Lulu-cause-that-door-to-be-open event, or
a proper part of a possible Lulu-cause-that-door-to-be-open event that ceases to
develop further towards completion). The infelicity of (20a) is due to the fact that
the second clause indicates that the event e described in the first one is not maximal
(since e is said to develop further towards a possible Lulu-cause-that-door-to-be-
open event). By contrast, (20b) is acceptable (despite that the reported event is not
complete with regard to the predicate used), because the maximality requirement is
not overtly violated.
(20) MANDARIN
a. #Lùlu kāi-le nèi-shàn mén, érqiě hái zài kai.
Lulu open-PFV that-CL door and still PROG open
Intended: ‘Lulu opened that door, and she is still opening it.’
b. Lùlu kāi-le nèi-shàn mén, dànshì mén gēnběn méi kāi.
Lulu open-PFV that-CL door but door at all NEG.PFV open
Literally: ‘Lulu opened that door, but it didn’t open at all.’
c. ∃e(MAX(e, λe .∃s(agent(e , lulu) ∧ cause(e , s) ∧
open(s) ∧ theme(s, that-door))))
On the basis of data involving stative predicates, Martin & Gyarmathy (2019) show
that in Romance and Germanic languages, event completion does not replace event
maximality, but has to be combined with it. Similarly, Altshuler & Filip (2014)
claim that the completion requirement has to be combined with the maximality
requirement for the Russian perfective. Following Martin & Gyarmathy’s (2019),
I call ‘weak perfective’ the perfective which requires event maximality only (see
(21)), such as the Hindi perfective, and ‘strong perfective’ the perfective requiring
event completion and event maximality (see (22)).
6 Altshuler’s (2014) definition is more elaborate, and also uses Landman’s (1992) stages, not just
event parts.
266 F. Martin
(21) PFVM = λP ∃e[MAX(e, P )]] (weak perfective)
(22) PFVC+M = λP ∃e[MAX(e, P ) ∧ P (e)] (strong perfective)
Some authors have suggested that the Mandarin perfective -le is not completive
either (see in particular Smith 1997; Soh & Gao 2007; Koenig & Muansuwan
2000; and Altshuler 2017), and that the Mandarin perfective may be similar
to the Hindi perfective (see also Martin et al. 2018a; Martin 2019; Martin &
Gyarmathy 2019).7 I follow these authors in the assumption that the perfective -
le in Mandarin is the source of the zero-change use of monomorphemic verbs in this
language.
8.3.2 Via Sublexical Modality
Unlike Mandarin or Hindi, Germanic and Romance languages do not have partitive
perfective aspectual operators, which explains why the Germanic and Romance
literal perfective counterparts of Mandarin, Korean or Hindi examples in Sect. 8.1
are all contradictory. To explain why verbs such as offer or teach nevertheless do
not entail the occurrence of a P-state they encode lexically, Koenig & Davis (2001)
introduce a sublexical modal component, which evaluates the relations between
participants and eventualities at various world indices, see e.g. the paraphrase (23b)
of (23a).
(23) a. Ivan taught the basics of Russian to Mary, but she did not learn anything.
b. Ivan caused Mary to know the basics of Russian in all worlds where the
goal of his teaching is achieved.
On this view, such verbs involve a cause relation exactly as Mandarin extensional
causative (monomorphemic) verbs. But contrary to what happens with the latter, the
encoded result state is in the scope of a sublexical modal operator (a modal base).
Since the world of evaluation is not necessarily included in the modal base, the
result state does not have to take place in the actual world (hence why Martin &
Schäfer (2017) call these verbs defeasible causatives). In the spirit of Koenig &
Davis (2001), Martin & Schäfer (2017) propose lexical representations such as (24)
for enseigner ‘teach’, where the encoded modal base includes all causally successful
worlds.
7 Interestingly,languages with a non-completive (weak) perfective often have serial verbs or verbal
compounds that do convey completion when combined with a weak perfective (Singh 1994;
Altshuler 2014; see Martin & Gyarmathy (2019) for an account). So there seems to be a division
of labor at play here between outer and inner aspect: in languages such as Hindi or Mandarin, the
perfective does not need to require completion because serial verbs or verbal compounds can take
this job when combined with this form.
(24) enseigner y à z ‘teach y to z/ z y’

λyλzλe.teach(e) ∧ theme(e, y) ∧ recipient(e, z) ∧
ρ ∃s(cause(e, s) ∧ know(s) ∧ theme(s, y) ∧ holder(s, z))
Observe that according to (24), defeasible causatives encode a manner property

associated to the causing event (the teach property in (24)), beyond the result
property associated to the result state.8 Thus when the causing event e is bound by
an aspectual operator requiring event completion, such as the English simple past
or the French passé composé, the causing event e has to be complete with respect
to this manner predicate. This is a welcome prediction; for instance, (23a) is false if
Ivan didn’t perform a complete teach-the-basics-of-Russian event.9 But of course,
an event which is complete with respect to the VP teach the basics of Russian may
nevertheless be unsuccessful, precisely because the result state encoded by teach
is under the scope of a modal operator. That is, an event e may be complete with
respect to the (manner) teach-the-basics-of-Russian property in the actual world
w0 even in situations where e does not cause the targeted result state in w0 . This is
why the zero-change use of teach is still acceptable in presence of an in-adverbial,
which explicitly indicates that the predicate encodes a telic endpoint (namely the
endpoint of the teach-the-basics-of-Russian event); see (25).
(25) Ivan a enseigné le matériel de base aux étudiants en trois

Ivan teach.PFV.3SG the material of basis to-the students in three
semaines, mais ils n’ont encore rien appris.
weeks, but they NEG=has still nothing learned
‘Ivan taught the basic material to the students in three weeks, but they still
don’t know anything.’
Section 8.6.2 shows how predicates such as (24) compose with the Voice head
introducing causer subjects, and argues that if the resulting structure tends to entail a
result state despite the fact that it is in the scope of the modal operator, it is because
the teaching event type is by default understood as tokenized by a learning event
taking place in the theme’s referent.
8 Note that enseigner ‘teach’ does not specify how exactly the manner is instantiated, but as
Rappaport Hovav & Levin (2010) observe about the verb exercise, manner verbs may remain
rather unspecified about the specific instantiations of the manner encoded. Thus exercise requires
an unspecified set of movements, whose only defining property is that they involve some sort
of activity. There are however conventional ways of teaching with an agent, and teach on its
agentive uses seems to get associated with them. More generally, actions are more indicative of
their potential effects than non-agentive events, see Martin (2015).
9 I use “complete” as in Zucchi (1999) to express that the event falls under the respective predicate:
e is complete wrt P iff P (e).

268 F. Martin
8.4 Shortcomings of Martin (2015): The Problem of

Action-Denoting Causers
In Martin (2015), I offered an account of the link between agenthood and deniability
of causal efficacy mainly capitalizing on two distinctive and related properties of
agentive causation events, typically not exhibited by non-agentive causation events.
Whereas I still believe that this analysis is on the right track, I think it does not pay
enough attention to the way the verbal predicate is combined with the functional
head introducing the external argument. That the properties by which agentive
causation events differ from non-agentive causation events cannot be the single
factor at play is well illustrated by the contrast between (6) and (12) repeated below.
(6) John taught Mary how to iron sheets (although in spite of the fact that she saw
him do it, she still doesn’t know how it is done).
(12) John’s careful demonstrations taught Mary how to iron sheets (#although in
spite of the fact that she saw him do it, she still doesn’t know how it is done).
Both (6) and (12) indicate that an agent is involved in the causation event reported.
However, the action is nominalized in a nominal description in subject position
in (12), but not in (6). And as the continuation in parenthesis indicates, a change
developing towards a state of knowledge is much more strongly implicated by (12)
than by (6).
The same contrast can be replicated in Mandarin, albeit action-denoting nominal
expressions in subject position of causative monomorphemic verbs are quite marked
(Jinhong Liu, Hongyuan Sun, p.c.). But to the extent that these sentences are
acceptable, they do not licence the zero-change use with predicates which licence
this use with a subject denoting an agent; see (26) vs. (2).
(2) MANDARIN
Lùlu guān-le nèi-shàn mén, dàn gēnběn méi guān-shàng.
Lulu close-PFV that-CL door but at-all NEG.PFV close-up
Literally: ‘Lulu closed that door, but it didn’t get closed at all.’
(26) MANDARIN, Jinhong Liu & Hongyuan Sun, p.c.

? Yuehan de dongzuo guān-le nèi-shàn mén, #dàn gēnběn méi
Yuehan DE movement close-PFV that-CL door but at-all NEG.PFV
guān-shàng.
close-up
Intended:‘Yuehan’s movement closed that door, but it didn’t get closed at all.’
So in summary, subjects denoting an action have the same effect as causer subjects:
they tend to strengthen the inference that a change occurs. In the following, I treat
action-denoting subjects as causer subjects, i.e. as eventuality-denoting subjects.
I argue below that the way the subject is semantically combined with the VP is
crucial for the inference of causal efficacy triggered by the sentence. The output of
this compositional step very much depends on the semantics of the functional head
introducing the subject, which differs with the type of subjects introduced—agents
or causers. The following sections address this point in detail, starting with the case
of extensional causative verbs such as kill.
8.5 A Closer Look at the Semantic Flavours of Voice
8.5.1 Basic Assumptions
Let me first spell out some basic assumptions on the syntax and semantics of lexical
causative verbs forming the background of this study. In the spirit of Distributed
Morphology, I adopt the idea that a derivation starts with a non-decomposable
root, which combines with functional categories to build words (Marantz 1997;
Embick & Noyer 2006). Voice is the functional category introducing the external
argument of the predicate it combines with (Kratzer 1996). Voice is ambiguous, and
has a different denotation depending on whether it introduces a causer or an agent
external argument (Harley 2007; Schäfer 2008). One of the key properties of the
functional head introducing agent subjects—Voiceag —is that it does not introduce
any further event, but only relates the external argument x it introduces to the event
e introduced by the predicate it combines with, and specifies that x is the agent
of e (Kratzer 1996). By contrast, the head introducing causer subjects—Voicec —
introduces a further eventuality v argument (saturated by the event description in
the subject position), as well as a relation R between this eventuality v and the event
e introduced by the predicate Voicec attaches to (Pylkkänen 2008). A key question
that probably did not receive the attention it deserves concerns the nature of the
relation R, to which I come back in Sect. 8.5.3.2.
Another assumption I adopt here is Kratzer’s (2005) idea—further elaborated on
in Schäfer (2008) and Alexiadou et al. (2006, 2015) among others—that we can
dispense with the BECOME predicate in the representation of lexical causatives, and
simply be left with a causing event e and a result state s. Under this view, causative
and anticausative predicates have exactly the same event structure, and semantically
differ only by the presence vs. absence of Voice (Kratzer 2005; Schäfer 2008). Take
e.g. shā ‘kill’ in Mandarin, which can be also used as an anticausative for a subset
of Mandarin speakers (Martin et al. 2018a).
(27) shā Fido ‘kill Fido/Fido die’

λe.∃s(cause(e, s) ∧ dead(s) ∧ theme(s, fido))
On its anticausative use, shā Fido receives the meaning (27), while on the
agentive causative use, it receives the meaning in (28b). On this view, the causative
alternation is essentially a Voice alternation; predicates have the same event
structure in both anticausative and causative VPs.
270 F. Martin
(28) a. Voiceag λP λxλe.agent(e, x) ∧ P (e)

b. Voiceag [shā Fido]
[λP λxλe.agent(e, x) ∧ P (e)]
(λe.∃s(cause(e, s) ∧ dead(s) ∧ theme(s, fido)) =
λxλe.∃s(agent(e, x) ∧ cause(e, s) ∧ dead(s) ∧ theme(s, fido))
8.5.2 Tokenizing Causative Event Types
What has perhaps been overlooked in this line of research is that crucially, the
event type, although identical in both anticausative and (agentive) causative uses,
is tokenized in a different way (i.e., is mapped with different event tokens in the
model), because the number of participants involved in the causation events in the
denotation of the VP is different.10
In the intransitive use, only one participant is involved in the denoted causation
events, namely the theme’s referent. Therefore, the event type λe . . . P (e) . . . is
tokenized by changes-of-state of the participant—aka BECOME events. For instance,
in its anticausative use, the causing events denoted by shā ‘kill/die’ in (27) are
dying events. That is, when causative predicates are used anticausatively, the causing
events they denote are changes-of-state. So in a sense, BECOME is in this framework
redefined as a hyponym of CAUSE11 : changes-of-state are a subtype of causing
events, namely proximate causes involving the theme of the result state only. Let
me emphasize that it is not so unnatural, and in fact quite meaningful, to conceive
a change developing towards a P-state as a cause of this state. A dying event is in
a sense a (proximate) cause of the death it develops into; an event of becoming
sick can be conceived as a cause of the state of sickness it develops into, or a
getting-opened-event can be seen as a cause of the state of being open. And in fact,
causative analyses have been proposed for inchoative verbs; see Kratzer (2005) on
English, or Piñón (2011) on Hungarian. The reluctancy to conceive BECOME-events
as CAUSE-events is probably partly rooted in the assumption that an event has a
single cause (and we do not want to identify a change developing into a P-state
as the cause of this state). But this assumption is wrong; as Lewis (1973), Kvart
(2001) and many others (e.g. Dowty (1979) and Neeleman & van de Koot (2012) on
the linguistic side) emphasize, a given eventuality often has many, many causes. As
Kvart (2001) notes, the selection of one cause as the cause is context-dependent and
interest-relative (see also Moens & Steedman 1988). That the event type denoted
by anticausatives is tokenized by the most proximate causes, namely a change in
the theme’s referent, does not preclude the existence of other external causes of
10 Ido not make use of event kinds (see Gehrke 2019 for an overview). A token of the event type
denoted by P is here simply understood as an event in the extension of P.
11 Thanks to Bridget Copley for pushing me to make this assumption explicit.
the encoded result state. These external causes can be denoted by adjuncts such as
from-PPs, for example.
Obviously, the causing event type is tokenized very differently when the
causative verb is used transitively. In the agentive transitive use, two participants are
necessarily involved in the denoted causing events e, namely the subject’s referent—
the agent of e—and the theme’s referent. As a result, the event type λe . . . P (e) . . .
is naturally tokenized by more complex event tokens in the model. That is, the events
e instantiating this type are necessarily understood as made of two subparts, namely,
an action performed by the subject’s referent (for x cannot be the agent of e without
doing anything), and an ensuing CoS in the theme’s referent (for y cannot be in a
result state s that it was not in before without changing its state).
Crucially, however, the two subparts of the complex event tokens instantiating the
agentive event type λe . . . P (e) . . . do not correspond to two different sub-events in
the event structure projected by causatives used with an agent subject. Rather, the
sum of these two sub-event parts instantiates a single causing event in the denotation
of the predicate, not decomposable at the semantic level. In particular, these two
parts cannot be separately accessed by modifiers, as Fodor (1970) was one of the
first to observe.
The proposal developed in Sect. 8.5.4 is that the story is entirely different when
the transitive use is built with the non-agentive Voicec , because Voicec has a
different semantics than Voiceag . Before that, building on Martin (2018), I develop
in the next section the idea that in the default case, Voicec introduces an eventuality
v understood as causing a causing event e in the denotation of the VP (rather than
being identified with e, as proposed by Pylkkänen 2008 or Alexiadou et al. 2015
a.o.). To address the semantic difference between the two voice heads, I start with a
famous observation of Fodor (1970) on lexical causative predicates.
8.5.3 The Semantics of Voiceag vs. Voicec

8.5.3.1 Fodor Revisited
Fodor (1970) argued against a decomposition of lexical causative predicates in a

CAUSE and a BECOME components, e.g., against the decomposition of kill into
cause to die. One of his arguments is that the CAUSE and BECOME events are not
accessible for separate adverbial modification by temporal or manner adverbials,
while they are with periphrastic causatives, see (29).
(29) a. *Floyd melted the glass on Sunday by heating it on Saturday.

b. Floyd caused the glass to melt on Sunday by heating it on Saturday.
However, it was observed in Martin (2018) that while it is true that separate
modification never seems possible with entity-denoting subjects, with eventuality-
denoting subjects, it is possible to modify separately the eventuality denoted by
the subject, and the (causing) VP-ing event, see the contrast in (30). Note that in
272 F. Martin
(30a), the reading where the second clause reports an action by Fred taking place on
December 25—e.g., the accidental shooting on December 23 inspired Fred to put an
end to the dog by poisoning him on December 25—is irrelevant for the judgement;
are only considered situations where the shooting on December 23 is the single
killing event performed by Fred.
(30) a. Fredi accidentally shot his dog on December 23! #Hei eventually killed
it on Dec. 25.
b. Fred accidentally shot his dog on December 23! This gunshot eventually
killed it on December 25.
Simply assuming with Fodor that the contradiction in (30a) arises as a consequence
of the fact that cause must relate temporally adjacent eventualities does not work,
for many authors have observed that lexical causative verbs may express a causation
relating temporally distant eventualities; see Danlos (2001), Rappaport Hovav &
Levin (2001), Neeleman & van de Koot (2012), Beavers (2012), a.o.
In Martin (2018), I proposed that the problem of (30a) is rather a direct
consequence of the fact that the temporal adverbial must scope on the single event
in the denotation of the lexical causative verb (namely the causing event), see (31)
(‘τ (e)’ gives the temporal trace of an event e).
(31) on December 25[kill Fido] [λP λe.P (e) ∧ τ (e) ⊆ dec.25]

(λe.∃s(cause(e, s) ∧ dead(s) ∧ theme(s, fido)) =
λe.∃s(cause(e, s) ∧ dead(s) ∧ theme(s, fido) ∧ τ (e) ⊆ dec.25)
When the verbal predicate (31) is combined with Voiceag in (32a), we obtain the
relation in (32b).
(32) a. Voiceag λP λxλe.agent(e, x) ∧ P (e)

b. Voiceag [on December 25[kill Fido]]
[λP λxλe.agent(e, x) ∧ P (e)]
(λe.∃s(cause(e, s) ∧ dead(s) ∧ theme(s, fido) ∧ τ (e) ⊆ dec.25)) =
λxλe.∃s(agent(e, x) ∧ cause(e, s) ∧ dead(s) ∧ theme(s, fido) ∧ τ (e) ⊆
dec.25)
This obviously accounts for why sentence (30a) is contradictory: given that (32b)
requires x to perform on December 25 an event causing a state of being dead, there
is no room left to identify this causing event with a previous action of x taking place
on December 23.
8.5.3.2 A New Semantics for Voicec
But then, what happens in (30b)? Pylkkänen (2008) assumes that event-denoting
subjects are introduced by another Voice head, that identifies the event introduced
by the subject e (e.g., the gunshot in (30b)) and the causing event introduced by the
verb (e.g., the killing event in (30b)). Pylkkänen’s (2008) Voice, that I call VoiceP ,
may be attributed the semantics in (33), where e is the event introduced by the
(eventuality-denoting) subject, which is identified with the event e’ introduced by
the predicate.12
(33) VoiceP λP λeλe .P (e ) ∧ e = e
If such a head was involved in the semantic composition of (30b), this sentence
should be contradictory, given that the gunshot would have to take place both on
December 23 and December 25. We therefore need another functional element than
Pylkkänen’s (2008) Voice.
I modify the semantics of this head that I will call Voicec in two respects. Firstly,
the eventuality v argument it introduces can be either an event or a state argument.
If the subject is fact-denoting in its literal reading (The fact that he came so late
surprised me), I assume that the fact is the theme of a covert eventuality (e.g., the
event of thinking about this fact). More importantly, the nature of relation R between
v and the VP-event e is underspecified; see (34a).13 However, it strongly tends to be
understood as a causal relation, although in marked cases, R can also be interpreted
as a relation of (partial or total) overlap, see (34b). I will leave aside this dispreferred
interpretation of R as the overlap relation in (34b) until Sect. 8.8, where it is argued
that this marked interpretation of R is chosen in the marginal cases where speakers
accept the zero-change use of causative predicates with causer subjects.
(34) a. Voicec λP λvλe.R(v, e) ∧ P (e)

b. R = cause ∨ ◦
Applying (34a) to (31c), and assuming that R is understood as meaning cause, we
obtain the verbal predicate (35), involving three different eventualities (the dotted
......
components are not entailed by the structure; they result from the choice of one of
the possible meanings of R in (34)).
(35) Voicec [on December 25[kill Fido]]

[λP λvλe.cause(v, . . ∧ P (e)]
. . . . . . . . . e)
(λe.∃s(cause(e, s) ∧ dead(s) ∧ theme(s, fido) ∧ τ (e) ⊆ dec.25)) =
λvλe.∃s(cause(v, . . ∧ cause(e, s) ∧ dead(s) ∧ theme(s, fido) ∧ τ (e) ⊆
. . . . . . . . . e)
dec.25)
Let us now apply the relation in (35) to the definite event description the gunshot,
and derive the predicate in (36).
12 Pylkkänen’s (2008) semantics for Voice is λxλe.e = x. In her system, Voice combines with a
VP by a rule called Event Identification (Kratzer 1996). I do not make use of this rule, and define
Voice as applying to a predicate and adding an external argument to it (see Bruening 2013 for a
similar approach).
13 On this point, I depart from Martin (2018), where I proposed that the head responsible for the
introduction of causer subjects necessarily encodes a causal relation.

274 F. Martin
(36) The gunshot[Voicec [On December 25[kill Fido]]]

λe.∃s(cause(the-gunshot, . . ∧ cause(e, s) ∧ dead(s) ∧ theme(s, fido) ∧
. . . . . . . . . . . . . . . . . . . . . e)
τ (e) ⊆ dec.25)
We can now understand why sentence (30b) is acceptable. Given that the
eventuality v denoted by the subject causes the causing event e leading to death
denoted by the verb (rather than being identified with it), v may, of course, take
place before the event e that must take place on December 25, e.g., on December
23. And observe that it is possible to add a temporal modifier within the subject DP
that refers to a time different from the modifier applying to the VP, see (37) ((37b)
is due to M. Rappaport Hovav, p.c.).
(37) a. Yesterday’s stabbing eventually killed him this morning.

b. The snow melt on Sunday eventually flooded the valley on Thursday.
Another argument in favour of the view that Voicec is by default understood

as introducing an eventuality causing the event denoted by the VP is provided by
progressive lexical causative sentences. Take, e.g., sentence (38).
(38) Fukushima nuclear accident is still destroying our planet. (uttered in 2019)
Although Fukushima nuclear accident happened (and culminated) in 2011, it

may still be destroying the planet today, 8 years later. In (38), the (past) accident
e culminated with regard to the nuclear accident description in 2011, but causes a
destroying event which is still ongoing today.
8.5.4 How (Non-)agentivity Drives the Tokenization of

Causative Event Types
The previous section argues that Voicec introduces an argument slot for an eventual-
ity causing a VP-event. A question arises at this point. When does the causing event
e denoted by the VP start, if e is understood as caused by the eventuality v denoted
by the subject, rather than identified with it? And what is the causing event, if it is
not the eventuality v? More concretely, in (30a) repeated below in (39), what is the
killing event taking place on December 25, if not the gunshot taking place two days
before?
(39) Fred accidentally shot his dog on December 23! This gunshot eventually
killed it on December 25.
I would like to argue that when the external argument is introduced by Voicec (and
R interpreted as cause as I assume to be by default the case), the causative event
type denoted by the VP is tokenized by changes-of-state of the theme referent, and
caused by the eventuality v introduced by the subject. So for instance, the causing
event type denoted by the VP kill Fido in (39) is tokenized by a dying event endured
by Fido. In other words, if we abstract away from the external argument, a non-
agentive causative VP is interpreted the same way as its anticausative counterpart.
The main difference between non-agentive causative VPs and anticausative VPs is
that in the former case, there is an external argument which introduces a slot for an
eventuality v causing the event e denoted by the VP, similar to a from-PP adjoined
to an anticausative, see (40)14 :
(40) a. The wind opened the window.

b. ≈ The window opened from the wind.
On this view, the causing event type denoted by lexical causatives is therefore
tokenized quite differently depending on whether the subject is an agent or a causer.
For as we saw in Sect. 8.5.2, when the external argument x is introduced by Voiceag ,
the causative event type denoted by the VP is ‘fleshed out’ by complex events
composed of an action of x and a CoS of the theme’s referent y. Voiceag does not
add any further event to the causal chain encoded by the VP. This is schematically
illustrated in Fig. 8.1a, where the circle in the right panel symbolises the causing
event introduced by the VP. In contrast, in the non-agentive use, the causing event
type denoted by the VP is fleshed out by CoS, themselves caused by the eventuality
denoted by the subject; see Fig. 8.1b.
The proposal is summarized in (41).
(41) a. The event type denoted by extensional causative VPs used agentively
is tokenized by complex events composed of an action of the subject’s
referent and an ensuing CoS of the theme’s referent as their parts.
b. The event type denoted by extensional causative VPs used non-agentively
is tokenized by changes-of-state of the theme’s referent (when the
relation R encoded by Voicec =cause)
The next subsections present in turn three arguments in favour of the proposal
(41).
8.5.4.1 In-Adverbials
A first argument supporting (41) concerns the interpretation of in-adverbials. An in-

adverbial measures the time span between the onset and the telos of the (complete)
events denoted by the predicate, i.e. causing events in the case of a causative
predicate. The telos of these events typically coincides with the onset of the result
14 There is an interesting similarity with Alexiadou & Anagnostopoulou (Forthcoming)’s claim that
object experiencer causative psych-verbs used non-agentively lack Voice altogether and simply
contain a VP, just like anticausative verbs.
276 F. Martin
x’s action
⊕y’s CoS
no eventuality denoted causing event denoted

by the subject DP by the vp
(a)
R=cause
v y’s CoS
eventuality denoted by causing event denoted

the subject DP by the vp
(b)
Fig. 8.1 Causal chains denoted by lexical causative statements (x stands for the subject’s referent,
and y for the theme’s referent). (a) Causal chain denoted by an agentive lexical causative statement.
(b) Causal chain denoted by a non-agentive lexical causative statement
state (but see the discussion in Martin 2018, section 4.2). Let us now compare the
interpretation of such adverbials when modifying causatives used non-agentively, as
in (42a), and agentively, as in (42b).
(42) a. The poison he swallowed this morning killed him in ten minutes in the
evening (#this being said, he died in less than a minute).
b. Mary killed him in ten minutes (this being said, he died in less than a
minute).
What we observe is that in (42a), the in-adverbial measures the theme’s CoS—
the dying event, exactly as in the anticausative counterpart of this example in (43).
(43) He died in ten minutes this evening from the poison he swallowed this
morning.
This explains why the continuation in parenthesis in (42a) is infelicitous: it

indicates that the CoS e culminated in less than a minute, whereas the first
clause specifies that the causing event e, which is by assumption identified with
e , culminated in ten minutes.
By contrast, in (42b), the in-adverbial measures the time span of the causing
event e this time fleshed out by one of x’s action e and y’s CoS e . Therefore, the
continuation in parenthesis is not contradictory, because it might be that the time
span of the CoS e is much shorter than the time span of the causing event e (of
which e is only a proper part).
I conducted an on-line truth-value judgement survey to probe this difference
in the tokenization of the causing event type. I used a Google questionnaire; the
participants were 36 native speakers of French (of which 6 were linguists) living
in Belgium, France or Germany. The test sentences were statements containing
an extensional causative verb modified by an in-adverbial, see (44b/d)–(45b/d).
The task consisted in judging whether these test sentences were true or false
in the given context, described in (44a/c) and (45a/c). In the test sentences, the
in-adverbial indicates that the causing event starts significantly before the change-
of-state endured by the theme. The prediction is therefore that these sentences
should be judged false with a causer subject, since by assumption, the causing event
is understood as starting when the CoS starts with such subjects.
(44) a. Causer-context: The dishwasher started running at 10.00. At 10.15, Paul

was awake, and it was because of the dishwasher. Paul started waking up
at 10.13.
b. Le lave-vaisselle a réveillé Paul en 15 minutes.
The dishwasher wake-up-PFV.3SG Paul in 15 minutes.
‘The dishwasher woke Paul up in 15 minutes.’
c. Agent-context: Ana has to wake up Paul and puts her plan into action at
10.00. At 10.15, Paul is awake (and this was because of Ana). He started
waking up at 10.13.
d. Ana a réveillé Paul en 15 minutes.
Ana wake-up-PFV.3SG Paul in 15 minutes
‘Ana woke up Paul in 15 minutes.’
(45) a. Causer-context: At 10.00, the wind starts blowing in the direction of the
window. The window remains closed but after 10 minutes, at one point,
the door suddenly opened (what took less than a minute).
b. Le vent a ouvert la fenêtre en 10 minutes.
the wind open-PFV.3SG the window in 10 minutes
‘The wind opened the window in 10 minutes.’
c. Agent-context: At 10.00, Sascha (a 3 year old) decides to open the
window which is one meter higher than the living-room table, and
immediately starts elaborating a strategy to achieve his plan. At 10.10,
the window is opened (because of Sascha). He needed less than a minute
for the opening of the window stricto sensu.
278 F. Martin
Table 8.2 SURVEY 2: réveiller ‘wake up’ With agent (44d) With causer (44b)
True/False/Undecided
answers to the agentive vs. False 36%(13/36) 56%(20/36)
non-agentive sentences in the True 50%(18/36) 19%(7/36)
given context Undecided 14%(5/36) 25%(9/36)
ouvrir ‘open’ With agent (45d) With causer (45b)
False 36%(13/36) 67%(24/36)
True 56%(20/36) 22%(8/36)
Undecided 8%(3/36) 11%(4/36)
d. Sascha a ouvert la fenêtre en 10 minutes.

Sascha open-PFV.3SG the window in 10 minutes
‘Sascha opened the window in 10 minutes.’
Three options were provided to the participants (true, false, undecided). The
results are summarized in Table 8.2. As they show, the test sentences are more often
judged true with an agent subject (50/56% of ‘Yes’ answers) than with a causer
subject (19/22% of ‘Yes’ answers). An χ -square goodness of fit test confirmed that
agents and causers have a distinct influence on the truth value chosen (for réveiller
‘wake up’, χ 2 = 21.513, p = .00002; for ouvrir ‘open’, χ 2 = 23.292, p < .00001).15
I take the results to indicate that the causing event (i) can be interpreted as starting
before the CoS starts with an agent subject (although it is not the default answer, a
point to which I return in Sect. 8.7), while (ii) this interpretation is dispreferred with
causers, although still possible for some of speakers (roughly 20%). I propose that
the speakers arriving at this interpretation with causers do so for they understand
the relation R encoded by Voicec as the overlap relation. As I mentioned earlier,
this interpretation of R is possible, remember (34), although dispreferred. I come
back to this interpretation of R in Sect. 8.8.
8.5.4.2 Begin-Sentences
A second argument in favour of the proposal in (41) concerns the interpretation of

sentences where the causative predicate is embedded under the aspectual adverb
begin/start. When the causative predicate has a causer subject, the begin-statement
requires the CoS to start. For instance, (46a) entails that the theme’s referent starts
getting an idea, and (46b) entails that the stone starts breaking. This is expected if
the causative event type denoted by the VP is tokenized by changes-of-state of the
theme when the predicate is combined with Voicec .
(46) a. The conversation started to give her an idea.

b. The heat started to break the stone.
15 Thanks to Nicolas Dumay for his help with the statistical analysis.
By contrast, when the causative predicate is used agentively, the begin-statement

entails that an action performed by the subject’s referent has started, because the
onset of the action is also the onset of the causing event when the causative predicate
is combined with Voiceag . And in an appropriate context, an action performed with
the goal of triggering a P-state may start although no change developing towards a
P-state has been initiated in the theme’s referent yet. This is the case in (47); for
instance, (47a) may be true while the theme’s referent has not started yet getting
an idea, in striking contrast with (46a). Similarly, (47b) is true if the workers start
getting to work, and this can happen while they haven’t triggered any change in the
stone yet. And (47c) is felicitous even if the book hasn’t started burning yet as long
as Lulu’s started burning the book.
(47) a. Paul started to give her an idea (but she is even not listening to him. . . )
b. The workers started to break the stone (but it’s so hard, it will take some
time before it starts breaking).
c. Lulu started to burn the book (but it’s so humid, it may take a lot of time
before it starts burning).
8.5.4.3 Progressive Sentences
A third argument in favour of (41) concerns the interpretation of progressive lexical

causative sentences. Let us look at the actions depicted in the three frames in
Fig. 8.2. Clearly, they are all seen as proper parts of (possible) ‘open the door’-
events, and this while the door didn’t start opening yet. For instance, on the basis
of the first frame in Fig. 8.2, we are typically ready to endorse the truth of a
progressive statement such as Alice is opening the door, and this well before the
door eventually starts opening. This supports the view that event types denoted by
causative predicates used agentively are tokenized by complex events starting with
the action performed by the subject’s referent.
Fig. 8.2 Examples of agentive (still not efficacious yet) openings

280 F. Martin
The pattern is very different with nonagentive causers. For instance, we typically
hesitate to endorse the claim that the wind is opening the window while it has not
affected the door yet. This is a well-known observation about progressive causative
sentences with inanimate subjects, see, in particular, Bonomi (1997); cf. also Martin
(2015).16 Suppose, for instance, that the water of a brook which has just been
diverted is approaching a little meadow. Bonomi (1997) observes that in this context,
sentence (48a), which contains a lexical causative, is clearly false. Bonomi suggests
that this is due to the fact that the event in progress e is not seen as part of a wetting-
that-meadow-event. This supports the proposal (41b) according to which event types
denoted by causative predicates used non-agentively are tokenized by changes-of-
state of the theme’s referent.
(48) a. The water is wetting that meadow.

b. I am wetting that meadow.
Interestingly, however, if we take an agentive version of (48b), the intuition

dramatically changes: if I am diverting the brook in order to wet the meadow, my act
is seen as a part of a wetting that meadow, and (48b) is felt to be true. This supports,
again, the proposal (41a).
Relatedly, Truswell (2011, 101–103) observed that in a context where the sea is
approaching a sandcastle, (49a) is felt to be false—and this even if it is very certain
that the sea will destroy the sandcastle, while (49b) is true while I’m gathering the
instruments I’ll be using to destroy the castle, although I haven’t touched it yet.17
(49) a. The sea is destroying the sandcastle.

b. I’m destroying the sandcastle.
As Despina Oikonomou (p.c.) observes, the same contrast obtains in (50), where
the causer subject in (50b) denotes an action (remember from Sect. 8.4 that action
nominalizations are a subtype of causer subjects): while (50a) can be judged true
although Daenerys Targarien hasn’t read the messages I am in the middle of writing,
it is not the case of (50b), which requires her to have started being affected.
(50) a. I’m destroying Daenerys Targarien.

b. The writing of these messages is destroying Daenerys Targarien.
I tested Bonomi’s (1997) and Truswell’s (2011) claims through truth-value

judgement tasks on French progressive sentences through a survey. The participants
were the same as for the previous survey (N = 36), and again, three options were
16 I
thank Zs. Gyarmathy for drawing my attention to Bonomi (1997) on this issue.
17 Note that (48b) and (49b) are not instances of the futurate progressive (Copley 2008). The
progressive in Romance languages does not have a futurate reading (Bertinetto 2000; Copley
2008), while Romance progressive counterparts of (48b) and (49b) are also acceptable in the given
contexts.
Table 8.3 SURVEY 3: N = 36 Sentence (51a) Sentence (51b)

True/False/Undecided
answers to the progressive False 86%(31) 64%(23)
sentences (51a/b) in the given True 3%(1) 22%(8)
context Undecided 11%(4) 14%(5)
provided (true, false, undecided). In the first task, subjects were firstly shown the
picture of a sandcastle distant approximately three meters from the sea, and were
told that the tide was rising in the direction of the sandcastle. The test sentence
is given in (51a). In the second task, the same subjects were shown a 30 seconds
long video of a house within a tornado, which ends up by a complete destruction
of it from the 20th second on (before that, the house was not affected in a visible
way yet). They were asked to judge the truth of the test sentence (51b). The results
summarized in Table 8.2 show that the test sentences are overwhelmingly judged
false in the provided context, although the subjects know that the ongoing event
ultimately develops into a complete VP-event.18 Again, this supports the proposal
(41b) that the event type denoted by causative predicates used non-agentively is
tokenized by CoS of the theme’s referent: since no ‘getting destroyed’ CoS is
ongoing in the depicted situation yet, the event type is felt not to be instantiated
yet, and the sentence therefore judged false (Table 8.3).
(51) a. Sur cette image, la mer est en train de détruire le château de sable.
on this picture the sea is destroying the sandcastle
‘On this picture, the sea is destroying the sandcastle.’
b. Dans les premières secondes de la vidéo, la tornade est
in the first seconds of the video the tornado is
en train de détruire la maison.
destroying the house
‘In the first seconds of the video, the tornado is destroying the house.’
8.6 Accounting for the Zero-Change Use of Causative

Predicates
8.6.1 Mandarin Extensional Causative Simple Verbs
This section shows that the proposal in (41) accounts for why in languages with
a weak (partitive) perfective such as Mandarin, zero-change uses of extensional
18 The fact that speakers judge (51b) more often than (51a) in the given context may be due to
the fact that it is plausible that the house already endures a change before being visibly destroyed
(while the sandcastle can’t be affected when untouched by the sea). The video used in the survey
is available at youtu.be/M77jJh6B4ok.
282 F. Martin
(non defeasible) causatives are possible when the external argument is introduced
by Voiceag , but much less so when introduced by Voicec .
The Mandarin perfective is a partitive aspectual operator, recall Sect. 8.3.1. Such
weak perfectives only require that there be a proper part of a VP-event in the world
of evaluation, and do not specify how large this part should be. When the causative
predicate is combined with Voiceag , the (causative) event type is tokenized in the
model by event tokens composed of an action of the subject’s referent and a change-
of-state of the theme’s referent. An action may itself have a causally inert initial
part (a part that does not trigger a theme’s CoS yet). The partitive perfective may
therefore quantify over a still causally inert part of an action by the subject’s referent.
Hence why denying the occurrence of any part of the change does not generate a
contradiction.
As an illustration, let us take the first clause of example (52a) again. (52b) pro-
vides the meaning of the untensed clause. Applying a weak perfective (expressing
PFVM as defined in (21)) to the agentive causative predicate in (52b) repeats the
event description in (52c). In (52c), the event e quantified over is either a (complete)
lulu-close-the-door event e , or a proper (maximal) part of such a possible event e .
In the latter case, e may very well correspond to an act fragment which still has not
triggered any change in the theme’s referent.
(52) MANDARIN
a. Lùlu guān-le nèi-shàn mén (dàn gēnběn méi guān-shàng).
Literally: ‘Lulu closed that door (but it didn’t get closed at all).’
b. Lulu[Voiceag [close the door]] λe.∃s(agent(e, lulu) ∧
cause(e, s) ∧ close(s) ∧ theme(s, the-door))
c. PFVM [Lulu[Voiceag [close the door]]]
∃e(MAX(e, λe .∃s(agent(e , lulu) ∧ cause(e , s) ∧
close(s) ∧ theme(s, the-door))))
Let us now turn to the non-agentive use of causatives. When the causative
predicate is combined with Voicec and the relation R it encodes interpreted as
cause, the causative event type λe . . . P (e) . . . denoted by the VP is by assumption
tokenized by changes-of-state of the theme’s referent in the model. Therefore, the
partitive operator existentially quantifying over the event variable introduced by the
VP must return a part of such a change.19 Denying the occurrence of any part of
these changes in the subsequent discourse therefore generates a contradiction.
Let us for instance take example (53a) again, for which I provide a partial
derivation in (53b/c). The crucial point is that the event e existentially quantified
over is now understood as a maximal part of a possible CoS of the door.
19 As for the event variable introduced by Voicec , I assume that it is bound at the level of the NP in
the specifier checked by Voicec when the NP is a definite.
(53) MANDARIN
a. Nà-zhen feng guān-le nèi-shàn mén (#dàn méi guān-shàng)
that-CL wind close-PFV that-CL door but NEG close-up
Intended: ‘The gust of wind closed that door (but it didn’t get closed at
all).’
b. Voicec [close the door]
λvλe.∃s(cause(v, . . ∧ cause(e, s) ∧ close(s) ∧ theme(s, the-door))
. . . . . . . . . e)
c. That gust of wind[Voicec [close the door]]
λe.∃s(cause(that-gust-of-wind, . . ∧ cause(e, s) ∧ close(s) ∧
. . . . . . . . . . . . . . . . . . . . . . . . . . . e)
theme(s, the-door))
d. PFVM [That gust of wind[Voicec [close the door]]]
∃e(MAX(e, λe .∃s(cause(that-gust-of-wind, . . ∧ cause(e , s) ∧
. . . . . . . . . . . . . . . . . . . . . . . . . . . e)
close(s) ∧ theme(s, the-door)))
That zero-change construals are always infelicitous with the intransitive use of
causative verbs (recall (13)–(17)) is due to the same reason.
8.6.2 Defeasible Causative Verbs
In this section, I show how the proposal extends to defeasible causative verbs such
as enseigner ‘teach’ or soigner ‘treat/ cure’. Again, the difference in the strength of
the change inference triggered by the agentive vs. non-agentive uses of these verbs
reflects the specific way the event type is tokenized in each use. An advantage of
this account over the one proposed by Martin & Schäfer (2012) is that the meaning
of the VP itself remains constant with causer and agent subjects, as are extensional
causative VPs across agentive vs. non-agentive uses. Martin & Schäfer (2012) have
to assume that the VP encodes a different type of modal base in the agentive vs. non-
agentive use; in the current proposal, the modal base is the same across all uses: it
contains all ‘causally successful worlds’.
Let us start with the agentive use, see (54).
(54) a. Ivan a enseigné à Marie les rudiments de la médecine (mais

Ivan teach.PFV.3SG to Mary the basics of the medicine (but
elle n’a encore rien appris du tout)
she NEG.has still nothing learned at all)
‘Ivan taught the basics of medicine to Mary (but she still hasn’t learned
anything at all).’
b. Voiceag λP λxλe.agent(e, x) ∧ P (e)
284 F. Martin
c. Voiceag [enseigner à Marie les rudiments de la médecine]

λxλe.teach(e) ∧
agent(e, x) ∧ theme(e, the-basics-of-med.) ∧ recipient(e, mary) ∧
ρ ∃s(cause(e, s) ∧ know(s) ∧ theme(s, the-basics-of-med.) ∧
holder(s, mary))
d. Ivan[Voiceag [enseigner à Marie les rudiments de la médecine]]
λe.teach(e) ∧ agent(e, ivan) ∧ theme(e, the-basics-of-med.) ∧
recipient(e, mary) ∧ ρ ∃s(cause(e, s) ∧ know(s) ∧
theme(s, the-basics-of-med.) ∧ holder(s, mary))
The predicate in (54d) denotes a set of events e such that e is a teaching of the basics
of medicine to Mary performed by Ivan, and such that in all causally successful
worlds, e causes Mary to know the basics of medicine.
The remarkable thing here is that differently from what happens with extensional
causative verbs, the causing event type denoted by defeasible causatives is tokenized
by actions of the subject’s referent only. Causing events denoted by defeasible
causatives do not necessarily include a change in the theme’s referent, precisely
because the targeted result state is in the scope of the modal. Remember that in the
case of extensional causatives, the causing event e has to include a change, for e
cannot cause a result state s in y without y enduring a change. But with defeasible
causatives, the higher event e is causally efficient in the worlds contained in the
modal base only. Thus e is not necessarily composed of a change in the theme’s
referent y. So for instance in (54a), the event type λe . . . .P (e) . . . . is tokenized by
teaching actions of Ivan, which may unfortunately culminate although the teachee
hasn’t made any step forward yet. Thus if we only consider the causing event and
abstract away from the result state under the modal operator, defeasible causatives
used agentively very much resemble manner (non-core transitive) verbs. This is
probably why these verbs have been classified as activities (see, e.g., Piñón 2014)
or are felt to be ‘less prototypically causative’ than extensional causatives such as
open (as an anonymous reviewer suggests). The difference in the way the causative
event type is tokenized by defeasible vs. extensional causative verbs is illustrated in
Fig. 8.3. The proposal (41a) has therefore to be slightly modified for this subclass of
causatives, see (55a).
(55) a. The event type denoted by defeasible causative VPs used agentively is
tokenized by actions of the subject’s referent.
b. The event type denoted by (defeasible or extensional) causative VPs used
non-agentively is tokenized by changes-of-state of the theme’s referent
(when the relation R encoded by Voicec =cause).
For the non-agentive use, however, the hypothesis remains the same ((55b) is
identical with (41b), modulo the first parenthesis). That is, defeasible causatives
VP s tend to entail a change when combined with Voicec because the causing event
type is then tokenized by changes-of-state of the theme’s referent. I illustrate the
idea through (56a).
x’action y ’s CoS
⊕x’s action
causing event denoted by teach causing event denoted by open
Fig. 8.3 Causing events denoted by defeasible vs. extensional causative verbs used agentively (x
stands for the subject’s referent, and y for the theme’s referent)
(56) a. Cette expérience a enseigné à Marie les rudiments de la

this experience teach.PFV.3SG to Mary the basics of the
médecine (#mais elle n’a encore rien appris du tout).
medicine (but she NEG.has still nothing learned at all)
‘This experience taught Mary the basics of medicine (but she still hasn’t
learned anything at all).’
b. Voicec λP λeλv.cause(v, . . ∧ P (e)
. . . . . . . . . e)
c. Voicec [enseigner à Marie les rudiments de la médecine] λvλe.
cause(v, .. ∧
. . . . . . . . . e)
teach(e) ∧ theme(e, the-basics-of-med.) ∧ recipient(e, mary) ∧
ρ ∃s(cause(e, s) ∧ know(s) ∧ theme(s, the-basics-of-med.) ∧
holder(s, mary))
d. Cette expérience[Voicec [enseigner à Marie les rudiments de la méd.]]
. . . . . . . . . . . . . . . . . . . . . . . . .e). ∧teach(e)∧theme(e,the-basics-of-med.)∧
λe.cause(this-experience,
recipient(e, mary) ∧ ρ ∃s(cause(e, s) ∧ know(s) ∧
theme(s, the-basics-of-med.) ∧ holder(s, mary))
The predicate in (56d) denotes a set of events e such that e is a teaching of the
basics of the medicine to Mary, e is caused by this experience, and in all causally
successful worlds, e causes Mary to know the basics of medicine.
In the default interpretation of (56a) captured by (56d), the experience referred
to by the subject is thus not identified with a teaching event introduced by the VP;
rather, the experience causes the teaching event, which is fleshed out by a learning
event in the theme’s referent. The change cannot be denied in the subsequent clause,
precisely because the teaching event e quantified over by the tense/aspect marker is
the learning event.
The arguments used in Sect. 8.5 support the proposal (55), too. Let us first
compare the interpretation of in-adverbial in agentive vs. non-agentive defeasible
causative statements, see (57)–(58).
286 F. Martin
(57) FRENCH
a. Dr Li m’a soigné en dix minutes, mais j’ai eu besoin d’une
dr Li me=has treated in ten minutes but I=need.PFV.1SG of one
semaine pour guérir.
week to recover
‘Dr. Li treated me in ten minutes, but I needed a week to recover.’
b. La voir sourire m’a soigné en dix minutes, #mais
her see smile me=has treated in ten minutes but
j’ai eu besoin d’une semaine pour guérir.
I=need.PFV.1SG of one week to recover
Intended: ‘Seeing her smiling cured me in ten minutes, but I needed a
week to recover.’
(58) FRENCH
a. Pierre lui a expliqué le problème en deux minutes, mais
Pierre him explain.PFV.3SG the problem in two minutes but
elle a eu besoin d’une heure pour le comprendre.
she need.PFV.3SG of an hour to it understand
‘Peter explained the problem to her in two minutes, but she needed an
hour to understand it.’
b. Le comportement de sa mère lui a expliqué le problème
The behavior of her mother him explain.PFV.3SG the problem
en moins d’une minute, #mais elle a eu besoin d’une heure pour
in less of a minute but she need.PFV.3SG of an hour to
le comprendre.
it understand
Intended: ‘Her mother’s behavior explained the problem to her in less
than a minute, but she needed an hour to understand it.’
In the (a)-examples, what is measured by the in-adverbial is the time interval

between the onset and the telos of the action performed by the subject’s referent,
while in the (b)-examples, it measures the time interval between the onset of the
change in the theme’s referent and the onset of the ensuing result state. This explains
the difference in the felicity of the second clause. In the (b)-examples, the event v
denoted by the subject may very well have culminated while the causing event e
introduced by the VP starts. This is another indication that v is not identical with e;
see also (59):
(59) FRENCH
Le traitement d’hier l’a soigné aujourd’hui.
The treatment of yesterday him.treat.PFV.3SG today.
‘Yesterday’s treatment cured him today.’
Let us now look at the interpretation of begin-statements:
(60) FRENCH
a. Pierre a commencé à enseigner le russe à Marie.
Pierre start.PFV.3SG to teach the Russian to Marie
‘Pierre started to teach Russian to Mary.’
b. Ce séjour linguistique a commencé à enseigner le russe à
this stay linguistic start.PFV.3SG to teach the Russian to
Marie.
Marie
‘This linguistic stay started teaching Russian to Mary.’
(61) a. Peter started to show them the problem.

b. These results started to show them the problem.
While (60a)/(61a) require the subject’s referent started acting in such or such
way, (60b)/(61b) strongly suggest that the theme’s referent started changing in the
direction of a P-state (e.g., started learning Russian in the case of (60b)). And note
that the showing- or teaching-change may begin well after the eventuality reported
by the subject has started. For instance, perhaps Mary had to get used to her new
environment one week long before starting learning any Russian.
Comparing progressive sentences with defeasible causatives used agentively vs.
non-agentively point to the same conclusion:
(62) a. Jean est en train de lui expliquer toute la vérité.

Jean is PROG him explain whole the truth
‘John is explaining the whole truth to her.’
b. Ce qui s’est passé hier est en train de lui
what REFL =happen. PFV.3 SG yesterday is PROG him
expliquer toute la vérité.
explain whole the truth
‘What happened yesterday is explaining the whole truth to her.’
Sentence (62a) only requires an action to be ongoing (the explaining event type is
instantiated in the model by actions of the subject’s referent), while (62b) describes
an ongoing change taking place in the theme’s referent (the explaining event type is
fleshed out in the model by the theme’s changes-of-state). That the (b)-sentence is
not contradictory again indicates that the eventuality v denoted by the subject does
not have to be identical with the ongoing event e introduced by the VP, since it is
made explicit that v and e have different spatio-temporal boundaries.
In summary, defeasible causative VPs have exactly the same semantics both with
agents and causers, but the very same event type is tokenized differently when the VP
combines with Voicec vs. Voiceag : when used non-agentively, the event satisfying
the encoded manner predicate is ‘fleshed out’ by a CoS rather than an action. This is
288 F. Martin
perhaps at the source of the feeling that under this latter use, defeasible causatives
are result verbs in a manner disguise: when used non-agentively, they ascribe to
changes-of-state the very same property satisfied by actions on the agentive use.
8.7 Accounting for the Cross-Linguistic Difference
In Sect. 8.2, I reported that the change inference of causatives licensing the zero-
change reading when used agentively is stronger in the case of Mandarin extensional
causative verbs than in the case of French, German or English defeasible causatives.
This section aims to account for this difference, to my knowledge not observed
before.
The zero-change/failed-attempt use of extensional causative VPs obtains through
outer aspect in languages such as Mandarin. To obtain this reading, two pragmatic
steps must be fulfilled. Firstly, the culmination inference which is by default trig-
gered by the perfective must obviously be cancelled. For on its default interpretation,
the first clause of a sentence such as (2) repeated below implies that the door is
closed, differently from its imperfective (progressive) counterpart.
(2) Lùlu guān-le nèi-shàn mén (dàn gēnběn méi guān-shàng).

Literally: ‘Lulu closed that door (but it didn’t get closed at all).’
The second clause of (2) cancels this inference; see Gyarmathy & Altshuler
(Forthcoming) and references therein, and Martin et al. (2018b) for the culmination
inference of the Mandarin perfective in particular. Gyarmathy & Altshuler have
argued that the culmination inference of perfective accomplishment sentences in
languages such as Hindi amounts to an abductive inference. And interestingly, the
inhibition of abductive inferences has been shown to require extra-effort, while in
the case of other defeasible inferences such as scalar implicatures triggered by items
such as some, the calculation of the inference generates extra-cost, see Noveck
et al. (2011). Secondly, zero-change readings of extensional causatives require to
conceive the causing event as having as its initial part an act of the subject’s referent
which is still causally inert, while already being a part of a VP-event. But this is
not trivial; in fact, it is often out of the blue challenging to find a context where
an act fragment is already a part of a VP-event while still being causally inert. On
this respect, it is interesting to note that some native speakers of Mandarin seem
to find the zero-result reading of (2) at first sight very marked, but then accept it
in a second phase, imagining a scenario where an obstacle prevents the closing of
the door. I propose that these speakers struggle to retrieve a context such that an
agentive closing event has a causally inert initial part.
In fact, assuming that a causing event has started without an ensuing change to
have started too is challenging in a default context in languages such as English or
French as well (remember the results of the second survey reported in Sect. 8.5.4.1).
Compare for instance the examples below, which are perfective statements contain-
ing an inchoative aspectual verb embedding an extensional causative VP.
(63) a. John started to burn the book, #but it hasn’t started burning yet.
b. John started to burn the book, but it’s so humid, it may take a lot of time
before it really starts burning!
(64) a. John started to open the door, #but it hasn’t started to open yet.
b. John started to open the safe, but the code is so complicated, it might
really take long!
In conclusion, obtaining the zero-change reading of extensional causatives via a

weak perfective morphology generates some costs, namely (i) the cancellation of
culmination inference obtained via abductive reasoning and (ii) the identification of
a causally inert proper part of a VP-event.
In contrast, none of these two steps are required to obtain the zero-change use of
defeasible causatives. Firstly, there is no need to cancel the inference of culmination
triggered by perfective sentences, since even on this use, the event denoted by the
VP does culminate with regard to the predicate. Recall for instance from Sect. 8.3.2.
the observation that the sentence (25) repeated below is in fact false if the reported
event is not a complete teaching-the-basics-of-Russian event, and this even on a
zero-change construal. As a result, (65) is contradictory.
(25) Ivan taught the basic material to the students in three weeks, but they still
don’t know anything.
(65) Ivan taught the basic material to the students in three weeks, #but he didn’t
finish (teaching it).
Secondly, differently from what happens in the case of the zero-change construal of
the Mandarin counterpart of open the door, no special context is required to identify
a non-efficacious part of the causing event e. In fact, events denoted by defeasible
causatives are complete while causally inert, in striking contrast with events denoted
by extensional causatives.
8.8 Why the Change Inference Is Sometimes Defeasible Even

with Causer Subjects
The previous sections aimed to explain why, across languages, zero-change readings
of causative predicates tend to require the external argument to be associated with
the agent rather than the causer role. However, exceptions to this generalization
have been observed here and there, and this section aims to explain them. I first
summarize the relevant observations.
290 F. Martin
Firstly, for Mandarin, Liu (2018) reports that in a zero-change situation, 7 of

the 30 adult speakers she tested occasionally accepted as true perfective statements
with some of the extensional causative simple verbs she tested even when used
with a causer subject. Secondly, for French, the first survey reported in Sect. 8.2.2.
reveals that 6 out of the 19 subjects tested tend to accept sentences where the change
is explicitly denied in a second clause even with a causer subject (3 out of 19
rated a sentence such as (11) with 3 on a [0–5] scale, and 3 with 5 on the same
scale). Thirdly, for French and German, Martin & Schäfer (2012) have argued that
adverbials clearly and objectively help to enhance the zero-change interpretation
of defeasible causatives used non-agentively. For instance, (66) below sounds less
contradictory than its counterpart without the adverbials.
(66) FRENCH, (Martin & Schäfer 2012)

Ce voyage leur a clairement et objectivement enseigné un peu
this trip them has clearly and objectively taught a little
de russe, tout de même! Il faut vraiment qu’ils soient idiots
of russian, though! It must really that they beSUBJ.3PL stupid
pour qu’ils n’aient rien appris.
for that they NEG-have.SUBJ.3PL nothing learned
‘This trip clearly and objectively taught them a little bit of Russian though!
They really must be idiots for not having learned anything.’
I argue that what is happening in this case is that the relation R encoded by Voicec
as defined in (34) repeated below is interpreted as the overlap relation, rather than
the cause relation, as the case by default.
(34) a. Voicec λP λvλe.event(v) ∨ state(v) ∧ R(v, e) ∧ P (e)

b. R = cause ∨ ◦
Under this marked interpretation of R, the eventuality v denoted by the causer
subject can be identified with the causing event e denoted by the VP. I propose that
in sentences such as (66), the adverbials clearly and objectively invite to identify v
as a VP-event e, by specifying that v clearly/objectively fulfills the manner property
of a VP- (here teaching) event e. In that case, v is the event tokenizing the event
type λe . . . P (e) . . . denoted by the VP. But in this interpretation, the event type
denoted by the VP is crucially not understood as instantiated by a CoS of the theme’s
referent! As a consequence, a CoS can be denied in the subsequent cause. So for
instance, in (66), the demonstrations denoted by the subject are not interpreted as
causing the teaching event. Rather, objectively/clearly indicates to the interpreter
that the demonstrations are the described teaching-how-to-iron-sheet event.
8.9 Conclusion
This paper concentrates on Mandarin, French and English data, but also leads to
testable predictions for other languages. In particular, we expect the zero-change
reading to be less challenging with defeasible causatives than with extensional

causatives via weak perfectivity (see Sect. 8.7). So if it turns out that weak perfective
languages such as Mandarin have defeasible causatives, those verbs should license
the zero-change use much more easily than extensional causatives. Martin et al.
(2018a) argue that xiū ‘fix/repair’ might be such a verb, and interestingly, in a
perfective sentence, it does not entail that the targeted result state is obtained even
in presence of an in-adverbial, just like what is observed with defeasible causative
verbs (recall (25)).
Korean is an interesting language to test in this respect. If the Korean aspectual
marker -ess- is a weak perfective similar in meaning to the Mandarin perfective
le, and if Korean causative predicates licensing zero-change uses are extensional
causative predicates, we expect this reading to raise difficulties and to obtain more
easily in facilitating contexts, where, for instance, something blocked the door while
the agent had been trying to close it (as Liu (2018) and Martin et al. (2018a)
observe for Mandarin). If, on the contrary, the past morphology -ess- requires
event completion with eventive predicates like the French perfective does, and the
causative predicates with failed-attempt uses encode a modal operator responsible
for this reading (Beavers & Lee Forthcoming), we expect failed-attempt uses to be
much easier to accept, as observed for French or English.
It may also be, however, that within the same language, speakers vary in the
semantics they attribute to some predicates. Verbs such as teach and explain seem
rather uniformly interpreted as defeasible causatives in English or French, but other
verbs, such as French réparer ‘repair/fix’ or English mend/repair, seem to give
rise to less homogeneous judgements; some speakers (e.g., Ryle 1949) consider
that mend/repair can be used to describe unsuccessful attempts, while for others,
a repairing event is by definition successful. So if the Korean marker -ess- turns
out to be a completive marker with eventive predicates and if the failed-attempt
use of Korean causative predicates reveals to be challenging for a significant set
of speakers, we could deal with a variation in the semantics speakers attribute to
the predicates at hand (as extensional vs. defeasible causatives). Such inter-speaker
variation is to be more expected when the sublexical operator is not spelled-out by
an overt morpheme (like it is in Salish languages, for instance; Jacobs 2011 a.o).
The analysis developed in the paper leaves some questions unanswered. In
particular, I have argued that the relation R encoded by Voicec between the
eventuality v denoted by the subject and the VP-event is preferably interpreted as
the causal relation (rather the relation of overlap), and provided arguments in favour
of this claim. But why the relation R encoded by Voicec should be preferrably
interpreted as a causal relation is still to be accounted for.
Acknowledgements I am very grateful to my three anonymous reviewers as well as to Daniel

Altshuler, Jacqueline Guéron, Louise McNally, Malka Rappaport Hovav, Despina Oikonomou,
Florian Schäfer and Giorgos Spathas for their valuable comments on a previous version of this
paper, as well as Artemis Alexiadou, Christopher Piñón and the participants to the RUESHEL
group meeting, the Types, tokens, roots, and functional structure OASIS 1 satellite workshop
(Paris, November 2018) and the (Non-)Agentivity in Natural Language workshop (Singapore, May
2019) for their feedback and suggestions. This work also finds its inspiration in a long-lasting
292 F. Martin
collaboration on Mandarin causative predicates together with Hamida Demirdache, Jinhong Liu
and Hongyuan Sun. I also thank Jinhong Liu, Hongyuan Sun and Jiyoung Choi for generously
providing me with Mandarin and Korean data, as well as the participants of the surveys. I am fully
responsible for any mistake or misunderstanding. This work is financially supported by DFG award
AL 554/8-1 (Leibniz-Preis 2014) to Artemis Alexiadou.
References
Alexiadou, A., & Anagnostopoulou, E. (Forthcoming). Experiencers and causation. In E. Bar-

Asher Siegal & N. Boneh (Eds.), Perspectives on causation. Berlin: Springer.
crosslinguistically. In M. Frascarelli (Ed.), Phases of interpretation (pp. 187–211). Berlin:
Mouton de Gruyter.
alternations: A layering approach. Oxford: Oxford University Press.
Altshuler, D. (2014). A typology of partitive aspectual operators. Natural Language & Linguistic
Theory, 32(3), 735–775.
Altshuler, D. (2016). Events, states and times. An essay on narrative discourse in English. Berlin:
De Gruyter.
Altshuler, D. (2017). Does viewpoint aspect make reference to time? Talk presented at the Beyond
Time workshop, University of Colorado, April 2017.
Altshuler, D., & Filip, H. (2014). Perfectivity as maximization over event stages. Talk presented
at the 11th International Conference on Actionality, Tense, Aspect, Modality/Evidentiality
(Chronos), Pisa.
Arunachalam, S., & Kothari, A. (2010). Telicity and event culmination in Hindi perfectives. In
P. M. Bertinetto, A. Korhonen, A. Lenci, A. Melinger, S. Schulte im Walde, & A. Villavicencio
(Eds.), Proceedings of Verb 2010, Interdisciplinary Workshop on Verbs: The Identification and
Representation of Verb Features (pp. 16–19).
Beavers, J. (2012). Resultative constructions. In R. Binnick (Ed.), The Oxford handbook of tense
and aspect (pp. 908–937). Oxford: Oxford University Press.
Beavers, J., & Lee, J. (Forthcoming). Non-culmination in Korean caused change-of-state
predicates. Linguistics.
Bertinetto, P. M. (2000). The progressive in Romance, as compared with English. In Ö. Dahl (Ed.),
Tense and aspect in the languages of Europe (pp. 559–604). Berlin: Mouton – De Gruyter.
Bonomi, A. (1997). The progressive and the structure of events. Journal of Semantics, 14, 173–
205.
Bruening, B. (2013). By phrases in passives and nominals. Syntax, 16(1), 1–41.
Chen, S. (2017). Initial stages of events: The Atayal unmarked predicates. In 34th West Coast
Conference on Formal Linguistics (pp. 107–114). Cascadilla Proceedings Project.
Copley, B. (2008). The plan’s the thing: Deconstructing futurate meanings. Linguistic Inquiry,
39(2), 261–274.
Danlos, L. (2001). Event coreference in causal discourses. In P. Bouillon & F. Busa (Eds.), The
language of word meaning (pp. 216–242). Cambridge: Cambridge University Press.
Demirdache, H., & Martin, F. (2015). Agent control over non culminating events. In E. Barrajón
López, J. Luis Cifuentes Honrubia, & S. Rodríguez Rosique (Eds.), Verb classes and aspect
Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht: D. Reidel.
Embick, D., & Noyer, R. (2006). Distributed morphology and the syntax/morphology interface.
In G. Ramchand & C. Reiss (Eds.), Oxford handbook of linguistic interfaces (pp. 289–324).
Fodor, J. (1970). Three reasons for not deriving kill from cause to die. Linguistic Inquiry, 1(4),
429–438.
Gehrke, B. (2019). Event kinds. In Oxford handbook of event structure (pp. 205–233). Oxford:
Gyarmathy, Z., & Altshuler, D. (Forthcoming). (Non-)culmination by abduction. Linguistics.
Harley, H. (2007). External arguments: On the independence of Voice◦ and v. Talk presented at
the 30th GLOW colloquium, University of Tromsoe.
Jacobs, P. (2011). Control in skwxwú7mesh. University of British Columbia dissertation.
Kazanina, N., Baker, S., & Seddon, H. (Forthcoming). Actuality bias in verb learning: The case of
sublexically modal transfer verbs. Linguistics.
Koenig, J.-P., & Davis, A. R. (2001). Sublexical modality and the structure of lexical semantic
representations. Linguistics & Philosophy, 24(1), 71–124.
Koenig, J.-P., & Muansuwan, N. (2000). How to end without ever finishing: Thai semi-perfectivity.
Journal of Semantics, 17(2), 147–182.
Kratochvíl, F., & Delpada, B. (2015). Degrees of affectedness and verbal prefixation in Abui
(Papuan). In S. Müller (Ed.), Singapore Proceedings of the 22nd International Conference on
Head-Driven Phrase Structure Grammar (pp. 216–233). Stanford: CSLI Publications.
Phrase structure and the lexicon. Dordrecht: Kluwer.
Kratzer, A. (2005). Building resultatives. In C. Maienborn & A. Wöllstein-Leisten (Eds.), Event
arguments in syntax, semantics and discourse (pp. 177–212). Berlin: De Gruyter.
Kvart, I. (2001). The counterfactual analysis of cause. Synthese, 127(3), 389–427.
Landman, F. (1992). The progressive. Natural Language Semantics, 1, 1–32.
Lee, J. (2015). An intention-based account of accomplishments in Korean. University of Texas at
Austin dissertation.
Lewis, D. (1973). Counterfactuals. Oxford: Blackwell.
Liu, J. (2018). Non-culminating accomplishments in child and adult Chinese. Université de Nantes
dissertation.
Lyutikova, E., & Tatevosov, S. (2010). Atelicity and anticausativization. In M. Duguine,
S. Huidobro, & N. Madariaga (Eds.), Argument structure and syntactic relations. A cross-
linguistic perspective (pp. 35–68). Amsterdam/Philadelphia: John Benjamins.
Marantz, A. (1997). No escape from syntax: Don’t try morphological analysis in the privacy of
your own lexicon. Penn Working Papers in Linguistics, 4(2), 201–225.
Martin, F. (2015). Explaining the link between agentivity and non-culminating causation. In
Proceedings of Semantics and Linguistics Theory (SALT) (Vol. 25, pp. 246–266).
Martin, F. (2016). Atypical agents and non-culminating events. Talk given to the workshop
Agentivity and Event Structure, DGFS 2016, Universität Konstanz, February 24–26.
verbs. In U. Sauerland & S. Solt (Eds.), Proceedings of Sinn und Bedeutung 22. Berlin:
ZASPIL. Vol. 2, ZASPiL 61.
Martin, F. (2019). Non-culminating accomplishments. Language and Linguistics Compass, 13(8).
Martin, F., & Gyarmathy, Z. (2019). A finer-grained typology of perfective operators. In C. Piñón
(Ed.), Empirical issues in syntax and semantics, vol. 12, pp. 187–216, www.cssp.cnrs.fr/eiss12/
index_en.html.
Martin, F., & Schäfer, F. (2012). The modality of ‘offer’ and other defeasible causative verbs.
In N. Arnett & R. Bennett (Eds.), West Coast Conference on Formal Linguistics (WCCFL)
(Vol. 30, pp. 248–258). Somerville: Cascadilla Press.
Martin, F., & Schäfer, F. (2013). On the argument structure of verbs with bi- and mono-eventive
uses. In S. Keine & S. Sloggett (Eds.), Proceedings of the North East Linguistic Conference
(NELS) (Vol. 42, pp. 297–308). Amherst.
Martin, F., & Schäfer, F. (2017). Sublexical modality in defeasible causative verbs. In A. Arregui,
M. L. Rivero, & A. Salanova (Eds.), Modality across syntactic categories (pp. 87–108). Oxford:
294 F. Martin
Martin, F., Sun, H., Demirdache, H., & Liu, J. (2018a). Monomorphemic verbs in Mandarin Chi-
nese: Lexical aspect, event structure and non-culminating construals. Manuscript, Université
de Nantes and Humboldt-Universität zu Berlin.
Martin, F., Sun, H., Demirdache, H., & Liu, J. (2018b). Why can one kill the cat twice in Mandarin
Chinese? Manuscript, Université de Nantes and Humboldt-Universität zu Berlin.
Moens, M., & Steedman, M. (1988). Temporal ontology in natural language. Computational
Linguistics, 14(2), 15–28.
Neeleman, A., & van de Koot, H. (2012). The linguistic expression of causation. In M. Everaert,
M. Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the interface (pp. 20–
Noveck, I., Bonnefond, M., & van der Henst, J.-B. (2011). Squib: A deflationary account of invited
inferences. Belgian Journal of Linguistics, 25, 195–208.
Oehrle, R. (1976). The grammatical status of the English dative alternation. MIT dissertation.
Park, K.-S. (1993). Korean causatives in role and reference grammar. University at Buffalo MA
thesis, Buffalo.
Piñón, C. (2011). Result states in Hungarian. In T. Laczkó & C. O. Ringen (Eds.), Approaches
to Hungarian: Volume 12. papers from the 2009 debrecen conference (pp. 109–134). John
Benjamins.
Piñón, C. (2014). Reconsidering defeasible causative verbs, Talk presented at the 11th Interna-
tional Conference on Actionality, Tense, Aspect, Modality/Evidentiality (Chronos 11), Pisa.
Rappaport Hovav, M., & Levin, B. (2001). An event structure account of English resultatives.
Language, 77(4), 766–797.
Rappaport Hovav, M., & Levin, B. (2008). The English dative alternation: The case for verb
sensitivity. Journal of Linguistics, 44, 129–167.
Rappaport Hovav, M., & Levin, B. (2010). Reflections on manner/result complementarity. In
E. Doron, M. Rappaport Hovav, & I. Sichel (Eds.), Syntax, lexical semantics, and event
structure (pp. 21–38). Oxford University Press: Oxford.
Ryle, G. (1949). The concept of mind. Harmondsworth: Penguin/Peregrine Books.
Sato, Y. (2019). How can one kill someone twice in Indonesian? Manuscript, Seisen University.
ling.auf.net/lingbuzz/004693.
Schäfer, F. (2008). The syntax of (anti-)causatives. External arguments in change-of-state contexts.
Singh, M. (1994). Perfectivity, definiteness, and specificity: A classification of verbal predicates in
Hindi. University of Texas at Austin dissertation.
Smith, C. (1997). The parameter of aspect. Dordrecht: Kluwer Academic Publishers.
Soh, H. L., & Gao, M. J. (2007). Verbal -le in Mandarin Chinese. In N. Hedberg & R. Zacharski
(Eds.), The grammar-pragmatics interface: Essays in honor of Jeanette K. Gundel (pp. 91–
109). Amsterdam: John Benjamins.
Tatevosov, S. (2008). Subevental structure and non-culmination. In O. Bonami & P. C. Hofherr
(Eds.), Empirical issues in syntax and semantics, vol. 7, pp. 393–422, http://www.cssp.cnrs.fr/
eiss7/.
Tham, S. W. (2019). Agents, causers, results, and contentfulness in mandarin expressions of
caused change. Talk given to the workshop Recent Approaches to (Non-)Agentivity in Natural
Language, National University of Singapore, May 3–4 2019.
Truswell, R. (2011). Events, phrases and questions. Oxford: Oxford University Press.
van Hout, A., Arche, M., Demirdache, H., García del Real, I., García Sanz, A., Gavarró, A.,
Gómez Marzo, L., Hommes, S., Kazanina, N., Liu, J., Lungu, O., Martin, F., & Strangmann, I.
(2017). Agent control and the acquisition of event culmination in Basque, Dutch, English,
Spanish and Mandarin. In M. LaMendola & J. Scott (Eds.), Proceedings of BUCLD 41.
Somerville: Cascadilla Press.
Zucchi, S. (1999). Incomplete events, intensionality and imperfective aspect. Natural Language
Part IV
Syntactic and Semantic Aspects of
Causation
Chapter 9
Experiencers and Causation
Artemis Alexiadou and Elena Anagnostopoulou
Abstract In this paper, we use the domain of object experiencer verbs in Greek
to discuss the behavior of non-agentive causative construals of this verb class
with clear implications for the syntax of causative predicates in general. We argue
that eventive causative object experiencer verbs are best analyzed as instances of
transitive internally caused change of state verbs. We then explore the consequences
of this analysis for a group of verbs that have been labeled in the literature defeasible
causative verbs. We substantiate the proposal that the layer introducing agents
as external arguments is distinct from the layer introducing causers as external
arguments. As a result, causers are conceived of as being part of the same event
structural component that contains the resultant state, while agents are separated
from it, being introduced in VoiceP.
Keywords Object experiencer · Causer · Agent · Internally caused change of

state verbs · Defeasible causatives · Dependent case
9.1 Introduction
In this paper, we investigate the properties of object experiencer verbs in Greek in

non-agentive causative construals. As is well known, the appropriate characteriza-
tion of object experiencer psychological verbs (henceforth EOs) such as fighten has
been in the center of discussion at least since Belletti & Rizzi’s (1988) seminal work.
A. Alexiadou ()
Humboldt Universität zu Berlin, Berlin, Germany
Leibniz-Zentrum Allgemeine Sprachwissenschaft, Berlin, Germany
e-mail: artemis.alexiadou@hu-berlin.de
E. Anagnostopoulou
University of Crete, Rethymno, Greece
e-mail: anagnostopoulou@uoc.gr

https://doi.org/10.1007/978-3-030-34308-8_9
298 A. Alexiadou and E. Anagnostopoulou
These verbs have an experiencer argument which appears in object position and a
stimulus argument that appears in subject position. The stimulus argument can be
animate or inanimate, as shown in (1), and can be interpreted agentively, as in (1a)
or non-agentively, as in (1b).
(1) a. Mary frightened. John
b. The earthquake frightened John.
Our main concern here is the question of whether the distinct semantic roles
associated with the stimulus argument signal a distinct syntax for the interpretations
associated with object experiencer verbs. Specifically, the issue here is whether (1b)
involves a change of state. (1a) illustrates an agentive causative use of the psych
verb, where the stimulus argument is undoubtedly causal, causing of a change of
mental state. Landau (2010) argues in detail that agentive variants of EO verbs are
accomplishments and should not be treated as instances of psych verbs. By contrast,
for Landau (1b) is stative and no change of state is involved. In this, EO verbs differ
from verbs that are labelled externally caused change of state in Levin & Rappaport
Hovav (1995) such as destroy or open. These verbs also allow external arguments
with different semantic roles, i.e. agents and causers, as we will see in Sects. 9.2 and
9.5, which are responsible for causing a change of state; the crucial difference is that
the change of state is entailed with a causer subject but not with an agentive one, as
argued in Martin (this volume). However, Arad (1998) and more recently Alexiadou
& Iordăchioaia (2014) argued that (1b) is eventive and causative, suggesting that a
change of mental state in the experiencer is also involved. From this perspective,
(1a) can be ambiguous between an agentive and a non-agentive causative reading.
This means that both subjects in (1) are causal and the difference between (1a)
and (1b) is the potential intentional involvement of the animate subject, leading to
a behavior quite similar to that of agentive vs. non-agentive causative non-psych
predicates.
In view of the fact that both agents and causers are involved in change of state
events, the question arises whether the grammar treats these two roles uniformly.
Authors such as Ramchand (2008) give a negative response and subsume both roles
under the general term initiator. Others, e.g. Schäfer (2012), argue that the two
have distinct thematic licensers, while they have an identical syntax. Martin (this
volume) argues that two distinct voice heads are responsible for the introduction
of causers as opposed to agents, while Rappaport Hovav (this volume) offers a
semantic/pragmatic explanation for the restrictions on external arguments. In this
paper, we argue that actually causers and agents have a distinct syntax. While we
offer a syntactic analysis for this distribution, as we will discuss in Sect. 9.5, the
syntax we will propose is compatible with Martin’s (this volume) proposal that the
basic difference between agentive and non-agentive causatives is that the former
involve an action of the subject’s referent which triggers a change-of-state of the
theme argument, while non-agentive causatives are actually VPs that signal the
change of state of the them. We will reach this conclusion by focusing on EO
verbs in Greek, which have causer subjects and whose experiencer arguments bear
9 Experiencers and Causation 299
accusative case morphology. We provide arguments that they pattern like transitive
internally caused verbs, e.g. wilt. Our analysis treats EO verbs with causer subjects
as having an anticausative structure, cf. Belletti & Rizzi (1988), Pesetsky (1995),
which in turn explains the fact that in Greek this verb class does not passivize,
Alexiadou (2018).
Our work is in line with other work on the causative alternation that assumes that
causative relations emerge as part of syntactic configurations. In particular, verbs
that encode an event which leads to a change of state are causative, irrespectively of
the type of argument or transitivity. We will substantiate this in Sect. 9.2. Causer
subjects are introduced in a different functional layer in the syntactic structure
from that of agents. While experiencers show subject properties in the case of
EO verbs, objects of internally caused verbs certainly lack such properties, unless
they are coerced into a psych interpretation. We account for this difference by
appealing to the association with TP of experiencer arguments, building on Landau’s
(2010) analysis of experiencers. We further propose that accusative experiencer
objects receive accusative dependent case in opposition to a higher argument, like
all accusative objects in Greek. As just outlined, our analysis of the exceptional
properties of EO predicates with causer subjects is based on the idea that causer
arguments are introduced in a layer distinct from Voice, which we take to be the
layer introducing agents. This particular analysis has consequences for the analysis
of so-called defeasible causative predicates discussed extensively by Martin &
Schäfer (2017) and Martin (this volume) among others.
The paper is structured as follows. In Sect. 9.2, we revisit the distinction between
externally caused and internally caused change of state predicates, and briefly
illustrate the analysis of transitive internally caused change of state verbs put forth
in Alexiadou (2014). In Sect. 9.3, we focus on causative EO verbs and show how
their properties can be explained by assuming that these are similar to internally
caused verbs. In Sect. 9.4, we present our analysis which combines the analysis of
internally caused verbs with Landau’s movement analysis of experiencer objects. In
Sect. 9.5, we explore how this extends to account for defeasible causative verbs. In
Sect. 9.6, we conclude.
9.2 The Syntax of Internally Caused Change of State Verbs
Levin & Rappaport Hovav (1995) and cf. Reinhart (2002) proposed that verbs of
change-of-state split into two groups: externally caused change-of-state verbs such
as open, and internally caused change- of-state verbs, such e.g. bloom, decay. Some
examples for each class are given in (2), from Wright (2002), based on Levin (1993):
(2) a. externally caused change-of-state verbs:

bake, boil, break, cool, crack, dry, freeze, lengthen, melt, open,
shatter, straighten, widen
.b internally caused change-of-state verbs:
bloom, blossom, corrode, decay, erode, ferment, germinate,
molt, rot, rust, sprout, stagnate, wilt, wither
The two classes have been argued to show important semantic and syntactic
differences. From the interpretational point of view, according to Levin & Rappaport
Hovav (1995), verbs of type (2a) necessarily imply an external argument that brings
about the event. By contrast, internally caused change of state verbs typically
involve properties that can be seen as inherent to the entity that undergoes the change
of state. Because of this semantic distinction, it follows that the former class will be
found in transitive construals, while the latter will lack such construals. In other
words, we expect that internally caused change of state of verbs will not figure in
transitive sentences, as the change of state does not require an external argument to
bring it about.
This semantic difference is reflected in the behavior of these two verbs in
the causative alternation. Externally caused change-of-state verbs participate in
the alternation,1 while internally caused change-of-state verbs fail to do so. An
important property characterizing externally caused change of state verbs is that
their external arguments can be agents, instruments as well as causers:
(3) a. The door opened.
b. Bill/the hammer/the lightning opened the door.
c. The roses bloomed.
d. *John bloomed the roses.
The group of internally caused change of state verbs and their proper classification
was revisited and refined in work by McKoon & Macfarland (2000), and Wright
(2001, 2002). On the basis of corpora as well as psycholinguistic experiments these
authors showed that internally caused change-of-state (henceforth ICCOS) verbs do
have transitive counterparts, see (4), taken from Wright (2002).
(4) a. Salt air rusted the metal pipes.
b. Early summer heat wilted the petunias.
Wright (2001, 2002) pointed out that in English ICCOS verbs differ from externally
cause alternating verbs in three main respects: type of subject they allow, gradience,
and subject modification. Our focus will here be on the first feature.
As we mentioned above, alternating verbs in English can take a variety of
arguments as subjects in their transitive uses. They allow agents (5a) as well as
1 Wenote here that Harley & Noyer (2000) and Alexiadou et al. (2006, 2015) use the term cause
unspecified to best characterize the class of alternating verbs across languages.
causers (this category subsumes natural forces (5b) and causing events (5c)), see
Levin & Rappaport Hovav (1995), Reinhart (2002), and Pylkkänen (2002):
(5) a. The earthquake broke the vase.

b. Will’s banging shattered the window.
c. A stone broke the window.
Wright and McKoon & Macfarland observe that the subjects of transitive ICCOS
verbs tend to be causers and are very rarely animate. McKoon and Macfarland in
fact clearly state that there is really no other type of subjects other than causers.
These considerations led Rappaport Hovav & Levin (2012) and Rappaport
Hovav (this volume) to revisit the class of ICCOS verbs; Rappaport Hovav (this
volume) actually deconstructs this classification and proposes (6), which explains
the restrictions on the ICCOS subject:
(6) For a given state and a given entity there is a default expectation of whether the
state (or the degree to which the state holds) will or will not change in the natural
course of events, i.e., whether the entity has the disposition to undergo a change
in state. The cause of a change of state is relevant only if for the given state
and the given entity, there is no default expectation of change.
According to Rappaport Hovav (this volume), what is special about ICCOS verbs is
that they express changes of state that are expected to happen in the natural course
of events. As a result, such verbs are best associated with intransitive construals.
Nevertheless, they might occur in transitive construals, but only if their subjects
are causers that facilitate or trigger the change of state. This is very different from
externally caused change of state verbs, which do not describe changes that arise
from properties of their theme argument, and generally allow a variety of causes as
subjects, as we saw above.
To this end, consider the examples in (7):
(7) a. *The farmer blossomed the fruit trees.
b. *The careless gardener decayed the fence with a misplaced sprinkler.
The examples in (7) are not acceptable, as they violate the condition in (6).
According to Rappaport Hovav & Levin (2012), agents normally precede causers in
the chain of causation, see e.g. Croft (1998). Typically, causers, e.g. natural forces
or ambient conditions cannot be under the control of an agent. For this reason,
causative uses with agent subjects are ruled out. As the authors point out, certain
kinds of event subjects, particularly those expressing the intentional action of an
agent, are excluded for the same reason: *The gardener’s careful digging blossomed
the trees early.
Alexiadou (2014) discusses a similar state of affairs for Greek, consider (8) taken
from Levin (2009: 22). Lavidas (2007) as well as Roussou & Tsimpli (2007) also
report several similar examples:
(8) a. O thalasinos areas skuriase to frahti.

the sea air rusted-3SG the fence
‘The sea air rusted the fence.’
b. I poli zahari sapizi ta dondia.
the much sugar rots-3SG the teeth
‘A lot of sugar rots the teeth.’
A comparison of ICCOS verbs in the two languages shows that verbs that
predominantly take causer subjects (blossom) never undergo passivization and thus
have to be set apart.2 The comparison further shows that verbs that in earlier work
have been classified as ICCOS are actually externally caused change of state verbs,
e.g. ferment and erode, see also Rappaport Hovav (this volume). This is shown in
(9) and (10) below, whether both in English and Greek these verbs can take agentive
subjects, suggesting that they are actually externally caused change of state verbs
and have been mis-classified as ICCOS verbs:
(9) I don't think I would ferment tomatoes nearly as long as I fermented cabbage
for sour kraut.
(10) a. I thalassa diavroni tis aktes tis Kritis.
the sea-NOM erodes the coast the Creta-GEN
‘The sea erodes the coast of Creta.’
b. diavronis to sistima ek ton eso.
erode-2SG the system-ACC from within
‘You erode the system from within.’
In order to capture this distinction, Alexiadou (2014) argued that transitive ICCOS
verbs that do not allow passivization involve causer subjects that are introduced at
the layer of vP and not in VoiceP, as in (11). By contrast, agents are introduced in
Voice, as in (12):
(11) vP
Causer v'
ResultP
theme
2 Rappaport Hovav (this volume) argues that the blossom class actually involves verbs of emission,
i.e. there is no change of state involved. Alexiadou (2014) offers an analysis of these verbs as verbs
of appearance, i.e. change of location verbs.
(12) VoiceP
agent Voice'
Voice vP
v ResultP
theme
Importantly, the structure in (11), as we will see in Sects. 9.4 and 9.5, is the structure
to assume also for transitive non-agentive EO verbs, which also resist passivization.
As we will discuss in Sect. 9.4, the structure in (11) suggests that the theme is co-
temporal with the causer argument, which is excluded from the structure in (12).
This representation builds on the view of the verbal decomposition of change of
state verbs put forth in Alexiadou et al. (2006, 2015) (AAS), according to which
these verbs are decomposed into a VoiceP layer introducing the external argument,
but no further event variable, crucially following Kratzer (1996), and a v layer
introducing event implications. Little v combines with a result phrase or a root,
and this particular combination yields causative semantics, involving a process that
leads to a result, building on Schäfer (2008), see also Ramchand (2008). Moreover,
AAS adopted the view in Solstad (2009) that causers are modifiers of events, and
thus best associated with the vP layer, while agentive external arguments are situated
in VoiceP.
An important point made in AAS (2015) was that ICCOS verbs contain the same
causative component as externally caused verbs, the difference being the presence of
an agentive external argument, i.e. vP-ResultP layers, as externally caused or cause
unspecified verbs of change of state. Evidence for this was provided by the fact
that ICCOS verbs across languages can be modified by the same causer PPs as the
other classes of change of state verbs. This is particularly clear in Greek, where PPs
headed by me ‘with’ introduce causers only in the context of causative predicates.
This is illustrated in (13):
(13) To fito anthise me ti zesti.
The plant-NOM blossomed with the heat-ACC
‘The plant blossomed with the heat.’
This particular structural analysis suggests that intransitive and transitive ICCOS
have exactly the same structure and more importantly that the causer is part of
the same vP as the undergoer/theme. Structurally this means that the causer can
be construed as directly being involved in the change of state, see also Rappaport
Hovav (this volume). We believe that this representation of causers accounts for the
fact that in the presence of a causer argument the change of state is always entailed,
as has been extensively discussed by Martin & Schäfer (2017), and we will turn to
this in Sect. 9.5.
While transitive ICCOS verbs which only admit causer subjects do not passivize,
the two DPs, the causer and the undergoer argument, surface with nominative and
accusative case respectively, a property that follows from Marantz’s (1991) and
Baker’s (2015) extension of dependent case, see also Baker & Vinokurova (2010):
the lower argument is assigned dependent accusative as it is c-commanded by a
higher DP, which is assigned unmarked nominative. Thus, the dissociation we find
between the absence of Voice (as reflected in lack of agentivity and passivizability)
and the presence of accusative case can be seen as an argument for a dependent
case approach towards accusative assignment, as opposed to the standard Voice-
based approach (Kratzer 1996; Chomsky 1995, 2000, 2001). While a dependent
case approach can be also assumed for the agentive transitive as well, the important
issue is that passivization relates to the presence of Voice.
With this background lets us now turn to Greek EO verbs.
9.3 The Properties of EO Verbs
As detailed in Anagnostopoulou (1999), where the following examples come from,

Greek has three classes of experiencer verbs. As in Italian, agapo (love), latrevo
(adore), antipatho (dislike), miso (hate) belong to Class 1 and they are associated
with an experiencer-subject and theme-object. We will have nothing to say about
this class here:
(14) O Petros agapai ta skilja.
The Peter-NOM loves the dogs-ACC
‘Peter loves dogs.’
Of particular interest here are the EO-verbs like anisixo (worry), provlimatizo
(puzzle), enoxlo (bother), diaskedazo (amuse), fovizo (frighten), endiafero (interest)
which belong to Class 2. In this class, the experiencer bears accusative case and the
theme/cause surfaces with nominative case and agrees with the verb:
(15) Ton Petro ton anisihi i katastasi.
The Peter-ACC cl-ACC worry-3SG the situation-NOM
‘The situation worries Peter.’
Greek also has a class corresponding to the Italian piacere-predicates (Class 3).
This includes verbs like aresi (like), ftei (bothers/matters) selecting for a dative
experiencer (PP as in (16a) or morphological genitive as in (16b)) and a nominative
agreeing theme:
(16) a. To krasi aresi ston Petro.

The wine-NOM like-3SG to-the Peter
‘Peter likes the wine.’
b. To krasi tu aresi tu Petru.
The wine- NOM cl- GEN like-3 SG the Peter- GEN
c. Tu Petru tu aresi to krasi.
The Peter-GEN cl-GEN like-3SG the the wine-NOM
Anagnostopoulou (1999) argued, as has been done by Belletti & Rizzi (1988)
and Masullo (1993) for Italian and Spanish, respectively, that the fronted datives
in psych verb constructions qualify as quirky subjects with respect to a num-
ber of criteria. Anagnostopoulou also provided a series of arguments that the
same also holds for accusative experiencers of Class 2 provided that they are
not associated with an agentive subject. We will briefly summarize some of
the arguments here (see Anagnostopoulou 1999 and Landau 2010 for further
discussion).
To begin with, the canonical word order with EO verbs is as in (15)–(16c),
i.e. the experiencer precedes the subject. Second, internal Clitic Left Dislocation
(CLLD) is generally possible in Greek, as shown in (17b–c) which involves CLLD
within a relative clause. However, in contexts where it is dispreferred, as in
(17a), presumably due to discourse factors, EO-fronting is felicitous, as shown in
(17b) and (17c).3
(17) a. #Ekini pu ton Petro ton fovunteine i mathites tu.

Those that the Peter-ACC cl-ACC fear are the students his
‘The ones that fear Peter are his students.’
b. Ekino pu ton Petro ton fovizi ine to mellon.
That that the Peter-ACC cl-ACC frighten is the future
‘What frightens Peter is the future.’
c. Ta vivlia pu tu Janni tu aresun ine ta loghotehnika.
The books that the John-GEN cl-GEN like-3PL are the literary
‘The books John likes are literature.’
Secondly, experiencers can act as binders for nominative anaphors in Greek,

suggesting that at some level of representation they c-command the stimulus
argument.
3 Greek has both clitic-doubling and CLLD. The two structures have been argued to differ from
one another on the basis of several criteria, see Anagnostopoulou (1994) for discussion. We will
discuss clitic doubling in detail in (23).
(18) a. Tin Maria tin provlimatizi/enoxli/anisihi o

The Mary-ACC cl-ACC puzzles/bothers/worries the
eaftos tis.
self-NOM her
‘Mary is puzzled/bothered/worried with/at/by herself .’
b. Tis Marias tis aresi o eaftos tis.
The Mary-DAT cl-DAT like-3G the self-NOM her
‘Mary likes herself.’
c. *Tin Maria den tin thavmazi/aghapai o eaftos tis.
The Mary-ACC not cl-ACC admires/likes the self her
‘*Herself doesn't admire/like Mary.’
We take non-agentive EO verbs to be bi-eventive. While Anagnostopoulou

(1999) did not focus on the fact that EO verbs of class 2 are ambiguous between
agentive and non-agentive as well as stative and eventive readings (see Landau 2010
for a comprehensive discussion and references), Alexiadou & Iordăchioaia (2014)
provide arguments that eventive EO verbs are actually similar to causative verbs
in having a bi-eventive structure of the type in (11) or (12). Recall that in (12)
an agent external argument does not introduce a further event. These authors used
a variety of tests to substantiate this point. Among other things, modification via
agent-oriented adverbs and in-X time and for-X time PPs disambiguates between
eventive vs. stative interpretations. For instance, (19a) contrasts with (19b) in that
the former is clearly agentive while the latter is non-agentivie but still involves an
eventive and telic interpretation, as is signaled by the well-formedness of the in-X
time PP. (19c) is non-agentive and stative as signalled by the well-formedness of
the for-X time PP. A further test Alexiadou & Iordăchioaia used involved resultative
readings with ksana ‘again’ with non-agentive and eventive EO verbs which shows
that a decomposition along the lines of the decomposition of causative verbs is on
the right track.
(19) a. O Janis enohlise ti Maria epitides/me ena bastuni. agentive
the John annoyed the Maria intentionally/with a stick
‘John annoyed Mary intentionally/with a stick.’
b. To pehnidi tin enohlise ti Maria se deka lepta. eventive
the game cl-ACC annoyed the Maria in ten minutes
‘The game annoyed Mary in ten minutes.’
c. To kurema tis Marias ton endiefere to Jani stative
the haircut the Maria-GEN cl-ACC interested the John-ACC
ja mia ora.
for an hour
‘Mary’s haircut interested John for an hour.’
Alexiadou & Iordăchioaia (2014) further noted that intransitive variants of this
class show the same morphological distinctions as anticausatives do in Greek.
While certain verbs, e.g. thimono ‘anger’ surface with active in both transitive and
intransitive variants, enohlo ‘bother’ is marked with non-active morphology in its
intransitive form. As shown in (20), intransitive EOs verbs productively allow for
causer PPs introduced by me ‘with’ (less readily combining with apo ‘from’) as well
as by-itself. As discussed in AAS (2006, 2015) and A&A (2009), me-PPs denote
facilitating causers and combine with internally caused verbs, see (13). Moreover,
as these authors have argued the by-itself phrase signals the lack of an external
argument triggering the change of state:4
(20) a. I Maria enohlithike me/?apo to pehnidi/apo moni tis.

The Maria got annoyed with/from the game/by self her
‘Mary got annoyed by the game/by herslf.’
b. *tu aresi me to krasi
cl- GEN like with the wine
Crucially then, eventive EOs of Class 2 qualify as causatives on the basis of

several criteria. By contrast, as shown in (20b), EO verbs of Class 3 are generally
considered stative/non-causative double object unaccusativs, and modification via
causer PPs is impossible.
Interestingly, EO predicates of both Class 2 and Class 3 show obligatory clitic-
doubling, as we had already seen in (17). The reason why doubling is required
with verbs of Class 3 and stative verbs of Class 2 (e.g. interest) is straightforward.
Being unaccusative, these statives require clitic doubling for the reasons outlined
in Anagnostopoulou (2003): to obviate intervention effects when the theme/target-
subject matter moves/enters Agree with T across the experiencer.
(21) Ta mathimatika *?(ton) endiaferun ton Petro.
The mathematics-NOM (cl-ACC) interest the Peter-ACC
‘Mathematics interest Peter.’
What is rather unexpected from the present perspective, however, is the fact that
clitic-doubling is also strongly preferred with eventive EO predicates with causer
subjects and animate experiencer objects (see Verhoeven 2009 for corpus data that
confirm this tendency):
4 AAS argue in detail that me PPs in Greek only receive a causative interpretation with anti-
causatives, they have a manner interpretation in other environments. Next to me, Greek uses the
prepositions apo ‘from’. As these authors detail, apo either introduces agents or causers, the former
construal emerges when it combines with ‘an animate DP, while the latter when it combines with
an inanimate DP. Levin (2009) calls me-PPs ‘facilitating causers’, and for this reason they are
strongly preferred over ‘apo-PPs’ with internally caused anticausatives.
(22) Ta mathimatika %(ton) enthusiasan ton Petro.

The mathematics-NOM (cl-ACC) excited the Peter-ACC
‘Mathematics excited Peter.’
Importantly, clitic doubling found with eventive/causative psych predicates seman-

tically differs from canonical (optional) clitic-doubling of direct objects. As Anag-
nostopoulou details, direct object (DO) clitic-doubling in Greek is felicitous only
with anaphoric definites, not with “novel” or “accommodative” definite (i.e. it is
subject to the Prominence Condition, Heim (1982), cf. Anagnostopoulou 1994, for
details). EO-doubling, on the other hand, violates the Prominence Condition when
the subject is a causer. This difference is exemplified in (23):
(23) a. Prin apo ligo kero eghrapsa mia vivliokrisia jia ena kenourjo vivlio
Before from some time wrote-1SG a review about a new book
pano sto clitic doubling.
on the clitic doubling
‘Some time ago, I reviewed a new book on clitic doubling.’
b. #Arghotera ton sinandisa ton sigrafea se ena taksidhi mu.
Later.on cl-ACC met-1SG the author-ACC in a trip my
‘Later on, I met the author during a trip of mine.’
c. I kritiki mu ton enohlise ton sigrafea
The criticism my cl-ACC bothered the-author-ACC
toso oste na paraponethi ston ekdhoti.
such that SUBJ complain to-the editor
‘My criticism bothered the author so much that he complained about
it to the editor.’
As (23b) shows, doubling of the direct object ton sigrafea in a canonical transitive
sentence is infelicitous in a context where the definite may satisfy the Prominence
Condition only via accommodation (i.e. linking of the index k of “the author” to the
index i of “the new book on clitic doubling that the speaker reviewed some time ago”
which has already been introduced in the discourse). The acceptability of (23c) in
the same context indicates that EO-doubling is not subject to this restriction. Note,
finally, that clitic doubling is completely optional and subject to the Prominence
Condition when the subject is an agent, as illustrated in (24), which behaves exactly
like the canonical transitive (23b):
(24) #Epitides ton enohlisa ton sigrafea.
Deliberately, cl-ACC bothered the-author-ACC
‘I deliberately irritated the author.’
Landau relates the syntactic differences between agentive and non-agentive EOs
to the aspectual difference achievement vs. accomplishment. In Alexiadou &
Anagnostopoulou (2019), we argue that this aspectual distinction in the cases at
hand correlates with a structural difference concerning the upper part of the vP
domain which provides the key for understanding the syntactic differences between
the two types of constructions. We will turn to this in the next section.
9.4 The Proposed Syntactic Analysis of Greek EO Verbs
Alexiadou & Anagnostopoulou (A&A 2019) argue, following Alexiadou (2014,

2018) that when the subject is a causer the structure is as in (25), i.e. a Voice layer
is missing, as in (11a) above proposed for internally caused verbs.5 As shown in
(25), the experiencer argument originates within the ResultP similarly to the theme
argument in the structures in (11).
(25) vP
Causer v'
v ResultP
experiencer √FEAR/ √WORRY/ √PUZZLE etc.
The assimilation of causative EO-verbs to internally caused change of state

verbs suggests that the constructions under discussion simultaneously have an
internal causer/undergoer argument (the experiencer argument) and an external
causer argument, unlike agentive experiencer verbs in which the experiencer is a
mere undergoer.6 Pesetsky (1995: 113–123) offered a similar analysis by linking the
internal causer role to a reflexive clitic bearing the role A(mbient)-Causer. Affixation
of a CAUS head introducing the external causer in experiencer object predicates led
to the suppression of the external argument bearing the A-Causer role and promotion
of the external causer DP to a derived subject position.
5 An anonymous reviewer asks whether a transitive ICCOS can be found with an animate theme
argument, and if so if it displays the subject properties identified for the object experiencer. The
anonymous reviewer suggests wither as a potential candidate for such a construal. Indeed, such
examples exist, (i), and the theme argument is interpreted as an experiencer, the experiencer is
clitic-doubled:
(i) Ta provlimata tin marazosan ti Maria
the problems cl withered-3PL the Maria-ACC
‘Problems depressed Mary.’
In turn this means that it behaves on a par with other EO predicates, as expected. Other verb classes
that can be coerced into experiencer readings are discussed in Alexiadou & Anagnostopoulou
(2019).
6 As an anonymous reviewer points out, this is exactly as with ICCOS verbs where the theme has a
double role, whereas in externally caused change of state verbs it is the true theme.
On the other hand, agentive EO verbs project an additional Voice layer, as in

(26):
(26) VoiceP
agent Voice'
Voice vP
v ResultP
experiencer √FEAR/ √WORRY/ √PUZZLE etc.
According to A&A (2019), it is precisely the presence vs. absence of Voice that
interacts in a crucial way with the special syntactic properties of experiencers in
eventive EO constructions. The presence of psych effects, i.e. backward binding,
is crucially related to the absence of Voice, which also explains the special
syntactic properties of accusative experiencer objects. Specifically, A&A (2019)
maintain the core intuition behind Landau’s locative inversion analysis, namely
that animate experiencer objects syntactically behave like subjects at some level
of representation, and this happens in EO constructions lacking Voice. The intuition
that EO-constructions are similar to multiple subject constructions is expressed in
Stowell (1986) and Cambell & Martin (1989) and relates to a semantic insight
expressed in different forms in Grimshaw (1990), Dowty (1991) and Reinhart
(2002), namely that there are two core properties behind subjecthood: causing
change and what Reinhart expresses with the feature [+ mental state involved].
In EO-constructions of the type discussed here the causer argument expresses the
former property and the experiencer argument expresses the latter.
When agentive Voice is present, however, as in (26), both properties are
associated with the agent argument. In structural terms, we take this to mean that the
agent has a privileged relationship with T, while the experiencer object is interpreted
in the vP like any other object. It may undergo clitic doubling like all DP objects in
Greek, animates and inanimates, falling under the Prominence condition, as shown
in (24). Assuming that clitic doubling always involves a movement relationship
between the vP-internal DP object and the doubling clitic in T, this means that
the object obligatorily reconstructs below the agent subject for phenomena like
e.g. anaphora, binding, adjunct control (see Anagnostopoulou 1999; Landau 2010
for the details of the systematic differences between these phenomena in agentive
vs. non-agentive experiencer object predicates in Greek, Hebrew and many other
languages).
On the other hand, EO-verbs with a causer subject lack a Voice layer, and the two
core subject properties are distributed over two different arguments, the causer and
the experiencer, respectively, as in structure (25). The experiencer must establish a
movement relationship with T at some point in the derivation (overtly or at LF), as
Stowell (1986), Cambell & Martin (1989) and Landau (2010) suggest. The strategy
to do this in Greek is via clitic doubling, which is obligatory in this case, not subject
to the Prominence Condition and leads to a subject-behavior of the experiencer with

respect to e.g. backward anaphora, binding and adjunct control, a fact suggesting
that it does not reconstruct below the causer subject. This explains why this type of
clitic doubling accusative construction in Greek is sensitive to animacy, unlike all
other instances of accusative clitic doubling.7
Assuming that Voice is a phase head (Chomsky 2000, 2001) can naturally explain
the asymmetry between agentive and causative EO-predicates. When VoiceP is
present all subject properties will be associated with the external argument in the
specifier of Voice and the internal argument will be too low to enter any relationship
with T. If Voice is a phase head it will trigger spell-out of its complement domain
(the vP in (26)) which includes the internal argument. In the absence of Voice, the
vP in (25) will be spelled out with T when C is introduced, explaining why the clitic
doubled T-related experiencer does not have to reconstruct below the subject.8
7A question that arises is whether clitic doubling effects are found also with ICCOS verbs.
Alexiadou (2014) points out that indeed in most examples involving transitive ICCOS verbs a clitic
is present and refers to work by Roussou & Tsimpli (2007), who establish a parallelism between
this class of predicates and EO verbs that take causer subjects. Roussou & Tsimpli (2007) analyze
the clitic as a sign that a causative structure is present, which is in line with the analysis offered
here.
8 We mentioned in Sect. 9.3 that Greek like other languages also has a Class 3 EO verbs (piacere
verbs) whose experiencers bear dative case morphology. We did not discuss them in this paper
as these are stative predicates. We assume that dative experiencers are introduced via applicative
heads and receive dependent dative case in opposition to a lower argument, like all applicative
arguments in Greek (Anagnostopoulou & Sevdali 2018). We further assume that these lack a v layer
introducing event implications and offer the applicative structure in (i) from Anagnostopoulou &
Sevdali (2018):
(i) vAPPLP
EXPERIENCER-GEN vAPPL’
vAPPL’ ROOTP
THEME-NOM
The two different structures proposed for the two EO classes enable us to explain why in the
one case the experiencer surfaces with accusative, while it surfaces with dative in (i). In (25) the
experiencer is assigned dependent accusative in opposition to the higher argument. On the other
hand, in (i) the experiencer is introduced by vAPPL. Following Anagnostopoulou & Sevdali (2018)
we assume that dative (morphologically genitive) in Greek is dependent case ‘upwards’ assigned
in opposition to a lower argument in the vAPPL domain, in accordance with the rule in (ii):
Dependent genitive case rule:
(ii) If DP1 c-commands DP2 in vAPPLP, then assign U (genitive) to DP1
Nevertheless, in both structures the experiencer and the theme arguments are in the same syntactic
domain and as a result behave as multiple subject constructions.
This particular analysis of causative EO verbs has larger implications for the
proper treatment of a group of verbs that have been labeled defeasible causatives in
Martin & Schäfer (2017), which we will briefly discuss in the next section.
9.5 Defeasible Causatives9
Oehrle (1976: 25) was perhaps the first to observe that verbs such as offer show a
meaning contrast depending on the thematic role their subject bears. While in the
case of an agentive subject it is not required that the internal argument has taken up
what was offered, this is not the case when the subject is a causer:
(27) Peter offered us a bed. But we didn’t want to lie there.
(28) Leaves, mingled with grass, offered us a bed. #But we didn’t want to lie
there.
A similar observation is made for the verb teach, while the example in (29) does not
imply a change of state, this is the case in (30) where some learning has taken place.
(29) Ivan taught me Russian, but I did not learn anything.
(30) Lipson’s textbook taught me Russian, # but I did not learn anything.
These and many other cases from other verb classes are discussed in Martin &
Schäfer (2017) who label these verbs defeasible causatives. According to Martin
& Schäfer (op.cit.), the difference between the French examples in (31a) and (31b),
involving an EO verb, has to do with the fact that with a causer subject the change
of state is strongly implicated if not entailed.
(31) a. Pierre l’a provoquée, mais cela ne l’a pas touchée du tout.
Pierre her has provoked but this NEG her has NEG touched at all
‘Pierre provoked her, but this didn’t touch her at all.’
b. Cette remarque l’a provoquée, #mais cela ne l’a pas touchée du tout.
This remark her has provoked but this NEG her has NEG touched at all
‘This remark provoked her, but this didn’t touch her at all.’
Martin & Schäfer (2017) point out that there is no difference in event complexity
between the two examples, both are bi-eventive verbs in the sense that they involve
a causative component leading to a change of state and offer a semantic analysis
thereof adopting a sub-lexical modal base. The reader is referred to their paper
for details. They also note that a similar contrast emerges between instrument and
causer subjects: when the subject is an instrument the result can be denied, when
the subject is a causer this is not possible:
9 All examples in this section taken from Martin & Schäfer (2017) and Martin (this volume).
(32) a. Le discours du recteur l’ a vraiment flatté à plusieurs reprises,

the discourse of the dean him has really flattered on several occasions
mais cela l’a laissé complètement indifférent.
but this him has left completely indifferent
‘The speech of the dean really flattered him on several occasions, but it
left him totally indifferent.’
b. Ce détail l’a vraiment flatté, # mais cela l’ a laissé complètement
this detail him has really flattered but this him has left completely
indifférent.
indifferent
‘This detail really flattered him, but it left him totally indifferent.’
Defeasible causatives have not been discussed specifically for Greek, but as Martin
& Schäfer (2017) point out the same verb classes yield similar effects across
languages, including Greek. The verbs under discussion include: agentive EO verbs
and verbs of social interaction (e.g. encourage, flatter), verbs of communication
(e.g. suggest), influence verbs (e.g. demand, urge), verbs of caused perception (e.g.
show), verbs of caused possession, (e.g. teach), epistemic verbs (e.g. verify, justify),
and a miscellaneous class (e.g. cure, clean).
The examples seem to suggest that in order to obtain the Oehrle effects, it is
important not only to have a causer but also an animate theme or an animate indirect
object.10 In other words, two ingredients seem necessary for defeasibility: the
inanimate property of the causer and the animate property of the theme. However,
this does not hold across verb classes. As Martin & Schäfer (2017) discuss, with
epistemic verbs as well as their miscellaneous class, animacy does not play a role,
i.e. the theme argument need not be animate. Nevertheless, as psych verbs must
be associated with animate arguments, while other predicates do not have to, it is
expected that EO verbs figure prominently in the discussion of defeasible causatives.
Focussing on Greek, all Greek EO verbs with causer subjects that figure in the
list provided by Martin & Schäfer (2017) have experiencers that show subject-
like properties. Moreover, beyond the EO class, as Alexiadou & Anagnostopoulou
(2019) detail, in configurations that involve causer subjects and animate arguments,
the animate argument is clitic-doubled and in certain cases the predicate is coerced
into an experiencer interpretation, as is the case in (33b). Thus, the requirement that
is special to Greek in the above configurations is that the animacy of the theme
in contrast to the [-animacy] of the causer is signalled by the presence of a clitic.
When both arguments are inanimate no special marking is required, though doubling
might occur. Crucially, both EO and coerced EO verbs show the behavior Martin &
Schäfer (2017) identify for the French and German verbs they discuss:
10 We are grateful to an anonymous reviewer for discussion on this section. The reviewer further
points out that in English The book/Her teacher angered/bored Anna, are equally non-defeasible
with the agent or the causer external argument. We believe this can be explained by assuming, as
we did in Sect. 9.1, that unintentional agents are actually causer arguments.
(33) a. To pehnidi tin enohlise ti Maria # ala emine anenohliti.

the game cl-ACC annoyed the Maria but remained unbothered
‘#The game annoyed Mary but she remained unbothered.’
b. I fori ton gonatisan to Jani #ala den ipoferi katholu.
the taxes him kneeled the John but not suffer at all
‘#Taxes cause suffering to John but he does not suffer at all.’
Martin & Schäfer (2017) do not offer a syntactic explanation of why the role of the
external argument influences the interpretation of these verbs. Martin (this volume)
revisits this discussion and offers an analysis that is very similar to our proposal
that agents and causers are introduced differently. According to Martin, there are
two types of Voice, Voiceag introducing Agents and Voicec introducing causers. In
the semantic characterization she proposes, given in (34), the two heads combine
differently with the vP, the causative one being very similar to the structure of
anticausatives. As Martin emphasizes, the difference between anticausatives and
transitives with non-agentive subjects is that the external argument introduces an
eventuality causing the event denoted by the VP. Agents, by contrast, introduced in
Voiceag do not introduce a further event.
(34) a. Event types denoted by causative VPs used agentively are tokenized
by event tokens having an action of the subject’s referent and an
ensuing change-of-state of the theme’s referent as their parts.
b. Event types denoted by causative VPs used non-agentively are
tokenized by change-of-state event tokens of the theme’s referent
(when R=cause)
Martin’s structure containing Voicec is close to our structure that lacks Voice
altogether and simply contains a vP; her overall analysis according to which non-
agentive causative VPs are similar to anticausative structures is compatible with our
structures in (11) and (25). In fact, we believe that our analysis captures her observa-
tion that non-agentive causatives are similar to anticausatives, as our structures are
in fact anticausative structures. From our perspective, a basic difference between
agent/instrument and causer subjects is that the latter are introduced within vP, i.e.
in Spec,vP, while the former are associated with Voice. Since both arguments are
contained within the same syntactic domain they are interpreted in the same phase,
and the change of state is entailed. When Voice is present, the vP is spelled-out
independently of the agent.
In support of her analysis, Martin employs a series of tests, e.g. the in-adverbial
test. As Martin points out, in-adverbials measure the span between the beginning
and the end of the complete eventuality denoted by the verbal predicates. In the
following examples, from Martin’s paper, we see, that the non-agentive variant
behaves like the anticausative counterpart. From our perspective this is so, as
anticausatives and transitive non-agentive verbs have the same structure:
(35) a. The poison he swallowed this morning killed him in ten minutes in the
evening (# this being said, he died in less than a minute).
b. Mary killed him this evening in ten minutes (this being said, he died in
less than a minute).
c. He died in ten minutes this evening because of the poison he swallowed
this morning.
Does this mean that all non-agentive causative verbs across languages, irrespectively
of their characterization in terms of external vs. internal causation, have a structural
representation along the lines of (11)? Here it seems to us that passivization truly
splits the two groups as well as the different languages. Transitive ICCOS verbs and
causative EOs do not passivize, while externally caused change of state verbs do
passivize, albeit this is severely restricted in Greek, see AAS (2015) for discussion.
As far as we can tell, Schäfer’s (2012) proposal that causers are syntactically
licensed in Voice, and Martin’s Voicec cannot explain why our verbs behave
differently with respect to passivization and why our EOs have subject properties.
This issue awaits further research.
9.6 Conclusions
In this paper, we argued that causative EO verbs can be treated similar to transitive
internally caused change of state verbs. Our analysis is built on the hypothesis that
agent and causers are introduced in distinct layers in the structural representation of
verbs, agents in VoiceP, while causers in vP. We explored the consequences of this
analysis for defeasible causatives and argued that the difference in representation
leads to distinct interpretations: in the presence of a causer argument, which is part
of the same phase as the undergoer argument, the change of state is entailed.
According to our analysis, there are two conceptions of transitivity and causation:
one related to the presence of Voice, which is associated with the presence of
agentive external arguments and allows for defeasible causative interpretations.
The second conception of transitivity, however, is associated with the presence
of a causer in vP: this particular configuration leads to entailments of change of
state. The fact that undergoer arguments bear accusative case in both environments
follows naturally from the theory of dependent case.
Acknowledgements We are grateful to three anonymous reviewers and the editors of this volume
for their comments. Special thanks to Fabienne Martin and Malka Rappaport Hovav. AL 554/8-1
(Alexiadou) is hereby acknowledged.
References
Alexiadou, A. (2014). The problem with internally caused change-of-state verbs. Linguistics, 52,
879–910.
Alexiadou, A. (2018). Able adjectives and the syntax of psych verbs. Glossa: a Journal of General
Linguistics, 3, 74. https://doi.org/10.5334/gjgl.498.
Alexiadou, A., & Anagnostopoulou, E. (2009). Agent, causer and instrument PPs in Greek:
Implications for verbal structure. MIT Working Papers in Linguistics, 57, 1–16.
Alexiadou, A., & Anagnostopoulou, E. (2019). Novel object experiencer predicates and clitic
doubling. Syntax. https://doi.org/10.1111/synt.1217.
Alexiadou, A., Anagnostopoulou, E., & Schäfer, F. (2006). The properties of anticausativs cross-
linguistically. In M. Frascarelli (Ed.), Phases of interpretation (pp. 187–212). Berlin: Mouton
de Gruyter.
alternations. Oxford: Oxford University Press.
Alexiadou, A., & Iordăchioaia, G. (2014). The psych causative alternation. Lingua, 148, 53–79.
Anagnostopoulou, E. (1994). Clitic dependencies in Modern Greek. PhD dissertation, University
of Salzburg.
Anagnostopoulou, E. (1999). On experiencers. In A. Alexiadou, G. Horrocks, & M. Stavrou (Eds.),
Studies in Greek syntax (pp. 67–93). Dordrecht: Kluwer.
Anagnostopoulou, E. (2003). The syntax of ditransitives: Evidence from clitics. Berlin: Mouton de
Gruyter.
Anagnostopoulou, E., & Sevdali, C. (2018). Two modes of dative and genitive case assignment:
Evidence from two stages of Greek. Ms. U Crete and U Ulster (submitted).
Arad, M. (1998). Psych-notes. UCL Working Papers in Linguistics, 10, 203–223.
Baker, M. (2015). Case: Its principles and parameters. Cambridge: Cambridge University Press.
Baker, M., & Vinokurova, N. (2010). Two modalities of case assignment in Shaka. Natural
Language and Linguistic Theory, 28, 593–642.
Belletti, A., & Rizzi, L. (1988). Psych-verbs and theta-theory. Natural Language and Linguistic
Theory, 6, 291–352.
Campbell, R., & Martin, J. (1989). Sensation predicates and the syntax of stativity. Proceedings of
WCCFL, 8, 44–55.
Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press.
Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels, & J.
Uriagereka (Eds.), Step by step. Essays on minimalist syntax in honour of Howard Lasnik (pp.
89–155). Cambridge, MA: MIT Press.
Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A life in language
(pp. 1–52). Cambridge, MA: MIT Press.
Croft, W. (1998). Event structure and argument linking. In M. Butt & W. Geuder (Eds.),
The projection of arguments: Lexical and syntactic constraints (pp. 21–63). Stanford: CLSI
Publications.
Dowty, D. (1991). Thematic proto-roles and argument selection. Language, 20, 547–619.
Harley, H., & Noyer, R. (2000). Formal vs. Encyclopedic properties of vocabulary: Evidence
from nominalization. In B. Peters (Ed.), The lexicon-encyclopedia interface (pp. 349–374).
Amsterdam: Elsevier Press.
Heim, I. (1982). The semantics of definite and indefinite noun phrases. PhD dissertation, Universtiy
of Massachusetts at Amherst.
Phrase structure and the lexicon (pp. 109–137). Kluwer: Dordrecht.
Landau, I. (2010). The locative syntax of experiencers. Cambridge, MA: MIT Press.
Lavidas, N. (2007). The diachrony of Greek anticausative morphology. In A. Alexiadou (Ed.),

Studies in the morpho-syntax of Greek (pp. 106–135). Cambridge: Cambridge Scholars
Publishing.
Levin, B. (2009). Further explorations of the landscape of causation: comments on the paper by
Alexiadou & Anagnostopoulou. MIT Working Papers in Linguistics, 49, 239–266.
Levin, B., & Hovav, M. R. (1995). Unaccusativity: At the syntax-lexical semantics interface.
Marantz, A. (1991). Case and licensing. Eastern States Conference on Linguistics (ESCOL ’91),
8, 234–253.
McCoon, G., & Macfarland, T. (2000). Externally and internally caused change of state verbs.
Language, 76, 833–858.
Martin, F. (This volume). Aspectual differences between agentive and non-agentive uses of
causative predicates. In Perspectives on causation. Cham: Springer.
Martin, F., & Schäfer, F. (2017). Sublexical modality in defeasible causative verbs. In A. Arregui,
M. L. Rivero, & A. Salanova (Eds.), Modality across categories (pp. 87–108). Oxford: Oxford
University Press.
Masullo, P. (1993). Two types of quirky subjects: Spanish vs. Icelandic. Proceedings of NELS, 23,
303–317.
Oehrle, R. (1976). The grammatical status of the English dative alternation. PhD thesis. MIT,
Cambridge, MA.
Pesetsky, D. (1995). Zero syntax: Experiencers and cascades. Cambridge, MA: MIT Press.
Pylkkänen, L. (2002). Introducing arguments. PhD dissertation, MIT.
Ramchand, G. (2008). First phase syntax. Cambridge: Cambridge University Press.
Reinhart, T. (2002). The theta system: An overview. Theoretical Linguistics, 28(3), 229–290.
Rappaport Hovav, M., & Levin, B. (2012). Lexicon uniformity and the causative alternation. In M.
Everaert, M. Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the interface
Roussou, A., & Tsimpli, I. (2007). Clitics and transitivity. In A. Alexiadou (Ed.), Studies in the
morpho-syntax of Greek (pp. 138–174). Cambridge: Cambridge Scholars Publishing.
Schäfer, F. (2008). The syntax of (anti-)causatives. External arguments in change-of-state contexts.
Schäfer, F. (2012). Two types of external argument licensing: The case of causers. Studia
Linguistica, 66, 128–180.
Solstad, T. (2009). On the implicitness of arguments in event passives. Proceedings of NELS, 38,
365–375.
Stowell, T. (1986). Psych-movement in the mapping between D- structure and LF. Paper presented
at GLOW 9.
Verhoeven, E. (2009). Experiencer objects and object clitics in Modern Greek: Evidence from a
corpus study. In Proceedings of the 8th international conference on Greek Linguistics (pp. 574–
588).
Wright, S. (2001). Internally caused and externally caused change of state verbs. Northwestern
University dissertation.
Wright, S. (2002). Transitivity and change of state verbs. Berkeley Linguistics Society, 28, 339–
350.
Chapter 10
“Agent Exclusivity” Effects in Hebrew
Nominalizations
Odelia Ahdout
Abstract This paper discusses restrictions on the type of external arguments

available in clauses headed by nominalizations in Hebrew. Previous work has
identified a bias against causers in English nominalizations corresponding to
transitive verbs (DP-causers), despite the congruence of both causers and agents
in the base verb. Most accounts of this bias have attributed it to the defective
nature of nominalizations compared to verbs, more specifically the lack of the
Voice projection. Based on the behaviour of two nominal structures in Hebrew,
one of which claimed here to contain Voice, it emerges that the presence of Voice
does not seem to alter the prevalence of this bias, as both structures – with and
without Voice – reject causers. An additional observation is that prepositional-
causers (comparable to English from-phrases), are perfectly grammatical in Hebrew
nominalizations based on anticausative verbs. This class of verbs, believed to
lack (active) Voice to begin with, suggests that the notions of nominalization and
causation are not in principle incompatible, and that the degraded nature of DP-
causers has to do with some other factor (possibly syntactic), but not the absence of
Voice.
Keywords Nominalization · Voice · DP-/PP-causers · Hebrew · Templatic

morphology · Agent Exclusivity
10.1 Introduction
In studies of English nominalizations, it has been observed that the nominal

counterparts to verbs that may have causer subjects (1a) are exclusively agentive
(1b)–(1c), a thematic restriction termed in the nominalization literature as “Agent
Exclusivity” (Lakoff 1970; Grimshaw 1990; Iwata 1995; Pesetsky 1995; Marantz
O. Ahdout ()
Humboldt Universität zu Berlin, Berlin, Germany
e-mail: odelia.ahdout@hu-berlin.de

https://doi.org/10.1007/978-3-030-34308-8_10
320 O. Ahdout
1997; Harley & Noyer 2000; Sichel 2010). This is the case for both “possessor”
subjects (pre-nominal, marked with genitive) (1d), and prepositional complements
(1e) (adapted from Harley & Noyer 2000: (14b)):
(1) a. The cold war/The Allies separated East and West Germany. verb
b. The Allies’ separation of East and West Germany Possessor agent
c. The separation of East and West Germany by The Allies PP-agent
d. #The cold war's separation of East and West Germany Possessor causer
e. The separation of East and West Germany #by the cold war PP-causer
This effect has been attributed by some to the lack of Voice in the nominalization
(Harley 2009; see also Marantz 1997), or to event structure restrictions which stem
from a common intuition that the nominalized form is structurally “smaller” com-
pared to the structure of the corresponding verb (Sichel 2010; see also Rappaport
1983; Kayne 1984; Abney 1987).
In Hebrew, focusing on transitive causative verbs, e.g. (2a)/(3a), causers are
similarly degraded in nominal clauses (3b)/(3c), compared to agents (2b)/(2c)
(Sichel 2010). As I show in this paper, this is the case for the two structural variants
associated with nominals of transitive verbs. The first, named here the GEN-OBJ
structure, is similar to the English clause in (1c), where the object is marked with
genitive, and the agent is optionally realized via a by-phrase (2b). As for English
(e.g. Harley 2009; Alexiadou 2017), this structure is believed to lack (active) Voice.
In contrast, the ACC-OBJ nominal, (2c) is distinct from the English clauses above
in arguably containing Voice, first and foremost by virtue of marking its object with
accusative case, and the obligatory status of the subject (see Ahdout In preparation
for a set of diagnostics, partially described here in Sect. 10.2). As such, the ACC-
OBJ nominal is closer to the syntactic structure of the base verb than the GEN-OBJ
nominal/English counterpart:
(2) a. ha-ʿiriya harsa et ha-mivne verb + agent

the-municipality destroyed ACC the-building
‘The municipality destroyed the building’.
b. harisat ha-mivne (al-jedej ha-ʿirija) GEN-OBJ nominal
the.destruction (of) the-building.GEN by the-municipality
‘The destruction of the building by the municipality’
c. harisat ha-ʿirija et ha-mivne ACC-OBJ nominal
the.destruction (of) the-municipality.GEN ACC the-building
‘The municipality’s destruction of the building’
10 “Agent Exclusivity” Effects in Hebrew Nominalizations 321
(3) a. ha-∫eme∫ ha-xazak-a harsa et verb + causer

the-sun the-strong-F.SG destroyed.F.3SG ACC
ha-rehitim b-a-mirpeset
the-fourniture in-the-balcony
‘The strong sun ruined the furniture in the balcony ’.
b. harisat ha-rehitim (al-jedej ha-∫eme∫) GEN-OBJ nominal
the.destruction (of) the-furniture.GEN by the-sun
c. ?harisat ha-∫eme∫ et ha-rehitim ACC-OBJ nominal
the.destruction (of) the-sun.GEN ACC the-furniture
Despite the incongruence of causers with nominalizations derived from transitive

causative verbs, Hebrew shows that the notions of causation or causer are not
in principle incompatible in the nominal domain, and that the causer role is
not categorically absent in these environments. Anticausative verbs allow PP-
causers, introduced by from-phrases (4a), and this realization pattern repeats itself
unrestrictedly in the corresponding nominalization (4b).
˘
(4) a. ha- or hitrakex/hitnape ax/hitjabe∫ me-ha-krem anticausative verb

the-skin softened/swelled.up/dried from-the-creme
‘The skin softened/swelled up/dried from the cream’.
b. hitrakxut/hitnapxut/hitjab∫ut ha- or me-ha-krem anticausative nominal
˘
the.getting.soft/swollen/dry (of) the-skin.GEN from-the-cream
The Hebrew data regarding the availability of PP-causers with anticausative

nominals adds to observations in Alexiadou et al. (2013a, b) on Greek, Romanian,
German, Spanish and French, languages which similarly license a causer as a
prepositional phrase. PP-causers have also been reported for nominalizations of
psychological verbs in Greek and Romanian, where it has been claimed that these
nominalizations are based on the anticausative alternant of causative-alternating
verbs (Alexiadou & Iordăchioaia 2014).
The behaviour of nominals produced from transitive causative verbs, namely the
picking out of the agent role over the causer role, provides another motivation to
consider causers and agents as distinct at some level of representation. One (of
the) option(s) discussed in the literature is that causers and agents are syntactically
distinct, e.g. in several works on causation (Pylkkänen 2002, 2008; Doron 2003;
Folli & Harley 2005; Kratzer 2005; Schäfer 2012; Alexiadou et al. 2015).
The implications drawn from behaviour of Hebrew nominalizations are two-
faceted; on the one hand, in contrast to previous findings, nominalization is shown
to be compatible with the notion of causation, as evident from the licensing of PP-
causers in anticausative environments in Hebrew. On the other hand, the behaviour
of ACC-OBJ nominals puts into question the association in the literature between
Voice and the causer role, as non-agentive causers are incompatible even in a
nominal structure which should contain all the necessary structural layers to license
non-agentive causers. From this, it emerges that the absence of causers in transitive
nominalizations cannot be a reflex of the absence of Voice.
In the following, I discuss the distribution of causers in verbal and nominal envi-
ronments in Hebrew. In Sect. 10.2, I introduce the two deverbal noun constructions
322 O. Ahdout
in the language. In Sect. 10.3, I review the literature on causation and causers in
verbal clauses, and on the “Agent Exclusivity” effect exhibited in nominal clauses.
In Sect. 10.4, I present novel data on Hebrew nominalizations and discuss the contri-
bution of these data to the understanding of the nature of the factors that may or may
not be related to the phenomenon of “Agent Exclusivity”. I conclude in Sect. 10.5.
10.2 The Structure of Hebrew Nominalizations Corresponding

to Transitive Verbs
In Hebrew, verbal templatic morphology has been claimed to mark Voice distinc-
tions (Doron 2003; Alexiadou & Doron 2012; Kastner 2016, 2018): active-marked
verbs are (mostly) transitive, middle-marked verbs are only intransitives (unac-
cusative/anticausative, unergative, reflexive and reciprocal). Nominalization pre-
serves these templatic distinctions, and nominals can also be argued to mark Voice
(Ahdout In preparation, on formal relations between verbal templates and nominal
derivatives, see Borer 2013).1 In Sect. 10.4.3 I discuss in more detail the relevance
of templatic morphology to the subject matter of this paper, namely the distribution
of non-agentive causer participants and the correlation with morpho-syntax.
Transitive verbs in Hebrew derive nominalizations which may head two distinct
types of clausal structures (for the different classes of intransitive verbs see Ahdout
In preparation, and Ahdout & Kastner to appear).
The first structural variant is referred to here as the ACC-OBJ structure/nominal
(6a). Most notably, this structure exhibits two properties which render it ‘verbal’:
the marking of accusative on the internal argument, and the obligatory realization of
the external argument. The structure referred to here as GEN-OBJ (6b) marks the
internal argument with the possessive preposition Sel ‘of’, which parallels genitive
marking in the nominal domain (see below (9b) for a variant for marking the
possessive relation). The external argument is optional, and may be realized with
a by-phrase, on a par with English, as apparent from the translation.
(5) sar ha-ʾotsar higdil et taktsiv verb
the.minister (of) the-finance increased.CAUS.ACT ACC the.budget
ha-revaxa.
the-welfare
‘The minister increased the welfare budget’.
1 Glosses for Hebrew verbs/nominals are based on Doron (2003). The mnemonics “simple”,
“causative” and “intensive” are used here descriptively (see the system in Doron for motivation for
the original classification to these three classes). ACT stands for a morphological active template,
MID for a middle template, PASS for a passive template. SMPL.ACT for the XaYaZ verbal templatic
form, CAUS for the hiXYiZ template, INTNS.ACT for XiYeZ, INTNS.MID for hitXaYeZ, SMPL.MID
˘ ˘
for niXYaZ.
(6) a. ha-hagdala ∫el ha-sar ACC-OBJ nominal

the-increasing.CAUS.ACT of the-minister.GEN
et taktsiv ha-revaxa
ACC the-budget (of) the-welfare
‘The minister’s increasing of the welfare budget’
b. ha-hagdala ∫el ha-taktsiv GEN-OBJ nominal
the-increasing.CAUS.ACT of the-budget.GEN
(al-jedej ha-sar)
by the-minister
‘The increasing of the budget (by the minister)’
The GEN-OBJ nominal implies an agent even in the absence of the by-phrase,
as is the case in English.2 This is reflected below in the congruency with a purpose
clause (see among others Grimshaw 1990; Alexiadou 2001, 2017):
(7) ha-hagdala ∫el taktsiv ha-revaxa kedey
the-increasing.CAUS.ACT of the.budget the-welfare in.order
le-fatsot al ha-haznaxa ∫el ha-mem∫ala ha-kodem-et
to-compensate on the-neglect of the-government the-former-F.SG
‘The increasing of the welfare budget in order to compensate for the neglect of the
former government’
Note that the post-nominal genitive DP is the “subject” in the former structure,
but the “object” in the latter. Importantly, the subject genitive-DP in (6a) differs
from the by-phrase realization of the subject in GEN-OBJ nominals (6b) in being
obligatory; the accusative marked DP in the ACC-OBJ nominal is dependent on
the presence of the genitive DP (Borer 2013) (8). In the absence of the genitive DP
subject, the only way to realize the object DP is as in (6b), i.e. with genitive/Sel ‘of’.
(8) *ha-hagdala et ha-taktsiv
the-increasing.CAUS.ACT ACC the-budget
A final issue concerns the morph-syntax of genitive marking/the possessive

relation in Hebrew. The so-called Construct State (CS) (Ritter 1991) is an alternative
strategy of realizing the possessive relation in noun-noun compounds, and in this
case, the nominal and post-nominal DP (be it the subject or object role). In the
CS, instead of the overt genitive-like preposition Sel ‘of’, the possessive relation is
usually achieved via the use of a stem allomorph of the first noun (here, a deverbal
2 See Chomsky (1970), Pesetsky (1995), Marantz (1997), Harley & Noyer (2000), and Sichel
(2010) for causative verbs which allow both a transitive and an intransitive/unaccusative construal
in the nominal, e.g. ‘the explosion of the balloon’. In Hebrew, the overt marking of Voice
alternations eliminates most cases of such ambiguity, see Sect. 10.4.
324 O. Ahdout
noun). Finally, the first noun in the CS does not take the definite article, but instead
is interpreted as definite by virtue of the overt definiteness of the second DP:3
(9) a. ha-hagdala Sel ha-taktsiv Sel + DP (‘Free State’)
the-increasing.SMPL.ACT of the-budget
b. hagdalat ha-taktsiv Construct State4
the.increasing.SMPL.ACT (of) the-budget
For sake of consistency, the examples in the next sections are all given in the CS
variant. I briefly return to this issue in Sect. 10.4.1.
The contrasts between the GEN-OBJ and ACC-OBJ constructions are informa-
tive ones. One of the most important differences between verbal and nominal clauses
is the non-obligatory nature of the external “argument” in the nominal clause,
whereas in the verbal clause it is always required. (Grimshaw 1990 and subsequent
literature). The ACC-OBJ structure sheds light on the interaction between the
morpho-syntactic properties of the nominal clause (namely case marking), and
argument realization: accusative marking is concomitant with the (obligatory)
realization of the subject, i.e. with (active) Voice (but see Siloni 1997 for the claim
that some verbal properties found in the nominal domain are defective compared to
the verbal clause, and reply in Borer 2013).
Following works which posit an embedded verbal structure within nominal-
izations (e.g. Hazout 1991, 1995; Fassi-Fehri 1993; Fu et al. 2001; Engelhardt
1998, 2000; Alexiadou 2001; Borer 2013), I assume that both structures contain
a little v layer. However, on the basis of case marking and the status of the
subject/external argument (obligatory vs. optional/implicit), I propose that ACC-
OBJ nominals necessarily contain (non-passive) Voice (see Ahdout In preparation)
(10a). In contrast, GEN-OBJ nominals pattern with English nominals of the type
(1c). Below are the structures for the two constructions:
(10) a. ACC-OBJ b. GEN-OBJ
3 The existence of systematic differences between the CS and the Sel + DP is not a clear matter,
and would require further investigation, not taken up here. See Rappaport & Doron (1990) for
discussion.
4 As mentioned above, usually the first noun in the CS shows allomorphy, in this case the addition
of stem-final [-t].
In both structures, genitive is proposed to be licensed by the n head, see recent

discussion in Alexiadou (2017) for the relevant English constructions. In the ACC-
OBJ structure, the object DP receives accusative case by virtue of Voice being
present. As a T head is absent in nominalization, genitive (instead of nominative)
case is assigned to the subject, being the unmarked case in the nominal domain
(Alexiadou 2017).
In the GEN-OBJ structure, which lacks Voice and consequently also accusative
marking, (e.g. Marantz 2000) it is the internal argument which receives genitive
marking. The ‘subject’, in turn, may only be realized as an oblique, on a par with
a passive clause. Note that the parallelism between passives and deverbal nominals
in English, and here, with GEN-OBJ nominals (10a), stems from the intuition that
in both passives and nominalizations, the external argument of the verb is no longer
obligatorily realized, but is still present in the semantics, e.g. with implicit control
into a purpose clause, as in (7), (see Grimshaw 1990; Borer 2013; Alexiadou 2017).5
Deverbal nouns in Hebrew do not fall under any of the three main categories
described for English: under deverbal nouns/nominalizations here I include -
ation nominals (Borer’s (2013) ATK-AS-nominals) (11a), nominal gerunds (‘ing-of’
gerunds) (11b), and verbal gerunds (‘ACC-ing’ gerunds) (11c) (Reuland 1983). In
Hebrew we see – within one form – a mix of verbal and nominal properties which
is unattested in other languages, deeming the Hebrew nominal harder to classify.
(11) a. The separation of students to different rooms in order to prevent chatting
b. The (teacher’s) separating of students to different rooms in order to prevent chatting
c. The teacher(‘s) separating students to different rooms in order to prevent chatting
I will not enter into a detailed discussion of these forms, but summarize
and note that the literature on deverbal nouns and gerunds in English seems to
converge upon the conclusion that deverbal nouns either lack Voice, or exhibit
defective/passive-like Voice (e.g. Abney 1987; Kratzer 1996; Alexiadou 2001, 2005,
2017; Harley 2009; Borer 2013). This view is mostly motivated by the absence of
accusative marking on the object/internal argument, the non-obligatory status of the
subject/external argument, as well as divergence in other verbal behaviours, e.g.
not allowing particle shift (Chomsky 1970; see discussions in Harley 2009; Sichel
2010). Verbal gerunds are usually believed to include Voice, first and foremost as
they obligatorily realize a subject and mark their objects with accusative (Alexiadou
2005). Nominal gerunds show mixed behaviour, but as they pattern more closely
to deverbal nouns in absence of accusative case and optionality of the subject, they
are taken to lack Voice or have passive-like Voice.6
5 See Sichel (2009) and Bruening (2013) for the view that this agent is a null argument projected in
the syntax.
6 Due to space limitations, I do not discuss in detail the set of nominal properties exhibited by
Hebrew deverbal nouns (for nominal and verbal scales in mixed-category nominalizations see
Borsley & Kornflit 2000; Ackema & Neelman 2004; Alexiadou et al. 2011; Panagiotidis 2015).
I note briefly that the inclusion of the nP and DP projections in the representations above is
326 O. Ahdout
In the following sections, I address the main issue dealt with in this paper, namely
the distribution and licensing of causer arguments in the two variants of Hebrew
nominal clauses described above, with a focus on the presence vs. absence of Voice
and its implications to the congruence of nominalizations with causers. First, I
discuss two syntactic types of causers in English, and their (limited) distribution
in the nominal domain. This discussion is followed by a set of novel generalizations
regarding the distribution of non-agentive causers in nominal clauses in Hebrew.
10.3 Causers in Verbal Constructions: DPs vs. PPs
Causer arguments may be realized via several morpho-syntactic means, two of

which will be considered here (see Schäfer 2012 for an overview and analysis of the
different subtypes of causers). The first instantiation is as a nominative subject DP,
syntactically identical to agentive subjects, and the second – as a (non-obligatory)
prepositional phrase (12a).
The latter kind of causers is discussed in Alexiadou et al. (2015: 31ff; see
also DeLancey 1984; Piñon 2001; Levin & Rappaport Hovav 2005; Kallulli
2006; Schäfer 2012), where anticausative verbs are compared and contrasted to
passives. Based on English, German and Greek, it is shown that anticausative verbs
(12c) – unlike passives (12b) – lack (implicit) external arguments and accordingly
disallow by-phrases. However, it is possible to specify causer participants in clauses
predicated by these verbs via from-phrases, complements describing external or
internal causes or sources (12d):
(12) a. The protestors/the immense pressure shattered the windows. agent/causer (active)
b. The windows were shattered by the protestors. agent (passive)
c. *The windows shattered by the immense pressure. *causer (anticausative)
d. The windows shattered from the immense pressure. causer (anticausative)
These data motivate the view in Alexiadou et al. (2006, 2015; see also Pylkkänen
2002, 2008; Kratzer 2005; among others) that ‘anti’-causatives include some
causative layer, and differ from their transitive-causative counterpart only in the
absence of the structural layer introducing external arguments, Voice.
The separation of causation from Voice embeds the view that causative semantics
is independent of the presence of an external argument in the underlying structure
of the (causative) event (e.g. Doron 2003; Folli & Harley 2005; Pylkkänen 2008).
The question of how to structurally represent the causative meaning component
has several implementations. A recent proposition is made in Alexiadou et al.
justified on the basis of the following behaviours: congruence with adjectival modification, with the
indefinite article, carrying a grammatical gender specification, and optionally marking agreement
and post-nominal DP (the subject for ACC-OBG, and the object for GEN-OBJ, see Engelhardt
1998, 2000), and congruence with the plural (contra Grimshaw 1990). I refer the reader again to
Ahdout (in preparation), as well as relevant discussions in Hazout 1995, Siloni 1997 and Borer
2013.
(2015), where the concept of causation is not taken to be a syntactic primitive,

but is instead interpreted post-syntactically and read-off of the resultative syntactic
structure which includes an unbounded event and a (result) state (see also Schäfer
2012). Under this account, agents and causers are syntactically distinct. Causers
denote a participant in a causative event; not a true causer argument, but rather a kind
of modifier (see also Solstad 2009). Alexiadou et al. (2015) claim that structurally,
PP-causers are adjuncts of the little v layer (13b). Subject DPs, in contrast, are
taken to reside in the specifier of Voice (as do agents) (13a):
(13) a. DP-causer b. PP-causer
(Adapted from Schäfer 2012, ex. 60)
In English, DP- and PP-causers do not only differ from each other in their
morpho-syntactic realization, as described above, but also in the specific nature of
the type of causation they imply. It has been noted (Schäfer 2012; Alexiadou et
al. 2015) that DP-causers (in lexical causatives) may only be direct causers of the
event denoted by the verb (e.g. Fodor 1970; Wolff 2003, but see Neeleman & van
den Koot 2012), whereas no such restriction is found with PP-causers. PP-causers
in English exhibit a wider range of relations to the event; however, the exact nature
of the causative relation as it is represented by PPs seems to be subject to cross-
linguistic variation, and might also depend on the exact preposition that is used (e.g.
in Greek and Romanian, Alexiadou et al. 2015; see also Maienborn & Herdtfelder
2017).
Next, the distribution of causers with nominalized predicates is shown to be
more restricted than the distribution with the corresponding verbal predicates.
I survey two possible sources for this effect that have been suggested in the
literature, both of which attribute the effect to the absence of structural layers in the
nominalization, compared to the base verb. I show that, based on data from Hebrew,
a re-consideration of the view that the ban on causers stems from lack of structure
may be required.
DP-causers in nominal clauses: in the examples below with nominalizations of
transitive verbs, causers are infelicitous:
(14) a. The authorities justified the rapid evacuation of the inhabitants.

b. The authorities' justification of the rapid evacuation of the inhabitants
c. The justification of the rapid evacuation of the inhabitants by the authorities
328 O. Ahdout
(15) a. The approaching hurricane justified the abrupt evacuation of the inhabitants.
b. #The approaching hurricane's justification of the abrupt evacuation of the inhabitants
c. #The justification of the abrupt evacuation of the inhabitants by the hurricane
(Sichel 2010: ex. 32)
This thematic restriction is a matter still unclear in the nominalization literature.

Sichel (2010) suggests that the restriction is in fact not a pure thematic one, but
is related to the relation between the event (the verb and its complements) and
the external/instigating argument (Sichel 2010; cf. Harley & Noyer 2000 on the
relevance of encyclopaedic knowledge). This effect is furthermore shown to be a
more general one, as implied by its manifestation in verbal predicates as well, see
(17) below.
Sichel (2010) claims that nominalizing affixes select for specific types of events.
Focusing on deverbal (-ation) nouns, the suggestion is that these nominals are
restricted to simple, single events, which may be either mono- or bi-eventive (com-
plex), but in the latter case both events must occur in parallel, i.e. be co-temporal,
see Levin & Rapapport Hovav’s (2004) Conditions on Event Identification. Co-
temporality between the two sub-events also means that the participation of the
external argument in the causing event must be co-temporal. If not – the event is
classified as complex, rather than a simple one.
The event structure restrictions under this account are reflected in the proper-
ties of the external argument: the surface phenomenon of “Agent Exclusivity”.
According to Sichel, the correct generalization should be framed not in terms of
agentivity, but rather in terms of directness of participation. The generalization is
that in nominal clauses, only direct participants are allowed, whereas in the verbal
clause the semantic range of causer participants may be wider.
First, unlike direct causers, direct participants are co-temporal (and perhaps co-
spatial) with the outset of the ensuing event. In contrast, direct causers can be non-
co-temporal, as exemplified in, e.g.
(16) The widow murdered the old man by putting arsenic in his coffee.
(Sichel 2010: ex. 52a, attributed to Levin & Rappaport Hovav 2004)
In the examples below, natural forces may be construed as a direct causer of

events such as illuminating and dispersing due to their inherent (physical) properties
(teleological capability in Folli & Harley 2008), but not cancelling or postponing.
For the latter type of events to come about, an intermediate (in this case, agentive)
participant is required, as the ability to cancel or postpone an event is not a part of
the range associated with the aforementioned natural forces.
(17) a. The sun illuminated the room / #The sun postponed the hike.
b. The wind dispersed the tear gas / #The wind cancelled the outdoor show.
(Sichel 2010: ex. 30)
In (18), ‘the war’ is only a direct cause, but not a direct participant, and therefor
is felicitous with the verb, but not with the nominal:
(18) a. The cold war/The Allies separated East and West Germany. agent/causer + verb
b. #The cold war's separation of East and West Germany #causer + nominal
(Sichel 2010: ex. 13)
In contrast, in (19) the semantic requirement on the external argument is met in

both the verbal and nominal clauses, as ‘the wind’ is a direct participant in the event
of dispersal, and as such automatically also a direct causer:
(19) a. The wind dispersed the tear gas.
b. The soldiers counted on the wind’s quick dispersal of the tear gas. (Sichel 2010: ex. 29)
To strengthen this claim, deverbal nominals are compared and contrasted to

nominal gerunds (20) (But see Alexiadou et al. 2013b and discussion in Sect.
10.4.2). The -ing affix in nominal gerunds is not restricted to single, simple events,
and may take a complex event as its complement. Thus, external arguments of the
latter group do not show any semantic restriction, and accordingly the appearance
of direct causers with nominal gerunds is accepted:
(20) The wind's eventual shutting of the door (Sichel 2010: ex. 53d)
A second approach to “Agent Exclusivity” is very briefly discussed by Harley

(2009: 335, fn. 16; see also Chomsky 1970; Marantz 1997; and a discussion in
Sichel 2010), who shares the intuition that the size of the nominalized structure is
correlated with the acceptability of DP causers in the nominal clause. The smaller
deverbal nouns show the strongest ban, followed by the nominal gerund in which it
is mellower, and finally the verbal gerund, in which it vanishes altogether.
However, unlike Sichel (2010), in Harley (2009), the association drawn between
the “size” of the structure and availability of causers does not hinge on event
structure in itself, but on the projections introducing external arguments. To licence
a causer, the nominal has to have a “true” external argument position, rather than
an (inherently ambiguous) possessor, as is found with non-derived nouns such as
“John’s book”.
When a possessor, as in the nominal gerund in (21), the genitive DP may
constitute one of an array of relations to the event denoted by the head nominal,
and is not exclusively construed as the volitional instigator of the event, e.g. “any
other suitable association [between possessor DP and the event], for example mixing
of drugs and alcohol carried out on Belushi’s behalf by some intermediary” (Harley
2009: 324). Importantly, causers are claimed not to be included in this range (see
also Marantz 1997; and Schäfer 2012; Alexiadou et al. 2015 against subsuming
both agent and causer roles under the Voice layer). Instead, causers are only licit in
constructions which contain Voice – be they verbal or nominal:
(21) (The/Belushi’s) mixing of drugs and alcohol (Harley 2009: ex. 4b)
Alexiadou et al. (2013a, b) reject both proposals described above, namely that
either [a] event structure/size of nominalized structure, or [b] the presence of Voice,
330 O. Ahdout
are the determining factors in the distribution of causers across the different nominal
forms in English, adding data samples from German, Romance and Greek. An
example is the nominalized infinitive in German, which only accepts agents, while
the “smaller” -ung nominalization shows no restriction whatsoever (in Spanish as
well there is a negative correlation between size of nominal and the array of external
arguments permitted, see Alexiadou et al. 2013a). As some of the nominal forms do
include Voice, the association between the presence of Voice and the acceptability
of indirect participants is also weakened.
As I show in the next sections, the Hebrew data corroborate the view which
dissociates the size of the nominalization from the congruence with non-agentive
casuers in general, and, in particular, the relevance of the Voice layer. Focusing on
causative verbs and their alternants (when available), it will be shown (Sect. 10.4.1)
that causers are usually degraded with nominalizations, regardless of the degree
of ‘verbiness’. More specifically, the presence of Voice does not seem to improve
the nominal clause with a causer, to be discussed in Sect. 10.4.2. To complete
the picture, it will be shown that in principle, no semantic incongruence exists
between the causer role and nominalized verbs, based on Hebrew anticausative
verbs and corresponding nominals (Sect. 10.4.3). These represent environments
lacking Voice/external arguments, and yet they do allow causer participants.
10.4 Causers in Hebrew Nominal Clauses
10.4.1 DP-causers
As in English, DP-causers in Hebrew appear with transitive verbs. As briefly

mentioned in Sect. 10.2, transitive causative verbs are always marked with active-
morphology:
(22) ha-∫arav ha-kitsoni hi∫mid et ha-jevulim
the-heat.wave the-extreme destroyed.CAUS.ACT ACC the-crops
‘The extreme heat wave destroyed the crops’.
To show that “Agent Exclusivity” is present in Hebrew as well, Sichel (2010)

provides examples of causative verbs (based on unergatives) and object-experiencer
verbs. An example of the former class is below (Sichel 2010, exx. 16–17):
(23) a. ha-magad hikpits et ha-xajal verb + agent
the-commander jumped.CAUS.ACT ACC the-soldier
me-ha-mita
from-the-bed
‘The commander made the soldier jump out of bed’.
˘
b. ha-de aga hikpits-a oto me-ha-mita verb + causer

the-worry jumped.CAUS.ACT-F.3SG him from-the-bed
‘His worries caused him to jump out of bed’.
(24) a. hakpatsat ha-xajal me-ha-mita GEN-OBJ + agent

the.jumping.CAUS.ACT (of) the-soldier.GEN from-the-bed
(al-jedej ha-magad)
by the-commander
‘His being made to jump from bed (by the commander)’
b. *hakpatsat-o me-ha-mita al-jedej ha-de aga GEN-OBJ + causer
˘
the.jumping.CAUS.ACT-his from-the-bed by the worry
This paper extends the pool of Hebrew verbs included in Sichel’s (2010) study
by examining predicates from the three active-marked templates (3 verbs per
template), with DP-agents/causers. Moreover, the acceptability of causers is also
checked for both structures described in Sect. 10.2, the GEN- and ACC-OBJ. The
anticausative verbs examined included PP-causers (Sect. 10.4.3). Results reported
here were collected based on a questionnaire filled out by 18 native speakers. The
questionnaire included 9 active verbs and 5 anticausative verbs7 .
I first note that judgments on the acceptability of causers in clauses with
nominalizations are complicated due to several reasons. The first is general; the
use of nominalizations is much more frequent in formal or written language than in
spoken language. Secondly, some speakers show a preference for the Construct State
(CS) (see Sect. 10.2) for the realization of the possessive/genitive relation between
the nominal and post-nominal genitive DP (subject in ACC-OBJ or object in GEN-
ACC, see (6)), while some do not. Still others (myself included) prefer the CS in
some clauses but not others.
Another trend that emerges from the questionnaire is a general dis-preference
reported by several speakers for the ACC-OBJ structure, which is possibly related
to a tendency described in the literature to interpret the post-nominal NP in the CS as
the direct object (25a), rather than the subject of the predication, as in (25b) (Rosen
1956). The fact that the post-nominal DP in the ACC-OBJ structure is a subject,
and not an object, interferes with this preference: the subject DP intervenes between
the nominalization and the accusative marked DP, the object, creating a “garden
path”-like effect (25b):8
(25) a. harisat ha-tsava […]
the-destruction.SMPL.ACT (of) the-army.GEN
Preferred reading: army is destroyed (object).
˘
b. harisat ha-tsava et hair

the-destruction.SMPL.ACT (of) the-army.GEN ACC the-city
[nominal – subject DP – ACC object DP]
7 Some examples with PP-causers were not included in the questionnaire, and corresponding
judgments are my own.
8 Borer (2013: 100–104) observes a dis-preference, not restricted to nominal clauses, for bare, light
direct objects to appear in a position non-adjacent to the predicate. It could be the case that the
distance between the predicate and object in ACC-OBJ constructions, contributes to the preference
of the GEN-OBJ realization even in cases when the object DP is not bare/light.
332 O. Ahdout
To try and minimize this effect, intervening DPs were kept as short as possible. In
light of this complication, in order nonetheless to determine the status of causers in
this construction, it was necessary to focus on contrasts between causers and agents,
and not on general acceptability.
The results of the survey regarding transitive causative verbs, described in detail
immediately below, replicate findings from other languages, and show that (even
direct participant) causers are less acceptable than agents in clauses headed by
nominalizations. Crucially, this is the case for both structural variants, with and
without Voice.
First, from all clauses examined, the highest rated ones are GEN-OBJ nom-
inalizations with agent arguments, with very little inter-participant variation in
grading:9
(26) a. ha-xaklaʾim hi∫mid-u et ha-jevulim ∫elahem
the-farmers destroyed.CAUS.ACT-3PL ACC the-crops their
kedey le-himana mi-hefsedim
in.order to-avoid from-losses
‘The farmers destroyed their crops in order to avoid losses’.
b. ha∫madat ha-jevulim al-jedej ha-xakla’im
the.destruction.CAUS.ACT (of) the-crops.GEN by the-farmers
‘The destruction of the crops by the farmers’
(27) a. bet ha-mi∫pat hixpil et ha-ʿone∫ ha-mekori

the.house the-law doubled.CAUS.ACT ACC the-punishment the-original
∫e-nigzar al ha-ne’e∫am
that-was.sentenced on the-defendant
‘The court doubled the original punishment the defendant was sentenced for’.
b. haxpalat ha-ʿone∫ al-jedej bet
the.doubling.CAUS.ACT (of) the-punishment.GEN by the.house
ha-mi∫pat
the-law
‘The doubling of the punishment by the court’
(28) a. ha-tabax himis et ha-xemʾa

the-cook melted.CAUS.ACT ACC the-butter
‘The cook melted the butter’.
b. hamasat ha-xemʾa al-jedej ha-tabax
the.melting.CAUS.ACT (of) the-butter.GEN by the-cook
‘The melting of the butter by the cook’
9A third verb in the active ‘intensive’ template (pizzer ‘scatter, disperse, distribute’, pizzur
‘scattering, dissolving, distributing’) scored low in all environments, and is not included here.
(29) a. ha-mi∫tara xasma et ha-kvi∫

the-police blocked.SMPL.ACT.F.3SG ACC the-road
‘The police blocked the road’.
b. xasimat ha-kvi∫ al-jedej ha-mi∫tara
the.blocking.SMPL.ACT (of) the-road.GEN by the-police
‘The blocking of the road by the police’
(30) a. ha-ʿiriya harsa et ha-mivne
the-municipality destroyed.SMPL.ACT.F.3SG ACC the-building
‘The municipality destroyed the building’.
b. harisat ha-mivne al-jedej ha- ʿirija
the.destruction.SMPL.ACT (of) the-building.GEN by the-municipality
(31) a. ha-tsoref jatsar et ha-tax∫it

the-jeweler created.SMPL.ACT ACC the-jewel
‘The jeweler created the jewel’.
b. jetsirat ha-tax∫it al-jedej ha-tsoref
the.creation.SMPL.ACT (of) the-jewel.GEN by the-jeweler
‘The creation of the jewel by the jewler’
(32) a. ha-bos tsimtsem et ha-nesiʿot le-xul

the-boss reduced.INTNS.ACT ACC the-trips to-abroad
‘The boss reduced the (amount of) trips abroad’.
b. tsimtsum ha-nesiʿot le-xul al-jedej ha-bos
shrinking.INTNS.ACT (of) the-trips.GEN to-abroad by the-boss
‘The reduction of the (amount of) trips abroad by the boss’
(33) a. an∫e miktsoʿa ∫ikmu et ha-ʾasirim

people profession rehabilitated.INTNS.ACT.3.PL ACC the-prisioners
‘Professionals rehabilitated the prisioners’.
b. ∫ikum ha-ʾasirim al-jedej an∫e miktsoʿa
the.rehabilitation.INTNS.ACT (of) the-prisoners.GEN by people profession
‘The rehabilitation of the prisoners by professionals’
Corresponding GEN-OBJ clauses with causers are marked low, and responses
show more variation. Note that causer participants in the following examples are
direct participants ((37) being an exception), and nonetheless their status in the
clause, compared to that of agents, is degraded:
(34) a. ha-∫arav ha-kitsoni hi ∫mid et ha-jevulim
the-heat.wave the-extreme destroyed.CAUS.ACT ACC the-crops
‘The extreme heat wave destroyed the crops’.
b. ?ha∫madat ha-jevulim al jedej ha- ∫arav
the.destruction.CAUS.ACT (of) the-crops.GEN by the-heat
334 O. Ahdout
˘
(35) a. ha- inflatsja hixpil-a et mexirej ha-jerakot
the-inflation doubled.CAUS.ACT-F.3SG ACC the.prices (of) the-vegetables.GEN
‘Inflation caused the prices of vegetables to double’.
b. #haxpalat mexirej ha-jerakot
the.doubling.CAUS.ACT (of) the.prices.GEN (of) the-vegetables.GEN
˘
al-jedej ha- inflatsja
by the-inflation
(36) a. ha-xom ha-gavoa b-a-xeder himis et ha-xemʾa
the-heat the-high in-the-room melted.CAUS.ACT ACC the-butter
‘The extreme heat in the room melted the butter’.
b. #hamasat ha-xemʾa al-jedej ha-xom
the.melting.CAUS.ACT (of) the-butter.GEN by the-heat
(37) a. ha-∫eme∫ ha-xazak-a harsa et

the-sun the-strong-F.SG destroyed.SMPL.ACT.F.3SG ACC
ha-rehitim b-a-mirpeset
the-furniture in-the-balcony
‘The strong sun ruined the furniture in the balcony’.
b. #harisat ha-rehitim al-jedej ha-∫eme∫
the.destruction.SMPL.ACT (of) the-furniture.GEN by the-sun
(38) a. ha-slaʿim xasmu et ha-∫vil

the-rocks blocked.SMPL.ACT.3PL ACC the-path
‘The rocks blocked the path’.
b. ?xasimat ha-∫vil al-jedej ha-slaʿim
the.blocking.SMPL.ACT (of) the-path.GEN by the-rocks
(39) a. ha-jove∫ jatsar sdakim b-a-delet

the-dryness created.SMPL.CAUS cracks in-the-door
‘The dryness created cracks in the door’.
b. ?jetsirat ha-sdakim b-a-delet al-jedej ha-jove∫
the.creation.SMPL.ACT (of) the-cracks.GEN in-the-door by the-dryness
(40) a. xomer ha-bidud tsimtsem et ha-ra'aS

the.material the-insulation reduced.INTNS.ACT ACC the-noise
‘The insulation material reduced the noise’.
b. #tsimtsum ha-ra'aS al-jedej xomer ha-bidud10
shrinking.INTNS.ACT (of) the-noise.GEN by the.material the-insulation
10 Thisexample is acceptable under a reading where the material is an instrument, i.e. used by an
(implicit) agent to reduce noise (in accordance with observations in Sichel 2010).
(41) a. ha-g∫amim ∫ikmu et ha-jeʿarot

the-rains rehabilitated.INTNS.ACT.3PL ACC the-forests
‘The rains brought about the rehabilitation of the forests’.
b. ∫ikum ha-jeʿarot al-jedej ha-g∫amim
rehabilitation.INTNS.ACT (of) the-forests.GEN by the-rains
‘The rehabilitation of the forests from the rains’
With ACC-OBJ nominals, ratings were generally lower. For some verbs, the
clause with an agent was overall rated higher than the clause with a causer, (43)–
(46). For the rest, both versions were judged as degraded, (42), (47)–(49).11
˘
(42) ha∫madat ?ha-xakla m/?ha-∫arav et ha-jevulim
the.destruction.CAUS.ACT (of) the-farmers.GEN/the-heat.GEN ACC the-crops
(43) a. haxpalat bet ha-mi∫pat et ha- one∫
˘
the.doubling.CAUS.ACT (of) the.house.GEN the-law ACC the-punishment
‘The court’s doubling of the punishment’
˘
b. ?haxpalat ha- inflatsja et mexirej

the.doubling.CAUS.ACT (of) the-inflation.GEN ACC the.prices (of)
ha-jerakot
the-vegetables.GEN
(44) hamasat ha-tabax/#ha-xom et ha-xemʾa

the.melting.CAUS.ACT (of) the-cook.GEN/the-heat.GEN ACC the-butter
‘The cook’s melting of the butter’
(45) a. xasimat ha-mi∫tara et ha-kvi∫

the.blocking.SMPL.ACT (of) the-police.GEN ACC the-road
‘The police’s blocking of the road’
b. ?xasimat ha-slaʿim et ha-∫vil
the.blocking.SMPL.ACT (of) the-rocks.GEN ACC the-path
(46) a. harisat ha-ʿirija et ha-mivne
the.destruction.SMPL.ACT (of) the-municipality.GEN ACC the-building
b. ?harisat ha-∫eme∫ et ha-rehitim
the.destruction.SMPL.ACT (of) the-sun.GEN ACC the-furniture
11 Inaccordance with an observation by Borer (2013: 550), verbs of the ‘intensive’ class (48)–
(49) were overall dis-preferred in the ACC-OBJ structure. Note also regarding this class that it
has been suggested (Doron 2003) that verbs belonging to it impose a thematic restriction on their
external argument: it must be a direct participant, cf. Sichel (2010), Alexiadou et al. (2013a). The
independent existence of such restriction should have deemed verbs of this class irrelevant to the
“Agent exclusivity” effect, under Sichel’s characterization, as they take a direct participant to begin
with. The data here, however, do not provide evidence for this, as verbs in this class also seem to
be prone to “Agent Exclusivity”.
336 O. Ahdout
(47) a. ?jetsirat ha-tsoref et ha-tax∫it

the.creation.SMPL.ACT (of) the-jeweler.GEN ACC the-jeweler
‘The creation of the jewel by the jeweler’
b. #jetsirat ha-jove∫ et ha-sdakim b-a-delet
the.creation.SMPL.ACT (of) the-dryness.GEN ACC the-cracks in-the-door
(48) a. #tsimtsum ha-bos et ha-nesiʿot le-xul

shrinking.INTNS.ACT (of) the-boss.GEN ACC the-trips to-abroad
b. #tsimtsum xomer ha-bidud et ha-raʿa∫
shrinking.INTNS.ACT (of) the.material.GEN the-insulation ACC the-noise
(49) a. ?∫ikum an∫e ha-miktsoʿa

the.rehabilitation.INTNS.ACT (of) the.people.GEN the-professionals.GEN
et ha-‘asirim
ACC the-prisoners
b. #∫ikum ha-g∫amim et ha-jeʿarot
the.rehabilitation.INTNS.ACT (of) the-rains.GEN ACC the-forests
10.4.2 Discussion: DP-causers in Hebrew
To account for the degraded status of causers in Hebrew nominalizations, one could
draw on the many works which have paralleled nominalization and passivization
(Grimshaw 1990; Alexiadou 2001, 2017; Borer 2013; Bruening 2013), as described
in Sect. 10.2. In the nominal as in the passive, the external argument is optional and
implicit, and when expressed overtly, it surfaces with a by-phrase. This is the case
in English and Hebrew, and in many other languages discussed in the literature (see
Alexiadou et al. 2013a, b).
Focusing on Hebrew, it has been shown (Doron 2003) that the passive in the
language allows only a subset of the range of thematic roles found with external
arguments, namely agents and instruments (cf. Icelandic, Jónsson 2003, 2009).
Arguably, if indeed nominalization is comparable to passivization, the same type
of thematic restrictions imposed on the passive head would be expected to appear in
nominalized form as well (for a similar effect in nominalization, see Alexiadou et
al. 2013b on Romanian).
Indeed, this might be the case for the GEN-OBJ nominals which pattern with
English deverbal nouns and do not require the obligatory realization of the subject
argument, nor mark their objects with accusative case, see (6b)/(7). ACC-OBJ
clauses, on the other hand, show exactly these properties, motivating the proposal
that they contain (non-passive) Voice (Sect. 10.2). The data on ACC-OBJ nominals
suggests that, despite the presence of Voice, the nominal is still defective compared
to the corresponding verb regarding causer participants. A view associating Voice
projections and availability of the full array of thematic roles in the nominal is thus
inadequate (see Sichel 2010 and Alexiadou et al. 2013b for a similar conclusion).
A different path of accounting for these data is to adopt the selectional restric-
tion view of nominalizing heads, according to which deverbal nouns in English may
only denote simple, single events (Sect. 10.3). Alexiadou et al. (2013a, b) phrase
this selectional restriction in event structure terms of ‘process’ and ‘result state’.
Translating Sichel’s account into event structure terminology (Levin & Rapapport
Hovav 1999), deverbal nouns in English denote only the process part of the event,
and lack the result state component that is included in the base verb structure.
As discussed in Sect. 10.3, it is the result state portion of the event which is
the necessary ingredient for the licensing of causers in both verbal and nominal
domains. Consequently, in its absence, “Agent Exclusivity” arises.12 Can this be
the case in Hebrew?
As shown in Sect. 10.2, Hebrew deverbal nouns do not correspond directly to
any of the three major groups of nominalizations in English, -(at)ion-type, nominal
gerunds or verbal gerunds. To find out whether Hebrew deverbal nouns pattern with
-(at)ion nominals, by virtue of imposing similar selectional restrictions, one should
first check the aspectual properties of these nominals. If the nominalized form is
imperfective/atelic, i.e. causes aspectual shift, then an event structure-type account
may be adopted for Hebrew as well (for aspectual shift in nominalization, see
Iordăchioaia & Soare 2008 on the Romanian Supine, Borer 2005, 2013 for nominal
-ing).
An aspectual shift, however, does not seem to take place in Hebrew (contra
Engelhardt 1998, 2000, who proposes that Hebrew nominalizations are inherently
imperfective). For example, if this were the case, punctual predicates, which denote
a binary change, should be incongruent in imperfective contexts with the original
punctual denotation. However, achievements produce comparable nominalizations
in the language:13
(50) a. hu higiʿa l-a-mesiba be-fitʾomiyut.
he arrived.CAUS.ACT to-the-party in-suddenness
‘He suddenly arrived at the party’.
b. hagaʿa pitʾomi-t ∫elo l-a-mesiba tihiye bilti-svira
arrival.CAUS.ACT sudden-F.SG his to-the-party be.FUT.F.3SG NEG-probable
‘His sudden arrival at the event is improbable’
12 These authors, however, note that in English, it is nominal gerunds, and not deverbal nouns,
which typically denote processes rather than complete, complex events. This view is based on
works that report an aspectual shift to an atelic event as a further effect of nominalization in nominal
gerunds, regardless of the aspectual class the corresponding verb belongs to (e.g. Borer 2005,
2013). If this is indeed the case, then gerunds – but not deverbal nouns – are the category which is
expected to show “Agent Exclusivity” effects (Alexiadou et al. 2013a,b).
13 The nominal in (50) is deliberately indefinite, as part of Engelhardt’s (1998, 2000) claim is that
definiteness is associated with perfectivity in nominals.

338 O. Ahdout
As Hebrew nominalizations do not show aspectual shift and may be either telic or
atelic14 – depending on the aspectual value of the base verb – an “Agent Exclusivity”
effect is not likely to stem from event structure differences between the verb and the
nominal.
This conclusion re-opens the question of the source of the restriction to agents
in nominalizations, as it appears that the nominalizations contain all necessary
ingredients for the licensing of causers. If one adopts the view that the source of
causers is in the event structure/result state part of the event (Folli & Harley 2005;
Schäfer 2012; Alexiadou et al. 2015), one would not expect “Agent Exclusivity” to
arise in Hebrew, since nominalization in the language does not affect event structure
or aspectual values of base verbs. Nonetheless, causers aren’t licensed.
In the next section, it will be shown that, despite the incompatibility of causers
in nominalizations derived from transitive causative verbs, anticausative verbs in
Hebrew do surface with non-agentive causers. As such, the ban on this thematic
role in nominal clauses is not categorical. The data, presented in the following
subsection, add to the conclusion reached in this subsection, that Voice is not a
sufficient condition for the licensing of causers in nominal clauses.
10.4.3 PP-causers
Anticausative verbs in several languages license prepositional phrases which specify

a causer-participant or a causing event, as shown in Sect. 10.3. In many of
these languages (e.g. Greek, Romanian, Romance), anticausative verbs are marked
morphologically, creating an overt differentiation between transitive and intransitive
causative verbs.
In Hebrew as well, by virtue of overtly marking transitivity distinctions in both
verbal and nominal domains (Sect. 10.2), transitivity, Voice-marking and the type of
preposition which surfaces in the verb and nominal align. PP-causers are restricted
to unaccusative structures, which in turn (usually) correlates with morphological
marking as middle (51c), while DP-causers are restricted to transitive structures,
where the verb is marked with active morphology (51a)/(51b) (for further examples
see Alexiadou & Doron 2012). By-phrases are only allowed with passive verbs (51d)
(for passives marked with middle morphology, see Ahdout & Kastner to appear).
14 Inaccordance, nominalizations of accomplishment verbs in Hebrew are perfectly grammatical

with ‘in X time’ adverbials, which diagnose telicity. For relevant examples, see e.g. Borer (2013:
97).
(51) a. ha-xajalim niku et ha-tender. transitive verb + agent

the-soldiers cleaned.INTNS.ACT.3PL ACC the-truck
‘The soldiers cleaned the truck’.
b. ha-ge∫em nika et ha-tender. transitive verb + causer

the-rain cleaned.INTNS.ACT ACC the-truck
‘The rain cleaned the truck.
c. ha-tender hitnaka *al-jedej/me-ha-ge∫em/ anticaus. verb + causer
the-truck cleaned.INTNS.MID by/from-the-rain/
*al-jedej/*me-ha-xajalim.
by/from-the-soldiers
‘The truck got cleaned by the rain’.
d. ha-tender nuka *me-/al-jedej ha-xajalim/ passive verb
the-truck was.cleaned.INTNS.PASS from/by the-soldiers/
*al-jedej ha-ge∫em.
by the-rain
‘The truck was cleaned by the soldiers’.
Nominalizations derived from anticausative verbs in Voice-marking languages

are not marked as such, and the morphological/syntactic causative-anticausative
alternation is neutralized in the nominal domain. What Hebrew allows us is
to check the behaviour of anticausative nominals in a system free of syntactic
ambiguities. Hebrew nominals retain the morphological marking of corresponding
verbs, therefore at least some properties the template is endowed with transfer to
the nominal.15 In this case, it is Voice-marking, compare (51a) and (52) for active-
Voice/transitive syntax, with (51c) and (53) for middle-Voice/unaccusative syntax:
(52) a. nikuj ha-tender al-jedej ha-xajalim GEN-OBJ
the.cleaning.INTNS.ACT (of) the-truck.GEN by the-soldiers
b. nikuj ha-xajalim et ha-tender ACC-OBJ
the.cleaning.INTNS.ACT (of) the-soldiers.GEN ACC the-truck
(53) hitnakut ha-tender me-ha-ge∫em anticausative

the.getting.clean.INTNS.MID (of) the-truck.GEN from-the-rain
As briefly mentioned in Sect. 10.3, it might be the case DP- and PP-causers
represent different types of causation which do not necessarily completely overlap.
The extent to which DP- and PP-causers are distinct is a manner which is still
unclear, and possibly subject to cross-linguistic variation. In Hebrew, the relevant
preposition me- ‘from’, is ambiguous between a location/spatial and a source-
like interpretation, which is also found with unergative verbs, e.g. English the boy
jumped for joy (see Alexiadou et al. 2015: 37–42).
15 See examples (61)–(63) for a subgroup of active-marked verbs in the hiXYiZ template which
form a causative alternation within the template. For these verbs, ambiguous between unaccusative
and transitive syntax, both DP- and PP-causers are allowed. The latter – despite the morphological
marking of these verbs as active.
340 O. Ahdout
In the realm of anticausative verbs, me- seems to encode a direct causation

relation when the DP denotes a participant (or event coercion, see Maienborn &
Herdtfelder 2017).16 This preposition introduces participants who are directly, and
oftentimes physically, involved in bringing about the event:17
(54) a. ha-ʾor hitrakex/hitnapeʾax/hitjabe∫ me-ha-krem. anticausative verb
the-skin softened/swelled.up/dried.INTNS.MID from-the-cream
‘The skin softened/swelled up/dried from the cream’.
b. ha-xalonot hitnaptsu me-ha-hedef.
the-windows shattered.INTNS.MID.3PL from-the-hedef
‘The windows shattered from the blast’.
c. ha-sdakim b-a-delet notsru me-ha-jove∫.
the-cracks in-the-door got-created.SMPL.MID.3PL from-the-dryness
‘The cracks in the door were created due to dryness’.
In other cases, only a DP-causer is felicitous, e.g. where both participants are
metonymically related (Edit Doron, p.c.):
(55) a. ha-gaʾava hix∫il-a oto. active/transitive
the-pride failed.CAUS.ACT-F.3SG him
‘His pride caused him to fail’.
b. #hu nix∫al me-ha-gaʾava. middle/anticausative

he failed.SMPL.MID from-the-pride
(Intended meaning: ‘He failed because of because of his pride’).
This preposition can also introduce events (via nominal clauses), and in this case
indirect causation is allowed. For example, (56) below may refer to a misuse of
the contact lenses, wherein they were not removed during showering, and a bacteria
(=the direct cause/participant) present in the water penetrated the eye and ultimately
caused it to get infected:
(56) ha- aj in
˘
hizdahama mi-∫imu∫ lo naxon

˘
the-eye got.infected.INTNS.MID.F.3SG from-use NEG correct

˘
˘
be- ad ∫ot maga /mi-bakterja.

in-lenses contact/from-becteria
‘The eye got infected due to incorrect use of contact lenses/from bacteria’.
16 To the extent that change of state psychological predicates also involve a similar type of direct
causation (Pesetsky 1995), the prominence of me- as the marker of the non-experiencer argument
in these predicates is also predicted:
(i) hu hit acben/hitrage∫/hitlahev
˘
me-ha-katava.
he became.annoyed/excited/thrilled.INTNS.MID from-the-article
17 The causative preposition biglal ‘because (of)’, which isn’t restricted to clauses with anti-
causative verbs, allows a wider range of causal relations between participant and event, including
indirect causation, cf. Maienborn & Herdtfelder (2017). See Sichel (2010: 168–171) for the
interpretation of nominal clauses with biglal.
Returning to the matter of the licensing of causers in nominal clauses, the result
obtained from anticausative verbs and PP-causers differs from the one described
in the previous sections, for transitive verbs and DP-causers. Unlike transitive verbs,
anticausative verbs produce nominalizations which preserve (whichever type of) PP-
causer that is available in the verbal clause. As such, no “Agent Exclusivity” is found
in this class.
Moreover, nominals derived from anticausative verbs with PP-causers were rated
as high as the highest-rated ‘transitive’ nominals, i.e. GEN-OBJ nominals with
agents ((26)–(33), b examples). Below, examples (57) and (58) show the nominals
corresponding to the verbal forms in (54) and (56), in accordance. Notice that in
example (58) both indirect causation as well as direct causation/participation are
exemplified:
(57) a. hitrakxut/hitnapxut/hitjab∫ut ha-ʾor me-ha-krem

the.getting.soft/swollen/dry.INTNS.MID (of) the-skin.GEN from-the-cream
b. hitnaptsut ha-xalonot me-ha-hedef
the.shattering.INTNS.MID (of) the-windows.GEN from-the-blast
‘The shattering of the windows because of the blast’
d. hitlakxut ha-ʾe∫ mi-∫iluv ∫el jove∫
the.ignition.INTNS.MID (of) the-fire.GEN from-combination of dryness
ve-ruxot
and-winds
‘The fire igniting due to a combination of dryness and winds’
(58) hizdahamut ha-ʿajin mi-∫imu∫ lo

the.getting.infected.INTNS.MID (of) the-eye.GEN from-use NEG
naxon be-ʿad∫ot magaʿ/me-ha-bakterja
correct in-lenses contact/from-the-bacteria
‘The eye getting infected due to incorrect use of contact lenses/the bacteria’
The examples below present nominalizations of anticausative alternants of

active/transitive forms in (47)–(48), which were judged as infelicitous with a causer
argument. Again, the nominalization of the anticausative is perfectly grammatical
with causers:
(59) hitstamtsemut jam ha-melax mi-maxsor be-melaxim
the.shrinking.INTNS.MID (of) the.sea.GEN the-salt.GEN from-lack in-salts
‘The shrinking of the dead sea due to lack of salts’
(60) hivatsrut sdakim b-a-delet me-ha-jove∫

the.getting-created.SMPL.MID (of) the-cracks.GEN in-the-door from-the-dryness
‘The creation of cracks in the door due to dryness’
A final note on the licensing of prepositional causers: despite the data just
presented, it is not the case that being marked as middle as such is a necessary
condition in licensing prepositional causers. Hebrew has a class of active-marked
342 O. Ahdout
verbs which are ambiguous between active and inchoative interpretations, on a par
with the non-marked causative/anticausative alternation in English, e.g. with break
(See Borer 1991; Lev 2016; Kastner 2019).
(61) a. ha-ximikalim/ha-madʿanim hi∫xir-u et ha-mi∫tax. trans. V
the-chemicals/the-scientists blackened.CAUS.ACT-3PL ACC the-surface
‘The-chemicals/the scientists blackened the surface’.
b. ha-mi∫tax hi∫xir me-ha-ximikalim. anticausative V
the-surface blackened.CAUS.ACT from-the-chemicals
‘The surface blackened because of the chemicals’.
Nominals produced from these verbs often preserve the inchoative interpretation,
despite the overt marking as active:
(62) heʿic ‘accelerated (intrans./trans.)’, heʿaca ‘acceleration (intrans.)’;
hirpa ‘to relax (intrans./trans.)’, harpaja ‘relaxing (intrans./trans.)’; hiktsin ‘get more
extreme/make more extreme’, haktsana ‘getting/making more extreme’;
hexmir ‘worsened (intrans./trans.)’, haxmara ‘worsening (intrans./trans.)’ […].
PP-causers are grammatical in the anticausative/inchoative context, in the same

manner as are middle-marked anticausatives discussed above:
(63) ha∫xarat ha-mi∫tax me- nominal + PP-causer/agent
the.blackening.CAUS.ACT (of) the-surface.GEN from
ha-ximikalim/al-jedej ha-mad’anim
the-chemicals/by the-scientists
‘The surface turning black because of the chemicals/by the scientists’
Previous literature on psychological predicates reports cases where PP-causers

appear with nominalizations. In Greek and Romanian psychological predicates
(Alexiadou & Iordăchioaia 2014), it has been claimed that the surfacing of prepo-
sitional causers implies an underlying unaccusative (Voice-less) syntax, compare
anticausative verb in (64a) and nominal with PP-causer in (65a). In contrast, under
the agentive construal, i.e. with an agentive by-phrase, Voice is included, compare
transitive/agentive verb in (64b) and nominal with by-phrase in (65b).
(64) a. i maria enohlithike me ta nea. anticaus./non-active verb
the Maria annoyed.NACT with the news
‘Maria got annoyed with the news’.
b. o janis enholise ti Maria. transitive/active verb
the John annoyed the Maria
(65) a. i enholisi tis Marias me ta nea nominal + causer

the bothering the Maria.GEN with the news
b. i enholisi tis Marias apo to janis nominal + agent
the bothering the Maria.GEN by the John
‘Maria getting annoyed from the news/by John’
(Greek, Alexiadou & Iordăchioaia 2014: exx. 11, 14)
The structures corresponding to (65a) and (65b) are as follows, in accordance:

(66) a. PP-causer (base verb is anticausative) b. Agent (base verb is transitive)
(Alexiadou & Iordăchioaia 2014, exx. 29, 30)
Importantly, the difference between the two proposed structures in (66) cannot
be reflected in morphology, as Greek (and Romanian) nominals do not retain the
verbal active-middle distinctions typical of the verbal alternation (64). As shown
above, in Hebrew this ambiguity is resolved by both prepositions and marking of
the alternation directly on the nominal (see Ahdout 2017 on this contrast in the
domain of psychological predicates):
(67) a. ha-xalonot hitnaptsu me-ha-hedef. middle V + causer
the-windows shattered.INTNS.MID.3PL from-the-hedef
‘The windows shattered from the blast’.
b. ha-mafginim niptsu et ha-xalonot. active V + agent
the-demonstrators shattered.INTNS.ACT.3PL ACC the-windows
‘The demonstrators shattered the windows’.
(68) a. hitnaptsut ha-xalonot me-ha-hedef middle N + cause

‘The windows shattered from the blast’
b. niputs ha-xalonot al-jedej active N + agent
the.shattering.INTNS.ACT (of) the-windows.GEN by
ha-mafginim
the-demonstrators
‘The shattering of the windows by the demonstrators’
To conclude the findings from previous sections, data from Hebrew show that [1]
DP-causers are usually infelicitous in nominals, but [2] PP-causers are available in
the nominal. A similar situation is found in German -ung nominals (Alexiadou 2001;
Alexiadou et al. 2013b), where causers may only surface as prepositional phrases
(69) (with the preposition durch ‘through’). In contrast, the “high” pre-nominal
possessor position (parallel to English example (1b)) is restricted to common names
(70).
344 O. Ahdout
(69) Die Bestätigung der ursprünglichen Diagnose durch die

the.GEN confirmation the.GEN initial diagnosis through the
Ergebnisse des Tests
results the.GEN test.GEN
‘The confirmation of the initial diagnosis by the results of the test’
(Alexiadou et al. 2013b, ex. 21)
(70) Attilas Zerstörung der Stadt
Attila-GEN destruction the city-GEN
‘Attila’s destruction of the city’ (Alexiadou 2001: 80, ex. 11)
Finally, an important implication of the findings presented here, namely the

absence of DP-causers in nominals, in contrast to the felicity of PP-causers,
motivates a characterization of the “Agent Exclusivity” effect as syntactic in nature.
In the nominal clauses corresponding to the verbal clause with a non-agentive causer
argument in (71), this argument is available in a PP instantiation alone (and with the
nominal derived from the anticausative verb) (73). The (identical) DP-causer (72a),
or a causer introduced with a by-phrase (72b) are both ruled out. Thus, it cannot
be the case that the ruling-out of the DP-causer stems from semantic grounds (cf.
agentive clause in (68b)).
(71) ha-hedef nipets et ha-xalonot. active verb + causer
the-blast shattered.INTNS.ACT ACC the-windows
‘The blast shattered the windows’.
(72) a. #niputs ha-hedef et ha-xalonot ACC-OBJ

the.shattering.INTNS.ACT (of) the-blast.GEN ACC the-windows
b. #niputs ha-xalonot al-jedejha-hedef GEN-OBJ
the.shattering.INTNS.ACT (of) the-windows.GEN by the-blast
(73) hitnaptsut ha-xalonot me-ha-hedef middle nominal

‘The windows shattered from the blast’
10.5 Conclusions and Open Questions
This article surveys the landscape of nominalizations of causative verbs in Hebrew,

focusing on the contrasts between agents and causers in nominal clauses. It was
shown that Hebrew exhibits the same bias against DP-causers as does English.
That causers are concomitant with the presence of the structural layer introducing
external arguments was put into question following an examination of the structural
variant named here the ACC-OBJ structure. This structure allows us to check
the acceptability of causer participants in a nominalized structure which arguably
contains Voice, and more generally, resembles the verbal structure by virtue of
licensing accusative case and the obligatory realization of the external argument.
The seeming lack of improvement in the acceptability of non-agentive causers in
this class of nominalizations, leads to the weakening of the view that relates these
verbal properties and the appearance of causers. Abandoning this view naturally
leaves open the question of a possible source of this ban in a language like Hebrew.
The Hebrew data presented here also raises an issue for studies which oppose
the view that causation is encoded in the structure of events as an atomic unit,
and instead suggest that this meaning component is post-syntactically interpreted
from the combination of event and result state. Assuming that the ACC-OBJ
nominalization does preserve the relevant structural layers, which bring about this
interpretation, there is no immediate reason as to why causative subjects cannot be
licensed by these layers and hosted in Voice suggests itself.
To this set of evidence, Hebrew Voice-marking properties add the possibility
of checking nominals which are unequivocally derived from anticausative verbs,
i.e. lack active Voice/an external argument. PP-causers typical of this structure are
perfectly acceptable in corresponding nominalizations, either as a direct participant
or as a (possibly indirect-causation-denoting) event. Again, these data raise the
issue of the source of the restriction in the domain of DP-causers; the rejection
of causer participants in one syntactic environment but not the other, motivates a
characterization of this restriction in syntactic terms.
More generally, the bias towards agents in nominal environments hints at a
possible distinction between these two theta-roles at some representational level –
even in transitive causative verbs, where both are realized as subjects. It could be
the case that the agent vs. causer roles are encoded separately in the syntax. If
nonetheless a view that takes the nominal to be “smaller” or “defective” compared
to the verbal structure is to be maintained, one possible way to translate the findings
here (as well as former findings to this effect) to syntactic terms, would be to
hypothesize that the causer role is hierarchically higher than the agent. As such,
the relevant structural layer hosting the causer is left out of some nominal structures
in the same manner as Voice, i.e. the external argument, is truncated in many classes
of nominalizations cross-linguistically.
Acknowledgements A special thanks to Ivy Sichel for extensive discussions of topics relating
to this study. Any errors remain my own. This work was funded by AL 554/8-1, DFG Gottfried
Wilhelm Leibniz Preis 2014 awarded to Artemis Alexiadou.
References
Abney, S. P. (1987). The English noun phrase in its sentential aspect. PhD dissertation, Mas-
sachusetts Institute of Technology.
Ackema, P., & Neeleman, A. (2004). Beyond morphology. Interface conditions on word formation.
Oxford/New York: Oxford University Press.
Ahdout, O. (2017). Eventive Object Experiencer nominalizations in Hebrew. In M. Bloch-Trojnar
& A. Malicka-Kleparska (Eds.), Aspect and valency in nominals (pp. 31–52). Berlin/New York:
Mouton de Gruyter.
346 O. Ahdout
Ahdout, O. (In preparation). Nominalizations in Hebrew and beyond: Between syntax and
competition. PhD dissertation, Humboldt Universität zu Berlin.
Ahdout, O., & Kastner, I. (to appear). Bases, transformations and competition in Hebrew niXYaZ.
In Nominalizations: 50 years on from Chomsky’s remarks, eds. Oxford University Press.
Alexiadou, A. (2001). Functional structure in nominals: Nominalization and Ergativity. Amster-
dam: John Benjamins.
Alexiadou, A. (2005). Gerund types, the present participle and patterns of derivation. In C.
Maienborn & A. Wöllstein (Eds.), Event arguments: Foundations and applications (pp. 139–
Alexiadou, A. (2017). Ergativity in nominalization. In J. Coon, D. Massam, & L. d M. Travis
(Eds.), Oxford handbook on Ergativity (pp. 355–372). Oxford: Oxford University Press.
Alexiadou, A., & Doron, E. (2012). The syntactic construction of two non-active voices: Passive
and middle. Journal of Linguistics, 48, 1–34.
crosslinguistically. In M. Frascarelli (Ed.), Phases of interpretation (pp. 187–211). Berlin/New
York: Mouton de Gruyter.
alternations: A layering approach. Oxford: Oxford University Press.
Alexiadou, A., Iordăchioaia, G., & Schäfer, F. (2011). Scaling the variation in Romance and
Germanic nominalizations. In P. Sleeman & H. Perridon (Eds.), Variation and change in the
Romance and Germanic noun phrase (pp. 25–40). Amsterdam/Philadelphia: John Benjamins.
Alexiadou, A., Iordăchioaia, G., Cano, M., Martin, F., & Schäfer, F. (2013a). ‘Direct participation’
and ‘agent exclusivity’ effects in derived nominals and beyond. In G. Iordăchioaia, I. Roy,
& K. Takamine (Eds.), Categorization and category change in morphology (pp. 153–180).
Cambridge: Cambridge Scholars Publishing.
Alexiadou, A., Iordăchioaia, G., Cano, M., Martin, F., & Schäfer, F. (2013b). The realization of
external arguments in nominalizations. The Journal of Comparative Germanic Linguistics,
16(2–3), 73–95.
Alexiadou, A., & Iordăchioaia, G. (2014). Causative nominalizations: Implications for the structure
of psych verbs. In A. Bachrach, I. Roy, & L. Stockall (Eds.), Structuring the argument: Mul-
tidisciplinary research on verb argument structure (pp. 119–140). Amsterdam/Philadelphia:
John Benjamins.
Borer, H. (1991). The causative-inchoative alternation: A case study in Parallel Morphology. The
Linguistic Review, 8, 119–158.
Borer, H. (2005). Structuring sense: The normal course of events. Oxford: Oxford University Press.
Borer, H. (2013). Taking form: Structuring sense (Vol. III). Oxford: Oxford University Press.
Borsley, R., & Kornfilt, J. (2000). Mixed extended projections. In R. Borsley (Ed.), The nature and
function of syntactic categories (pp. 101–131). San Diego: Academic Press.
Bruening, B. (2013). By phrases in passives and nominals. Syntax, 16, 1–41.
Chomsky, N. (1970). Remarks on nominalization. In R. A. Jacobs & P. S. Rosenbaum (Eds.),
Readings in English transformational grammar (pp. 184–221). Boston: Ginn.
DeLancey, S. (1984). Notes on agentivity and causation. Studies in Language, 8(2), 181–213.
Doron, E. (2003). Agency and voice: The semantics of the Semitic templates. Natural Language
Semantics., 11(1), 1–67.
Engelhardt, M. (1998). The syntax of nominalized properties, PhD dissertation, The Hebrew
University of Jerusalem.
Engelhardt, M. (2000). The projection of argument-taking nominals. Natural Language &
Linguistic Theory, 18, 41–88.
Fassi-Fehri, A. (1993). Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer.
Fodor, J. A. (1970). Three reasons for not deriving “kill” from “cause to die”. Linguistic Inquiry,
1(4), 429–438.
Folli, R., & Harley, H. (2005). Flavours of v: Consuming results in Italian and English. In P.
Kempchinsky & R. Slabakova (Eds.), Aspectual Enquiries (pp. 95–120). Dordrecht: Springer
Science & Business Media.
Folli, R., & Harley, H. (2008). Teleology and animacy in external arguments. Lingua, 118, 190–
202.
Fu, J., Roeper, T., & Borer, H. (2001). The VP within process nominals: Evidence from adverbs
and the VP anaphor do-so. Natural Language & Linguistic Theory, 19(3), 549–582.
Grimshaw, J. (1990). Argument structure. Cambridge, MA: Massachusetts Institute of Technology
Press.
Harley, H. (2009). The morphology of nominalizations and the syntax of vP, in M. Rathert & A.
Giannakidou (Eds.), Quantification, Definiteness, and Nominalization (pp. 321–343). Oxford:
Harley, H., & Noyer, R. (2000). Formal versus encyclopedic properties of vocabulary: Evidence
from nominalizations. In B. Peters (Ed.), The lexicon-encyclopedia interface (pp. 349–374).
Amsterdam: Elsevier Press.
Hazout, I. (1995). Action nominalizations and the lexicalist hypothesis. Natural Language &
Linguistic Theory, 13(3), 355–404.
Hazout, I. (1991). Verbal nouns: Theta theoretic studies in Hebrew and Arabic. Ph.D. dissertation,
University of Massachusetts at Amherst.
Iordăchioaia, G., & Soare, E. (2008). Two kinds of event plurals: Evidence from Romanian
nominalizations. Empirical Issues in Syntax and Semantics, 7, 193–216.
Iwata, S. (1995). The distinctive character of psych-verbs as causatives. Linguistic Analysis, 25(1–
2), 95–120.
Jónsson, J. G. (2003). Not so quirky: On subject case in Icelandic. In E. Brandner & H. Zinsmeister
(Eds.), New perspectives on case theory (pp. 127–163). Stanford: CSLI Publications.
Jónsson, J. G. (2009). The new impersonal as a true passive. In A. Alexiadou, J. Hankamer, T.
McFadden, J. Nuger, & F. Schäfer (Eds.), Advances in comparative Germanic syntax (pp. 281–
306). Amsterdam: John Benjamins.
Kallulli, D. (2006). A unified analysis of passives, anticausatives and reflexives. Empirical issues
in formal syntax and semantics, 6, 201–225.
Kastner, I. (2016). Form and meaning in the Hebrew verb. PhD dissertation, New York University.
Kastner, I. (2018). Templatic morphology as an emergent property: Roots and functional heads in
Hebrew. Natural Language and Linguistic Theory.
Kastner, I. (2019). Inchoatives in causative clothing: Change of state in Modern Hebrew heXYiZ.
The Linguistic Review.
Kayne, R. (1984). Connectedness and binary branching. Dordrecht: Foris.
Kratzer, A. (1996). Severing the external argument from the verb. In J. Rooryck & L. Zaring (Eds.),
Phrase structure and the lexicon. Dordrecht: Kluwer.
Kratzer, A. (2005). Building resultatives. In C. Maienborn & A. Wöllstein-Leisten (Eds.), Event
arguments in syntax, semantics, and discourse (pp. 178–212). Tübingen: Niemeyer.
Lakoff, G. (1970). Irregularity in syntax. New York: Holt, Reinhart and Winston.
Lev, S. (2016). Hebrew labile alternation. Master’s thesis, Tel Aviv University.
Levin, B., & Rappaport Hovav, M. (1999). Two structures for compositionally derived events.
Proceedings of SALT, 9, 199–223.
Levin, B., & Rappaport Hovav, M. (2004). The semantic determinants of argument expression: A
view from the English resultative construction. In J. Guéron, J. Lecarme, & A. Lecarme (Eds.),
The syntax of time (pp. 477–494). MIT Press.
Levin, B., & Rappaport Hovav, M. (2005). Argument realization. Cambridge: Cambridge Univer-
sity Press.
Maienborn, C., & Herdtfelder, J. (2017). Eventive versus stative causation: the case of German
causal von-modifiers. Linguistics and Philosophy, 40(3), 279–320.
Marantz, A. (1997). No escape from syntax: Don’t try morphological analysis in the privacy of
your own lexicon. University of Pennsylvania Working Papers in Linguistics, 4(2), 201–225.
Marantz, A. (2000). Case and licensing. Arguments and case: Explaining Burzio’s generalization,
11–30.
348 O. Ahdout
Marelj, & T. Siloni (Eds.), The theta system: Argument structure at the interface (pp. 20–51).
Oxford: OUP.
Panagiotidis, P. (2015). Categorial features. Cambridge: Cambridge University Press.
Pesetsky, D. M. (1995). Zero syntax: Experiencers and cascades. Cambridge, MA: Massachusetts
Institute of Technology Press.
Piñón, C. (2001). A finer look at the causative-inchoative alternation. Semantics and Linguistic
Theory, 11, 346–364.
Pylkkänen, L. (2002). Introducing arguments. PhD dissertation, Massachusetts Institute of Tech-
nology.
Rappaport, M. (1983). On the nature of derived nominals.. Papers in lexical-functional grammar
(pp. 113–142).
Rappaport, M., & Doron, E. (1990). Semantic aspects of the deverbal noun [Hebetim semantiyim
šel šem hape’ula]. In Paper presented at the workshop on Hebrew grammar (4.2.90). The
Hebrew University of Jerusalem.
Ritter, E. (1991). Two functional categories in noun phrases: Evidence from Modern Hebrew.
Syntax and semantics, 25, 37–62.
Reuland, E. J. (1983). Governing-ing. Linguistic Inquiry, 14(1), 101–136.
Rosen, H. (1956). Ivrit Tova. Jerusalem: Kiryat Sefer Publishers.
Schäfer, F. (2012). Two types of external argument licensing–the case of causers. Studia Linguis-
tica, 66(2), 128–180.
Sichel, I. (2009). New evidence for the structural realization of the implicit external argument in
nominalizations. Linguistic Inquiry, 40(4), 712–723.
Sichel, I. (2010). Event structure constraints in nominalization. In A. Alexiadou & M. Rathert
(Eds.), The syntax of nominalizations across languages and frameworks (pp. 151–190).
Berlin/New York: Mouton de Gruyter.
Siloni, T. (1997). Noun phrases and nominalizations: The syntax of DPs. Dordrecht: Springer
Science & Business Media.
Solstad, T. (2009). On the implicitness of arguments in event passives. In A. Schardl, M. Walkow,
& M. Abdurrahman (Eds.), Proceedings of NELS 38 (pp. 365–374). Amherst, MA: GLSA
Publications.
Chapter 11
Causees are not Agents
Léa Nash
Abstract This article explores the structure of morphological causatives. Cross-

linguistically, the subject of the embedded intransitive verb surfaces as the second
argument akin to direct object in these constructions while the subject of transitive
verbs surfaces either as the third argument, comparable to an indirect object, or as
an oblique adjunct. I argue that in spite of various ways to encode the causee, which
semantically corresponds to the agent of embedded verb, it is not structured as the
embedded agent in morphological causatives. This claim entails that the structure
of agentive unergative and transitive predicates in root and causative contexts
is not the same. When they are embedded in/under morphological causatives,
these predicates are agentless, and the causee, if present, is not introduced by
agentive Voice. I further argue that the causee in causatives of unergatives and
in causatives of transitives is not structured in the same way. In the former, the
causee is the argument of the embedded VoiceP predicate, but structured as the
Holder of the process/state. In causatives of transitives, however, the causee is
not the argument of the embedded VoiceP predicate: it is either not present at
all structurally, or is introduced by an Applicative head above the domain of the
embedded predicate, and interpreted as an associate/accompanying argument, a
secondary agent, an animate instrumental of the morphological causative. While the
embedded unergative predicate conserves its argument in causative configurations,
the embedded transitive predicate is systematically deagentivized. The present
analysis is based on evidence from Georgian, where two morphological strategies
of causativisation are attested: causatives of unergatives, just as causatives of all
monovalent and stative predicates carry the affix a-, while causatives of agentive
transitives are formed by the circumfix a- . . . -in-. Morphological causativisation is
analysed in Georgian as adding of agentive/transitive Voice, spelled out as a-, to
any type of embedded predicate that does not contain or require an agent, while -in-
marks deagentivisation of the embedded transitive predicate.
L. Nash ()
Department of Language Sciences, Université Paris Lumières-Saint Denis/CNRS, Paris, France
e-mail: lea.nash@univ-paris8.fr

https://doi.org/10.1007/978-3-030-34308-8_11
350 L. Nash
Keywords Morphological causatives · Causee · Agent · Holder · Secondary

agent · Voice · Deagentivisation · Mediopassive · Applicative · Dative
argument · Unergatives · Achievements · Ingestion/perception verbs · Direct
causatives · Indirect causatives
11.1 The Issue
This article explores the structure of configurations where an agentive verbal

predicate is causativized by means of affix(es). Cross-linguistically, the subject
of the embedded intransitive verb surfaces as the second argument akin to direct
object in these constructions while the subject of transitive verbs surfaces either
as the third argument, comparable to an indirect object, or as an oblique adjunct.
I argue that in spite of various encoding options, the causee, which semantically
corresponds to the agent of embedded verb, is not structured as the embedded agent
in morphological causatives cross-linguistically, and within the same language. This
claim entails that the structure of agentive unergative and transitive predicates in
root and causative contexts is not the same. When they are embedded in/under
morphological causatives, these predicates are agentless, and the causee, if present,
is not introduced by agentive Voice. I further argue that the causee in causatives
of unergatives and in causatives of transitives is not structured in the same way.
In the former, the causee is the argument of the embedded VoiceP predicate, but
structured as the Holder of the process/state. In causatives of transitives, however,
the causee is not the argument of the embedded VoiceP predicate: it is either
not present at all structurally due to deagentivisation, or is introduced by an
Applicative head above the deagentivised domain of the embedded predicate. In the
latter scenario, the causee is interpreted as an associate/accompanying argument, a
secondary agent, an animate instrumental of the morphological causative. While the
embedded unergative predicate conserves its argument in causative configurations,
the embedded transitive predicate systematically lacks the agent. The present
analysis is based on evidence from Georgian, where two morphological strategies
of causativisation are attested: causatives of unergatives, just as causatives of all
monovalent and stative predicates carry the affix a-, while causatives of agentive
transitives are formed by the circumfix a- . . . -in-. Morphological causativisation is
analysed in Georgian as adding of agentive/transitive Voice, spelled out as a-, to
any type of embedded predicate that does not contain or require an agent, while -in-
marks deagentivisation of the embedded transitive predicate.
11.1.1 Introduction and Preview of Analysis
Cross-linguistically, causative predicates can be composed analytically (e.g.

English) and morphologically, via an affix, (e.g. Turkish). A considerable body
of research has addressed the question whether both types have the same underlying
structure. Since the early 70s, a line of thought that can be traced back to generative
11 Causees are not Agents 351
semanticists, and is determinant for Baker’s (1988) analysis of morphological

causatives in terms of incorporation, treats analytic and morphological causatives
alike. This parallelism particularly holds for those analytic causatives that involve
an embedded complement with reduced functional structure, lacking temporal
specification of its own and incompatible with negation. The only difference
between this type of analytic and morphological causatives resides in the shape
of the causativizer: an independent light verb in analytic causatives and an affix in
morphological causatives. Regardless of their shape, causativizers are categorized
as verbal categories with identical selectional properties in both types of causatives.
The central common property concerns restructuring, or clause union, of the main
causative verb and of the non-finite embedded verb into one predicate domain.
(Rizzi 1978; Rouveret & Vergnaud 1980; Burzio 1986; Manzini 1983; Wurmbrand
2001, a.o.). Clause-union accounts for case distribution on main arguments in a
causative configuration, and especially for the form of the subject of the embedded
verb, the causee. When the embedded verb is transitive, the causee may surface
under two guises: it can be marked as a second internal argument, or as an optional
adpositional adjunct (Chichewa)–(French).1
(1) a. Nungu i-na-phik-its-a kadzidzi maungu
9 porcupine 9 S-PS-cook-CAUS-FV 1a owl 6 pumpkins
‘The porcupine made the owl cook the pumpkins.’
b. Niungu i-na-phik-its-a maungu (kwa kadzidzi)
9 porcupine 9 S-PS-cook-CAUS-FV 6 pumpkins to 1a owl
‘The porcupine had the pumpkins cooked by the owl.’
[Chichewa, Alsina 1992:518]
(2) a. Zoé a fait écrire une lettre à Emma FAIRE-INFINITIVE (FI)

Zoé CAUS write a letter to Emma
‘Zoé made Em ma write a letter.’
b. Zoé a fait écrire une lettre (par Emma) FAIRE PAR (FP)
Zoé CAUS write a letter by Emma
‘Zoé had a letter written (by Emma).’ [French]
The variation between two realisations of the causee in (1)–(2), i.e. between
obligatoriness and direct case-marking versus optionality and oblique case-marking,
depends on the properties of the non-finite embedded complement restructured with
the causative verb, especially on its ability to contain the external argument of the
embedded transitive predicate. The causee either surfaces as the external argument
of the embedded predicate and gets assigned case by the complex causative
predicate, (1a)–(2a), or optionally surfaces as an adpositional phrase adjoined to
an agentless truncated verb constituent, (1b)–(2b). For the ease of exposition, I
1 Thefollowing abbreviations are used: 1,2,3=person markers; ACC=accusative; AOR=aorist;

CAUS=causative; DAT=dative; ERG=ergative; FM=fientive marker; FV=final vowel;
GEN=genitive; NACT=middle voice; NOM=nominative; prev=perfectivizing preverb;
pl=plural; PS/PAST=past; S=subject; sg=singular; TS=thematic suffix; VAM=voice-applicative
marker.
352 L. Nash
will refer to the configuration with obligatory causee as FI (faire-infinitive) and to

the configuration without one as FP (faire-par), following Romance tradition since
Kayne (1975).
Yet, there are languages with morphological causatives that do not show variation
as in (1–2): the causee in causatives of transitives (henceforth, COT) in Japanese
can only be marked as the dative argument, while its homologue in Malayalam can
only be shaped as an instrumental PP. The question is whether this lack of variation
in shaping the causee corresponds to limiting restructuring in each language to
one type of embedded complement: richer structure in Japanese that includes the
external argument of the embedded transitive verb (FI), and truncated agentless
constituent in Malayalam (FP). However, this lack of variation may turn out not
to be indicative of properties of the embedded clause. If the causee, regardless of its
shape, can be dropped or, inversely, must always surface, the correlation between the
shape of the causee in COTs and the presence of the embedded external argument
in morphological causatives ceases to hold.
In this respect, investigation of morphological causative constructions in Geor-
gian directly concerns the correlation between the shape of the causee and projection
of the embedded agent in COTs and can contribute to a better understanding of
structural mechanisms underlying causativisation. The causee in COTs can only
bear dative case, which might suggest that it occurs in FI-type configurations akin
to Japanese sase constructions. I show that in spite of its dative marking shared with
indirect objects, the causee in COTs behaves differently than the third argument in
ditransitive constructions. The indirect object is obligatory in all ditransitives but
the dative causee is generally optional. Concretely, given that Georgian is a pro-
drop language, omitted dative argument must be interpreted as discourse-specific in
a ditransitive construction, but in COTs can be also interpreted as existential/non-
specific.
(3) a. keti-m surat-i ga=u-gzavn-a

Keti-ERG picture-NOM prev=VAM-send-AOR.3sg
‘Keti sent him/her/them a picture.’
=‘Keti sent someone a picture.’ =‘Keti sent a picture away.’
b. keti-m surat-i da=a-xat’-in-a

Keti-ERG picture-NOM prev=CAUS-draw-CAUS-AOR.3sg2
‘Keti made him/her/them draw a picture.’
‘Keti had the picture drawn.’
Morphologically, COTs are formed by the circumfix a- . . . -in-, while causatives

of intransitive, stative and a small part of transitive verbs (perception or ingestion)
are formed just with the prefix a- I argue that the two pieces, a- and -in- are inde-
pendent morphemes with distinct functions. The prefix a- signals the introduction of
agent/causer/initiator argument in the extended projection of the verb predicate, and
2 At this point, I gloss the two affixes a- and -in- as CAUS. In the following sections, each affix will
be analysed at length and a different gloss will be proposed for each.

can be found not only on all clearly causative verbs, such as a-q’vir-a (make scream)
but also on many transitive accomplishments, such as a-cx-o (bake). As this function
has been standardly attributed to a specialised functional head Agentive Voice since
Kratzer (1996), a- is analysed as its morphological realisation in Georgain: a- spells
out agentive active Voice. The suffix -in- is analysed as a deagentivizing morpheme;
it is necessary in causatives of transitive verbs to mark the eventuality that involves
another agent, which is not structurally realised. Deagentivisation in Georgian is
carried out by mediopassive non-active Voice (Doron 2003; Alexiadou & Doron
2012), and embedded constituents in COTs are headed by this category. When the
embedded predicate in causatives is not agentive, there is no need to deagentivize it
and to mark the verb with -in-. Indeed, causative counterparts of anticausatives and
of stative verbs lack this morpheme and only comprise a-. But so do causatives of
unergatives, contrary to expectations. I claim that unergative predicates in Georgian
are not inherently agentive, and their argument is introduced via stative Voice,
following Nash (2018). Agentive semantics of unergatives is structurally built on
top of the stative VoiceP core. When unergatives are causativised, agentive Voice
realised as a- embeds the core unergative structure with its sole holder argument,
interpreted as the causee.
The present analysis differs from many accounts of COTs which correlate
deagentivisation of the embedded verb to the availability of optional adpositional
phrases. Georgian a- . . . -in- causative verbs, with deagentivized embedded pred-
icate, parallel FP causatives in (1b–2b) because of the optionality of the causee,
but unlike Romance FPs, the optional causee surfaces with the direct dative case.
These optional dative causees are arguments of the causative verb, introduced
via an Applicative head, on top of the embedded deagentivized VoiceP. They are
interpreted as a secondary agent or associate argument (cf. Shibatani & Pardeshi
2002 on sociative causatives). The main conclusion of the present study is that
morphological COTs do not embed a complement with agent irregardless of the
case-marking of the causee. If the causee can be dropped, it is not the structural
argument of the embedded clause even if it is interpreted as agent.
The article is structured as follows: Sect. 11.2 briefly introduces basic facts about
Georgian verbal morphology and Georgian (anti)causativisation strategies. Section
11.3 presents an analysis of the marker a- and of transitive verbs that carry this
affix. Section 11.4 provides an account of the affix -in- and of a- . . . -in- causatives,
which are all derived from transitive agentive predicates. Section 11.5 is devoted to
the study of causatives of unergatives and of a subclass of transitive verbs denoting
ingestion and perception. I show that these two verb classes differ from agentive
transitives in a significant way: in root contexts they involve a holder argument
coindexed with agent. Section 11.6 discusses a- . . . -in- causatives of achievements
and verbs of propositional attitude, which differ from standard COTs as they involve
an obligatory dative causee. I show that Romance languages, exemplified by French,
manifest the same behaviour when it comes to achievements. I also discuss other
contexts where COTs must combine with a causee in Georgian and in French. I
argue that in both languages, the presence of the obligatory dative causee in COTs
is correlated with the presence of an anaphoric argument in the agentless embedded
354 L. Nash
clause. The dative causee is mandatory in such cases in order to provide a structural
antecedent for the anaphor. Section 11.7 presents a conclusion.
11.2 Basic Facts About Georgian And Georgian Causatives
Georgian is a South Caucasian SOV language, characterized by free word-order,

main argument pro-drop, variable case-marking sensitive to tense-aspect properties
of the clause and to argument structure of the main predicate. Georgian is also
a split ergative language: it exhibits ergative case alignment in the aorist and
subjunctive, while in other simple tenses, the main arguments are case-aligned
according to nominative-accusative schema (4a–b). In addition to aspect split,
Georgian manifests intransitive split, whereby unergative and unaccusative verbs
do not follow the same case-alignment in ergative case environments: unaccusatives
mark their sole argument as nominative, while unergatives, like transitives, have
ergative subjects, (5a–b). In perfective tenses, transitive and unaccusative verbs are
generally preceded by a perfectivizing preverb (separated in glosses from the rest of
the verbal form by =), which indicates the telicity of the predicate. (cf. Nash 2017)
(4) a. keti-m hamlet’-i gada=targmn-a
Keti-ERG Hamlet-NOM prev=translate-AOR.3sg
‘Keti translated “Hamlet”.’
b. keti hamlet’-s targmn-is
Keti.NOM Hamlet-ACC translate-3sg
‘Keti is translating “Hamlet”.’
(5) a. keti ezo-ši da=vard-a

Keti.NOM yard-in prev=fall-AOR.3sg
‘Keti fell in the yard.’
b. keti-m i-varjiš-a
Keti-ERG VAM-exercise-AOR.3sg
‘Keti exercised.’
Georgian verbal morphology is notorious for its complexity. For the purposes
of the present analysis, I expose ordering of functional morphemes that can in
principle surround a root in finite verbs (for a more detailed study of Georgian
verb morphology, see Harris 1981; Hewitt 2005; Nash 1995; Wier 2011, a.o.). The
markers in bold in (6) are directly relevant for the present study and will be discussed
in great detail in what follows. In (7), a verbal form with 9 out of the 10 markers is
presented.
[1]
(6) preverb (prev=)—[2]1/2p subject or object agreement (1/2S/O)—[3]argument structure
modifying voice marker (VAM)—[4]ROOT—[5]fientive marker (FM)—[6]thematic suffix
(TS)—[7]deagentivising marker (NACT)—[8]thematic suffix (TS)—[9]tense (AOR,
PAST)—[10]subject or object number agreement (agr).
(7) ga-v-a-supta(v)-eb-in-eb-d-i
prev=1S-VAM-clean-TS-NACT-TS-PAST-agr
‘I would have someone / him clean it.’
The order in (6) shows how different markers can be aligned around the root in a
verbal form, but does not imply that all the morphemes can appear simultaneuously.
Concretely, if VAM signals addition or suppression of agent, it is incompatible
with postradical fientive marker [7] which marks anticausatives. Additionally, the
rightmost thematic suffix (TS) [8] is incompatible with the aorist tense (Nash 1995,
2017).
Before presenting how active transitive verbs are formed in Georgian, some
words on unaccusatives are in order. Suffixing the fientive marker -d- to the root
is the most standard way to construe unaccusative verbs. The vast majority of –d-
unaccusatives belong to the class of anticausatives/inchoatives derived from nominal
and adjectival roots (lengthen, soften, become night), (8). Two other strategies of
building unaccusatives are also available in the language: the unmarked strategy and
mediopassive strategy. Unmarked unaccusatives are not productive and constitute
the smallest class (e.g. ga=šr-a “dry”, ga=tb-a “warm up”, ga=dn-a “melt”).
Unaccusatives that look like mediopassives are derived by prefixation of the agent
suppressing marker i- [marker 3] to the root, (e.g. da=i-xrč-o “drown”, ga=i-zard-a
“grow”, da=i-mal-a “hide”). Unlike true mediopassives, such as nonactive variants
of Georgian transitives, e.g. da=i-c’er-a “write.nact”, še=i-k’er-a “sew.nact”,
da=i-xarj-a “spend.nact”, verbs in (10) do not obligatorily entail the implicit
agent. The fact that the same morphological marker yields mediopassive verbs with
implicit agent and unaccusative verbs without one is typologically common, cf.
Romance verbs with se, Greek -t-, and Hebrew nifal template (Alexiadou & Doron
2012, cf. also Schäfer (2008) on anticausatives).
(8) a. k’arak-i da=rbil-d-a
butter-NOM prev=soft-FM-AOR.3sg
‘The butter melted.’
b. rezo da-k’ac-d-a
Rezo.NOM prev=man-FM-AOR.3sg
‘Rezo became a man/turned into a young man.’
(9) a. msxal-i ga=xm-a

pear-NOM prev=dry-AOR.3sg
‘The pear dried.’
b. naq’in-i ga=dn-a
ice-cream-NOM prev=melt-AOR.3sg
‘The ice-cream melted.’
356 L. Nash
(10) a. rezo ga=i-zard-a

Rezo.NOM prev=VAM-grow-AOR.3sg
‘Rezo grew up.’
b. k’at’a da=i-xrč-o
cat.NOM prev=VAM-drown-AOR.3sg
‘The cat drowned.’
Transitive counterparts of unaccusative verbs in (8–9) and of some mediopassive-

like unaccusatives (10b) are formed by affixation of a-, one of VAMs [3] in (6). The
marker is incompatible with the fientive -d- and replaces another VAM i-, (11).
(11) a. keti-m msxal-i ga=a-xm-o TRANSITIVE COUNTERPART
Keti-ERG pear-NOM prev=VAM-dry-AOR.3sg OF ANTICAUSATIVE
‘Keti dried the pear.’
b. mze-m k’arak-i da=a-rbil-a
sun-ERG butter-NOM prev=VAM-soft-AOR.3sg
‘The sun softened the butter.’
c. rezo-m k’at’a da=a-xrč-o
Rezo-ERG cat.NOM prev=VAM-drown-AOR.3sg
‘Rezo drowned the cat.’
The prefix a- is also employed to form causatives of unergatives, statives, psych

verbs, verbs of perception and verbs of ingestion, (12a–c).
(12) a. keti-m gogo a-varjiš-a CAUSATIVE OF UNERGATIVE
Keti-ERG girl.NOM VAM-exercise-AOR.3sg
‘Keti made / had the girl exercise.’
b. keti-m megobar-s brok’ol-i še=a-q’var-a CAUSATIVE OF PSYCH -VERB
Keti-ERG friend-DAT broccoli-NOM prev=VAM-love-AOR.3sg
‘Keti made her friend love broccoli.’
c. keti-m gogo-s papa gada=a-q’lap’-a CAUSATIVE OF INGESTIVE
Keti-ERG girl-DAT porridge.NOM prev=VAM-swallow-AOR.3sg
‘Keti made the girl swallow some porridge.’
Other verbs, which can be roughly defined as agentive transitive verbs, are
causativized by a- . . . -in- strategy. The material inserted between a- and -in-
consists of verbal root but can also include functional morphemes, such as Thematic
Suffix (marker [6] in (6)); this indicates that a verbal constituent rather than a root
undergoes causativisation, (13).
(13) keti-m gogo-s otax-i da=a-lag-eb-in-a CAUSATIVE OF TRANSITIVE
Keti-ERG girl-DAT room-NOM prev=CAUS-tidy-TS-CAUS-AOR.3sg
‘Keti made the girl tidy up the room.’
Georgian does not have other causativisation strategies. The lexeme cause does
not exist in the language. But it is always possible to convey the meaning of
causative verbs in (11)–(13) with periphrasis involving the main verb order, help,
force, according to the context, as examples in (14) show. The embedded predicate
surfaces as a nominalisation in non-finite contexts, usual state of affairs in a

language without infinitives.
(14) a. keti-m gogo-s varjišoba da=a-dzal-a
Keti-ERG girl-DAT exercising.NOM prev=VAM-force-AOR.3sg
‘Keti forced the girl to exercise.’ (lit. Keti forced exercising on the girl)
b. keti-m gogo-s papis gadaqlap’-a u-brdzan-a
Keti-ERG girl-DAT porridge.GEN swallowing.NOM VAM-order-AOR.3sg
‘Keti ordered the girl to swallow porridge.’
c. keti gogo-s otaxis dalageba-ši da=e-xmar-a
Keti-NOM girl-DAT room.GEN tidying-in prev-VAM-help-AOR.3sg
‘Keti helped the girl in tidying up the room.’
In the next sections, affixes a- and -in- that participate in causativisation will be
discussed in great detail.
11.3 The Affix a-
All morphological causatives in Georgian involve the affix a-. In this section,
I consider its properties and show that this morpheme cannot be analysed as a
causative verb, like have, make in English, or faire in French. Rather, it functions in
Georgian as the spell-out of agentive Voice, the category that introduces an agent or
a causer of an active transitive predicate. I show that many agentive transitive verbs
formed from a stative root are endowed with a-, while those which are based on
manner roots do not occur with a-. I therefore propose that agentive Voice has two
morphological realisations depending on the semantics of its complement: when its
complement has a stative meaning that does not entail an agent, Voice is realised
as a- and when it entails one, Voice is null. Georgian evidence does not justify
treating animate agents and non-animate agents, referred as causers in literature
on causativisation, differently (cf. Alexiadou & Anagnostopoulou, this volume).
Transitive and causative verbs may have animate or inanimate external argument,
which is always marked as ergative in the aorist. Therefore, throughout this study
I use the terms agent and causer interchangeably; both terms refer to the external
argument of an active transitive verb in Georgian introduced by agentive Voice.
The prefix a- is one of the four argument-structure modifying verbal prefixes
in Georgian (marker [3] in (6)), traditionally refered to as versionizers (or version
markers) (Shanidze 1973; Boeder 1968, a.o.). They can only appear on finite
verbs, and they signal argument structure modification in the clause. Versionizers
express valency reduction or valency extension. Importantly, the added/suppressed
argument cannot be a theme and may be a dative argument or an agent.3 In
3 Apart from VAM a-, Georgian has a benefactive/possesseve VAM u- (cf. Sect. 11.4.4.1), reflexive-
mediopassive i- (cf. Sects. 11.2, 11.5.1, 11.5.2, and 11.6), and non-active VAM e- which adds
358 L. Nash
current neo-Davidsonian syntactic approaches to argument structure (Marantz 2013;

Pylkkänen 2008; Boneh & Nash 2017, a.o.), agents and dative arguments – goals,
holders, experiencers, beneficiaries, possessors – are added to the verbal structure
by specialized categories: agents and holders of state by Voice, and the others
by an Applicative head. Wood & Marantz (2017) propose to conflate these two
argument-introducing categories under a common label of i-heads (introducer
heads). Versionizers are these types of heads, and will be referred and glossed
hereafter as VAM, Voice-Applicative Marker.
Traditionally, the VAM a- is labelled as a neutral or as a locative/superessive
versionizer, as it has a double function (i) to signal the transitivity of verb, i.e. the
presence of agent, (15), or (ii) to add a locative argument marked with dative case
to a verb that does not require one, (16).
(15) a. keti-m saxl-i a=a-šen-a
Keti-ERG house-NOM prev=VAM-build-AOR.3sg
‘Keti built a house.’
b. keti-m k’argi sakme ga=a-k’et-a
Keti-ERG good deed.NOM prev=VAM-do-AOR.3sg
‘Keti did a good deed.’
c. keti-m k’edel-i ga=a-tetr-a
Keti-ERG wall-NOM prev=VAM-white-AOR.3sg
‘Keti whitened the wall.’
d. keti-m gogo-s leks-i gada=a-targmn-in-a
Keti-ERG girl-DAT poem-NOM prev=VAM-translate-CAUS-AOR.3sg
‘Keti made the girl translate a poem.’
(16) a. vano zi-s

Vano.NOM sit-3sg
‘Vano is sitting.’
a’. vano a-zi-s surat-s
Vano.NOM VAM-sit-3sg picture-DAT
‘Vano is sitting on the picture.’
b. vano-m xe da=xat’-a
Vano-ERG tree.NOM prev=draw-AOR.3sg
‘Vano drew a tree.’
b’. vano-m kva-s xe da=a-xat’-a
Vano-ERG stone-DAT tree.NOM prev=VAM-draw-AOR.3sg
‘Vano drew a tree on to the stone.’
What distinguishes locative vs. neutral functions of a- is its obligatory vs.

optional presence on a finite verb. When a- is obligatory on transitive and ditran-
sitive verbs with a dative argument, a- is neutral. When a verb can exist without
a dative argument interpreted as benefactive/locative/experiencer to mediopassive verbs (Sects.

11.4.3.1, 11.4.3.2). VAMs cannot be stacked, only one VAM can appear on the finite verb:
if the neutral a- is in competition with a more specific benefactive/possessive u- or reflexive-
mediopassive i-, the latter VAM wins. (for more details, see Nash 2018; Marantz 1989).
a-, e.g. da = xat’a “paint” in (17a), its addition to the verbal stem involves the
addition of the dative argument with locative reading; in this context a- has a locative
function.4
(17) a. da=xat’-a —> da=a-xat’-a LOCATIVE a-
prev=draw-AOR.3sg prev=VAM-draw-AOR.3sg
‘draw’ ‘draw onto X’
b. *ga=did-a —> ga=a-did-a NEUTRAL a-
prev=big-AOR.3sg prev=VAM-big-AOR.3sg
‘enlarge, make bigger’
c. *čvena —> a-čven-a NEUTRAL a-
show-AOR.3sg VAM-show-AOR.3sg
‘show’
As versionizers do not occur in non-finite contexts, locative semantics added

by a- in (17a) is lost on the gerund da=xat’va ‘painting’. Gerunds derived from
verbs with netural a- (deadjectival verbs) are ambiguous between inchoative and
causative readings. For example, ga=tetreba “whitening” in (18), corresponding to
the verb “whiten” in (15c), can denote both a causative or inchoative eventuality.
It is felicitous in the context when hair turns white due to age, as well as in the
situation when a hairdresser changes the color of someone’s hair.
(18) tm-is ga=tetr-eb-a
hair.GEN prev=white-TS-NOM
‘whitening of hair’
In this work, I only focus on VAM a- in neutral function.5 Can a- be compared

to a causative verb in languages with analytic causatives? The answer is positive,
sentences (19a–b), analytic make melt and transitive melt can describe similar, albeit
non-identical situations, they are both rendered by morphological causative with a-
in Georgian in (19c). In other words, the situation where Mary does not directly
4 There are exceptions to this rule. In verbs of perception, expressing forgetting-remembering, such
as axsovs “remember.3sg”, agondeba “recall.3sg”, avic’q’deba “forget.3sg”, a- is obligatory. Yet,

these are stative/unaccusative verbs where a- functions as the licenser of the dative experiencer
subject.
5 As shown in (17c), a number of core ditransitive verbs contain a compulsory a-:
(i) a-čuk-a (ii) a-čven-a

VAM-gift-AOR.3sg VAM-show-AOR.3sg
‘He gave it as a gift to her.’ ‘She showed it to him.’
The VAM a- is also found on the present tense form of the highly irregular triadic verb give in (iiia)
decomposable as CAUSE+power “empower”.
(iii) a. a-dzl-ev-s b. mi=(s)-c-a
VAM-power-TS-3sg prev=(3sg)-give-AOR.3sg
‘give’ (present tense) ‘gave’ (past tense)
360 L. Nash
manipulate the butter (and no one else besides her does) and a situation where Mary
directly acts on the butter, are equally felicitously expressed by a morphological
causative in Georgian.
(19) a. Mary made the butter melt (by leaving it out of the fridge in a sunny morning).
b. Mary melted the butter.
c. meri-m k’arak-i ga=a-dn-o
Mary-ERG butter-NOM prev=VAM-melt-AOR.3sg
‘Mary melted the butter.’, ‘Mary made the butter melt.’
Neutral a- appears on all deadjectival/denominal transitive verbs that denote

change of state and are compatible with animate agents or inanimate causers.
(20) a. vano-m / axal-ma k’anon-ma p’at’imar-i ga=a-tavisupl-a
Vano-ERG / new-ERG law-ERG prisoner-NOM prev=VAM-free-AOR.3sg
‘Vano / the new law freed the prisoner.’
b. vano-m / usamartloba-m mezobel-i ga=a-marksis’t-a
Vano-ERG / injustice-ERG neighbour-NOM prev=VAM-marxist-AOR.3sg
‘Vano / injustice turned the neighbour into a Marxist / marxized the neighbour.’
As mentioned in Sect. 11.2, most unaccusative counterparts of transitive dead-

jectival verbs are formed via suffixation of the fientive marker -d-. As a- does not
co-occur with –d-, it is unwarranted to analyse a- causatives to be derived from
non-active verbs. Rather, causative-inchoative derivation in Georgian is equipollent.
I therefore conclude that neutral a- has a transitivizing function. It spells out a
functional category that selects verbs with the change of state meaning and adds
an initiator/causer of that change.
However, in Georgian, not all transitive verbs denoting a change of state and
sharing the semantics of accomplishments with a-causatives are marked with VAM
a-. For example, Georgian has two verbs for open (21)–(22), and two verbs for
break, (23)–(24). Only one member of each pair is marked with a-.
(21) a. ga=xsn-a open1 a box/a business
b. nino-m q’ut-i ga=xsn-a
Nino-ERG box-NOM prev=open-AOR.3sg
‘Nino opened the box.’
(22) a. ga=a-ġ-o open2 a door/a window

b. nino-m k’ar-i ga=a-ġ-o
Nino-ERG door-NOM prev=open-AOR.3sg
‘Nino opened the door.’
While open1 has a more abstract meaning of unveiling, freeing, exposing

something whose content was not available, open2 , with a-, denotes an act that
makes some space available. In this sense, we can open1 a shop by creating a
business, and then open2 the shop in the morning to let clients in. The same contrast
can be observed for break. A more concrete/physical action of breaking that entails
disintegration of the whole into many solid pieces, like breaking of a glass or of a
wall is conveyed by break1 in (23) which literally means to pulverize/crumble/make

dust, while break2 is used to convey a more general act of breaking, that of a wall,
of an egg, of a word or of a heart, (24).
(23) a. da=a-mt’vri-a break1 (a solid whole into pieces)
b. nino-m satamašo da=a-mt’vri-a
Nino-ERG toy.NOM prev=break-AOR.3sg
‘Nino broke a toy.’
(24) .a ga=t’ex-a break2 (a heart/an egg)

b. nino-m p’iroba ga=t’ex-a
Nino-ERG promise.NOM prev=break-AOR.3sg
‘Nino broke a promise.’
The contrast between synonymous transitive verbs with or without a- leads to a

conclusion that accomplishments in Georgian are construed from two types of roots,
roots that name the final state reached by the object and roots that denote the manner
of action that leads to the final state. In this respect, open2 and break2 with neutral a-
contain respectively the stative adjectival root ġ(ia) “open”, and the noun mt’v(e)r(i)
“dust”, while the roots of synonymous open1 and break1 , –xsn- and -t’ex-, denote
manners of acting that lead to change.
Following much recent research on the decomposition of verbal constituents, I
propose that the structure of core transitive verbs involves a verbalizer v that fuses
with the root and selects the theme (Marantz 2013). The external argument is added
by an independent category, Voice. (Kratzer 1996). The VAM a- is the realisation of
Voice that introduces agents/initiators of the event (25).
(25) Accomplishments, transitive change of state verbs
VoiceP
DPagent
Voice vP
a-/Ø-
v+Root DPtheme
Why don’t all transitive verbs that are combined with agentive Voice appear with
a-? I contend that the reason lies in the nature of the verbalized root, whether it
denotes a manner of action or the final result. Transitive verbs based on manner
roots such as write, plough, paint, tear, etc., entail initiation, and this information is
passed from the root to v to Voice. In such a context, Voice is null. On the other hand,
roots denoting states and properties do not provide any information about intrinsic
agentivity and require a morphologically spelled out Voice. A similar situation is
found in Karachai (Lyutikova & Tatevosov 2014) where some accomplishments
carry the causative affix and others do not. This does not mean that their syntactic
event structures are different. What this implies is that simple causative verbs can
be built from different types of roots, and less “verby” roots need a morphological
support to express agentivity.
362 L. Nash
I conclude that VAM a- positively signals the presence of an external agent/causer

argument in those transitive predicates where the root, or more generally, the
complement of agentive Voice does not entail initiation. This conclusion suggests
that in all morphological causatives the complement of Voice spelled out as a- is
non-agentive in the sense that it does not require an agent. In the next sections,
I provide arguments in support of this statement. I first analyse causative of
transitive verbs and show that -in- suffix functions as a deagentiviser of the transitive
predicate embedded under VAM a-, and then I provide an analysis of causative of
unergative verbs and a small class of transitive verbs where the key idea is that the
complement of a- is a stative predicate and involves the external holder argument.
I show that morphological closeness of transitive accomplishments with a- and of
causatives of unergatives, also with a-, blurs important structural differences. In
accomplishments, the complement of a- is a vP that contains the theme, while
in causatives of unergatives, the complement of a- is another VoiceP, which is
stative and contains an external holder argument. In this respect, the present analysis
resonates with Pylkkänen’s (2008) account of causativisation patterns according to
which a causativizer can combine with complements of different structural sizes
cross-linguistically.
11.4 The Suffix -in-
Causativisation of transitive predicates is a very systematic process, whereby the

transitive verb appears between the a- . . . -in- circumfix. I show in this section
that COTs are instances of indirect causation in Georgian in the sense of Cruse
(1972) and Wolff (2003), where an intermediary agent/causer is involved in the
causal chain initiated by the main agent/causer. I argue that COTs contain two
agentive Voice categories that are distinguished in their capacity to introduce a
referential argument: the higher Voice spelled out by a- introduces the external
agent/causer and is identical to agentive Voice in direct a- causatives, while the
lower agentive Voice, spelled out by -in-, fails to introduce a referential agent
and hence functions as Middle Voice, glossed henceforth as NACT (non-active)
(marker [7] in (6), (cf. Doron 2003; Alexiadou & Doron 2012). VoiceMiddle shares
its semantic features with agentive active Voice, but is distinct from the former
in its ability to syntactically licence a referential expression in its specifier (cf.
Alexiadou et al. 2015; Embick 1998; Schäfer 2008). The analysis is motivated
by two factors: firstly, there is evidence that -in- marks agentless transitives in
other configurations than COTs in Georgian, and secondly, Georgian COTs, just
like their FP homologues in Romance, do not have to contain a causee. When the
dative causee is present in COTs, it is not generated as an embedded agent, but is
introduced above the deagentivised embedded clause by an Applicative head. The
skeleton of COTs is summarized in (26):
(26) Voice > (Appl) > VoiceMiddle > vP > v
11.4.1 COTs: Basic Facts
In addition to the prefix a-, COTs contain the suffix -in-:

(27) a. keti-m gogo-s k’ar-i da=a-k’et’-in-a
Keti-ERG girl-DAT door-NOM prev=VAM-close-NACT-AOR.3sg
‘Keti made the girl close the door.’
b. keti-m st’udent’-s leks-i gada=a-targmn-in-a
Keti-ERG student-DAT poem-NOM prev=VAM-translate-NACT-AOR.3sg
‘Keti made the student translate a poem.’
c. keti-m ert kal-s at’am-i da=a-č’r-ev-in-a
Keti-ERG one woman-DAT peach-NOM prev=VAM-cut-TS-NACT-AOR.3sg
‘Keti made one woman cut the peach.’
d. keti-m mezobel-s otax-i da=a-lag-eb-in-a
Keti-ERG neighbour-DAT room-NOM prev=VAM-tidy-TS-NACT-AOR.3sg
‘Keti made the neighbour tidy up the room.’
In (27c–d) we observe an extra suffix between the verb stem and -in- glossed
as TS (Thematic Suffix). It is a categorial marker of verbs in those contexts when
the verbal constituent (VoiceP) is not directly selected by T, and signals that the
linearly preceding root is verbalized (cf. Nash 2017 for tense-dependent distribution
of TS, see also McGinnis 2016). Not all verbs appear with TS before -in- in COTs.
When the embedded transitive verb has a manner root, its occurrence with TS is
not systematic in such contexts, and has a tendency to disappear in contemporary
Georgian (da=a-xat’-v-in-a vs. da=a-xat’-in-a “have X draw Y”). However, when
a denominal/deadjectival transitive verb with a state root is causativized, TS -eb- is
obligatory.
11.4.2 -in- is a marker of indirect causation
Georgian is not unique in displaying a more complex morphology in causatives

of agentive transitive verbs than in causatives of intransitives. In Japanese, the mor-
pheme -sase, employed to causativize transitive verbs, looks like an enriched variant
of -ase, used as a causative affix of intransitives (Oseki 2017). In Hindi/Urdu –aa is
used in causatives of intransitives and its enriched variant -vaa in COTs (Ramchand
2014). In the three languages at hand, COTs engage an extra morpheme, but only in
Georgian, the extra part is linearly non-adjacent to the morpheme that is employed in
all morphological causatives. Our task is to determine the role of this extra marker.
364 L. Nash
Nash (1994) analyses the morpheme -in- as a causative predicate (à la Romance

faire) that selects VP, (cf. also McGinnis 2016). But contrary to faire, which is
categorially a verb, -in- is analysed as a noun that needs verbal support of a- to
function as a causative verb. Georgian verbal morphology does not endorse such
a hypothesis though: in Sect. 11.3, I showed that verbs based on agent entailing
manner roots, which the causative root certainly is, are selected by agentive Voice
realised as a null morpheme. Furthermore, taking -in- for a causative verb/root
calls for an additional explanation for its systematic absence in causatives other
than COTs. In languages with analytic causatives the causative verb combines with
unaccusatives and transitives alike, e.g. faire mourir (cause to die) vs. faire tuer
(cause to kill) in French. Why would the putative causative root -in- be restricted for
causativisation of transitive verbs in Georgian?
Semantically, Georgian COTs with -in- are monoeventive and contain
one structural agent. Agent-oriented adverbs may only modify the causing
eventuality: the dative causee in (28) cannot control adverbs with plea-
sure/intentionally/intelligently. Georgian causatives share this property with Turkish
and Hungarian and differ from Japanese, where the causee controls agent-oriented
adverbs (Legate 2014; Harley 2017).
(28) keti-m gogo-s k’at’a č’k’vianurad / ganzrax / siamovneb-it
Keti-ERG girl-DAT cat.NOM intelligently / intentionally / pleasure-INSTR
da=a-mal-v-in-a
prev=VAM-hide-TS-NACT-AOR.3sg
‘Keti made the girl hide the cat intelligently / with pleasure/intentionally.’
[Keti did this intelligently / intentionally / with pleasure
≠the girl did this intelligently / intentionally / with pleasure]
An objection can be made concerning the failure of dative causees to control

agent-oriented adverbs in (28): maybe for some reason these adverbs are sensitive
to the case of agent and can be related to ergative/nominative agents but not to
dative arguments. However, this objection can be easily countered. In Georgian,
semantic agents are marked as dative in the evidential mood, (29a). The sentence in
(29b) clearly shows that the dative agent can control agent-oriented adverbs in the
evidential, unlike the dative causee in (28).
(29) a. gogo-s k’at’a da=u-mal-av-s
girl-DAT cat.NOM prev=VAM-hide-TS-3sg
‘The girl has apparently hid the cat.’
b. gogo-s č’k’vianurad / siamovenbit / ganzrax k’at’a
girl-DAT intelligently / intentionally / pleasure-INSTR cat.NOM
da=u-mal-av-s
prev=VAM-hide-TS-3sg
‘The girl has apparently intelligently / intentionally / with pleasure hid the cat.’
Systematic monoeventivity of Georgian causatives argues against the existence

of an affixal verb cause with its own temporal specification. Rather, following Ritter
& Rosen’s (1997) analysis of the contribution of causative have in English, I claim
that Georgian COTs do not express a situation where the action initiated by one
participant causes another participant to initiate an event, but rather the point of
initiation of one event is pushed back by virtue of adding another initiator, (cf. also
Bjorkman & Cowper 2013). I argue that this operation, adding an initiator to the
core representation of a transitive event can only be possible if the original initiator
is demoted. This function is performed in Georgian by -in- which signals that the
embedded predicate is deagentivized.
While a- adds an agent/causer, -in- signals that the caused eventuality is not
directly initiated by that causer but by another participant. In other words, -in- is
the sign of indirect causation. According to Ramchand’s (2014) analysis of indirect
causation marker -vaa in Hindi COTs, this marker expresses that the initiation of the
eventuality does not immediately cause the (resulting) state in indirect causatives.
There is no incremental relation or temporal co-extensivity between initiation and
result in these configurations.
Likewise, I contend that -in- signals the existence of intermediary initiation in
Georgian COTs, structurally represented by the embedded agentive Voice. Yet, this
agentive Voice does not introduce an argument. This function of -in- is typical of
mediopassive voice: the referential agent is suppressed but semantic agentivity of
the predicate is preserved. Technically this can be achieved by a specific feature
make-up of Voice. Active agentive Voice is endowed with the feature [Agent], which
denotes semantic agentivity, and with the feature [D], which denotes active nature
of the predicate and the projection of a referential expression in the specifier of
Voice. Mediopassive Voice, on the other hand, is endowed with feature [Agent]
and with feature [-D] which disallows the projection of a referential expression in
the specifier of Voice, encoding its non-active property (cf. Embick 1998; Schäfer
2008).
If -in- is analysed as a mediopassive voice morpheme that suppresses the agent
in eventuality, Georgian COTs are parallel to FP causatives in Romance, where the
embedded agent is also analysed as suppressed by several accounts mentioned in
Sect. 11.1.1 (cf. also Sect. 11.6.1).
Analysing the embedded constituent in FP and in a- . . . -in- causatives as a
projection of VoiceMiddle clarifies why these causatives have different properties
than passives, (Kayne 1975). Passives allow agent-oriented adverbs, but middle
constructions in English (30a), embedded clause in FP (31) and in Georgian COTs
(32), do not.
(30) a. *This book translates on purpose.
b. This book was translated on purpose.
(31) Marie a fait détruire l’immeuble exprès

‘Mary had the building destroyed on purpose.’
[=only Mary acts on purpose
≠Someone destroys on purpose]
366 L. Nash
(32) nino-m ganzrax da=a-c’er-in-a leks-i

Nino-ERG intentionally prev=VAM-write-NACT-AOR.3sg poem-NOM
‘Nino made Vano write the poem intentionally.’
[=only Nino acts on purpose
≠ Someone writes on purpose]
In the next sections, I show that a- . . . -in- causatives indeed embed an agentless
structure just like FP-type causatives. Firstly, I present independent evidence in
favour of treating -in- as a marker of deagentivisation. I show that -in- is not only
used in COTs in Georgian but in other contexts too, where its role is to deagentivize
a transitive predicate. Then I proceed to demonstrate that a- . . . -in- causatives share
with FPs an essential syntactic trait, namely, they can surface without a causee. To
anticipate the discussion in Sect. 11.6, while Georgian does not have the equivalent
of an optional oblique adjunct causee of par-DP type available in FPs, I show
that the dative causee in COTs is not obligatory, unlike dative arguments of other
ditransitive predicates in Georgian, including triadic causatives with a-.
When the dative argument is present in COTs, it is introduced by an Applicative
head generated above the lower deagentivized structure and below the upper
agentive Voice. The dative causee is structurally an associate argument, comparable
to animate instrumental, and semantically approaches the meaning of the agent of
the embedded clause. Under this account a- . . . -in- causative verbs are sometimes
dyadic (when the dative causee is not projected in the structure) and sometimes
triadic (when the dative causee is present).
11.4.3 -in- is a deagentivizer
I present some evidence that confirms the structural property of -in- to signal the
agentive property of the predicate it attaches too in contexts where the agent must
be non-realised.
11.4.3.1 -in- in the Pluperfect
The strongest piece of evidence for deagentivizing property of -in- comes from its
use in the pluperfect. The affix -in- has a larger function in the language and does
not occur uniquely in COTs: the suffix also occurs in pluperfect forms of a subpart
of transitive verbs.
In Georgian, pluperfect forms are used in the past subjunctive and conditional
tenses. The transitive verb in the pluperfect is formally non-active/unaccusative
in spite of the fact that it denotes an agentive eventuality. The form involves
VAM e-, whose general property is to introduce a dative argument in the agentless
template in simple tenses. To illustrate this point, consider first (33) which contains
a mediopassive form of give/pass: e- licenses the dative goal and the agent is
suppressed.
(33) c’eril-i keti-s gada=e-c-a
letter-NOM Keti-DAT prev=VAM-give-AOR.3sg
‘The letter was passed / was given to Keti.’
In the pluperfect, the same e- licenses the semantic agent of the transitive
verbs give/pass and of and unergative verb exercise in (34), marked with dative
case. Although the verbs in (34a) and in (33) are formally identical, the former is
interpreted as agentive.
(34) a. (mindoda rom) keti-s c’éril-i gada=e-c’-a
I.wanted that Keti-DAT letter-NOM prev=VAM-give-3sg
‘I wanted that Keti pass a letter.’
b. (mindoda rom) keti-s e-varjiš-a
I.wanted that Keti-DAT VAM-write-3sg
‘I wanted that Keti exercise.’
An interesting contrast is observed between two classes of transitive verbs in

the pluperfect: those that occur in simple tenses without VAM a- and are built on
manner roots, and those that appear with VAM a-, i.e. deadjectival causatives. While
both occur with VAM e- in the pluperfect, the latter, which is not a COT, carries an
additional -in-. This is a relatively new but pervasive strategy in Georgian, where
two pluperfect forms of accomplishment verbs coexist, with -in- and the archaic
one, without, (35)–(36).
(35) da=a-lag-a ‘tidy-AOR.3sg’ AORIST
a. da=e-lag-a (110 hits on Google) PLUPERFECT
b. da=e-lag-eb-in-a (6800 hits on Google) PLUPERFECT
c. mindoda rom keti-s saxl-i da=e-lag-eb-in-a
(I.wanted that) Keti-DAT house-NOM prev=VAM-tidy-TS-NACT-3sg
‘I wanted that Keti tidy up the house.’
(36) ga=a-k’et-a ‘make / fabricate-AOR.3sg’ AORIST

a. ga=e-k’et-a (657 hits on Google) PLUPERFECT
b. ga=e-k’et-eb-in-a (207000 hits on Google) PLUPERFECT
c. (mindoda rom) keti-s sakme ga=e-k’et-eb-in-a
I.wanted that Keti-DAT business.NOM prev=VAM-make-TS-NACT-3sg
‘I wanted that Keti do the business.’
How to account for the presence of -in- in the pluperfect forms of (35c) and
(36c)? I contend that -in- indicates that the pluperfect form is derived from an
agentive verb with stative root, which is not agent-entailing. The stative root is
inserted in an agentive template that must be subsequently deagentivised to derive
a non-active pluperfect form. The affix -in- flags this structural agentivity of the
lexical predicate. It signals that the complement of VAM e- is structurally agentive,
akin to the complement of a- in COTs. Other verbs, built from manner roots, do not
368 L. Nash
need -in- to indicate the basic agentivity of the predicate; the agent-entailing root
suffices to express the agentivity of the predicate embedded under e- in pluperfect.
11.4.3.2 -in- in modal contexts “feel like V-ing”
Another piece of evidence for deagentizing property of -in- concerns feel-like

constructions. Like many Slavic languages, Georgian builds modal contexts with
dispositional readings of type “(not) feel like V-ing” (cf. Rivero 2009). Most
frequently used in progressive tenses, underlyingly transitive verb is formally a
bivalent unaccusative in these configurations. The thematic agent surfaces in dative
case and is licenced by VAM e-, similarly to pluperfect forms. But unlike pluperfect
forms, the affix -in- occurs on all verbs.
(37) a. q’ovel dġe botleb-s q’ri-s, dġes k’i ar e-q’r-ev-in-eb-a
each day bottles-ACC throw-3sg, today but not VAM-throw-TS-NACT-TS-3sg
‘Every day she throws bottles, but today she doesn’t feel like throwing.’
b. vano-s lekseb-i e-c’er-in-eb-a
Vano-DAT poems-NOM VAM-write-NACT-TS-3sg
‘Vano feels like writing poems.’
c. vano-s didi sakmeeb-i e-k’et-eb-in-eb-a
Vano-DAT big deeds-NOM VAM-make-TS-NACT-TS-3sg
‘Vano feels like doing great deeds.’
d. msaxiob-s sisuleleeb-i e-tkm-ev-in-eb-a
actor-DAT stupidities-NOM VAM-say-TS-NACT-TS-3sg
‘The actor feels like saying silly things.’
In these constructions, the semantic agent is an experiencer in a situation where

it is caused by some ambient force to act (write a poem, say silly things).
11.4.3.3 Deponent verbs with -in-
Lastly, Georgian has some deponent verbs that are semantically agentive but
formally unaccusative. For example, the verb write has two deponent variants:
standard non-active variant in (38a) and a more colloquial non-active variant with
-in- in (38b–c) (Tuite 2002).
(38) a. nino i-c’er-eb-a rom mo=v-a
Nino.NOM VAM-write-TS-3sg that prev=go-3sg
‘Nino is writing that she will arrive.’
b. nino i-c’er-in-eb-a rom mo=v-a
Nino.NOM VAM-write-NACT-TS-3sg that prev=go-3sg
‘Nino is writing that she will arrive.’
c. p’at’rul-i št’rapeb-s i-c’er-in-eb-a
policeman-NOM fines-DAT VAM-write-NACT-TS-3sg
‘The policeman keeps on issuing fines.’
Even when these deponent verbs are used as true mediopassives, as in (39), the
variant with -in- is also attested. The phenomenon is lexically restricted.
(39) a. k’od-i ase i-c’er-eb-a
code-NOM this way VAM-write-TS-3sg
‘The code is / should be written this way.’
b. k’od-i sad i-cer-in-eb-a?
code-NOM where VAM-write-NACT-TS-3sg
‘Where is / should the code be written?’
An extensive analysis of modal contexts and deponent verbs is beyond the scope
of this article, but I conclude that the affix -in- “pops” up in non-active variants
of transitive verbs, without necessarily contributing the meaning of (indirect)
causation. What -in- expresses in all these contexts is that a transitive agentive
predicate is turned into a non-active form as a result of deagentivisation, which is
considered in this study to be the structural failure to license a referential argument
by a semantically agentive Voice.
11.4.4 Optional dative causee in COTs
Besides the morphological evidence in favour of treatment of -in- as the marker

of deagentivisation, Georgian causative configurations present a strong syntactic
evidence corroborating the claim that COTs involve a deagentivised embedded
clause, which concerns non-obligatoriness of the dative causee. In this property,
the causee in a- . . . -in- COTs differs from obligatory arguments and resembles an
adjunct akin to a by-phrase in English passive constructions.
Subject, direct object and dative argument can be pro-dropped in Georgian
finite clauses, as shown in Sect. 11.1.1. In (40), the sentence containing just the
verb is only interpreted as its English translation, i.e. with three discourse-specific
pronominals.
(40) (man) (mas) (is) ga=u-gzavn-a
(s)he.ERG her / him.DAT it / her / him.NOM prev=VAM-send-AOR.3sg
‘(S)he sent it / her / him to her / him.’
As COTs standardly occur with dative causees, it is expected that these can also
be pro-dropped and interpreted as pronouns, like the missing dative argument in
(40). However, surprisingly, omitting of the dative causee in (41) does not neces-
sarily entail the presence of a silent third person pronoun. Another interpretation is
available here, where the causee is non-specific and refers to some vague individual.
This second interpretation is unavailable in (40). This ambiguity implicates that two
structures may underlie the sentences in (41) when the omitted dative is specific, it is
structurally present in the structure of COTs, and when it is non-specific and vague,
the sentence lacks a causee phonologically and structurally, like what is observed in
Romance FPs.
370 L. Nash
(41) a. keti-m iat’ak’-i ga=a-c’mend-in-a

keti-ERG floor-NOM prev=VAM-clean-NACT-AOR.3sg
i) ‘Keti had the floor cleaned by him / her / them.’
ii) ‘Keti had the floor cleaned.’
b. keti -m pankar-i ga=a-tl-ev-in-a
Keti-ERG pencil-NOM prev=VAM-sharpen-TS-NACT-AOR.3sg
i) ‘Keti had the pencil sharpened by him / her / them.’
ii) ‘Keti had the pencil sharpened.’
c. keti-m roman-i gada=a-targmn-in-a
Keti-ERG novel-NOM prev=VAM-translate-NACT-AOR.3sg
i) ‘Keti had him / her / them translate the novel.’
ii) ‘Keti had the novel translated.’
d. keti-m k’at’a da=a-mal-v-in-a
Keti-ERG cat.NOM prev=VAM-hide-TS-NACT-AOR.3sg
i) ‘Keti had him / her / them hide the cat.’
ii) ‘Keti had the cat hidden.’
The reflexive possessive determiner tavisi “self’s” in (42) can have two potential
binders, as in (i), or only one binder, the causer, as in (ii). Under this second reading,
the sentence does not contain a silent dative pro and only the causer/agent can
function as the antecedent of the possessive determiner.
(42) vano-m tavisi cerili gada=a-targmn-in-a
Vano-ERG self.GEN letter.NOM prev=VAM-translate-NACT-AOR.3sg
i) ‘Vanoi had himj translate hisi/j own letter.’
ii) ‘Vanoi had hisi/*j letter translated.’
On the basis of the evidence in (41)–(42), and anticipating Sect. 11.6, where the
parallelism will be dealt with in more detail, COTs with a structurally present dative
causee are parallel to Romance FIs where the dative causee is obligatory, while
COTs where a causee is a non-specific existentially asserted individual are parallel
to Romance FPs.
11.4.4.1 Benefactives in a- and a- . . . -in- causatives
Another piece of evidence that argues in favour of non-obligatoriness of the dative

causee in COTs formed by a- . . . -in- affixes comes from the distribution of dative
benefactive arguments. I show in this section that dative benefactive arguments are
licit in a- . . . -in- causatives and banned in other triadic verbs, as Georgian prohibits
double datives in the same clause.
In Georgian, dative benefactive arguments are introduced by a benefactive-
possessive VAM u- (for 3rd person dative arguments) in finite transitive and
unaccusative clauses. I focus here on the behaviour of benefactives in transitive
clauses.
(43) a. keti-m mankana ga=recx-a

Keti-ERG car.NOM prev=wash-AOR.3sg
‘Keti washed the car.’
b. keti-m vano-s mankana ga=u-recxa
Keti-ERG Vano-DAT car.NOM prev=VAM-wash-AOR.3sg
‘Keti washed a car for Vano.’
When a benefactive argument is added to a transitive verb with a-, u- wins over
a-, (44). Recall from Sect. 11.3 that two VAMs cannot occur on the same verb.
(44) a. keti-m k’ar-i ga=a-ġ-o
Keti-ERG door-NOM prev=VAM-open-AOR.3sg
‘Keti opened a door.’
b. keti-m ert kal-s k’ar-i ga=u-ġ-o
Keti-ERG one woman-DAT door-NOM prev=VAM-open-AOR.3sg
‘Keti opened a door for one woman.’
Unlike transitives with a- in (44a), triadic verbs with a- are incompatible with
benefactives. Georgian disallows two dative arguments in the single clause, (45b).
Adding a benefactive dative direk’t’ors “director” by means of VAM u- to a triadic
predicate, and “silencing” the dative goal in (45c) does not save the structure.
(45) a. keti-m mezobel-s ekim-i ga=a-cn-o
Keti-ERG neighbour-DAT doctor-NOM prev=VAM-know-AOR.3sg
‘Keti presented (make-know) a doctor to the neighbour.’
b. *keti-m direk’t’or-s mezobel-s ekim-i
Keti-ERG director-DAT neighbour-DAT doctor-NOM
ga=u-cno
prev=VAM-know-AOR.3sg
‘Keti presented a doctor to the neighbour for the director.’
c. *keti-m direk’t’or-s ekim-i ga=u-cn-o
Keti-ERG director-DAT doctor-NOM prev=VAM-know-AOR.3sg
‘Keti presented the doctor (to him) for the director.’
However, a- . . . -in- causatives do not behave as triadic verbs with a-. The
benefactive argument can be added to COTs without a dative causee, via VAM u-,
(46). This shows that in (46b), the causee is non-specific and structurally absent.
Hence, a- . . . -in- causatives behave like transitive verbs with two arguments rather
than like triadic verbs. Notice however, that COTs with benefactive u-, albeit
grammatical, are marked forms and rarely used.
(46) a. keti-m vano-s mankana ga=u-recx-in-a
Keti-ERG Vano-DAT car.NOM prev=VAM-wash-NACT-AOR.3sg
‘Keti had the car washed for Vano.’
b. keti-m ert kal-s leks-i gada=u-targmn-in-a
Keti-ERG one woman-DAT poem-NOM prev=VAM-translate-NACT-AOR.3sg
‘Keti had a poem translated for one woman.’
372 L. Nash
I conclude that benefactive datives are only available in those constructions

where another dative argument is not an obligatory argument. Georgian COTs
occur in such configurations: as the dative causee does not have to be syntactically
projected, they constitute a context where benefactives are licit.
11.4.5 Semantic and syntactic properties of the dative causee

in COTs
Having shown that the dative causee does not have to be generated in a- . . . -in-
causatives, the next question to be addressed is: how do a- . . . -in- configurations
with a dative cause differ from those without?
I contend that the dative causee in a- . . . -in- causative configurations is intro-
duced by an Applicative head that relates an individual to an eventuality with
a implicit agent. The dative causee is not interpreted as an indirect object of
ditransitives: it is not a location/goal/possessor/benefactor/experiencer, but rather
functions as a “second agent”, in semantic sense, and an associate. Although until
now, I proposed English translations of sentences with COTs with the causative
make, Georgian COTs with dative causees often denote events done collectively
by two (groups of) agents where the causer/agent enables the second agent to act.
For example, sentence in (47a) felicitously expresses a situation where Thea gives
an order to Keti to carry up the bicycle but it also denotes a context where Thea
helps/enables Keti carry up a bicycle to the fifth floor. Another situation of collective
action expressed by a causative verb is presented in (47b). Here the sentence can
mean that the mother ordered/forced/incited the girls to bake a cake, but can also
describe an event when the mother and the girls bake a cake together. In fact, during
this baking process, the girls do not have to be principal actors and the mother may
even do more baking-related actions, but the girls must be present and involved in
the baking process and considered by the speaker as main agents. Situations in (47)
may be also expressed by a non-causative predicate and a comitative PP: Thea and
Keti carried the bicycle together, the mother and the girls baked the cake together.
But such sentences are devoid of the nuance present in causative configurations
where the causee is considered to bear principal responsibility for carrying out the
event denoted by the verb.
(47) a. tea-m keti-s velosip’ed-i amo=a-t’an-in-a
Thea-ERG Keti-DAT bycicle-NOM prev=VAM- carry-NACT-AOR.3sg
‘Thea made / enabled / let / helped Keti carry the bicycle.’
b. deda-m gogoeb-s t’ort’-i gamo=a-cx-ob-in-a
mother-ERG girls-DAT cake-NOM prev=VAM-bake-TS-NACT-AOR.3sg
‘Mother made / enabled / let / helped the girls bake a cake.’
The type of causation where there is no strict distinction in actions of the causer
and the causee is known as sociative causation (Dixon 2000; Shibatani & Pardeshi
2002) and involves supervision, joint action and assistance. Shibatani & Pardeshi
(op.cit) claim that sociative causation is a mid-way between direct causation and
indirect causation as the causee is as involved in the action as the causer. I assume
that this sociative reading is a direct consequence of the structural origin of the
dative causee as the argument of the causative verb rather than as the agent of
the embedded predicate. Introduced higher and outside of the embedded event,
it can be interpreted as an associate of the causer in the joint action that brings
about the result denoted in the embedded clause. It is important to emphasize that
the sociative reading is context-dependent and not obligatory, whereas the neutral
reading, where the causer acts such that the event is initiated without the causer’s
actual participation in the event is the default one. Therefore, the applicative head
in (48) is not a carrier of a specific lexical meaning such as “with”, it just relates
the causee to the caused event, leaving the exact reading of the configuration to
pragmatic and context-depending factors. In some sense, the semantic role played
by Appl in (48) is comparable to that of prepositions par in French, or by in English,
when they combine with a semantic agent in passive and causative constructions.
(48) VoiceP
DP agent
Voice ApplP
a-
DPcausee
Appl Voice Middle P
Voice Middle vP
-n-
v DP theme
In this section, I showed that Georgian COTs contain agentless embedded
transitive predicate headed by mediopassive VoiceMiddle and spelled out by suffix
-in-. When the dative causee occurs in COTs, it is introduced by an Applicative
head above the embedded VoiceMiddle P. The dative causee is interpreted as a second
agent, i.e. semantic agent of the embedded clause, without being its syntactic
agent. In the next section I turn to causatives of unergatives, which are analysed as
configurations that do not contain an agentive causee either. The obligatory causee
in these structures is not the structural agent of the embedded unergative predicate
but rather the structural holder argument. This derivation is a direct consequence of
the central property of Georgian unergatives to introduce their sole argument as a
holder of process, by stative Voice.
11.5 Causatives of unergatives
The principal tenet of the present work is to show that the causee in morphological
causatives is not structured as an embedded agent. While it is trivial why causees
in causatives of unaccusatives and stative verbs are not structured as embedded
agents, – non-causative variants of these predicates do not have agents in the
374 L. Nash
first place, – extending the same conclusion to causatives of unergative verbs

is not straightforward. Unlike unaccusatives and statives, the external argument
of unergative verbs is interpreted as agent. Moreover, since Perlmutter’s (1978)
division of intransitive verbs into unaccusatives and unergatives, the claim that the
subject of unergative verbs is structured differently than the subject of unaccusatives
constitutes an uncontroversial theoretical postulate. Nash (2018) shows that not
only the structures of unergatives and unaccusatives are distinct universally, but the
external argument of transitive verbs and the external argument of unergatives are
not structured alike either. While the agent of transitives is introduced by agentive
Voice, the external argument of unergatives can be mapped in two different sites
(two different Voice projections) in the unergative predicate: it must be mapped
first as the holder of unergative process/state (undergoer in Ramchand 2008, actor
in Massam 2009, cf. also Tollan 2018; Tollan & Oxford 2018) and can be also
mapped as the agent that initiates this process by agentive VoiceP. In this sense,
unergative verbs express activities where the initiator and the experiencer/holder are
co-arguments.
Building on this insight, I argue in this section that causatives of unergatives,
which are morphologically identical to direct causatives with a-, do not involve an
embedded agent. The lower of the two arguments in (49) is structured as the holder
of the embedded state/process predicate while the upper agent/causer is introduced
by agentive VAM a-, which is a common trait of all morphological causatives in
Georgian.
(49) a. keti-m gogo a-varjiš-a
Keti-ERG girl.NOM VAM-exercise-AOR.3sg
‘Keti made the girl exercise.’
b. keti-m gogo a-cek’v-a
Keti-ERG girl.NOM VAM-dance-AOR.3sg
‘Keti made the girl dance.’
c. keti-m gogo a-cur-a
Keti-ERG girl.NOM VAM-swim-AOR.3sg
‘Keti made the girl swim.’
11.5.1 Structure of Unergatives in Georgian
In order to understand how causatives of unergatives are formed in Georgian, it

is important to look more closely into the structure of unergatives themselves.
Morphologically, Georgian unergatives look like monovalent verbs in imperfective
tenses (50a), and as bivalent verbs in perfective tenses (50b). Namely, in perfective
tenses, unergative verbs are marked with an extra morpheme, the agent-suppressing
VAM i-, which behaves as Romance se/si and which occurs on a class of unac-
cusative verbs, as shown in Sect. 11.2.
(50) a. vano cek’v-av-s / lap’arak’-ob-s / mogzaur-ob-s

Vano.NOM dance-TS-3sg / talk-TS-3sg / travel-TS-3sg
‘Vano dances / talks / travels.’
b. vano-m i-cek’v-a / i-lap’arak’a-a /
Vano-ERG VAM-danse-AOR.3sg VAM-talk-AOR.3sg
i-mogzaur-a
VAM-travel-AOR.3sg
‘Vano dansed / talked / travelled.’
Se/si and i- do not only express mediopassive/non-active voice, but also signal
reflexivity, i.e. co-reference between the agent and a lower argument in a transitive
configuration, (51) (cf. Wood 2014). Although examples in (51) are in the perfective
tense-aspect, VAM i- in its mediopassive and reflexive functions occurs on verbs in
both perfective and imperfective tenses, unlike i- on unergatives.
(51) a. keti-m da=i-ban-a
Keti-ERG prev=VAM-wash-AOR.3sg
‘Keti washed (herself).’
b. keti -m simind-i mo=i-xarš-a
keti-ERG corn-NOM prev=VAM-cook-AOR.3sg
‘Keti cooked some corn for herself . ’
While VAM a- adds an argument to the verbal structure, VAM i- indicates that
an argument required by the structure is not realised as an independent referential
expression. Another property that shows that unergative forms in the aorist are
bivalent concerns case-marking: the subject of unergative predicates is marked as
ergative, like the subject of a transitive predicate.
According to Nash (2018), unergative predicates in Georgian have different
templates in perfective and imperfective tenses. In the imperfective tense, these
predicates have one external argument. This argument is not introduced by agen-
tive Voice but by stative Voice because agentivity in Georgian is contingent on
transitivity: only those external arguments that c-command another argument may
be interpreted as agents. As unergative predicates lack the internal argument,
their external argument fails to be introduced as agent, by agentive Voice, and is
hence introduced as holder, by stative Voice. Nash (2018) provides ample evidence
that shows that unergatives and statives are the only two predicate classes that
show unstable morphosyntactic behaviour across tenses, and differ in this respect
from unaccusatives and transitives, which do not change their argument structure
depending on Viewpoint Aspect. In spite of the fact that unergatives are structured as
statives in imperfective tenses, their external argument is interpreted as agent. This
is due to the process of semantic repackaging of a stative eventuality into a dynamic
one under the scope of Imperfective Viewpoint Aspect, as proposed by Rothstein
(1999) for English active be in constructions such as Kim is being naughty. The
tree in (52) shows that repackaging is represented as syntactic reanalysis: stative
Voice and Aspect specified as [imperfective] are bundled together as one syntactic
category and conjointly license the argument in the specifier as agentive.
376 L. Nash
(52) VoiceState /AspectP UNERGATIVES IN IMPERFECTIVE TENSES
DPagent
Aspectimperf /VoiceState vP
As in perfective tenses repackaging is unavailable, the language has to have

recourse to another structural strategy to ensure agentivity of unergatives. This is
obtained by a process, identical to causativisation – the agent is added to the core
stative structure. This entails adding another Voice category to the existing stative
VoiceP. But as the agent and the holder are the same individual in unergatives, the
upper argument is coindexed with the lower holder argument – this reflexivisation
operation is signalled by VAM i-. In (53a), the verbal skeleton expresses semantic
and syntactic reflexivity: the higher argument and the lower argument refer to the
same individual, the girl is the initiator and the holder of dancing/travalling in (53b)
(cf. Ramchand 2008 for the proposal that the agent of unergatives hold a hybrid
initiator/undergoer role.)
(53) a. VoiceP UNERGATIVES IN PERFECTIVE TENSES
DPagent-i
Voice VoiceStateP
i-
<holder>i
Voice vP
v
b. gogo-m i-cek’v-a / i-mogzaur-a

girl-ERG VAM-dance-AOR.3sg / VAM-travel-AOR.3sg
‘The girl danced / travelled.’
The structure in (53a) shows that in perfective tenses unergative configurations

are reflexive causative configuratons. The only difference between this structure and
the one in (54a) resides in VAM, reflexive vs. agentive, and the only difference
between i-marked unergatives in (53b) and their causatives with VAM a- in (54b)
is that in the latter, the initiator of the event and the holder of the process/state are
referentially disjoint: Keti initiates a process of dancing/travelling that has the girl
as the holder.
(54) a. VoiceP CAUSATIVES OF UNERGATIVES
DPagent-i
Voice VoiceStateP
a-
DPholder-k
Voice vP
v
b. keti-m gogo a-cek’v-a / a-mogzaur-a
Keti-ERG girl.NOM VAM-dance-AOR.3sg / VAM-travel-AOR.3sg
‘Keti made the girl dance / travel.’
In principle, nothing bans that unergatives in imperfective tenses be also struc-

tured as reflexive causatives, but as the language has a more economical way to
agentivize a non-agentive structure through repackaging of Voicestate and Aspect,
the derivation in (52) wins over a more costly reflexive-causative derivation in (54a).
In this sense, paradoxically, basic unergatives in perfective tenses are a subtype of
causatives of unergatives in Georgian.
The difference between transitive change of state verbs with a- and causatives of
unergatives consists in the structural representation of the embedded predicate. In
the former, what is initiated/caused is an event resulting in the change of state of the
theme argument. In causatives of unergatives, what is caused is a state/atelic process
that holds of the embedded non-agentive causee. Causatives of unergatives involve
two external arguments, while transitive change of state verbs involve one external
and one internal argument, sister to v.6
When an unergative verb containing a cognate object (55a) is causativized, the
causee is marked as dative and the cognate object carries the same case as thematic
objects, (55b–c).
6 An anonymous reviewer asks whether the analysis of Georgian causatives of unergatives can
provide insights on the difference between English construction in (i-ii). For which of the two can
the causee be claimed as non-agentive?
(i) Mary caused/made the kids (to) dance.
(ii) Mary danced the kids to the room.
If (ii) means that Mary changed the state of the children by placing them into the room by dancing,
the children are interpreted as the theme that changes a location, and in this sense is closer to a
simple transitive change of state verb than to a causative of unergative. Therefore, the Georgian
counterpart in (54b) is closer to the meaning of (i) than of (ii).
378 L. Nash
(55) a. gogo-m t’ango i-cek’v-a / č’adrak’-i

girl-ERG tango.NOM VAM-dance-AOR.3sg / chess-NOM
i-tamaš-a / simartle i-lap’arak’-a
VAM-play-AOR.sg / truth.NOM VAM-speak-AOR.3sg
‘The girl danced tango / played chess / spoke truth.’
b. keti-m gogo-s t’ango a-cek’v-a / č’adrak’i
Keti-ERG girl-DAT tango.NOM VAM-dance-AOR.3sg / chess.NOM
a-tamaš-a
VAM-play-AOR.sg
‘Keti made the girl dance a tango / play chess.’
c. dost’oevsk’i-m mišk’in-s simartle a-lap’arak’a
Dostoyevsky-ERG Mishkin-DAT truth.NOM VAM-speak-AOR.3sg
‘Dostoyevsky made Mishkin speak the truth.’
In this section, I have shown that causatives of unergatives do not contain an

embedded agent. The lower argument in these constructions is the external argument
of the embedded unergative predicate, structured as the holder of the process/state.
It is introduced by stative Voice. As the argument of the embedded VoiceState P, it is
obligatory in causatives and hence differs from the optional dative causee in COTs
introduced by Applicative outside the embedded predicate domain. The causee in
causatives of unergatives constitutes the lower of two arguments and is marked as
the direct object, without being one, while the agent/causer introduced by agentive
Voice is marked as the subject of a transitive verb, according to the algorithm of
Dependent Case (Marantz 1991; Baker & Vinokurova 2010; Nash 2017). In the
next section, I show that a class of transitive verbs denoting perception and ingestion
builds its causatives similarly to causatives of unergatives with cognate objects. In
these triadic a- causatives, the causee is obligatory and marked as dative, while the
theme behaves as a regular direct object. I show that the unusual behaviour of this
transitive class with respect to causativisation is due to the presence of the holder
argument, coindexed with the agent, in its argument structure.
11.5.2 Causatives of transitive verbs denoting ingestion

and perception
Although the majority of transitive verbs are causativised by circumfix a- . . . -in-,

some transitive verbs that do not denote change of state and take non-affected
objects, such as perception and ingestion verbs, form causatives with only a-, like
unergatives. (cf. Guasti 1996 on causatives of perception verbs in Romance).
(56) a. nino-m keti-s musik’a ga=a-gon-a

Nino-ERG Keti-DAT music.NOM prev=VAM-hear-AOR.3sg
‘Nino made Keti hear the music.’
b. nino-m keti-s rusul-i a-sc’avl-a
Nino-ERG Keti-DAT Russian-NOM VAM-learn-AOR.3sg
‘Nino taught (=made learn) Keti Russian.’
c. nino-m keti-s c’ign-i c’a=a-k’itx-a
Nino-ERG Keti-DAT book-NOM prev=VAM-read-AOR.3sg
‘Nino made Keti read a book.’
d. nino-m keti-s k’eks-i ga=a-sinj-a
Nino-ERG Keti-DAT cake-NOM prev=VAM-taste-AOR.3sg
‘Nino made / let / helped Keti taste a cake.’
e. nino-m keti-s bal-i a-č’am-a
Nino-ERG Keti-DAT cherry-NOM VAM-eat-AOR.3sg
‘Nino made Keti eat cherries.’ / ‘Nino fed Keti cherries.’
As for case-distribution on main arguments and the verbal form, causatives in

(56) are similar to causatives of unergative verbs with cognate object in (55), to
causatives of stative Experiencer-Theme verbs in (12b) and to ditransitives with a-
in (18c).
The absence of -in- in these causatives suggests that they do not denote indirect
causation: the embedded predicate is not deagentivised and the dative causee is not
optional. Indeed, unlike a- . . . -in- COTs, causatives of perception and ingestion
verbs cannot be “stripped” off the dative argument. When it is absent, the causee
is still present as a discourse-specific null pronoun in (57). Semantically, the dative
causee is not interpreted as the initiator of the embedded predicate, but as a recipient
or holder of the ingested or perceived entity.
(57) a. nino-m musik’a ga=a-gon-a
Nino-ERG music.NOM prev=VAM-hear-AOR.3sg
‘Ninoi made her / him / them hear music.’
b. nino-m rusul-i a-sc’avl-a
Nino-ERG Russian-NOM VAM-learn-AOR.3sg
‘Nino made her / him / them learn Russian.’
e. nino-m bal-i a-č’am-a
Nino-ERG cherry-NOM VAM-eat-AOR.3sg
‘Nino made him / her / themeat a cherry.’
If the dative causee in (56) is structurally distinct from the dative causee argument
in COTs with -in-, the question is whether it is generated as an embedded agent
of perception/ingestion verbs, or whether it structurally behaves as the causee
in causatives of unergatives and statives, i.e. as an experiencer/holder argument
introduced by VoiceState P. Under the latter scenario, the external argument of inges-
tion/perception verbs occupies different positions in causative and non-causative
pairs: in the former, it is in Spec,VoiceState P and in the latter, in the specifier of
agentive VoiceP, mirroring the behaviour of the external argument of unergative
verbs in perfective tense-aspects, as discussed in Sect. 11.5.1.
380 L. Nash
Indeed, morphological evidence supports the claim that perception and inges-
tion verbs, albeit transitive, have common traits with unergatives in Georgian.
Concretely, most perception/ingestion verbs occur with reflexive mediopassive
VAM i- in perfective tense-aspects, just like unergatives. The external argument of
perception and mental ingestion verbs in (46) is not only the initiator of the event but
is also interpreted as the holder or recipient of the theme. In these eventualities, the
theme is “placed”, physically or mentally, by the agent in its own body or mind. In
this sense, verbs in (58) are structured as reflexive causative verbs: the agent causes
itself to be the holder of the theme in the eventuality.
(58) a. vano-m musik’a ga=i-gon-a7

Vano-ERG music.NOM prev=VAM-hear-AOR.3sg
‘Vano heard music.’
b. vano-m c’ign-i c’a=i-k’itx-a
Vano-ERG book-NOM prev=VAM-read-AOR.3sg
‘Vano-ERG read a book.’
c. nino-m algebra i-sc’avl-a
Nino-ERG algebra.NOM VAM-learn-AOR.3sg
‘Nino learned algebra.’
d. nino-m mankana da=i-nax-a
Nino-ERG car.NOM prev=VAM-see-AOR.3sg
‘Nino saw the car.’
Causative variants of (58) in (56) are minimally different: instead of instantiating

a structure where the initiator and the holder of ingestion/perception is the same
individual, these roles are assigned in non-reflexive standard causatives to different
individuals. This operation is morphologically reflected by replacement of VAM i-
by VAM a-.
In (59a), I provide the structure of perception/ingestion verbs, which parallels the
structure of reflexive causative unergatives in (53a), but also includes the obligatory
theme of ingestion. The holder of ingestion is introduced by VoiceState , as the
holder of process/state in unergative constructions. The structure in (59b) is the non-
reflexive, i.e. causative, variant of (59a), which involves removing reflexive index
and spelling out the holder argument.
7 Georgian has two verbs for “hear” depending on tense-aspect of the sentence: ga=i-gon-a (AOR),
used in perfective tenses, and e-smi-s (PRESENT), employed in imperfective tenses. The latter
variant combines with a dative experiencer subject and its root sm is closer to the stem smen
“hearing”. The root gon in ga=i-gon-a has the meaning close to understanding, consciously
receiving a signal.
(59) a. VoiceP INGESTIVE/PERCEPTION VERBS
DPagent-i
Voice VoiceStateP
i-
<holder> i
Voice vP
v DPtheme
b. VoiceP CAUSATIVES OF INGESTIVE/PERCEPTION VERBS
DPagent-i
Voice VoiceStateP
a-
DPholder-k
Voice vP
v DPtheme
Unlike verbs of mental ingestion, verbs denoting physical ingestion such as taste,
swallow, eat do not occur with reflexive mediopassive VAM i- in the aorist in (60a).
Yet, their causative forms in (60b) are similar to causatives of mental ingestion and
perception verbs.
(60) a. ia-m nuš-i gada=q’lap’-a / še=č’am-a /
Ia-ERG almond-NOM prev=swallow-AOR.3sg / prev=eat-AOR.3sg /
ga=sinj-a
prev=taste-AOR.3sg
‘Keti swallowed / ate / tasted an almond.’
b. deda-m ia-s nuš-i gada=a-q’lap’a /
Mother-ERG Ia-DAT almond-NOM prev=VAM-swallow-AOR.3sg /
še=a-č’am-a / ga=a-sinj-a
prev=VAM-eat-AOR.3sg / prev=VAM-taste-AOR.3sg
‘Mother made Ia swallow / eat / taste an almond.’
In spite of the absence of i- in verbs of ingestion in (60a), their agent is still

understood as the initiator of the eventuality and as the holder of the ingested
material. Therefore, in causative counterparts, these two roles, of initiator and of
holder, are carried by different individuals: X acts so that Y be a holder of ingestion.
In their studies on parallel structures in Hindi-Urdu, Ramchand (2014), Bhatt &
Embick (2003) treat causatives of ingestives as ditransitives where the causee is
understood as the recipient.8
8 An alternative analysis of a- causatives of physical ingestion verbs can be proposed. If mor-
phological cues are taken seriously, the language does not treat mental ingestion and physical
ingestion verbs alike. The former need VAM i- in the perfective while the latter do not. This could
indicate that physical ingestion verbs do not involve the holder argument coreferent with the agent.
Causatives of these verbs may contain the locative a-, rather than neutral a-, which adds a locative
argument to a predicate (see Sect. 11.3). Under such analysis, causatives of physical ingestion verbs
382 L. Nash
When physical ingestion verbs have figurative meanings and are not literally
interpreted to denote ingestion, their causative counterparts are construed as regular
COTs, by means of a- . . . -in- strategy. For example, some idiomatic expressions
involve ingestion verbs in Georgian: eat someone means “nag/annoy”, and swallow
tongue means “shut one’s mouth up”. Their causatives with a- are at best infelicitous
because they are interpreted literally.
(61) a. am ambav-ma mat ena gada=a-qlap’#(-in)-a
this news-ERG they.DAT tongue.NOM prev=VAM-swallow-NACT-AOR.3sg
‘This news made them swallow their tongues (=shut up).’
b. me čems švil-s am mezobel-s ar še=v-a-č’m-ev#(-in)-eb
I my child-ACC this neighbour-DAT not prev=1-VAM-eat-TS-NACT-TS
‘I will not let my neighbour eat up/nag my child (=nag/annoy).’
The same ill-formedness as in (61) results when a causative of ingestion verb

describes a situation of indirect causation where the causer does not act directly
on the theme and the recipient of ingestion. In such cases, the ingestion process
has to be initiated by another participant. The sentence in (62) describes a life-
attested situation where a scientist working on claustrophobia and dressed in special
protection clothes provokes an anaconda to swallow him for purposes of experiment.
The causative variant with a- is highly infelicitous in this context and can only
describe a macabre situation where the scientist feeds the anaconda with his own
body.
(62) mk’vlevar-mai anak’onda-s tav-ii gada=a-q’lap’-#(in)-a
scientist-ERG anaconda-DAT self-NOM prev=VAM-swallow-NACT-AOR.3sg
‘The scientisti had a anaconda swallow himi.’
(https://www.palitravideo.ge/index.php?option=com_content&view=article&id=51888&
catid=21&Itemid=30)
To summarize, transitive verbs denoting perception and ingestion are

causativised in Georgian by means of VAM a-. These causatives are triadic verbs,
where the obligatory dative argument is not structured as the embedded agent but
rather as the embedded experiencer/holder/recepient of a directly caused eventuality.
The main reason for this type of causative formation is due to the argument structure
of perception/ingestion predicates in Georgian, which, like unergative verbs and
unlike standard unergatives, contain two coindexed external arguments, agent and
holder, each introduced by a distinct Voice category. Their causative variants are
structurally identical and also contain the same two Voice categories, but with
referentially disjoint agent and holder arguments.
have the semantics of a simple ditransitive predicate with agent, theme and location. Although
such an analysis is morphologically sounder, it loses parallelism between ingestion and physical
ingestion verbs as a semantically natural class.
11.5.3 Interim Conclusion
All Georgian morphological causatives contain the prefix a-. It fleshes out the
category of agentive Voice that introduces the agent/causer argument that initiates a
causative eventuality. A complex event can be directly caused when an individual or
an entity initiates a dynamic change of state that affects the theme (make X soften)
or holds of an individual (make X love Y, make X work, make X learn Y).
On the other, a- . . . -in- causatives describe indirectly caused events: the causer
does not act directly and the eventuality must involve another initiator that directly
brings about that change. While semantically the intermediary agent is a necessary
participant in indirect a- . . . -in- causatives, structurally this argument is suppressed.
In this property, Georgian follows a well-attested typological pattern of building
indirect causatives of transitive verbs, known in Romance as FP strategy whereby
the expression of the causee is not obligatory. While in FPs the causee may surface
as an oblique by-phrase, the causee in Georgian surfaces under a different syntactic
guise in COTs, namely as a dative argument introduced by Applicative head. I
analyse its role to be that of secondary agent, instrument, associate. I show that the
properties of the causee in a- causatives and in a- . . . -in- causatives are different:
in the former, the (dative) causee belongs to the domain of the embedded predicate,
does not hold an agent role and is obligatory, in the latter the dative argument is
introduced by Applicative head, does not belong to the embedded predicate domain
and is optional. The most significant conclusion reached so far is that the structure
of morphological causative predicates in Georgian never involves an embedded
referential agent.
However, the claim that the dative causee is always optional in a- . . . -in-
causatives must be weakened in face of the evidence presented in the next
section. We will see that a- . . . -in- causatives of achievement verbs, transitive
idioms, and causatives that combine with inanimate causer or an anaphoric theme
must contain an obligatory dative causee. A parallel phenomenon is observed in
Romance languages where a subclass of transitive verbs or configurations involving
a transitive verb may only be causativised by FI strategy, i.e. with obligatory à-DP
causee (cf. Folli & Harley 2007). I put forth an account for these restrictions, both
in Georgian and Romance, based on the idea that the dative causee is obligatory in
indirect causatives when it serves as an antecedent/binder of an anaphoric index or
expression in the causative configuration.
11.6 COTs with Obligatory Dative Causee
The present analysis of Georgian causatives of agentive verbs establishes a corre-

lation between indirect causation, i.e. involvement of the intermediary agent in the
causative construction, and non-obligatoriness of the structural causee semantically
corresponding to the intermediary agent.
384 L. Nash
In this section, I show that indirect a- . . . -in- causatives do not necessarily have
optional causees, so the correlation between the presence of -in- and the omission of
the causee does not hold. A class of transitive verbs in Georgian, namely achieve-
ments and verbs of propositional attitude, manifest a behaviour in causative con-
structions which places them mid-way between causatives of perception/ingestion
verbs and causatives of standard transitives. As the latter, their causatives are built
via a- . . . -in- affixation, and as the former, they must combine with the dative
causee. I will argue in what follows that the structure of achievements and verbs
of propositional attitude is similar to that of perception/ingestion verbs and involves
a holder argument in addition to the initiator. While causatives of these verbs denote
indirect causation that requires the deagentivized predicate, severing of the initiator
leaves the anaphoric holder argument unbound by a referential expression. This is
the reason why the dative causee, and its licencer, the Applicative head are required
in these structures – the causee serves as an antecedent to the anaphoric holder.
Consider causatives of achievements in (63):
(63) a. keti-m mosc’avleeb-s tamaš-i mo=a-g-eb-in-a
Keti-ERG pupils-DAT game-NOM prev=VAM-win-TS-NACT-AOR.3sg
‘Keti made the pupils win the game.’
b. keti -m gogo-s k’at’a a-p’ovn-in-a
Keti-ERG girl-DAT cat.NOM VAM-find-NACT-AOR.3sg
‘Keti made the girl find the cat.’
c. vano-m megobar-s saxl-i a-q’id-in-a
Vano-ERG friend-DAT house-NOM VAM-buy-NACT-AOR.3sg
‘Vano made the friend buy the house.’
In (64), the missing dative causee is structurally present and interpreted as

discourse-specific. This state of affairs contrasts with the readings of sentences
containing a causative of accomplishments in (41), where the missing causee could
be understood as either discourse-specific, and structurally present, or existentially
asserted, and not projected.
(64) a. keti-m tamaš-i mo=a-g-eb-in-a

Keti-ERG game-NOM prev=VAM-win-TS-NACT-AOR.3sg
‘Keti made him / her / them win the game.’
b. keti-m k’at’a a-p’ovn-in-a
Keti-ERG cat.NOM VAM-find-NACT-AOR.3sg
‘Keti made him / her / them find the cat.’
c. vano-m saxl-i a-q’id-in-a
Vano-ERG house-NOM VAM-buy-NACT-AOR.3sg
‘Vano made him / her / them buy the house.’
This atypical behaviour of causatives of achievements stems from Aktionsart

properties of the embedded verb. The salient feature of achievements as an is that
the result and the initiation cannot be temporally teased apart: although these verbs
are clearly agentive, the agent participant is physically involved in each stage of
the temporal development of the event. The object, caused to undergo a change,
is in the possession of the agent. Therefore it seems reasonable to claim that the
initiator is also the experiencer of the result, and is structurally projected twice, as
the argument of agentive Voice and as the argument of stative Voice that selects
vP. In Georgian, this meaning is transparently expressed by morphological means:
achievements are marked (in perfective tenses) with the reflexive-mediopassive
VAM i-, like unergatives and perception/mental ingestion verbs.
(65) a. mosc’avle -m tamaš-i mo=i-g-o

student-ERG game-NOM prev=VAM-win-AOR.3sg
‘The student won the game.’
.b nino-m k’at’a i-p’ovn-a
Nino-ERG cat.NOM VAM-find-AOR.3sg
‘Nino found the cat.’
c. deda-m saxl-i i-q’id-a
Mother-ERG house-NOM VAM-buy-AOR.3sg
‘Mother bought a house.’
The structure in (66) proposed for achievements resembles the ones put forth for
perception verbs in (59), however their causative counterparts differ: achievements
in (66b) are deagentivised and comprise VoiceMiddle P that dominates VoiceState P,
while embedded perception verbs are structurally smaller, VoiceState P. When an
achievement predicate is deagentivized, its lower substructure VoiceState P still
contains the anaphoric holder argument that must be bound by the causee introduced
above the deagentivised predicate by the applicative.
(66) a. VoiceP ACHIEVEMENTS (same as PERCEPTION VERBS)
DPagent-i
Voice VoiceStateP
i-
<holder>i vP
v DPTheme
b. VoiceP CAUSATIVES OF ACHIEVEMENTS
DPagent
Voice ApplP
a-
DPi
Appl VoiceMiddleP
VoiceMiddle VoiceStateP
-in-
<holder>i
Voice vP
v DPtheme
Furthermore, a- . . . -in- causatives of propositional attitude verbs also require a

dative causee. As in achievements, the agent is mentally involved throughout the
386 L. Nash
entire event in standard verbs of propositional attitude in spite of the fact that not all
of them are marked with the reflexive VAM i-. For example, in (67) i-pikra “think”
is, but tkva “say” is not.9,10
(67) a. vano-m simartle tkv-a / es i-pikr-a
Vano-ERG truth.NOM say-AOR.3sg / this.NOM VAM-think-AOR.3sg
‘Vano said the truth / thought this.’
b. keti-m vano-s simartle a-tkm-ev-in-a /
Keti-ERG Vano-DAT truth.NOM VAM-say-TS-NACT-AOR.3sg
es a-pikr-eb-in-a
this.NOM VAM-think-TS-NACT-AOR.3sg
‘Keti made Vano say the truth / think this.’
The present study leads to a conclusion that transitive verbs in Georgian fall
into two classes: those where the agent is temporally dissociated from the result –
accomplishments—, and those where the agent initiates and holds / experiences the
result – perception and ingestion verbs, achievements and verbs of propositional
attitudes. When accomplishments are deagentivized in causatives, the structure
of the embedded predicate does not have to contain any co-argument of the
implicit agent, while when achievements and verbs of propositional attitude are
deagentivised, the embedded structure keeps the index of the lower co-argument
of the implicit agent unbound. This forces the projection of the dative causee that
functions as the binder thereof.
11.6.1 Obligatory causees in COTs in Romance languages:

*FP contexts
Non-unitary behaviour of COTs (causatives of transitives) has been observed in

Romance languages, too. Kayne (1975), and subsequently Burzio (1986), Guasti
(1996), a.o., have argued that two causative configurations faire-infinitif (FI)
and faire-par (FP), have different structural properties and consequently different
distribution. Importantly, FI with obligatory à-DP causee is the only option in a
number of circumstances in French, which I refer to as *FP-contexts in (68).11
9 The verb say in Georgian is irregular, and undergoes root suppletion in each main tense. While in
the aorist its form is without i-, in the perfective future it does occur with reflexive-mediopassive
i-: i-t’q’v-is (VAM-say-3sg).
10 The verb think in Georgian can be causativised as a direct causative, with VAM a-: a-pikr-
a. Although a- and a- . . . -in- variants can be used interchangeably in many contexts, a- . . . -in-
variants are more felicitous when the causer intentionally manipulates the causee such that (s)he
starts thinking. In other words, the a- . . . -inform expresses indirect causation of thinking.
11 *FP contexts should be understood here as contexts where causative configurations must include
a causee. In FP causatives, the causee is never obligatory, while in FI configurations, it is

mandatory.
Each context is exemplified in (69). What all these contexts have in common is
the presence of a referentially dependent expression or of the embedded verb other
than a standard transitive change of state predicate.
(68) *FP contexts: FI is the only option: (cf. Folli & Harley 2007)
a) when the embedded transitive verb is a nonpassivizable idiom: casser la croute – “lit.
break the crust, eat”; (69a)
b) when the embedded direct object is an inalienable expression linked to the
embedded subject; (69b)
c) when the embedded direct object contains a bound variable pronoun; (69c)
d) when the embedded direct object is not affected (is not the argument of a change
of state verb): perception verbs, achievements (cf. Alsina 1992, Guasti 1996). (69d)
e) when the transitive verb cannot passivize; (69e)
f) when the causer is non-animate (pure cause); (69f)
(69) a. Paul a fait casser la croute *(aux enfants)

Paul made “break the crust” to-the children
‘Paul made the children eat.’
b. Paul a fait lever la maini *(aux élèvesi)
Paul made raise the hand to-the students
‘Paul made students raise a hand.’
c. Paul a fait réparer soni jouet *(à chaque enfanti)
Paul made repair his toy to each child
‘Paul made each childi repair hisi toy.’
d. Paul a fait entendre la chanson *(à Marie)
Paul made hear the song to Mary
‘Paul made Mary hear the song.’
d’. Paul a fait trouver une robe *(à Marie)
Paul made find a dress to Mary
‘Paul made Mary find a dress.’
e. Paul a fait quitter la maison *(à Marie)
Paul made leave the house to Mary
‘Paul made Mary leave the house.’
f. La famine a fait voler les poules *(à la pauvre femme)
The famine made steal the chickens to the poor woman
‘The famine made the poor woman steal chickens.’
Accounting for structural differences of FI and FPs has been one of the most
important challenges in scholarly work on Romance causatives. For Kayne (1975),
FPs involve an embedded transitive predicate with a removed external argument,
comparable to passives. Yet, as already mentioned in Sect. 11.4.2, differences exist
between passives and the embedded clause in FP, e.g. agent-oriented adverbs are
illicit in FPs but licit in passives. This led researchers to consider the embedded
transitive predicate in FPs under a different angle. Zubizarreta (1985) argues that
the syntactic realisation of the external argument is blocked in FPs while FIs involve
the internalization of the agent into an indirect object. Ippolito’s (2000) analysis
of FIs is similar to Zubizarreta’s: in FIs, the causee, similar to benefactives, is
388 L. Nash
introduced by an applicative category rather than by agentive Voice (cf. also Schäfer
2008; Pitteroff & Campanini 2013).12 Guasti (1996) proposes that FPs involve
bare VPs while FIs contain small clause VPs where the agent of the embedded
verb receives a composite theta-role: agent theta-role from the embedded predicate
and a benefactive theta-role from the causative verb (cf. also Alsina 1992). Folli
& Harley (2007) claim that the transitive verb in FIs is structured as in other
embedded configurations, with one difference: the causee is projected in the right-
hand specifier of v (which corresponds to Voice in the present analysis) and hence
receives exceptional dative case, (cf. Rouveret & Vergnaud 1980). As for FPs, the
authors argue that the embedded transitive VP is nominalised; it hence lacks v
(=Voice) and consequently the agent argument.
While these analyses succeed to account for the structural absence of the causee
in deagentivised FPs and its presence in FIs as the agent of the embedded predicate,
they cannot account for the pattern in (68). Why do some transitive verbs, e.g.
perception verbs and achievements, require the presence of the dative causee? Why
do only transitive agentive accomplishments appear in FPs?13
The present analysis of causatives of transitive predicates has addressed precisely
this issue. Georgian, like Romance languages, requires the presence of the dative
causee in causatives of perception and ingestion predicates, as well as in causatives
of achievements. I argued that obligatory dative causee is present in each class
for different structural reasons. Causative of perception and ingestion verbs are
built as direct triadic causatives where the upper agentive Voice embeds a stative
VoiceP; hence these causatives do not structurally represent the involvement of
an intermediary agent. On the other hand, causatives of achievements are built as
indirect causatives and involve an embedded deagentivized VoiceP. As achievements
have two external co-arguments, agent and holder, severing of the former as a result
of deagentivisation removes the antecedent of the latter. The “secondary agent”,
introduced by Applicative hence serves as an antecedent to the anaphoric holder
argument of the embedded clause. I propose to extend this reasoning to account for
point (68d) in *FP contexts for Romance languages as well.
12 In the present analysis, the causee in COTs is also introduced by Applicative. However, this
account, where Appl is generated above a deagentivised VoiceP, significantly differs from Ippolito-
style analyses where Appl replaces agentive Voice. In the latter analyses, Appl is endowed with
agentive semantics and directly selects vP only in FIs. It is unclear how this head differs from a
standard agentive Voice and why the embedded predicate in both causative types, FI and FP, is
differently structured. In my account, the embedded transitive predicate conserves its non-active
agentive template Voice>vP in FI and FP causatives alike. The two causativisation strategies differ
by the presence of ApplP on top of the embedded VoiceP only in FI type.
13 Guasti (1996) addresses the issue of asymmetric distribution of FPs and FIs and establishes
a generalisation according to which only verbs that take an affected object can be deagentivized.
Yet, the author does not explore further why this correlation should hold in causative constructions.
11.6.2 *FP contexts in Georgian
In this section, I show that Georgian exhibits *FP behaviour not only when it
comes to cauativisation of achievements and perception verbs, but also in those
contexts when the embedded clause contains an anaphoric expression. Causative
configurations in (70) contain a referentially dependant anaphoric expression in
the embedded clause bound by the dative causee: a body part in (70a) and a
bound variable in (70b). The idiomatic expression in (70c) can be also counted as
anaphoric, following Burzio (1986) analysis of idioms as anaphorically bound to
the agent (cf. Folli & Harley 2007).
(70) a. keti-m tea-si xel-ii a=a-c’ev-in-a INALIANABLE OBJECTS
Keti-ERG Tea-DAT hand-NOM prev=VAM-rise-NACT-AOR.3sg
‘Keti made Thea i raise heri hand.’
b. keti-m q’ovel gogo-si tavisii pankar-i BOUND PRONOUNS
Keti-ERG each girl-DAT self.GEN pencil-NOM
ga=a-tl-ev-in-a
prev=VAM-sharpen-TS- NACT-AOR.3sg
‘Keti made each girli sharpen heri own pencil.’
c. keti-m gogo-si [kva ga=a-xetk-in-a]i IDIOMS
Keti-ERG girl-DAT stone.NOM prev=VAM-split-NACT-AOR.3sg
‘Keti made the girl split a stone.’
=Keti made the girl do the impossible (to achieve something)
What unifies the configurations in (70) with achievement predicates is the

presence of an argument which in root contexts is referentially dependent on the
agent. As embedded clauses are deagentivised in causative constructions, the causee
needs to be projected in order to function as the potential binder of these anaphoric
constituents in the absence of the structural agent.
Finally, dative causees are obligatory with a- . . . -in- causatives when the causer
is inanimate (context 68f):
(71) a. sircxvil-ma keti-s iat’ak’-i ga=a-c’mend-in-a INANIMATE CAUSER
shame-ERG Keti-DAT floor-NOM prev=VAM-clean-NACT-AOR.3sg
‘Shame made Keti clean the floor.’
b. am ambav-ma tea-s gancxadeba da=a-c’er-in-a
this news-ERG Thea-DAT announcement.NOM prev=VAM-write-NACT-AOR.3sg
‘This news made Thea write an announcement.’
The reason for the obligatory presence of the causee in these contexts is different
than in all previous configurations where the anaphoric argument belonged to the
embedded clause. Here, I hypothesize that it is the inanimate causer that manifests
referential dependence on the secondary agent. Structurally, this is reminiscent
of backwards binding when the anaphoric index occurs on a structurally higher
bindee (Pesetsky 1995). Concretely, albeit being the source of a causative event,
the inanimate causer does not initiate the event independently of the causee, its
force, i.e. the property to cause an eventuality, is effective only if perceived as
390 L. Nash
such in the mind of the causee, (at least from the speaker’s point of view). When
we say ‘Unemployment/financial crisis made Mary sell the house’, unemployment
must be Mary’s, the financial crisis is also appropriated/internalised by Mary. So
it is not unemployment as an abstract notion but Mary’s mental representation of
unemployement that makes her sell the house. Likewise, in the sentence “John’s
article made Mary write a book”, the article must be appropriated by Mary(’s mind)
to cause her to write the book. I therefore conclude that pure causers must directly
affect the (mind of the) causee in order to provoke their actions. In some sense,
in these contexts, we get the strongest instance of sociative causation where the
inanimate causer and the causee act temporally undissociably together to bring
about the eventuality denoted by the embedded verb.
To conclude, in *FP contexts in Georgian and in Romance COTs, the dative
causee is obligatory due to binding requirements in the embedded clause or to
referential dependency imposed by the inanimate causer. Although dative causees
in a- . . . -in configurations can be present as an optional argument introduced by
ApplP, in *FP contexts, ApplP must introduce the dative argument that serves as
an antecedent to an anaphoric index: such an index appears on inanimate causers,
on an embedded theme denoting a body part or a bound pronominal variable, on an
implicit lower holder argument, or on embedded idiomatic expression.
11.7 General Conclusion
The primary goal of this study is the investigation of the syntactic status of
the causee in causatives of agentive predicates. The main conclusion is that in
morphological causatives, and at least in a subset of analytical causatives, e.g.
Romance causatives, the causee of the embedded predicate is not structured as
the embedded agent, even if it is interpreted as such. The study of Georgian
causatives of transitive verbs reveals that causativisation involves two steps: firstly,
the embedded transitive predicate is deagentivised, and secondly, the deagentivised
predicate is selected by agentive Voice that introduces the causer/agent. Each
operation is spelled out by a distinct morpheme: deagentivisation by middle voice
morpheme -in-, and adding a causer/agent by active voice morpheme a-. The
two operations yield a semantic interpretation of indirect causation: participant X
causes an eventuality initiated by participant Y. The intermediary agent Y cannot
be structurally present but is existentially asserted, as in mediopassive sentences.
However, a participant semantically corresponding to the intermediary agent can,
and sometimes must (cf. *FP contexts) be introduced in a causative configuration,
via ApplP, generated above the embedded deagentivised VoiceP and below the main
agentive VoiceP.
Causativisation of unergative verbs follows a different pattern in Georgian, and
morphologically involves only a- affixation. Unlike COTs, the causee is obligatory
in causatives of unergatives. I claim that in these cases, causation is not indirect, the
causee does not initiate the embedded process/activity but is in the process/state.
The causee is the core holder argument of unergative predicates projected as the
specifier of stative VoiceP (Nash 2018); this constituent is embedded under the
upper agentive Voice introducing agent/causer. As a consequence, neither causatives
of transitive verbs nor causatives of unergative verbs involve an embedded agent
licensed by agentive VoiceP.
I also showed that causatives of transitives do not present a unified class. Firstly,
some transitives are causativised by a- morpheme only. These are perception and
ingestion verbs, which I claim to be structurally close to unergatives as they
involve a holder argument. Their causatives express direct causation where the
agent/causer acts directly on the holder and the theme and hence does not entail
the presence of the intermediary agent. Secondly, causatives of achievements and
propositional attitude verbs, causativised by a- . . . -in- affixes, require the presence
of the dative causee. The embedded predicate in these causatives is deagentivised
but the embedded clause retains the unbound index of the lower holder coargument
of the severed agent. The dative causee must be instantiated in these configurations
in order to serve as the antecedent for this anaphoric index. I extend this reasoning to
account for other *FP contexts, which are syntactic environments where the dative
causee must be present, both in Romance and in Georgian.
The following table summarizes causatives of all agentive verbs explored in the
study.
(72) AGENTIVE CAUSATIVE LOCUS OF CAUSEE EMBEDDED

PREDICATE TYPE MARKING PREDICATE
accomplishments a-…-in- Spec,ApplP – optional VoiceMiddleP
achievements a-…-in- Spec,ApplP – obligatory VoiceMiddleP
perception/ingestion a- Spec,VoiceStateP – obligatory VoiceStateP
unergatives a- Spec,VoiceStateP – obligatory VoiceStateP
The present analysis has typological ramifications. In many languages, e.g. Hindi
and Japanese, causatives of transitive and intransitive verbs do not involve the
same morpheme. Usually, COTs are more complex morphologically. The extra
morpheme, which is not shared by causatives of intransitives, is the sign of
deagentivisation of the embedded transitive predicate. If in such languages, the
causee in COTs surfaces as a direct argument, case-marked with accusative or dative
cases, it is important to investigate whether its behaviour is parallel to Georgian
or Romance dative causees. Are these direct arguments exceptionally optional in
COTs? Is their presence required in *FP contexts? I leave these typological inquiries
for future research.
392 L. Nash
References
and middle. Journal of Linguistics, 48(1), 1–34.
alternations: A layering approach (Vol. 55). Oxford: Oxford Studies in Theoretical Linguistics.
Alsina, A. (1992). On the argument structure of causatives. Linguistic Inquiry, 23, 517–555.
Baker, M. (1988). Incorporation: A theory of grammatical function changing. Chicago: University
of Chicago Press.
Baker, M., & Vinokurova, N. (2010). Two modalities of case assignment in Sakha. Natural
Language and Linguistic Theory, 28, 593–642.
Bhatt, R., & Embick, D. (2003). Causative derivations in Hindi. Ms. University of Pennsylvania.
Bjorkman, B., & Cowper, E. (2013). Inflectional shells and the syntax of causative have. In
Proceedings of the 2013 annual conference of the Canadian Linguistic Association.
Boeder, W. (1968). Über die Versionen des georgischen Verbs. Folia Linguistica, 2(1–2), 82–152.
Boneh, N., & Nash, L. (2017). The syntax and semantics of dative DPs in Russian ditransitives.
Natural Language and Linguistic Theory, 35(4), 899–953.
Burzio, L. (1986). Italian syntax: A government and binding approach. Dordrecht: Reidel.
Cruse, D. A. (1972). A note on English causatives. Linguistic Inquiry, 3, 522–528.
Dixon, R. M. W. (2000). A typology of causative: Form, syntax and meaning. In R. M. W.
Dixon & A. Y. Aikhenvald (Eds.), Changing valency: Case studies in transitivity (pp. 31–83).
Doron, E. (2003). Agency and voice: The semantics of the Semitic templates. Natural Language
Semantics, 11(1), 1–67.
Embick, D. (1998). Voice systems and the syntax/morphology interface. In H. Harley (Ed.),
MITWPL 32: Papers from the UPenn/MIT roundtable on argument structure and aspect (pp.
41–72). Cambridge, MA: MITWPL.
Folli, R., & Harley, H. (2007). Causation, obligation and argument structure: On the nature of little
v. Linguistic Inquiry, 38(2), 197–238.
Guasti, M. T. (1996). Semantic restrictions in Romance causatives and the incorporation approach.
Linguistic Inquiry, 27, 294–313.
Harley, H. (2017). The “bundling” hypothesis and the disparate functions of little v. In R.
D’Alessandro, I. Franco, & A. J. Gallego (Eds.), The verbal domain (pp. 3–28). Oxford: OUP.
Harris, A. (1981). Georgian syntax: A study in relational grammar (p. 33). London: Cambridge
Studies in Linguistics.
Hewitt, G. (2005). Georgian: A learner’s grammar. London/New York: Routledge.
Ippolito, M. (2000). Remarks on the argument structure of Romance causatives. Cambridge, MA:
Ms., MIT.
Kayne, R. (1975). French syntax. Cambridge, MA: MIT Press.
Phrase structure and the lexicon (pp. 109–137). Dordrecht: Kluwer.
Legate, J. A. (2014). Voice and v: Lessons from Acehnese. Cambridge, MA: MIT Press.
Lyutikova, E., & Tatevosov, S. (2014). Causativization and event structure. In B. Copley & F.
Martin (Eds.), Causation in grammatical structures (Oxford Studies in Theoretical Linguistics)
(Vol. 52, pp. 279–327).
Manzini, R. (1983). On control and control theory. Linguistic Inquiry, 14(3), 421–446.
Marantz, A. (1989). Relations and configurations in Georgian. Chapel Hill: Ms., University of
North Carolina.
Marantz, A. (1991). Case and licensing. In G. Westphal, B. Ao, & H.-R. Chae (Eds.), Proceedings
of the 8th Eastern States Conference on Linguistics (ESCOL 1) (pp. 58–68). Ithaca: CLC
Publications.
Marantz, A. (2013). Verbal argument structure: Events and participants. Lingua, 130, 152–168.
Massam, D. (2009). The structure of (un)ergatives. In S. Chung, D. Finer, I. Paul, & E. Potsdam
(Eds.), Proceedings of AFLA (Vol. 16, pp. 125–135).
McGinnis, M. (2016). The morphosyntax of thematic suffixes in Georgian. Talk given at
The South Caucasian Chalk Circle Workshop. Paris. http://paris2016.mariapolinsky.com/wp-
content/uploads/2015/11/McGinnis_ThemSuf.pdf
Nash, L. (1994). On the categorial specification of causative morphemes. Proceedings of ELS, 24,
411–425.
Nash, L. (1995). Portée argumentale et marquage casuel dans les langues SOV et dans les langues
ergatives: l’example du géorgien. Doctoral dissertation, Université de Paris VIII.
Nash, L. (2017). The structural source of split ergativity and ergative case in Georgian. In J. Coon,
D. Massam, & L. Travis (Eds.), The Oxford handbook of ergativity. Oxford: OUP.
Nash, L. (2018). Non-unitary structure of unergative verbs: From monovalent statives to bivalent
reflexive causatives in Georgian. Ms. Université Paris 8 & CNRS.
Oseki, Y. (2017). Voice morphology in Japanese argument structures. NYU: ms. lingbuzz/003374.
Perlmutter, D. (1978). Impersonal passives and the unaccusative hypothesis. In Proceedings of the
4th annual meeting of the Berkeley Linguistics Society (pp. 157–189). Berkeley: UC.
Pesetsky, D. (1995). Zero syntax. Cambridge, MA: MIT Press.
Pitteroff, M., & Campanini, C. (2013). Variation in analytic causative constructions: A view on
German and Romance. The Journal of Comparative Germanic Linguistics, 16(2–3), 209–230.
Ramchand, G. (2008). Verb meaning and the lexicon: A first phase syntax. Cambridge: Cambridge
University Press.
Ramchand, G. (2014). Causal chains and instrumental case in Hindi/Urdu. In B. Copley &
F. Martin (Eds.), Causation in grammatical structures,. Chapter 10 (pp. 245–278). Oxford:
Ritter, E., & Rosen, S. T. (1997). The function of have. Lingua, 101(3–4), 295–321.
Rivero, M. L. (2009). Intensionality, high applicatives and aspect: Involuntary state constructions
in Bulgarian and Slovenian. Natural Language and Linguistic Theory, 27, 151–196.
Rizzi, L. (1978). A restructuring rule in Italian syntax. Recent Transformational Studies in
European Languages, 3, 113–158.
Rothstein, S. (1999). Fine-grained structure in the eventuality domain: The semantics of predicate
adjective phrases and ‘b’. Natural Language Semantics, 7, 347–420.
Rouveret, A., & Vergnaud, J.-R. (1980). Specifying reference to the subject: French causatives and
conditions on representation. Linguistic Inquiry, 11(1), 97–202.
Schäfer, F. (2008). The syntax of (anti-)causatives (External arguments in change-of-state con-
texts). Amsterdam/Philadelphia: John Benjamins.
Shanidze, A. (1973). Kartuli enis gramatikis sapudzvlebi (Grammar of the Modern Georgian
Language), Tbilisi.
Shibatani, M., & Pardeshi, P. (2002). The causative continuum. Typological studies in language,
48, 85–126.
Tollan, R. (2018). Unergatives are different: Two types of transitivity in Samoan. Glossa: A Journal
of General Linguistics, 3(1).
Tollan, R., & Oxford, W. (2018). Voice-less unergatives: Evidence from Algonquian. Proceedings
of WCCFL, 35, 399–408.
Tuite, K. (2002). Deponent verbs in Georgian. In W. Bublitz, M. von Roncador, & H. Vater (Eds.),
Philologie, Typologie und Sprachstruktur: Festschrift für Winfried Boeder zum 65 Geburtstag
(pp. 375–389). Frankfurt am Main: Peter Lang Verlag.
Wier, T. (2011). Georgian morphosyntax and feature hierarchies in natural language. PhD
dissertation, University of Chicago.
Wood, J. (2014). Reflexive -st verbs in Icelandic. Natural Language & Linguistic Theory, 32(4),
1387–1425.
394 L. Nash
Wood, J., & Marantz, A. (2017). The interpretation of external arguments. In R. D’Alessandro,
I. Franco, & A. J. Gallego (Eds.), The verbal domain (pp. 255–278). New York: Oxford
University Press.
Wurmbrand, S. (2001). Infinitives: Restructuring and clause structure. Berlin/New York: Mouton
de Gruyter.
Zubizarreta, M. L. (1985). The relation between morphophonology and morphosyntax: The case
of Romance causatives. Linguistic Inquiry, 16, 247–289.
Chapter 12
The Causative Component
of Psychological Verbs
Edit Doron†
Abstract The paper distinguishes two different subclasses of psychological verbs

(psych verbs), each associated with a different way for expressing the causes of
the mental state described by the verb. (1) Relational psych verbs describe the
mental state as a two-place relation between an Experiencer (Exp) and the mental
representation of a Target/Subject Matter (T/SM); e.g. Exp relates to a mental
representation of the target of Exp’s love/curiosity. These verbs allow the expression
of a third argument, a Cause, which brings about the relation between Exp and
T/SM. (2) Property psych verbs describe the mental state as a one-place property
of Exp; e.g. Exp is in the state of fear/anger. These verbs too may encode a Cause
argument, but here the causal relation is evaluated differently depending on the
construction. The paper shows that the interaction between the two subclasses of
verbs and the category of causation is not special to the mental domain but also
holds for stative physical verbs such as locative verbs, where the arguments Location
and Theme replace Exp and T/SM. Hence this is a general distinction among
verbs denoting stative relations between two arguments: verbs denoting two-place
relations vs. verbs denoting one-place properties attributed to a Cause.
Keywords Agent · Cause · Force · Target matter · Psychological predicates ·

Linking problem · Templates · Hebrew
Edit Doron submitted this chapter a few weeks before she passed away. The editors took liberty to
correct typos and insert slight editorial modifications.
E. Doron
Department of Linguistics and Language, Logic and Cognition Center, The Hebrew University of

https://doi.org/10.1007/978-3-030-34308-8_12
396 E. Doron
12.1 Introduction
The nature of the fundamental category of causation is still under debate. Here I
will examine it by studying its linguistic contribution to the semantics of natural
language verbs. In general, verbs linguistically describe events (including states
as well as dynamic events) by specifying both their temporal profile and their
participants. The participants are linguistically represented as arguments of the verb,
classified according to their thematic roles (such as Agent, Experiencer, Location,
Theme). Thematic roles are a designated set of linguistically significant relations
(first introduced by Fillmore 1968 and Jackendoff 1972) in which arguments stand
to the described event. These relations are significant in determining the aligning of
arguments to grammatical functions such as subject, object, indirect object etc.
The received view in linguistics since Dowty (1979) and Parsons (1990) has been
that the meaning of a causative verb includes a component, represented as CAUSE,
which is a relation between a pair of events. An alternative view (Doron 1999, 2003;
Neeleman & van de Koot 2012; Aimar 2018) is that a causative verb describes a
single event, which includes a participant with the thematic role of Cause. Hence,
causation as encoded by causative verbs is not a relation between events but a
relation between a participant and the event it participates in, i.e. a thematic role.
Clearly, if the participant itself happens to be an event, then Cause indeed relates
two events. But this is a special case, and in general the Cause participant may be
an entity of various different types, not necessarily an event.
The present paper studies the thematic role of Cause in its interaction with
the other thematic roles that characterize a particular class of verbs – the so-
called psychological verbs (psych verbs). Psych verbs are a recognized subclass
of mental verbs – alongside perception verbs, mental-state verbs (also called
propositional attitude verbs), emotive-factive verbs (propositional attitude verbs
involving emotions such as surprise, happiness, or regret), and mental-act verbs
(verbs of apprehending, deciding, choosing, calculating, reasoning etc). Mental
verbs have been distinguished from physical verbs like physical-action verbs,
motion verbs, verbs of spatial configuration, verbs of locative placement, verbs of
emission, and others studied by lexical semanticists and philosophers.
There has been a long debate on the argument structure of psych verbs (e.g.
Belletti & Rizzi 1988; Pesetsky 1995; Arad 1999; McGinnis 2000; Reinhart 2002;
Landau 2010) and on the alternation in the aspectual categories of these verbs
(Marin & McNally 2011; Alexiadou & Iordăchioaia 2014). Here I would like to
discuss the different thematic roles of the arguments of psych verbs. A famous
puzzle associated with psych (and locative) verbs is the “linking problem”: the
alternation in the alignment of thematic roles to grammatical functions. I will
suggest a solution to the linking problem, which is valid both for psych and locative
verbs.
Another issue raised by psych verbs is the interaction of their canonical
arguments with the argument bearing the thematic role of Cause. I will contrast
the thematic role of Cause with the thematic role of Force (Talmy 2000; Croft 1991;
12 The Causative Component of Psychological Verbs 397
Copley & Harley 2015; Copley et al. 2016), and show that it is the same contrast,
which also plays a role in the lexical semantics of locative verbs.
The thematic role of Cause is special in that it is typically the role of a supplemen-
tary argument, one which is not necessarily part of the basic characterization of the
event described by the verb. The verb always has another argument with a variety of
possible roles, none of them Cause. Verbs which allow a Cause argument sometimes
describe events, which can alternatively be viewed as spontaneous (in which case
the verb is alternatively unaccusative). Other verbs require the Cause argument as an
obligatory participant. The different role of Agent is an obligatory participant of all
the events encoded by the verb. Verbs with an Agent participant, unlike verbs with
a Cause participant, are never interpreted as unaccusative. Moreover, the two roles
affect the aspectual class of a verb in different ways. Verbs with a Cause argument
may be stative, but not verbs with an Agent argument, unless the Agent is a Force.
Force is like an Agent in being an obligatory participant of the event, but it does not
involve action. Hence verbs with a Force argument may be stative. Still, Force is not
a type of Cause, as will be shown in Sect. 12.2.
Sect. 12.2 shows how in Semitic, causative morphosyntax differs from agentive
morphosyntax. Sect. 12.3 discusses relational psych verbs, psych verbs which
denote a stative relation between two arguments, the experiencer (Exp) and the
Target/Subject Matter (T/SM) toward which emotion or evaluation is directed by
Exp. I show that despite what is mostly assumed in the literature, these verbs do no
exhibit the linking problem. Sect. 12.4 shows that the causative morphosyntax of
Hebrew allows the addition of a third argument, a Cause, to relational psych (and
locative) verbs. Sect. 12.5 rejects the implicit assumption in the linguistic literature
that all psych verbs are relational. I argue that some psych verbs are property psych
verbs, i.e. they denote a state which is a one-place property of Exp. I show that the
other argument often found with these verbs is not T/SM but Cause. Sect. 12.6 is
the conclusion.
12.2 Cause vs. Agent/Force
The difference between Cause and Agent as thematic roles is illustrated here for
physical verbs. Causative physical change of state verbs such as destroy and kill in
(1) may take abstract causes as their subjects. This is different for the agentive verbs
in (2).
(1) Physical verb with a Cause subject
a. Military losses destroyed the empire.
b. The inappropriate use of the drugs killed the patient.
(2) Physical verb with an Agent subject

a. The lion hunted a big bison.
b. The wind slammed the windowblinds.
398 E. Doron
Agents are often intentional entities, as in (2a). Verbs such as hunt require their
subject to be an intentional Agent. Other verbs which select for Agent subjects, such
as slam, do not require intentional entities, but allow Agents which are elements of
nature, as in (2b), or artifacts constructed to engage in certain actions, e.g. Levin &
Rappaport Hovav’s 1995 example The teapot whistled. Crucially, subjects which are
abstract causes cannot fulfill the role of Agent, hence (3) is unacceptable, in contrast
both to (1b) and (2b):
(3) *Inappropriate use of the cord slammed the window blinds.
Cause and Agent are not arguments of the verb’s roots (Kratzer 1996). The
differences between them have been attributed to the nature of the functional v
head, which verbalizes the root. Folli & Harley (2007) distinguish vCAUS vs. vDO
for languages such as English and Italian. In the Semitic languages, the difference
between these different v heads is expressed by the morphology of the verb (Doron
2003; Kastner 2018). Semitic verbs are constructed by intertwining a consonantal
root morpheme with the v head (whose exponent is traditionally called template)
that derives the actual verb by determining its syllabic structure. Hebrew has three
different templates that derive verbs in the active voice. The marked templates are
the CAUSATIVE and the INTENSIVE, exponents of vCAUS and vINTNS respectively.1
The difference in form between them correlates with the role assigned by the v head
to the external argument of the verb (the argument of the verb which functions as
subject in the active voice). Hence, Semitic verbal morphology is special not only
in having roots, which are consonantal morphemes, but also in morphologically
marking whether the external argument of the verb is a Cause – in the CAUSATIVE
template, or an Agent in the INTENSIVE template. Verbs constructed from the
root with the unmarked functional v are derived by the default SIMPLE template,
the verbalizer vSMPL . In (4), three equi-rooted verbs derived by the three different
templates are shown for the same root ‘ripe/cook’. The contrast between (4b) and
(4cii) again shows that Causes, but not Agents, may be abstract2 :
1 The system contains a lot of noise due to phonological considerations and lexical idiosyncrasy,
which brings about a reversal of the exponents in the environment of many roots. Systematicity is
only guaranteed when the different templates are contrastive – i.e. derive equi-rooted verbs as in
(4) in the text (cf. Doron 2003). But see Sect. 12.5 for a novel environment of systematicity.
2 In the transcription of the examples, the pairs of allophones b-v, k-x, p-f are rendered according
to the Hebraist tradition b-b, k-k, p-p̄. Glosses use the following abbreviations: ACC – accusative
case; ADJ – adjective; CAUS¯ – CAUSATIVE
¯ template; INTNS – INTENSIVE template; MID – middle
voice; PASS – passive voice; SMPL – SIMPLE template; SUPR – superlative.
(4) √bšl ~> ripe/cook

a. Simple template
bašlu ha-tna’im le-heskem ezori kolel
ripened.SMPL the-conditions for-agreement regional comprehensive
‘The conditions have ripened for a comprehensive regional agreement.’W
b. Causative template
… ma še-ke.kol.ha.nir’e hibšil et-ha-‘isqa
… what that-probably ripened.CAUS ACC-the-deal
‘The companies worked together before, which probably cooked the deal.’W
c. Intensive template
i. hu bišel et-ha-‘isqa ha-gdola be.yoter
he cooked.INTNS ACC-the-deal the-big SUPR
He cooked up the biggest deal in the history of Israeli high-tech.’W
ii. *… ma še-ke.kol.ha.nir’e bišel et-ha-‘isqa
… what that-probably cooked.INTNS ACC-the-deal
‘*The companies worked together before, which probably cooked up the deal.’
I now introduce the distinction between Agent and Force formulated by Scott
DeLancey as
. . . the distinction between active (prototypically, moving) participants in the event and
inactive entities which somehow produce their effect simply by being in the right place at
the right time. (DeLancey 1983: 61)
The Agent/Force distinction can be represented through the theoretical tools

developed by Schäfer (2008) and Alexiadou et al. (2015). These authors suggest
that an Agent argument is always introduced by an additional functional Voice
head (additional to the v verbalizer). This ascertains that verbs with Agents can
passivize (since active/passive is a distinction encoded in the Voice head). Example
(4ci) above, with an Agent subject, may indeed passivize as in (5):
(5) ha-‘isqa ha-gdola be.yoter … bušla al.yedey E.M.
the-deal the-big SUPR cooked.INTNS.PASS by E.M.
‘The biggest deal in the history of Israeli high-tech had been cooked up by E.M.’
If no Voice head is selected, the thematic role assigned to the subject by vINTNS is
actually not Agent, but its subtype Force. In such a case, the verb does not passivize.
Parallel facts for vCAUS without a Voice head will be shown later in Sect. 12.5.
(6) a. gal ha -ħom bišel l-o et-ha-móaħ
wave(of) the-heat cooked.INTNS to-him ACC-the-brain
‘The heat-wave cooked his brain.’W
b. *ha-móaħ bušal l-o al.yedey gal ha-ħom
the-brain cooked.INTNS.PASS to-him by wave(of) the-heat
‘His brain was cooked by the heat-wave.’
The examples in (7) and (8) similarly illustrate the difference between Agent and
Force as arguments of the intensive template:
400 E. Doron
(7) Intensive template with Agent

a. ha-cava ha -adom šiħrer et-asirey ha-maħane
the-army the red released.INTNS ACC-inmates(of) the-camp
‘The Red Army released the camp’s inmates.’W
b. asirey ha-maħane šuħreru al.yedey ha-cava ha -adom
inmates(of) the-camp released.INTNS.PASS by the-army the red
‘The camp’s inmates were released by the Red Army.’
(8) Intensive template with Force

a. ha-ħómer šiħrer edim re‘ilim bi.zman ha-apiya
̄
the-substance released.INTNS fumes toxic during the-baking
‘The substance released toxic fumes during the baking.’W
b. *edim re‘ilim šuħreru al.yedey ha-ħómer bi.zman ha -a p̄ iya
fumes toxic released.INTNS.PASS by the-substance during the-baking
‘Toxic fumes were released by the substance while baking.’
The distinction between Agent and Force solves a puzzle which I left open
in Doron (2011), where I noted a particular subclass of psych verbs which was
systematically in the INTENSIVE template. Yet the existence of psych verbs in the
INTENSIVE template is very mysterious under the assumption that this template
is the exponent of vINTNS , which introduces an Agent argument. How is this
consistent with the fact that psych verbs are non-agentive stative verbs? But if
vINTNS assigns the different role of Force, which is compatible with stative verbs,
then the INTENSIVE verbalizer is actually compatible with the semantics of psych
verbs. The lack of a Voice head in psych verbs moreover explains why Hebrew
psych verbs do not passivize.
12.3 Relational Psych Verbs
Psych verbs have been taken to denote a relation between two arguments, the
experiencer (Exp) and the argument toward which emotion or evaluation is directed
by Exp, entitled “object of emotion” in the philosophical literature (Kenny 1963 and
Nissenbaum 1985) or Target of Emotion/Subject Matter of Emotion (T/SM) in the
linguistic literature (Pesetsky 1995). Grammatically, it appears that either argument
may be subject. Exp may be the subject, as illustrated in (9a), and in this case the
psych verb is called a SubjExp verb. Or Exp is the object of the verb, as in (9b), and
such a verb is called an ObjExp verb:
(9) a. SubjExp verb
We admire science.
Mary didn’t care for the play.
b. ObjExp verb
Science fascinates us.
The play didn’t appeal to Mary. (Pesetsky 1995:52)
This section discusses relational psych verbs such as the ones in (9), psych verbs
which indeed denote a relation between Exp and the target of emotion. I leave until
Sect. 12.5 the discussion of property psych verbs – verbs which denote a one-place
property of Exp which is not directed toward a second argument.
The examples in (9) illustrate the famous “linking problem” presented by psych
verbs. Whereas for other verbs the thematic role of an argument relative to other
arguments determines its grammatical function, here we seem to find an alternation
in the alignment of thematic roles to grammatical functions. Psych verbs appear
to allow an alternation in the alignment of Exp and T/SM to their grammatical
function – subject vs. object.
I will argue that the alternation is only apparent. Of the roles Exp and T/SM, Exp
always receives the function of subject, as in (9a). Where Exp is object, as in (9b), it
is because the other argument has the thematic argument of Force. This conception
is in accordance with the view voiced by Scott DeLancey:
A situation in which a person experiences some cognitive or emotional state can be
construed . . . as a state which the individual enters into, parallel to sick or grown-up; or as
a force which enters into the individual, as a disease. The first of these is grammaticalized
as dative-subject predicates like like; the second is grammaticalized as a species of change-
of-state predicate like please. (DeLancey 2001)
The same solution for the linking problem is also valid for locative verbs.
12.3.1 Relational SubjExp Psych Verbs
What characterizes relational SubjExp verbs is that both Exp and T/SM are
arguments of the root, i.e. these verbs have binary roots, which denote locative
relations. These verbs are locative in the sense that they describe feelings and
emotions as located within Exp: “Experiencers are mental locations (containers)
in which the mental state resides” (Landau 2010:11). A similar point was made in
DeLancey (2001). Hence the Exp role parallels the role of Location in locative verbs.
The thematic roles of the arguments are determined by “inherent prepositions”
which surface in positions where the argument lacks grammatical case such as
nominative.3
The Hebrew examples in (10) are in the simple template and do not passivize.
Their adjectival passives in (11) reveal the inherent locative preposition underlyingly
attached to the locative experiencer (Doron 2003):
3 Inherent
prepositions are the assigners of the so-called “inherent case” which differs from
“grammatical case” in being dependent on thematic roles (Emonds 1985).
402 E. Doron
(10) Relational psych verbs in the SIMPLE template

a. ha-talmid ahav et-ha-ši‘ur
the student loved.SMPL ACC-the-class
b. ha-talmid sana et-ha-ši‘ur
the student hated.SMPL ACC-the-class
c. ha-talmid zaḵar et-ha-ši‘ur

the student remembered.SMPL ACC-the-class
d. ha-talmid ma’as b-a-ši‘ur
the student loathed.SMPL at-the-class
(11) Locative preposition surfaces on non-nominative Exp

a. ha-ši‘ur ahuv al ha-talmid
the-class love.SMPL.ADJ.PASS on the-student
‘The class is pleasing to the student.’
b. ha-ši‘ur sanu al ha -talmid
the-class hate.SMPL.ADJ.PASS on the-student
‘The class is detestable to the student.’
c. ha-ši‘ur zakur
- l-a-talmid
the-class remember.SMPL.ADJ.PASS to-the-student
‘The class is borne in the student’s mind.’
d. ha-ši‘ur ma’us al ha-talmid
the-class loathe.SMPL.ADJ.PASS on the-student
‘The class is loathsome to the student.’
The same distribution is found with physical locative verbs:

(12) Relational locative verbs in the SIMPLE template
a. ha-qarqa‘ sapg - a ħomer radioaqtivi
the-ground absorbed.SMPL substance radioactive
b. ha-smika -
atpa et-ha-tinoq
the-blanket wrap.SMPL ACC-the-baby
(13) Locative preposition surfaces on non-nominative Loc

a. ħomer radioaqtivi adayin sapu - g b-a-qarqa‘
substance radioactive still absorb.SMPL.ADJ.PASS in-the-ground
‘Radioactive substance is still soaked up in the ground.’W
b. ha-tinoq atup- b-a-smika
the-baby wrap.SMPL.ADJ.PASS in-the-blanket
‘The baby is wrapped up in the blanket.’
12.3.2 Relational ObjExp Psych Verbs
The relative prominence of the two arguments of relational psych verbs is reversed
in verbs which present the T/SM as a Force. In Hebrew, the functional verbalizer
vINTNS , when it does not introduce an additional Agent argument through the
use of a Voice head, assigns the thematic role of Force to the active participant
which forcefully penetrates the location, a construal already described by DeLancey
(2001). DeLancey speaks of a “Force which enters the individual”. The following
are Hebrew examples of such verbs:4
(14) a. ha.balšanut inyena ota b. ha.i.cédeq qomem ota
linguistics interested.INTNS her injustice revolted.INTNS her
c. ha-maxaze ye’eš ota d. ha-nosé riteq ota
the-show dismayed.INTNS her the-topic riveted.INTNS her
e. ha-signon ikzeb ota f. ha-nosé iyef ota
the-style disappointed.INTNS her the-topic tired.INTNS her
g. ha-ši‘ur ši‘amem ota h. ha-tašlum rica ota
the-class bored.INTNS her the-payment gratified.INTNS her
i. šmo siqren ota j. ha-macab dike ota
his.name intrigued.INTNS her the-situation depressed.INTNS her
k. ha-šókolad pita ota l. ha-mišpat iyem aleyha
the-chocolate seduced.INTNS her the-trial threatened.INTNS her
As follows from the lack of a Voice head, the verbs in (14) do not passivize, as
already noticed by Landau (2010: 60–63).
(15) a. *hi unyena al.yedey ha.balšanut
she interested.INTNS.PASS by linguistics
b. *hi qumema al.yedey ha.i.cédeq
she revolted.INTNS.PASS by injustice
Some roots allow a Voice head to be added to the derivation after all. In such a
case, an agentive, non-psych action verb is derived:
(16) a. yedidah pita ota b. ha-biryon iyem aleyha
her.friend seduced.INTNS ACC-her the-thug threatened.INTNS her
‘Her friend seduced her.’ ‘The thug threatened her.’
For such verbs, passive is possible. The by-phrase then denotes either the Agent,
or an instrument deployed by an implicit Agent:
4 Insome circumstances, its is actually a CAUSAITVE template exponent which surfaces as the
exponent of the functional head vINTNS . An example is hirtía‘deter.CAUS. The converse is true too:
rigeš ‘excite.INTNS actually belongs to verbs with vCAUS . As already mentioned, such occasional
lack of transparency in the correspondence between the syntactic make-up of the verb and its
exponent template is to be expected in the system.
404 E. Doron
(17) a. hi puteta al.yedey yedid.ah / al.yedey ha-šókolad

she seduced.INTNS.PASS by her.friend / by.means.of the-chocolate
b. hi uyma al.yedey ha-biryon / al.yedey ha-xipus
she threatened.INTNS.PASS by the-thug / by.means.of the-search
What is striking in all the examples in (14) is that if the subject is not Agentive,
then it is always the argument toward which the mental state is targeted, since the
latter is an argument of the relational psych verb.
Parallel locative verbs are shown in (18):
(18) Relational locative verbs in the INTENSIVE template (Force subject)
a. ha.mayim mil’u et-ha-‘émeq
water filled.INTNS ACC-the-valley
b. ha.šéleg kisa et-ha-har
snow covered.INTNS ACC-the-hill
c. ha.krazot qištu et-ha-‘ir
posters decorated.INTNS ACC-the-town
d. qtoret ha -mor bisma et-beit.ha.miqdaš
incense(of) the-myrrh perfumed.INTNS ACC-the.temple
The structure of relational psych verbs is schematically shown here:

(19) a. Relational SubjExp b. Relational ObjExp
SIMPLE INTENSIVE
vSMPL vINTNS
vSMPL √ vINTNS √
T/SM √ Force √
Exp √ Exp √
ahav ‘love’ inyen ‘interest’
12.4 Causativization of Relational Psych Verbs
Both SubjExp and ObjExp relational psych (and locative) verbs can be causativized,
i.e. acquire a new subject, the Cause argument. Adding the Cause argument requires
“demotion” of the verb’s original subject (Keenan & Comrie 1977). The original
subject is Exp in the case of SubjExp psych verbs, and Force in the case of ObjExp
psych verbs. Demotion consists of allowing the demoted argument to surface with
its inherent preposition.
12.4.1 Causativization of Relational SubjExp Verbs
Relational SubjExp verbs were introduced in Sect. 12.3.1 above, where it was also
shown that Exp’s inherent prepositions were ‘al ‘on’ and l- ‘to’. These indeed
surface when we add the Cause argument in (20), whereas the T/SM remains the
direct object of the verb:
(20) Causative relational SubjExp verbs
a. ha-more he’ehiv al ha-talmid et-ha-ši‘ur
the teacher loved.CAUS on the student ACC-the class
‘The teacher made the student love the class.’
b. ha-more hisni’ al ha-talmid et-ha-ši‘ur
the teacher hated.CAUS on the-student ACC-the-class
‘The teacher made the student hate the class.’
c. ha-more hizkir l-a-talmid et-ha-ši‘ur
the teacher remembered.CAUS to-the-student ACC-the-class
‘The teacher reminded the student of the class.’
d. ha-more him’is al ha-talmid et-ha-ši‘ur
the teacher loathed.CAUS on the-student ACC-the class
‘The teacher made the student loathe the class.’
The same distribution is found with physical locative verbs, with the inherent
preposition b- ‘in’:
(21) Causative locative verbs
a. hem hispigu ħomer radioaqtivi b -a-qarqa‘
they absorbed.CAUS substance radioactive in-the-ground
‘They made radioactive substance seep into the ground.’
-
b. hem atpu et-ha-tinoq b-a-smika
they enveloped.SMPL ACC-the-baby in-the-blanket
‘They enveloped the baby in the blanket.’
12.4.2 Causativization of ObjExp Verbs
The most surprising property of causative ObjExp verbs is their systematic INTEN-
SIVE form, the same as that of the basic ObjExp verb. Yet the additional argument
is interpreted as a Cause despite the INTENSIVE morphology. This is due to the fact
that it is the local functional head (vINTNS in this case) which determines the spellout
of the root, rather than the higher functional head (vCAUS ). The relevant functional
heads are shown in the causatives structure in (27b) below.
406 E. Doron
In order to uncover the different inherent prepositions, which introduce the

Force argument, we check the corresponding middle-voice verbs.5 The choice of
prepositions marking Force depends on the root, and varies quite a bit: b- ‘in’,
néged ‘against’, me/mi ‘from’, klapey ‘towards’, l- ‘to’. The same point was made
for English in Levin (1993:190).
(22) a. ha.balšanut inyena ota a’. hi hit‘anyena be-balšanut
linguistics interested.INTNS her she interested.INTNS.MID in-linguistics
b. ha.i.cédeq qomem ota b’. hi hitqomema néged ha.i.cedeq
injustice revolted.INTNS her she revolted.INTNS.MID against injustice
c. ha-signon ikzeb
- - ota c’. hi hit’akzeba
- - me ha-signon
the-style disappointed.INTNS her she disappointed.INTNS.MID from the-style
d. ha-ši‘ur ši‘amem ota d’. hi hišta‘amema me ha-ši‘ur
the-class bored.INTNS her she bored.INTNS.MID from the-class
e. šmo siqren ota e’. hi histaqrena klapey šmo
his.name intrigued.INTNS her she intrigued.INTNS.MID towards his.name
f. ha-šókolad pita ota f’. hi hitpateta l-a.šókolad
the-chocolate seduced.INTNS her she tempted.INTNS.MID to-chocolate
We indeed find the same inherent preposition surfacing on the Force argument
when a Cause is added to the verb:
(23) a. ha-marce inyen ota be-balšanut
the lecturer interested.INTNS her in-linguistics
b. ha-séret qomem ota néged ha.i.cédeq
the film revolted.INTNS her against injustice
c. hitnahaguta -
iypa oto mim.éna
her.behaviour tired.INTNS her him from.her
‘Her behaviour made him tired of her.’
d. hitnahaguta ye’aša oto mim.éna
her.behaviour dismayed.INTNS him from.her
‘Her behaviour made him dismayed at her.’
e. ha-marce riteq ota l-a-nosé
the-lecturer riveted.INTNS her to-the-topic
f. ‘azibata
- ikzeba
- - oto mim.éna
her.leaving disappointed.INTNS him from.her
‘Her leaving made him disappointed in her.’
g. šmo siqren ota klapav
his.name intrigued.INTNS her towards.him
‘His name made her intrigued about him.’
With locative verbs, the inherent preposition is always be- ‘with’:
5 Onthe middle voice as a non-active voice different from the passive voice see Doron (2003),
Alexiadou & Doron (2012).
(24) a. ha-mayim mil’u et-ha-‘émeq a’. ha-‘émeq hitmala be-mayim

water filled.INTNS ACC-the-valey the-valey filled.INTNS.MID with-water
b. ha-šéleg kisa et-ha-har b’. ha-har hitkasa be-šéleg
snow covered.INTNS ACC-the hill the-hill covered.INTNS.MID with-snow
c. ha-krazot qištu et-ha-‘ir c’. ha-‘ir hitqašta be-krazot
posters decorated.INTNS ACC-the town the-town decorated.INTNS.MID with-posters
d. ha-qtoret bisma et-ha-miqdaš d’ ha-miqdaš hitbasem be-qtoret
the-incense perfumed.INTNS the-temple the-temple perfumed.INTNS.MID with-incense
(25) a. ha-sufa mil’a et-ha-breka

- be-mayim
the-storm filled.INTNS ACC-the pool with-water
b. ha-sufa kista et-ha-har be-šéleg
the-storm covered.INTNS ACC-the-hill with-snow
c. hu qišet et-ha-‘ir be-krazot
he decorated.INTNS ACC-the-town with-posters
d. ha-kohanim bismu et-beit.ha.miqdaš be-qtoret ha -mor
the-priests perfumed.INTNS ACC-the.temple with-incense(of) the-myrrh
The basic structures in (26) repeat (19), and (27) are the causative versions:
(26) a. Relational SubjExp b. Relational ObjExp
SIMPLE INTENSIVE
vSMPL vINTNS
vSMPL √ vINTNS √
T/SM √ Force √
Exp √ Exp √
ahav ‘love’ inyen ‘interest’
(27) a. Causativized Relational SubjExp b. Causativized Relational ObjExp

CAUS INTENSIVE
v v
Cause v Cause v
vCAUS √ vCAUS √
T/SM √ vINTNS √
PEXP √ PFORCE √
he’ehiv ‘cause to love’
Exp PEXP Force PFORCE Exp √
inyen ‘make
interested in’
408 E. Doron
12.5 Causative Property Psych Verbs
The ObjExp verbs below differ in several respects from the relational ObjExp verbs,
which have the structure (26b) above. First, they are typically in the CAUSATIVE
template, whereas the relational ObjExp verbs are in the INTENSIVE template. I
will argue that the CAUSATIVE template of these verbs indicates that their subject
has the thematic role of Cause rather than T/SM or Force. These psych verbs do not
denote a relation between Exp and the target of the mental state, but a one-place
property of Exp brought about by a Cause.
(28) Causative property ObjExp verbs
a. ha-ma’amar hirgiz ota b. ha-ma’amar hik‘is
- ota
the-article angered.CAUS her the-article annoyed.CAUS her
c. ha-televízya hip-ħida ota d. ha-televízya hid’iga ota
the-TV frightened.CAUS her the-TV worried.CAUS her
e. ha-doħ heħerid ota f. -
ha-siyur hip‘im ota
the-report appalled.CAUS her the-trip thrilled.CAUS her
g. ha-nisuy hidhim ota h. ha-sipur hibhil ota
the-experiment astounded.CAUS her the-story alarmed.CAUS her
i. ha-sipur hib‘it ota j. ha-maħaze he‘elib ota
the-story horrified.CAUS her the-show insulted.CAUS her
k. ha-macab- hisbía‘ et.recon.a l. ha-maħaze hiršim ota
the-situation satisfied.CAUS her the-show impressed.CAUS her
m. ha-maħaze hišpil ota n. ha-ma’amar hip-tía‘ ota
the-show humiliated.CAUS her the-article surprised.CAUS her
o. ha-sipur hitrid ota p. ha-maxaze hiqsim ota
the-story bothered.CAUS her the-show charmed.CAUS her
q. ha-nose hik’ib la r. ha-sipur his‘ir ota
the-topic distress.CAUS her the story agitated.CAUS her
s. ha-nose hip-li ota t. ha-maxaze he‘esiq ota
the-topic amazed.CAUS her the-show preoccupied.CAUS her
u. ha-nose hitrip- ota v. ha-dox hitmía ota
the topic incensed.CAUS her the report puzzled.CAUS her
w. ha-nisayon he‘ecim ota x. ha-mar’e hig‘il ota
the-experience empowered.CAUS her he-sight disgusted.CAUS her
y. ha-kanábis hirgía‘ ota z. ha-mar’e hid’ib ota
the-Cannabis calmed.CAUS her the-sight hurt.CAUS her
As was shown in (15) above, INTENSIVE relational ObjExp do not passivize.

If a passive form exists, it is actually the passive of a corresponding agentive, non-
psych action verb, as in (16)–(17). This may happen with CAUSATIVE ObjExp psych
verbs too, but here, most passive forms are actually exponents of SubjExp middle-
voice verbs (Landau 2010: 62). Since the CAUSATIVE template has no middle-voice
exponent, in some cases the passive-voice exponent serves as a middle-voice verb
(Doron 2008). The preposition in this case is me ‘from’ rather than al.yedey ‘by’.
(29) a. yedidah / ha-macav hip-tía‘ ota

her. friend / the-situation surprised.CAUS her
b. hi hup-te‘a al.yedey yedidah / me ha-macav
she surprised.INTNS.PASS by her.friend / from the-situation
(30) a. yedidah / ófen.ha.dibur šelo hišpil ota

her.friend/ speech.style his humiliated.CAUS her
b. hi hušpela al.yedey yedidah / me ófen.ha.dibur šelo
she humiliated.CAUS.PASS by her.friend / from speech.style his
But most SubExp verbs corresponding to the CAUSATIVE ObjExp verbs in

(28) are SIMPLE active-voice verbs. When we search for the relevant inherent
preposition, we mostly find causative prepositions (PCAUS ): mi/me ‘from/of’, al
‘for, on account of, about’. This is very different from the various prepositions we
found in (22) above for the SubjExp INTENSIVE.MID template verbs (be- ‘in’, néged
‘agaist’, me/mi ‘from’, klapey ‘towards’, l- ‘to’).
(31) a. hi ragza al ha-šħitut
she angered.SMPL at the-corruption
b. hi ka‘asa al ha-ha’ašamot
she annoyed.SMPL at the-accusations
c. hi paħada me ha.mávet
she feared.SMPL from death
d. hi da’aga l-a-yalda
she worried.SMPL for-the-girl
e. hi tamha al ha-toca’ot
she puzzled.SMPL for the-results
f. hi ħarda me ha-macab-
she apalled.SMPL from the-situation
g. hi nip-‘ama me ha-eru‘im
she thrilled.SMPL.MID from the-events
h. hi hitpal’a al ha-eru‘im
she amazed.INTNS.MID from the-events
i. hi sab‘a.racon me ha-macav
she satisfied.SMPL from the-situation
j. hi ka’aba al ha-macav
she distressed.SMPL from the-situation
k. hi da’aba al ha-macav
she hurt.SMPL from the-situation
410 E. Doron
The following examples show that the Hebrew causative prepositions PCAUS are
indeed mi/me ‘from/of’, al ‘for, on account of, about’, quite independently of psych
verbs. The examples in (32) below are all from the web.
(32) a. hu hištolel mi zá‘am
he went-wild from rage
b. ha-débeq
- namas me ha-ħom
the-glue melted from the-heat
c. hem he‘eníšu ota al de‘otéha
they punished her for opinions.her
d. libam gas ba al ki he‘éza le-harim roš me-ašpatot
their.heart rough at.her for that she.dared to-raise head from-dumps
‘They disparage her for having dared to raise from the dump.’
Here too, the alternation in psych verbs (between the CAUSATIVE template and
PCAUS ) is found with locative verbs (Doron 2005):
(33) a. ha-‘ec hišir et-ha-‘alim
the-tree shed.CAUS ACC-the-leaves
b. ha-kéleb- hidip- réaħ ra‘
the-dog emanated.CAUS smell foul
‘The dog emitted a foul smell.’
(34) a. ha-‘alim našru me ha-‘ec

the-leaves shed.SMPL from the-tree
b. réaħ ra‘ nadap- me ha-kéleb
smell foul emanated.SMPL from the-dog
An additional contrast between the INTENSIVE relational verbs and the

CAUSATIVE property verbs is attested for the nominalized versions of these verbs.
Ahdout (2016) shows that the nominalization of INTENSIVE psych verbs can be
stative (35a), whereas the nominalization of the CAUSATIVE verbs is dynamic only
(35b).
(35) a. ha-ye’uš šelahem me ha-séret
the-dismay.INTNS theirs from the-film
‘their dismay at the movie.’
b. ha-ha‘alaba šelahem al.yedey ha-bamay / *me ha-séret
the-insulting.CAUS theirs by the-director / *from the-film
Their insulting by the director’s / *the film’s
I refer the reader to Ahdout (2016) for a full account of (35), but here suffice it to
say that this contrast follows from the difference in structure between INTENSIVE
relational verbs (26b) and CAUSATIVE property verbs (36b). The relation denoted by
the root in (35a) may remain stative when nominalized, whereas the nominalization
of a causative relation is interpreted as dynamic (Grimshaw 1990).
I suggest the following structures, where the Cause role is assigned in two
different ways, but in neither case is it an argument of root:
(36) a. Property SubjExp b. Property ObjExp

SIMPLE CAUSATIVE
vSMPL vCAUS
PCAUS vSMPL Cause vCAUS
Cause PCAUS vSMPL √ vCAUS √
Exp √ √
ka‘as ‘be angry’ hik- ‘is ‘annoy’
Pesetsky (1995) discusses in detail the semantic differences between SubjExp

and ObjExp versions of parallel English verbs:
(37) a. SubjExp verb
Bill was angry at the article in the Times.
b. ObjExp verb with Cause subject
The article in the Times angered Bill. (Pesetsky 1995: 56)
(37b) has a reading that (37a) does not have, where Bill does not find anything
objectionable about the article in the Times, he thinks it is splendid. His anger is not
directed at the article, but maybe he is angry at the government for the corruption
revealed by the article. (37a) cannot be interpreted in this way, but only means that
Bill finds the article itself objectionable in some respect. The same holds in (38).
John worried about the television set.
b. ObjExp verb with Cause subject
The television set worried John. (Pesetsky 1995: 57)
As in example (37) above, (38b) has a reading that (38a) does not have, where
John does not worry about the television set, but where he worries about something
else, and his worrying is caused by the television set. For example, because the TV
set is not in its usual place, he may worry that his baby son pushed it and got stuck
underneath. Thus the television set is the Cause of John’s worry in (38b). He is
not worrying about the TV set, but because of it. In (38a), on the other hand, the
television set is the object of John’s worry. Similar contrasts were also noted in Croft
(1993).
Pesetsky attributes this semantic difference to a split between the roles of T/SM
and Cause. In the (a) examples, the non-Exp argument is a T/SM, whereas in the
(b) examples – it is a Cause. This split in thematic roles explains the semantic
differences, but generates a puzzle, which Pesetsky called the “T/SM restriction”.
Psych verbs can take a T/SM argument as in (37a) and (38a), and a Cause argument
as in (37b) and (38b), but not both in the same sentence, as shown by (39):
412 E. Doron
(39) a. *The article in the Times angered Bill at the government.

b. *The television set worried John about the whereabouts of his baby son.
The restriction is clearly not semantic, since the three arguments can be expressed
together in a periphrastic construction, as in (40):
(40) .a.. The article in the Times caused Bill to be angry at the government.
.b. The television set caused John to worry about the whereabouts of his baby son.
In Hebrew, we find that CAUSATIVE psych verbs abide by the T/SM restriction:
(41) a. *ha-šmu‘ot hirgizu ota al ha -šħitut
the rumours angered.CAUS her at the corruption
b. *ha-ne’um hik‘is ota al ha-ha’ašamot
the speech annoyed.CAUS her at the-accusations
c. *ha-ma’amar hip-ħid ota me-ha-mávet
the article frightened.CAUS her from-death
d. *ha-ma’amar hip-‘im ota me ha-eru‘im
the article excited.CAUS her from the events
e. *ha-dox heħerid ota me-ha-macav
the report appalled.CAUS her from the situation
f. *ha-sipur hibhil ota me-ha-‘alila
the story scared.CAUS her from-the-plot
g. *ha-tipul hirgía‘ ota me-ha-kanábis
the treatment calmed.CAUS her from-the-Cannabis
Nevertheless, I propose to reject the “T/SM restriction”. It does not hold for
relational ObjExp verbs, as we saw above in (23). Hence, a Cause is in general
compatible with a T/SM or Force argument. Instead, the ungrammaticality of the
examples in (39) and (41) can simply be attributed to the fact that both arguments
are assigned the same thematic role of Cause, once by the verbalizer vCAUS and once
by PCAUS . This contradicts the well-known requirement that in a non-periphrastic
construction, each thematic role can only be assigned once.
But what accounts for the semantic differences uncovered by Pesetsky regarding
(37) and (38)? I attribute it to the difference in the transparency of the two
environments where Cause is assigned. Vendler (1962) and Davidson (1967),
following a long tradition, argue that causality is transparent:
If it was a drying she gave herself with a coarse towel on the beach at noon that caused
those awful splotches to appear on Flora’s skin, then it was a drying she gave herself that
did it; we may also conclude that it was something that happened on the beach, something
that took place at noon, and something that was done with a towel, that caused the tragedy.
(Davidson 1967: 698)
Yet an environment with PCAUS is not transparent, as shown in (42), it seems to

involve explanation beyond mere causation:
(42) a. Flora contracted those awful splotches from drying herself with a coarse towel.
b. #Flora contracted those awful splotches from drying herself with a towel.
The contrast between the intensionality of the verb complement and the exten-
sionality of the verb subject was also noted by Levin & Grafmiller (2013). They
note a contrast in acceptability between a SubjExp verb, which they found in their
corpus, and the corresponding unacceptable ObjExp verb:
Did you fear a negative response from fans?
b. ObjExp verb
??Did a negative response from fans frighten you? (Levin & Grafmiller (2013: ex. 8)
. . . the frighten variant [43b] can only be understood as presupposing that a negative
response has in fact happened, while the fear example [43a] carries no such presupposition.
In (43a) the experiencer fears merely the possibility of something happening. That is, there
was no specific event that happened to cause him or her to become afraid . . . (Levin &
Grafmiller 2013: 24)
Another demonstration of the lack of intensionality of the causative subject is its

resistance to logophors. Logophoric elements can appear in causal environments
(Charnavel 2018), but when they do, they reflect a mental representation by an
Exp. This is natural for relational psych verbs as in (44a), but much less natural
for causative property psych verbs as in (44b), where the subject is an environment,
which does not represent the perspective of Exp:
(44) a. ha-biqóret al acma qomema ota
the-criticism of herself revolted.INTNS her
b. ?ha-biqóret al acma hišpila ota
the-criticism of herself humiliated.CAUS her
The following table summarizes the contrasts uncovered in the present section
between relational and property ObjExp verbs:
Relational psych verbs Property psych verbs

ObjExp verbs INTENSIVE template CAUSATIVE template
Force subject Cause subject
violate T/SM restriction uphold T/SM restriction
allow logophors disallow logophors
Passive is Agentive Passive expones the middle voice
Stative nominalization Dynamic nominalization
Corresponding INTENSIVE middle voice SIMPL active voice
SubjExp verbs
Force argument with varied P Cause argument exclusively with PCAUS
I find it striking that the distinction between relational and property psych verbs
systematically corresponds to a form distinction between the templates, which
derive these verbs in Hebrew. This form distinction is not an accidental phonological
414 E. Doron
fact. I must admit that so far I believed that finding equi-rooted verbs was the only
way to demonstrate the systematic semantic contrast encoded by the templates. But,
here is a novel environment which demonstrates this systematicity, though the verbs
are not equi-rooted. We have two subclasses of ObjExp verbs which differ in the
semantics of their subject argument: Force vs. Cause. What could be more natural
than deriving these verbs by the templates which signify these meanings!
12.6 Conclusion
One construal of psych verbs is that of a locative state: a mental representation of

the T/SM is located within the experiencer’s mind (DeLancey 1983; Landau 2010).
Two additional psych verb construals were proposed by DeLancey (1983). Under
the second construal, the psych verb describes a Force entering the Experiencer’s
consciousness. This is represented in Hebrew by the INTENSIVE template, which
typically describes agentive dynamic events, but here reflects the presence of
inactive but effective Force. This construal reflects the involvement of Force, not
the involvement of change. Under the third construal, the psych verb denotes a state
of mind of the experiencer, which may be viewed as caused, and is expressed in
Hebrew by dedicated causative morphology, the CAUSATIVE template.
The difference between the construals shows that psych verbs do not after all
suffer from a “linking problem”: the thematic role of Force which is assigned to the
subject of psych verbs differs from the thematic role of T/SM assigned to objects.
The categories of Cause and Force play a crucial role in the representation of
verbs. The present study has shown this for psych verbs. The centrality of these
categories is demonstrated by the fact that they are encoded in the morphosyntax of
Hebrew verbs: Some psych relations form a subtype of the causative relation and
are expressed in the CAUSATIVE template, whereas others describe the presence of
a force, and are expressed in the INTENSIVE template (which surfaces even when
an eventual Cause is added).
The existence of psych verbs in the INTENSIVE template is mysterious under the
assumption that this template introduces an Agent argument, whereas psych verbs
are non-agentive stative verbs. But if the INTENSIVE template is actually associated
with the different role of Force, which is compatible with stative verbs, then the
INTENSIVE verbalizer is actually compatible with the semantics of psych verbs.
The present study has also shown that psych verbs, though describing the mental
rather than the physical domain, actually do not lexicalize new types of thematic
roles. Psych verbs can be construed as different relations, but the participant roles in
these relations are parallel to the ones found in the physical realm of locative verbs.
Acknowledgements The paper has so far existed as a lecture handout from the 2011 Roots III
workshop in Jerusalem. I am grateful to the organizers of the 2017 Perspectives on Causation
workshop in Jerusalem for giving me the opportunity to finalize the written version. This research
has received funding from the Israel Science Foundation grant No. 1296/16 and from the European
Research Council H2020 Framework Programme No. 741360.
References
Ahdout, O. (2016). The syntax-semantics interface in Hebrew psychological nominalizations.

Hebrew University MA thesis.
Aimar, S. (2018). Causal claims and the ontology of causation. UCL ms.
and middle. Journal of Linguistics, 48(1), 1–34.
Alexiadou, A., & Iordăchioaia, G. (2014). The psych causative alternation. Lingua, 148, 53–79.
alternations. Oxford: OUP.
Arad, M. (1999). What counts as a class? The case of psych-verbs. MITWPL, 35, 1–23.
Belletti, A., & Rizzi, L. (1988). Psych-verbs and theta-theory. Natural Language & Linguistic
Theory, 6, 291–352.
Charnavel, I. (2018). Perspectives in causal clauses. Natural Language & Linguistic Theory, 36,
1–36.
Copley, B., & Harley, H. (2015). A force-theoretic framework for event structure. Linguistics and
Philosophy, 38(2), 103–158.
Copley, B., Wolff, P., & Shepard, J. (2016). Force interaction in the expression of causation. In
S. D’Antonio, M. Moroney & C. Rose Little (Eds.), Proceedings of the 25th semantics and
linguistic theory conference, pp. 433–451.
Croft, W. (1993). Case marking and the semantics of mental verbs. In J. Pustejovsky (Ed.),
Semantics and the lexicon (pp. 55–72). Dordrecht: Kluwer.
DeLancey, S. (1983). Agentivity and causation: Data from Newari. BLS, 9, 54–63.
DeLancey, S. (2001). LSA Summer Institute, UC Santa Barbara, 2001. Lecture, 8. https://
darkwing.uoregon.edu/~delancey/sb/LECT7-8.htm
Doron, E. (1999). The semantics of transitivity alternations. In P. Dekker (Ed.), Proceedings
of the twelfth Amsterdam colloquium (pp. 103–108). Amsterdam: Universiteit van Amster-
dam/Institute for Logic, Language and Computation.
Doron, E. (2003). Agency and voice: The semantics of the semitic templates. Natural Language
Semantics, 11(1), 1–67.
Doron, E. (2005). The aspect of agency. In N. Erteschik-Shir & T. Rapoport (Eds.), The syntax of
aspect (pp. 154–173). Oxford: Oxford University Press.
Doron, E. (2008). The contribution of the template to verb meaning. In G. Hatav (Ed.), Modern
linguistics of Hebrew (pp. 57–88). Jerusalem: Magnes Press. [in Hebrew].
Doron, E. (2011). The causative component of psych verbs. Paper Presented at the Roots III
Workshop, Jerusalem, June 2011. Also presented at Universitat Pompeu Fabra, Barcelona, May
2012.
Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht: Reidel.
Emonds, J. (1985). A unified theory of syntactic categories. Dordrecht: Foris.
Fillmore, C. J. (1968). The case for case. In E. Bach & R. T. Harms (Eds.), Universals in linguistic
theory (pp. 1–88). New York: Holt, Rinehart & Winston.
Folli, R., & Harley, H. (2007). Causation, obligation, and argument structure: On the nature of little
v. Linguistic Inquiry, 38(2), 197–238.
Jackendoff, R. (1972). Semantic interpretation in generative grammar. Cambridge, MA: MIT
Press.
Kastner, I. (2018). Templatic morphology as an emergent property: Roots and functional heads in
Hebrew. Natural Language & Linguistic Theory, 37, 1–49.
Keenan, E. L., & Comrie, B. (1977). Noun phrase accessibility and universal grammar. Linguistic
Inquiry, 8(1), 63–99.
416 E. Doron
Kenny, A. (1963). Action, emotion and will. London: Routledge.

Kratzer, A. (1996). Severing the external argument from its verb. In J. Rooryck & L. A. Zaring
(Eds.), Phrase structure and the lexicon (pp. 109–137). Dordrecht: Kluwer.
Landau, I. (2010). The locative syntax of experiencers. Cambridge, MA: MIT Press.
Levin, B., & Grafmiller, J. (2013). Do you always fear what frightens you? In T. H. King & V. de
Paiva (Eds.), From quirky case to representing space (pp. 21–32). Stanford: CSLI Publications.
Levin, B., & Hovav, M. R. (1995). Unaccusativity. Cambridge, MA: MIT Press.
Marín, R., & McNally, L. (2011). Inchoativity, change of state, and telicity: Evidence from Spanish
reflexive psychological verbs. Natural Language & Linguistic Theory, 29(2), 467–502.
McGinnis, M. (2000). Event heads and the distribution of psych-roots. In A. Williams & E. Kaiser
(Eds.), Current work in linguistics: University of Pennsylvania working papers in linguistics
6.3 (pp. 107–144). Philadelphia: University of Pennsylvania.
Marelj, & T. Siloni (Eds.), The Theta system: Argument structure at the Interface (pp. 20–51).
Oxford: OUP.
Nissenbaum, H. (1985). Emotion and focus. Stanford: CSLI Publications.
Parsons, T. (1990). Events in the semantics of English: A study of subatomic semantics. Cambridge,
MA: MIT Press.
Pesetsky, D. (1995). Zero syntax: Experiencers and cascades. Cambridge, MA: MIT Press.
Reinhart, T. (2002). The Theta system – An overview. Theoretical Linguistics, 28, 229–290.
Schäfer, F. (2008). The syntax of (anti-) causatives. External arguments in change-of-state contexts.
Talmy, L. (2000). Toward a cognitive semantics. Cambridge, MA: MIT press.
Vendler, Z. (1962). Effects, results and consequences. In R. J. Butler (Ed.), Analytic philosophy
(pp. 1–15). New York: Barnes & Noble.
Chapter 13
Linguistic Perspectives in Causation
Isabelle Charnavel
Abstract The relation of causation is a mental construct that must be established

by a reasoning individual. The aim of this chapter is to show that this property
of causation is reflected in at least some linguistic structures, namely in adjunct
causal clauses. Specifically, the behavior of English because-clauses and since-
clauses supports the hypothesis that causal clauses introduce a judge argument that
is syntactically represented. The main argument – based on data experimentally
collected – relies on contrasts in the availability and interpretation of logophoric
anaphors in causal clauses.
Keywords Adjunct clause · Causal clause · Because/since · Logophoricity ·

Perspective · Attitude · Anaphor/reflexive · Binding
Whatever relation the notion of causation is assumed to involve (e.g. regularity or

counterfactuality, see Lewis 1973, i.a.), it crucially requires a mental agent: the
relation between causes and effects is a mental construct that must be established
by a reasoning individual. The aim of this paper is to show that this characteristic
of causation is reflected in (at least some) linguistic structures expressing causality.
Specifically, we will see that the syntactic and semantic properties of adjunct causal
clauses reveal that they are relativized to a judge that is syntactically represented.
This chapter is a condensed version of an article recently appeared in Natural Language and
Linguistic Theory (Charnavel 2019a): it presents the same analysis, but incorporates new examples
and new quantitative results shown in the Appendix.
The following standard abbreviations are used in the example glosses: IND: indicative, LOG:
logophoric, PRO: pronoun, REFL: reflexive, SUBJ: subjunctive.
I. Charnavel ()
Department of Linguistics, Harvard University, Cambridge, MA, USA
e-mail: icharnavel@fas.harvard.edu

https://doi.org/10.1007/978-3-030-34308-8_13
418 I. Charnavel
The primary empirical motivation for this claim is the observation that per-
spectival elements such as logophoric pronouns or exempt anaphors (i.e. reflexives
escaping the locality conditions imposed by Condition A under perspective-based
conditions) can be found in causal clauses. This is illustrated below in Gokana and
Mandarin, and is documented in various other languages such as Japanese (Kuroda
1973; Sells 1987), Ewe (Culy 1994; cf. Clements 1975), Tamil (Sundaresan 2012;
cf. Jayaseelan 1998), Latin (Solberg 2017) or French (Charnavel 2019b), among
others.
(1) Lébàreè dìv im bòò beè kOO mm de-è a gíá [Gokana]
Lebare hit me because that I ate-LOG PRO yams
‘Lebare hit me because I ate his yams.’ Hyman & Comrie (1981), Sells (1987)
(2) Yinwei Lisi piping zijii , suoyi Zhangsani hen shengqi. [Mandarin]
because Lisi criticize REFL so Zhangsan very angry
‘Because Lisi criticized himi , Zhangsani was very angry.’ Huang & Liu (2001)
Logophoric elements like the verbal affix è in (1) and non-locally bound
reflexives like ziji in (2) are usually assumed to be subject to interpretive constraints
related to perspective: they typically occur in attitude contexts where they induce
reference to the attitude holder (Clements 1975; Sells 1987, i.a.). However, nothing
indicates in (1)–(2) that the referents of a1 and ziji contained in the causal clauses are
attitude holders. In particular, they do not seem to be in the scope of any intensional
operator.
The goal of the chapter is to solve this puzzle: I will show that causal clauses
in fact qualify as attitude contexts licensing such logophoric elements because they
introduce a judge evaluating the causal relation (the causal judge).
It is not the case that all causal clauses license this type of elements though. For
example, the logophoric reflexive sig in Icelandic can only occur in causal clauses
that are embedded under an attitude verb (Thráinsson 1976; Maling 1984, i.a.).
(3) a. Jóni kemur fyrst María elskar {hanni /∗ sigi }. [Icelandic]
John comes since Mary loves-IND PRO REFL
‘Johni comes since Mary loves himi (∗ self). Thráinsson (1976)
b. Jóni segir að hann komi fyrst María elski {hanni /sigi }.
John says that PRO comes since Mary loves-SUBJ PRO REFL
‘Johni says that he comes since Mary loves himi (self).’ Thráinsson (1976)
This suggests that the antecedent of sig in (3)a, unlike that of ziji in (2), does not
qualify as a causal judge.
To achieve the goal of the paper, it will thus be crucial to explain such referential
restrictions on the causal judge. In particular, the interaction between the syntactic
1 Gokana does not have logophoric pronouns per se, but logophoric verbal suffixes, which make
the pronouns of their clause logophoric: in (1), a is a regular pronoun, but it should be interpreted
as logophoric given that it is associated with the logophoric verbal suffix è.
13 Linguistic Perspectives in Causation 419
conditions and the referential possibilities of the causal judge will motivate the
hypothesis that the causal judge is syntactically represented as a variable that must
be locally bound.
For ease of presentation, I will henceforth restrict my empirical basis of
investigation to English causal clauses introduced by because and since, given that
the relevant contrasts between (1)–(2) and (3) are also observed in English. As
detailed in the Appendix, an online questionnaire revealed a significant contrast
between sentences like (4a), which involve an exempt reflexive in a because-clause,
and sentences like (4b), where exempt herself occurs in a since-clause.
(4) a. Lizi left because there was an embarrassing picture of heri (self) going around.
b. Lizi left since there was an embarrassing picture of heri (∗ self) going around.
The paper is organized as follows. In Section 13.1, I will show that because-
clauses are ambiguous attitude contexts: they cannot only express the speaker’s atti-
tude, but they can also present the attitude of a participant in the situation described
in the superordinate clause. This case arises when this participant can claim the
causal relation expressed by because. Section 13.2 will present and justify the
analysis of this observation, which makes use of two main ingredients. First, causal
connectors like because take an implicit causal judge as argument, which is syntacti-
cally represented as a variable that must be locally bound; this binding requirement
will not only be motivated by the behavior of because-clauses, but also by their
contrast with since-clauses, which is supported by the quantitative results presented
in the Appendix. Second, causal clauses like because-clauses contain a logophoric
operator in their left periphery, which determines their perspectival orientation.
13.1 Perspectival Effects in Causal Clauses
In a sentence like (5), the speaker makes three assertions: (i) that the tree fell, (ii)
that the tree was struck by lightning, and (iii) that the latter caused the former. In
other words, (s)he holds three types of attitude: one towards the matrix clause, one
towards the subordinate clause and one towards the causal relation between them.
(5) The tree fell because it was struck by lightning.
The same can hold in sentences like (6) that involve an event participant with
mental properties (Liz). But such cases give rise to a further possibility: the speaker
can present the causal clause from Liz’s perspective, who has a privileged access
to the reason for her leaving since she made the decision. Under this interpretation,
Liz holds an attitude towards the subordinate clause (she was tired) and towards its
relation with the matrix clause (her tiredness caused her departure).
(6) Liz left because she was tired.
This suggests that because-clauses are not only doubly, but also ambiguously
attitudinal: both the content of the causal clause and its relation with the matrix
420 I. Charnavel
Table 13.1 The perspectival possibilities of sentences involving because-clauses

Who evaluates the Who evaluates the Who evaluates the
superordinate clause causal relation subordinate clause
Case #1 AH AH AH
Case #2 AH AH + EP EP
Case #3 AH AH + EP AH + EP
clause are subject to an evaluation; and either the speaker or a participant in the
situation described can be responsible for these two types of evaluation. The goal
of this section is to argue in favor of this hypothesis and to document the possible
perspectival patterns more precisely as previewed in Table 13.1. For convenience, I
use the terms eventuality participant (EP) and attitude holder (AH) to refer to Liz
and the speaker, respectively. This is based on their status in the superordinate clause
(in 6, Liz is the event participant in the matrix clause and the speaker is the attitude
holder).2 But the point will be to show that each of them (and both) can be attitude
holders in the causal clause.
13.1.1 The Simple Case (Case #1)
In the default case, causal clauses are speaker-oriented, just like any clause that is not
embedded under any intensional operator. This is necessarily the case of sentences
like (5) above, which do not involve any mental event participant. As evidenced
by (7), the infelicitousness of continuations that contradict the matrix clause (e.g.
(7a)), the subordinate clause (e.g. (7b)), or the causal relation between them (e.g.
(7c)) shows that the speaker is committed to the content of all three parts of the
sentence.
2 Also, the neutrality of the term eventuality reflects the fact that the difference between events and
states is not relevant for our purposes: both types of eventualities give rise to the same kinds of
perspectival effects. For instance, example (4)a (which will be further examined in the text and in
which the matrix clause describes an event) does not differ from (i) (in which the matrix clause
describes a state) in any relevant way for our purposes.
(i) Liz hated the party because there was an embarrassing picture of herself going around.
Furthermore, the generality of the term attitude holder is meant to include cases like (ii) in which
the causal clause does not modify a matrix clause but a clause embedded under an attitude verb: in
such cases, the counterpart of the speaker is the subject of the attitude verb (John).
(ii) John thinks that Liz left because she was tired.
(7) The tree fell because it was struck by lightning.

a. #But the tree did not fall.
b. #But there was no lightning.
c. #But it only fell because it was old.
The speaker can similarly be a triple attitude holder in sentences like (6),
which involve a rational event participant. In particular, the fact illustrated in (8)
that epithets and epistemic modals in the causal clause can be speaker-oriented
demonstrates that the causal clause can express the speaker’s attitude.
(8) a. Lizi left because [the poor woman]i/k was tired.

b. Liz left because she must have been tired.
In (8a), reference of the epithet the poor woman to Liz guarantees that the
speaker, not Liz, is the attitude holder in the because-clause. Indeed, an epithet
cannot refer to the attitude holder of its context (independently of Condition C)
as shown in (9a) (cf. Ruwet 1990; Dubinsky & Hamilton 1998; Patel-Grosz 2012,
i.a.).3 This is further supported by the fact that the epistemic modal must can be
anchored to the speaker (i.e. relativized to the speaker’s epistemic state) in (8b):
when occurring in an embedded attitude clause as in (10), it must be anchored
to the closest attitude holder, that is, not the speaker, but Liz (cf. Hacquard 2006;
Stephenson 2007).
(9) a. Lizi thinks that [the poor woman]∗ i/k was tired.
b. According to Lizi , [the poor woman]∗ i/k was tired.
(10) Liz thinks that she must have been tired.
13.1.2 The Interesting Case (Case #2)
Unlike (5), (6) is compatible with another interpretation, under which the causal
clause presents Liz’s attitude instead of the speaker’s. At least three facts reveal the
existence of this interpretation. First, epistemic modals in because-clauses do not
have to be anchored to the speaker, but can be anchored to the relevant eventuality
participant (Liz in (11a), the editor in (11b), John in (11c)): in such cases, the
speaker need not believe the situation in the causal clause to be possible.
3 Formost speakers, this is the case even if the epithet is intended to be evaluated by the speaker
(pace Dubinsky & Hamilton 1998).
422 I. Charnavel
(11) a. Liz left because things might have spiraled out of control.
b. The editor reread the manuscript because there might have
been a mistake. von Fintel & Gillies (2007)
c. Airplanes frighten John because they might crash. Stephenson (2007)
Second, evaluative adjectives like embarrassing in (12a) or great in (12b) can
express Liz’s evaluation instead of the speaker’s (who may or may not agree).
(12) a. Liz left because there was an embarrassing picture of her going around.
b. Liz voted for Trump because he was going to be a great president.
This further supports the hypothesis that Liz can be construed as an attitude
holder in the causal clause as adjectives can only be evaluated by someone other
than the speaker in contexts like (13) introducing another attitude holder (Liz).
(13) a. Liz thought that there was an embarrassing picture of her going around.
b. Liz thought that Trump was going to be a great president.
Third, this hypothesis is corroborated by the observation mentioned in the
introduction that third-person exempt anaphors are licensed in because-clauses.4
(14) Liz left because there was an embarrassing picture of herself going around.
It has been observed that an anaphor can only be exempt from Condition A
(Chomsky 1986; Charnavel & Sportiche 2016, i.a.) if it is logophoric, namely if it
occurs in a clause expressing the perspective of its antecedent (Clements 1975; Sells
1987; Kuno 1987; Pollard & Sag 1992; Huang & Liu 2001; Charnavel 2020, i.a.).
The contrasts shown in (15)–(16) provide evidence for this generalization: himself
is acceptable in (15a) (vs. (15b)) because according to (vs. speaking of ) makes its
non c-commanding antecedent John the attitude holder of its clause; unlike himself,
inanimate itself can never take a non-local antecedent, as shown in (16), because
inanimates are not capable of perspective (Charnavel & Sportiche 2016).
(15) a. According to Johni , the article was written by Ann and himi (self).
b. Speaking of Johni , the article was written by Ann and himi (∗ self).
Kuno (1987) inspired by Ross (1970)
4 At least for native English speakers who accept exempt anaphors in general. Some dialectal
variation is indeed observed in this respect, but the questionnaire detailed in the Appendix shows
that in any case, the relevant contrasts are robust.
(16) a. Winston Q. Felixi insisted that that book had been written by Ann and
himi (self).
b. The Nature of It Alli insisted that those ideas had been simultaneously
revealed in my article and iti (∗ self).
c. Winston Q. Felixi praised Anne and himselfi .
d. The Nature of It Alli praised both Gone with the Wind and itselfi .
Postal (2006)
The acceptability of herself in (14) thus implies that its non-local antecedent Liz
is the perspective center in the because-clause. More specifically, Liz must be the
attitude holder in the causal clause, because herself must be construed de se.
(17) Context: At a party, Liz mistakes a circulating nude picture of herself

(showing her back) for a picture of her friend. She says: “this picture of my
friend is embarrassing, I am going to leave.”
Lizi left because there was an embarrassing picture of heri (#self) going
around.
In the scenario described in (17), Liz does not recognize herself in the picture so
that the long distance bound reflexive herself cannot be construed de se. Under this
interpretation, the reflexive is infelicitous. Given that de se readings are only relevant
in attitude contexts, this confirms that Liz is interpreted as the attitude holder in the
because-clause.
This is further corroborated by the continuation test in (18)a: a continuation
implying that Liz does not believe the content of the because-clause is contradictory.
However, the speaker need not endorse this content as shown in (18b).5
a. #But she thought the picture going around was not embarrassing.
b. But I think the picture going around was not embarrassing.
Furthermore, a similar test reveals that Liz is also committed to the causal
relation: the continuation in (19a) implying that she endorses a different reason for
her leaving sounds contradictory.
a. #But she thought she left because she was bored.
b. #But I think she left because she was bored.
When Liz is interpreted as the attitude holder of the causal clause, she is therefore
also construed as judge of the causal relation (i.e. as causal judge). But as shown in
(19b), the speaker must agree with her in this regard. This reveals that in cases in
5 As suggested by an anonymous reviewer, because-clauses thus provide a new type of false belief
attribution contexts, which could be used for testing theories about the development of false belief
in children.
424 I. Charnavel
Table 13.2 Mandatory Causal judge Attitude holder of causal clause

plural causal judge in case #2
Case #1 AH AH
Case #2 AH + EP EP
∗ EP EP
∗ AH EP
which the causal clause presents the eventuality participant’s perspective, the causal
judge must include both this eventuality participant EP and the speaker (i.e. the
attitude holder of the superordinate clause AH) as indicated in Table 13.2.
Under case #2 illustrated in (18)–(19), the speaker thus presents the participant’s
internal reason for the eventuality (Liz’s thinking that there was an embarrassing
picture of herself) as the cause of the eventuality (the cause of Liz’s leaving). But
even if (s)he endorses the causal relation, the speaker does not commit to the content
of the cause (that there was an embarrassing picture of Liz going around). Such
cases arise when the eventuality described in the superordinate clause involves
a participant that is capable of determining a reason for this eventuality (s)he is
involved in (cf. Hara 20086 ). This is clearly the case of volitional agents like Liz
in (18)–(19), whose intentional action is the caused eventuality (cf. Solstad 20107 ).
More generally, this is the case of any human participant that can determine the
reason for the eventuality they participate in: in particular, experiencers like John in
(11c) have a more direct access than the speaker to the reason for their experience
since it lies in their own mental state. However, inanimates like the tree in (7) or
John in (20b) cannot determine what caused the eventuality they are involved in
since they lack a mental state; sentient eventuality participants that do not have
access to the relevant cause (for instance because it was initiated by another agent
unbeknownst to them, as in (20c)) cannot either.
6 In a similar vein, Hara (2008) observes that in Japanese, contrastive marking with wa and
evidential marking with soona/soda is available in because-clauses (introduced by node), but
not in temporal clauses or if -clauses. She derives this fact from the hypothesis that because-
clauses, unlike these other adjunct clauses, do not necessarily express a relation between events (cf.
‘singular causal statement’ in Davidson 1967, ‘transparent because’ in Kratzer 1998), but can also
express a relation between propositions (‘causal explanation’ in Davidson 1967, ‘opaque because’
in Kratzer 1998); in the latter case, because can thus express the speaker’s or some attitude bearer’s
inference about the connection between two propositions, which is crucial for the availability of
wa and soona/soda.
7 Solstad (2010) claims that because is ambiguous between a ‘reason’ interpretation, under which
the caused entity is an attitudinal state (e.g. the intention to pick out the painting in (iiia)), and
a ‘plain cause’ interpretation, under which the caused entity is a (non-intentional) event or state
(e.g. the crash of the stunt plane in (iiib)). Our case #2 includes Solstad’s ‘reason’ case, but also
other cases (like (11c) or (20a)) where the eventuality participant can determine the reason for the
eventuality (s)he is involved in, even if it is not the result of his/her intention.
(iii) a. I picked out the painting because it matches my wall.
b. The stunt plane crashed because it ran out of petrol. Solstad (2010)
(20) a. John is mad because there was a picture of himself in the post office.
b. ∗ John is dead because there was a picture of himself in the post office.
c. ∗ John was arrested because there was a picture of himself in the post
office. Williams (1974)
13.1.3 The Plural Case (Case #3)
Table 13.2 does not exhaust all logical possibilities. In particular, the question arises
as to whether the causal clause can present a plural attitude, given that the causal
judge can be plural. Examples (21a–b) show that this is possible under specific
conditions
(21) a. ∗Liz left because there was an embarrassing picture of myself and
herself going around.
b. Liz left because there was an embarrassing picture of ourselves going
around.
The unacceptability of disjoint exempt anaphors co-occurring in the causal clause
in (21a) indicates that there can only be one perspective center in it: because-clauses
disallow mixed perspective. But the perspective center can be plural and include
both the eventuality participant and the speaker as shown by the acceptability of
the plural exempt anaphor in (21b). Similarly, the epistemic modal in (11a) can be
linked to the epistemic state of both the speaker and Liz, and in (12a), the picture
can be evaluated as embarrassing by both the speaker and Liz.
Under such interpretations, the causal judge must remain plural as shown by the
continuation tests in (22) and schematized in Table 13.3.
(22) Liz left because there was an embarrassing picture of ourselves going around.
a. #But she thought she left because she was bored.
In sum, because-clauses constitute a complex type of attitude report that can
involve two types of attitude holders (i.e. the participant EP in the eventuality
described in the superordinate clause, and the individual AH – typically the
speaker – holding the attitude in the superordinate clause) and two types of attitudes
(i.e. the attitude towards the causal relation and the attitude towards the content of
the cause). This gives rise to nine logical possibilities presented in Table 13.4 out of
which three are available.
Table 13.3 Mandatory Causal judge Attitude holder of causal clause

plural causal judge in case #3
Case #3 AH + EP AH + EP
∗ EP AH + EP
∗ AH AH + EP
426 I. Charnavel
Table 13.4 Possible and Causal judge Attitude holder of causal clause
impossible perspectival
effects in because-clauses Case #1 AH AH
Case #2 AH + EP EP
∗ EP EP
∗ AH EP
Case #3 AH + EP AH + EP
∗ EP AH + EP
∗ AH AH + EP
∗ EP AH
∗ AH + EP AH
The exclusion of the last two possibilities is justified by the status of the
continuations in example (23) (cf. (8a)). The contradictoriness of (23a) shows that
the speaker has to be a causal judge when (s)he endorses the content of the causal
clause. The acceptability of (23b) is not sufficient to rule out the last possibility as it
is compatible with case #1, but intuitions about (23) strongly suggest that Liz cannot
be a causal judge when she does not hold an attitude towards the causal clause.
(23) Lizi left because [the poor woman]i was tired.

a. #But I think she left because she was bored.
b. But she thought she left because she was bored.
13.2 Judge-Based Analysis
The goal of this section is to account for the possible and impossible perspectival
effects in because-clauses reviewed in the previous section. Typically, attitudinal
effects are triggered by overt attitude verbs like think or other types of attitudinal
expressions like according to or opinion (see Charnavel 2020; Pearson to appear,
i.a.). In the case of causal clauses, I will hypothesize that they are induced by
the causal connector (e.g. because), which introduces two implicit elements shown
in Fig. 13.1 representing case #2. First, the syntactic conditions required for case
#2–3 will motivate the hypothesis that causal connectors take an implicit causal
judge argument j that is syntactically represented and must be locally bound
(Sect. 13.2.1). Second, we will see that distributive and interpretive constraints
on perspectival elements in causal clauses require the existence of a mediating
element between j and the causal clause, namely a logophoric operator OP in the
left periphery of the causal clause (Sect. 13.2.2).
Under this hypothesis, the causal judge j is thus responsible for the perspectival
orientation of the causal relation, while the logophoric operator OP is responsible for
the perspectival orientation of the causal clause. Given that these two perspectives
cannot be independent as shown in Sect. 13.1 (see Table 13.4), I will further assume
Fig. 13.1 The two main ingredients of the analysis in case #2
that OP must be (partially) controlled by j. This entails that perspectival effects in the
causal clause can constitute (indirect) evidence for both j and OP as I now explain
in detail.
13.2.1 Arguments for Positing j
We have seen that the causal relation between the content of a because-clause and
that of its superordinate clause must be endorsed by a reasoning individual, namely
the attitude holder of the superordinate clause (typically the speaker) as well as the
relevant eventuality participant in some cases. I propose that this (potentially plural)
causal judge is an implicit argument of because (cf. Stephenson 2007).8 Because is
therefore similar to a complex attitude verb that would take three arguments: a judge
j, a subordinate clause B and a superordinate clause A, so that A because B basically
means that j believes that B is the cause of A.
(24) [[because (j)]] w = λB.λA.∀w’ compatible with j’s mental state in w, B

is the cause of A in w’
8 In
all diagrams, I represent j as the subject of because for simplicity, but I do not take a stand on
which argumental position j has.
428 I. Charnavel
Furthermore, I hypothesize that j is syntactically represented. The motivation for

this hypothesis is twofold. First, I will show (in Sect. 13.2.1.1) that cases #2–3 (in
which the causal clause involves the eventuality participant’s perspective) require
that the causal clause be in the scope of this participant. This argues for an analysis
under which j is a syntactic variable that must be bound. Second, we will see (in
Sect. 13.2.1.2) that in all cases, j must be anteceded by the closest attitude holder.
This supports and specifies the binding hypothesis as this observation directly
follows if we assume that j must be locally bound.
13.2.1.1 Cases #2–3: EP Binding Requirement
The first part of the argument for the binding requirement of j by EP in cases #2–3
consists in showing that because-clauses are sufficiently low to be in the scope of an
eventuality participant in the superordinate clause. The facts in (25) make the point
(cf. Rutherford 1970; Groupe Lambda-1 1975; Sæbø 1991; Iatridou 1991; Johnston
1994, i.a.): a pronoun in a because-clause can be bound by a matrix quantifier (25a);
coreference between an R-expression in a because-clause and a matrix pronoun
triggers Condition C effects (25b); a pronoun in a because-clause can give rise to a
sloppy reading in a VP-ellipsis context (25c).
(25) a. [No girl]i left because there was a picture of heri going around.
b. ∗Shei left because there was a picture of Lizi going around.
c. Lizi left because there was a picture of heri going around, and Lucyk
did too [leave because there was a picture of herk going around].
[sloppy reading]
These interpretive effects can only obtain if because-clauses can be (or must
be, in the case of (25b)) outscoped by the matrix subject. Moreover, the fact that
because-clauses can be retrieved in VP-ellipsis contexts suggest that they are VP
modifiers. We can thus assume that the matrix eventuality participant can bind into
because-clauses because it can move out of the VP that they modify (see Fig. 13.1).
The second part of the argument consists in showing that if the eventuality
participant EP cannot bind into the because-clause, that clause cannot present the
participant’s attitude. This is illustrated by the interaction between perspective and
Condition C in (26).
(26) a. Ian thinks that shei left because there was an embarrassing picture of
Lizi going around.
b. Ian thinks that Lizi left because there was an embarrassing picture of
heri going around.
In (26a), coreference between Liz and she guarantees (due to Condition C) that
she cannot bind into the because-clause (which therefore modifies the matrix, rather
than the embedded VP). In that case, crucially, the picture cannot be evaluated
as embarrassing by Liz, but only by Ian. This interpretation is however available
in (26b), where Condition C does not impose any constraint on the height of the
because-clause.
Such facts (see further cases in Charnavel 2019a) demonstrate that a because-
clause must be in the scope of an eventuality participant EP to present its referent’s
attitude. This constraint leads me to hypothesize that j must be bound by EP in such
cases (cases #2–3) as represented in Fig. 13.1 above.
The reverse, however, does not hold: a because-clause in the scope of an
eventuality participant EP does not necessarily express its referent’s perspective. For
instance, pronominal binding in (27a) ensures that no tree is in a position to bind j;
but as an inanimate, a tree cannot be a causal judge. Similarly, the licensing of the
NPI anything in the because-clause by the matrix negation in (27b) guarantees that
Liz can in principle bind j; but the coreferring epithet the poor woman shows that
Liz is not the causal judge here.
(27) a. [No tree]i fell because iti was struck by lightning.

b. Lizi did not leave because [the poor woman]i had anything to
do (but . . . ).
This implies that the causal judge j need not be bound by the eventuality
participant in general (it is bound by the speaker in such cases, as we will see in
the next subsection). Such binding is only required under the interpretation where
the causal clause presents its referent’s perspective (cases #2–3).
This requirement, which justifies the syntactic representation of j as a bound
variable, is further corroborated by the behavior of since-clauses. Unlike because-
clauses that usually express the cause of an event or state, since-clauses typically
provide evidence for the truth of the matrix proposition (as in (28a)) or a reason for
the matrix speech act (as in (28b)).
(28) a. Liz left, since her coat is not on the rack.

b. Liz left, since you must know everything.
Furthermore, since-clauses attach higher than because-clauses as shown by
(29), which contrasts with (25) (see Appendix for some quantitative results9 ):
pronominal binding into since-clauses is precluded (e.g. (29a)); Condition C effects
are alleviated (e.g. (29b)); sloppy readings are unavailable – in fact, since-clauses
cannot even be retrieved in VP-ellipsis sites.
9 The results of the Appendix also exclude an alternative hypothesis that is sometimes proposed
(see Iatridou 1991, i.a.) to explain why binding into since-clauses is impossible in this type of
cases, namely the fact that the content of since-clauses is not at-issue.
430 I. Charnavel
(29) a. *[No girl]i left, since heri coat is not on the rack.
b. ?Shei left, since Lizi ’s coat is not on the rack.
c. Lizi left, since heri coat is not on the rack, and Lucyk did too
[leave (*since herk coat is not on the rack]. [*sloppy reading]
The combination of the observations in (28) and (29) suggests that since-clauses
modify high projections of the left periphery, namely Speech Act Phrases (SAP) or
Evidential Phrases (EvidP), which scope higher than VP (Cinque 1999, i.a.).
Crucially, this syntactic height correlates with the unavailability of any eventual-
ity participant’s perspective occurring in since-clauses. For instance, (30) contrasts
with (14) (see Appendix for quantitative results) in that exempt herself is not
licensed in the causal clause and the picture cannot be evaluated as embarrassing
by Liz, but only by the speaker. Since-clauses thus always fall under case #1.10
(30) Liz left, since there was an embarrassing picture of her(*self) going around.
This confirms the hypothesis defended above that the interpretation of a causal
clause as presenting an eventuality participant EP’s attitude only obtains under
binding of j by EP. In other words, the correlation between the syntactic height of
causal clauses and their perspectival effects is accounted for if we assume that the
argument j of causal connectors is a bound variable and that just like the subject of
an attitude verb, it determines the attitudinal orientation of the subordinate clause.
13.2.1.2 General Binding Requirement of j
So far, I have provided evidence for the hypothesis that j must be bound by EP
in cases #2–3, which justifies the syntactic representation of j. This hypothesis is
further motivated by a general binding requirement of j: in all cases #1–3, j must be
bound by the closest attitude holder.
Recall from Sect. 13.1 that the speaker must be included in the causal judge in all
three cases observed. When causal clauses are embedded under attitude verbs, the
relation expressed by because or since must however be endorsed by the subject of
the (closest) attitude verb. For instance, it is Paul, not the speaker, that commits to
the causal relation expressed by because in (31). Similarly, it is Paul, not the speaker
that must believe the existence of the evidential relation expressed by since in (32).
(31) Paul thinks that [every plant]i died because he forgot to water iti .
a. #But he thinks the reason why they died is that they needed more light.
b. But I think that the reason why they died is that they needed more light.
10 The same probably holds of fyrst-clauses in Icelandic: the ungrammaticality of sig in (3)a may be
due to the fact that fyrst-clauses scope too high for their causal judge to be bound by an eventuality
participant in the superordinate clause. The application of scopal tests like (29) to Icelandic would
be necessary to confirm this hypothesis.
(32) Paul believes that since their radio is off, the neighbors must have left.
a. #But he believes that the neighbors turn their radio on when they leave.
b. But I believe that the neighbors turn their radio on when they leave.
In both cases, the causal clause is embedded under the attitude verb (as evidenced
by pronominal binding by the embedded quantifier in (31) and by fronting of the
since-clause within the embedded clause in (32)). Assuming that the speaker is
represented in the left periphery of root clauses (cf. Ross 1970; Speas & Tenny
2003; Haegeman & Hill 2013; Zu 2018, i.a.), Paul is consequently the attitude holder
closest to the causal connector in both (31) and (32). What these examples show is
therefore that the causal judge must be (or at least include11 ) the closest attitude
holder.
This observation suggests that the obligatory inclusion of the speaker in the
causal judge observed in Sect. 13.1 can be generalized as obligatory inclusion of the
local attitude holder in the causal judge. Such a locality requirement further supports
the j binding hypothesis: under this hypothesis, one only needs to assume that the
binding must be local to derive the generalization.12 In sum, j is exclusively bound
by the local attitude holder AH (the speaker or the subject of the lowest attitude
verb) in case #1, and in cases #2–3, j is co-bound by the local attitude holder AH
and the eventuality participant EP as represented in Fig. 13.1 and in (33) (where log
stands for any perspectival element in the causal clause).
(33) Case #1: AH [ . . . ][ jAH because/since [ ... logAH ]]

Case #2: AH [ EP . . . ][ jAH+EP because [ ... logEP ]]
Case #3: AH [ EP . . . ][ jAH+EP because [ . . . logAH+EP ]]
This assumption is further motivated by the obligatoriness of sloppy readings
with respect to the causal judge, which is illustrated in (34): in both (a) and (b), the
causal relation between lightning and the utility pole fall must be endorsed by Mark,
not by Paul. This interpretive constraint directly follows from the assumption that j
is bound by the closest attitude holder.
11 (31)and (32) are embedded counterparts of case #1, where the causal judge is singular. The
embedded counterparts of cases #2 and #3, which involve a plural judge, would similarly involve
a plural judge including the closest attitude holder and the relevant eventuality participant (see
Charnavel 2019a for illustrations).
12 In Charnavel (2019a), the observation that the local binder of j must be an attitude holder is
derived from the hypothesis that j is a logophor: j is in fact not directly bound by the local attitude
holder, but by a logophoric operator present in the superordinate clause. This hypothesis further
explains why j can take a split antecedent (the attitude holder and the eventuality participant).
432 I. Charnavel
(34) a. Paul: “The tree fell because it was struck by lightning.”

Mark: “The utility pole did too [fall because, according to {Mark/
*Paul}, it was struck by lightning].”
b. Paul said that the tree fell because it was struck by lightning, and Mark
said that the utility pole did too [fall because, according to {Mark/
*Paul}, it was struck by lightning].
13.2.2 Arguments for Positing OP
We have just seen that several arguments motivate the syntactic representation of
the causal judge. Similarly, several facts support the syntactic representation of the
attitude holder of causal clauses as a logophoric operator in their left periphery (cf.
Hara 200813 ): I am now going to show that the constraints on perspectival elements
licensed in causal clauses require this additional element to mediate between the
causal judge and causal clauses.
First, recall from (21a) (repeated below) that two exempt anaphors co-occurring
in a causal clause cannot be disjoint.
(35) a. *LizEP left [jAH + EP because there was an embarrassing picture of

myselfAH and herselfEP going around].
b. LizEP left [jAH + EP because there was an embarrassing picture of
ourselvesAH + EP going around].
Given that we have established in Sect. 13.2.1.1 that the perspectival orientation
of the causal clause is determined by the causal judge j, we could reasonably assume
that j binds the perspectival exempt anaphors in (35). The contrast between (35a)
and (35b) should thus imply that j cannot partially bind these anaphors. But case
#2 contradicts this assumption as shown by (4) repeated below: exempt herself
referring to Liz is acceptable while j must also include the speaker (see (19)).
(36) LizEP left [jAH + EP because there was an embarrassing picture of herselfEP
going around].
As shown in (37), this puzzle is however solved if we posit the presence of a
logophoric operator OP (partially) controlled by j14 at the periphery of the causal
13 Similarly, Hara (2008) argues on the basis of Japanese facts (see fn. 6) that there is some
representation of point of view in the complement of because: specifically, she argues that because
can shift the context of utterance just like attitude predicates.
14 The hypothesis that OP is (partially) controlled by j is motivated by the observation that the
attitude holder of the causal clause is always included in the causal judge, but need not be
exhaustively coreferent with it (see Sect. 13.1, Table 13.4). The reason why logophoric operators in
causal clauses are subject to this constraint (unlike those in clausal complements of attitude verbs,
see Charnavel 2020) remains to be further understood.
clause, which represents the perspective center of the causal clause and exhaustively
binds the anaphors.
(37) a. LizEP left [jAH + EP because [OP EP there was an embarrassing picture of
herselfEP going around]].
b. *LizEP left [jAH + EP because [OP EP/AH/AH + EP there was an embarrassing
picture of myselfAH and herselfEP going around]].
This solution is inspired by previous proposals (Koopman & Sportiche 1989;
Huang & Liu 2001; Anand 2006) that assume that logophoric elements such as
logophoric pronouns or exempt reflexives are licensed by logophoric operators
and that the presence of at most one logophoric operator per clause derives the
impossibility of disjoint logophoric elements within a clause.
The extension of this approach to causal clauses combined with the new hypoth-
esis that logophoric operator binding must be exhaustive has a further consequence:
as detailed in Charnavel (2020), it explains why anaphors can apparently be exempt
from Condition A when logophorically interpreted (see Sect. 13.1.2). As under
this hypothesis, these logophoric anaphors are locally and exhaustively bound by
a logophoric operator, they are in fact reduced to plain anaphors obeying Condition
A, which explains why they are morphologically identical to them in so many
languages.15 The illusion that they are exempt is created by the implicitness of
their binders, which need not be exhaustively or locally bound themselves. This
argument provides further support for the existence of OP in addition to j as j could
neither qualify as local binder of the anaphors since as an argument of because, it
sits outside the causal clause, nor as exhaustive binder because of case #2. In sum,
the exhaustive coreferential constraint on exempt anaphors in causal clauses is due
to their anaphoric requirement (local and exhaustive binding), which can be satisfied
by OP, but not by j.
The presence of the logophoric operator does not only explain why exempt
anaphors in causal clauses seem to escape Condition A, but also why they are
logophorically interpreted. Recall from Sect. 13.1.2 that exempt anaphors in general
must occur in clauses expressing the perspective of their antecedent. Furthermore,
we have observed that exempt anaphors in causal clauses must be read de se (see
example (17)). This interpretive constraint would remain mysterious if exempt
anaphors were bound by j (cf. binding by the subject of an attitude verb does
not entail de se reading). But binding by the logophoric operator derives this
interpretation as long as we assume that the role of the operator is to impose the first-
personal perspective of the logophoric center on its complement (cf. Anand 2006;
Charnavel 2020).16 In sum, positing a logophoric operator OP at the periphery of
15 More precisely, I hypothesize in Charnavel (2020) that the logophoric operator OP is a head
taking a silent logophoric pronoun pro as subject, which is the actual binder of exempt anaphors.
As a phrase, it qualifies as A-binder.
16 This effect of the logophoric operator further explains why the perspectival orientation of other
types of logophoric elements in causal clauses must be harmonized. For instance, fragile and might
434 I. Charnavel
causal clauses solves three issues: it accounts for why exempt anaphors are licensed
in causal clauses (they are in fact plain anaphors locally bound by a silent binder
OP ), why they must corefer (like plain anaphors, they must be exhaustively bound
and there is only one possible binder for them in the clause, i.e. OP), and why they
have a specific perspectival interpretation (they inherit their interpretation from their
logophoric binder OP).
This logophoric operator hypothesis finally sheds light on a conceptual problem.
Based on examples like (18)–(19) (repeated below), we have concluded that in case
#2, the speaker does not endorse the content of the causal clause (only the relevant
eventuality participant – e.g. Liz – does), but only the causal relation between the
matrix and the causal clause.
a. But I think the picture going around was not embarrassing.
But to believe that some eventuality B caused some eventuality A, it seems
necessary to believe the existence of B. For instance, how can the speaker believe
that the presence of an embarrassing picture of Liz going around caused Liz’s
departure if (s)he does not believe that there was an embarrassing picture of Liz
going around? The answer to this question lies in the presence of the operator.
What the speaker in fact believes caused Liz’s departure is not that there was an
embarrassing picture of her going around, but the fact that Liz thought so. This
relativization of the content of the causal clause to the eventuality participant’s
mental state is precisely what the logophoric operator codes.
Interestingly, this seems to be morphologically reflected in languages like
Gokana (see example (1)) in which because-clauses licensing logophoric elements
are not only introduced by the causal connector, but also by the complementizer kOO
(derived from a verb meaning say) which generally serves as complementizer for
attitude clauses containing logophoric elements.
To wrap up, the perspectival patterns observed in Sect. 13.1 can be explained
if we posit two silent elements in the structure of causal clauses: a causal judge j
(referencing the reasoning individual(s) endorsing the causal relation) that must be
locally bound and a logophoric operator OP (partially) controlled by j (referencing
the attitude holder(s) of the causal clause) that must locally bind logophoric
elements in its scope.
in (iv) must be anchored to the same individual (John alone, the speaker alone, or both the speaker
and John). That’s also why in (14), in which the causal clause contains the exempt anaphor herself
referring to Liz, the adjective embarrassing must also be evaluated by Liz.
(iv) Airplanes frighten John because the fragile machines might crash.
Note that this property of OP is allowed by the hypothesis that unlike j (a simple pronoun), OP is
a head (see fn. 15) and is thus similar to a Free Indirect Discourse operator (see further details in
Charnavel 2020).
(39) Case #1: AH [ . . . ][ jAH because/since [OP AH . . . logAH ]]

Case #2: AH [ EP . . . ][ jAH+EP because [OP EP . . . logEP ]]
Case #3: AH [ EP . . . ][ jAH+EP because [OP AH+EP . . . logAH+EP ]]
13.3 Conclusion
The licensing of logophoric elements such as exempt reflexives in causal clauses

thus reveals that the intrinsically mental nature of causal relations is linguistically
reflected. It is because causal clauses require a judge for the causal relation that they
qualify as attitude contexts that can host logophoric elements.
The licensing of logophoric elements in causal clauses remains nevertheless
restricted because the referential possibilities of the judge are syntactically con-
strained. In particular, they depend on the height of causal clauses. Perspectival
effects in causal clauses therefore provide indirect information about their structural
position.
This may be exploited in other linguistic causal environments such as subjects
of causative verbs.17 For instance, if it turned out that in (40), exempt herself
is acceptable and embarrassing can be evaluated by Liz, this could suggest that
causative verbs are also relativized to a causal judge j that is bindable by an element
within their object. The examination of such perspectival effects in future research
could thus shed new light on the structure of causative constructions.
(40) The {idea/fact} that an embarrassing picture of her(?selfi ) was circulating

made Lizi leave the party earlier than planned.
Acknowledgements For their sharp comments, judicious questions and helpful suggestions, many
thanks to the alert and friendly audience of Linguistic Perspectives on Causation as well as three
anonymous reviewers. I am also very grateful to the organizers of this workshop, thanks to whom
I was not only exposed to many interesting linguistic perspectives on causation, but also to many
fascinating aspects of Jerusalem. Moreover, this work benefited from fruitful discussions I had
17 Many other causal environments could be investigated from this perspective, such as other causal
clauses in English (e.g. clauses introduced by given that or as) and other languages, as well as
causal prepositional phrases (e.g. phrases introduced by because of or due to). For example, Solstad
(2010) observes that (va), unlike (vb) is ambiguous: while the because-clause in (va) can either
specify Bill’s motive for going back home or the speaker’s evidence for inferring that Bill must
have gone back home, the because of -phrase in (v)b only exhibits the former interpretation. Solstad
(2010) attributes this difference in interpretation to a difference in adjunction site. For our purposes,
this hypothesis would imply that case #2 could in principle be available for because of -phrases if
logophoric operators are not restricted to propositions as proposed in Charnavel (2020). This type
of prediction would be worth testing in future research.
(v) a. Bill must have gone back home because the jacket is missing.
b. Bill must have gone back home because of the missing jacket.
436 I. Charnavel
with the linguistics departments of Stony Brook, Rutgers, UMass Amherst, NYU, USC and my
own (Harvard), as well as the conference audiences at NELS46, GR30, LSRL47 and SALT27.
Last but not least, many thanks to the participants of the experiment presented in the Appendix and
to Gunnar Lund who played an essential role in the running of the experiment. This experiment was
supported by a Harvard grant under the Junior Faculty Research Assistant program. The material
of the whole paper is based upon work supported by the National Science Foundation under grants
1424054 and 1424336.
Appendix
This appendix presents the results of experimental work done in collaboration with
Gunnar Lund, which confirm some crucial judgments of the paper.18 Specifically,
we obtained quantitative data corroborating the difference between because-clauses
and since-clauses in their licensing of exempt reflexives and pronominal binding.
First, we tested the contrast between the availability of exempt anaphors in
because-clauses and in since-clauses as illustrated in (41)–(42) (cf. (4a–b)).
(41) Alice sued the newspaper because it published an embarrassing photo

of herself. [condition mean: 4.7 out of 6; standard deviation: 1.15]
(42) Tom went on vacation since there was a picture of himself at a beach
on Facebook. [condition mean: 3.5 out of 6; standard deviation: 1.38]
Each condition included 3 items. The sentences were randomly ordered and
presented one at a time without any previous context. A total of 90 native speakers
were asked to perform grammaticality judgment tasks online based on a 6-point
Likert scale (the survey was run on Qualtrics via the Amazon Mechanical Turk
website).19
The results were calculated using the R-software and t-tests revealed the
existence of a significant contrast between the two conditions (p < 0.001).20 This
corroborates the observation detailed in the paper that because-clauses, unlike since-
clauses, can present the eventuality participant’s perspective.
The second part of the experiment consisted in examining whether the accept-
ability of exempt reflexives correlates with the syntactic height of the causal clauses
containing them. To this end, we tested the availability of pronominal binding in
sentences like (43)–(44).
18 This questionnaire included other types of adjunct clauses (concessive clauses), but only the
relevant results on causal clauses are presented here.
19 As is standard, the questionnaire included a consent, instructions, practice sentences and
attention checks to ensure that participants understood and paid attention to the task. Inattentive
participants were excluded from the survey.
20 As is standard, I consider that only contrasts (when statistically significant) are informative,
unlike absolute scores.

(43) Congressmen Smith, Jones, and Johnson hate their jobs. However, they feel
a sense of duty to their citizens and go to work every day for that reason.
No congressman goes to work because he loves his job.
a. Intended true interpretation (pronominal binding):
[No congressman]i goes to work because hei loves hisi job.
b. Intended false interpretation (no pronominal binding):
[No congressman]i goes to work, because hek loves hisk job.
[condition mean: 64% true]
(44) The headmaster at a boarding school wants to make sure that all the boys
in the dorm are at dinner. Walking by one room, he hears someone talking
on the phone very loudly. The rest of the rooms seemed totally empty.
No schoolboy is in his dorm since his light is on.
[No schoolboy]i is in his dorm since hisi light is on.
[No schoolboy]i is in his dorm, since hisk light is on.
Like above, each condition included three items and the sentences were randomly
ordered. But this time, the participants (the same as above) were asked to perform
a truth value judgment task: they had to decide whether the sentence was true or
false in the scenario indicated. As illustrated in (43)–(44), we guaranteed in each
case that only the narrow scope reading of the pronoun with respect to the quantifier
could make the sentence true (as shown in a). A true answer thus indicated that the
reading with pronominal binding was available, and a false answer indicated that it
was not (assuming that participants considered all possible readings because they
are generally biased towards giving true answers whenever possible).
The results confirmed the structural difference between because-clauses and
since-clauses discussed in the paper (see (25a) and (29a)): while most participants
judged sentences like (43) true (average across sentences and participants: 64%
of true answers), they almost never judged sentences like (44) true (average
across sentences and participants: 2% of true answers). This significant difference
(p < 0.001) supports the hypothesis defended in the paper that because-clauses
attach lower than since-clauses and can thus be outscoped by matrix elements.
The third part of the experiment aimed to discard a possible alternative analysis
of this result. The unavailability of pronominal binding in since-clauses like (44)
could arguably be due to the fact that since-clauses can never be bound into because
they are presupposed or not-at-issue (see Iatridou 1991; cf. Potts 2005). Under
such an analysis, the previous results do not provide evidence for any difference
in syntactic height between since- and because-clauses. To dismiss this possibility,
we tested sentences like (45)–(46) where the causal clauses modify an embedded
attitude clause. Under our hypothesis based on syntactic height, pronominal binding
by the matrix quantifier is this time predicted to be available with both because- and
438 I. Charnavel
since-clauses. Under the alternative analysis based on not-at-issueness, the same

difference between because- and since-clauses is predicted to obtain.
(45) Three mothers are talking about their diets. As it turns out, they eat beets
with nearly every meal. They agreed that the health benefits outweigh the
mediocre taste.
No mother claims that she eats beets because she finds them tasty.
[No mother]i claims that she eats beets because shei finds them tasty.
[No mother]i claims that she eats beets, because shek finds them tasty.
(46) Three mailmen always park in the same places, despite the fact that they are
no parking zones. One day, all three of them got a call from their manager
telling them that their trucks were towed.
No postman said that his mail truck was towed since it’s not in his usual
parking spot.
[No postman]i said that his mail truck was towed since it’s not in hisi
usual parking spot.
[No postman]i said that his mail truck was towed, since it’s not in hisk
usual parking spot.
Like above, each condition included three items, the sentences were randomly
ordered, and the participants (still the same ones21 ) performed a truth value
judgment task. Again, a true answer meant that pronominal binding was available
(as shown in a) and a false answer that it was not (as shown in b). Indeed, the
scenarios were only compatible with the narrow scope of the pronoun in the causal
clause with respect to the quantifier and they only made relevant the interpretation
under which the causal clause modifies the embedded, not the matrix clause. The
results were similar to the previous cases in the case of because-clauses (p = 0.66),
but they were significantly different in the case of since-clauses (p < 0.001). Thus,
since-clauses can in fact be bound into when they are low enough to be outscoped
by a quantifier.22 This argues against the analysis based on not-at-issueness and
21 The truth value judgment task was randomly divided into two lists so that each participant only
had to judge the same number of sentences as in the grammaticality judgment task. The results
about the truth value judgment task are therefore based on a total number of 45 answers.
22 Note that the result for (46) cannot be considered as relying on chance: if participants answered
at chance level each time binding into a since-clause is intended, the same would hold in (44). The
contrast between (45) and (46) nevertheless raises an interesting question: why does binding into
corroborates the hypothesis defended in the paper that since-clauses attach higher
than because-clauses.
In sum, the significant difference between since-clauses and because-clauses
with respect to both exempt reflexives and pronominal binding (in non-embedded
cases) supports the hypothesis that there is a correlation between the two facts. This
correlation is explained by the analysis presented in the paper: exempt reflexives in
causal clauses must be bound by the logophoric operator controlled by the causal
judge that must itself be bound; by transitivity, exempt reflexives in causal clauses
must therefore be bindable by their antecedent to be acceptable.
References
Anand, P. (2006). De De Se. Ph.D. Dissertation. MIT.

Charnavel, I. (2019a). Perspectives in causal clauses. Natural Language and Linguistic Theory,
37(2), 389–424.
Charnavel, I. (2019b). Point of view on causal clauses: The case of French parce que and puisque.
In I. Feldhausen, M. Elsig, I. Kuchenbrandt, & M. Neuhaus (Eds.), Romance languages and
linguistic theory 15. Selected papers from ‘Going Romance’ 30, Frankfurt (pp. 93–112).
Amsterdam: John Benjamins Publishing Company.
Charnavel, I. (2020). Logophoricity and locality: A view from French anaphors. Linguistic Inquiry,
1–53. https://doi.org/10.1162/ling_a_00349.
Charnavel, I., & Sportiche, D. (2016). Anaphor binding – What French inanimate anaphors show.
Linguistic Inquiry, 47(1), 35–87.
Chomsky, N. (1986). Knowledge of language: Its nature, origin, and use. New York: Praeger.
Cinque, G. (1999). Adverbs and functional heads: A cross-linguistic perspective. New York:
Clements, G. N. (1975). The Logophoric pronoun in ewe: Its role in discourse. Journal of West
African Languages, 10, 141–177.
Culy, C. (1994). Aspects of Logophoric marking. Linguistics, 32, 1055–1094.
Dubinsky, S., & Hamilton, R. (1998). Epithets as Antilogophoric pronouns. Linguistic Inquiry,
29(4), 685–693.
Groupe Lambda-1. (1975). Car, parce que, puisque. Revue Romane, 10, 248–280.
Hacquard, V. (2006). Aspects of modality. Ph.D. Dissertation. Massachusetts Institute of Technol-
ogy.
Haegeman, L., & Hill, V. (2013). The Syntacticization of discourse. In R. Folli, R. Truswell, & C.
Sevdali (Eds.), Syntax and its limits (pp. 370–390). Oxford: Oxford University Press.
Hara, Y. (2008). Evidentiality of discourse items and Because-clauses. Journal of Semantics, 25,
229–268.
since-clauses seem to be more difficult than binding into because-clauses? This apparent effect
may be due to the fact that the reading under which the sentence is false (i.e. under which the
causal clause is not embedded and cannot therefore be bound into) is more easily accessible with
since-clauses than with because-clauses. The exact reason for this contrast remains unclear (and
may be related to (non)-at-issueness), but in any case, the fact that since-clauses can be bound into
in at least some cases shows that not-at-issueness does not entail unbindability.
440 I. Charnavel
Huang, C.-T. J., & Liu, C.-S. L. (2001). Logophoricity, attitudes and ziji at the Interface. In P. Cole
et al. (Eds.), Long distance reflexives, syntax and semantics (Vol. 33, pp. 141–195). New York:
Academic.
Hyman, L. M., & Comrie, B. (1981). Logophoric reference in Gokana. Journal of African
Languages and Linguistics, 3, 19–37.
Iatridou, S. (1991). Topics in conditionals. Ph.D. Dissertation. MIT.
Jayaseelan, Karattuparambil A., 1998: Blocking effects and the syntax of Malayalam taan. In R.
Singh (ed.), The Yearbook of South Asian Languages and Linguistics, 11–27. New Delhi: Sage.
Johnston, M. J. R. (1994). The Syntax and Semantics of Adverbial Adjuncts. Ph.D. Dissertation.
University of California Santa Cruz.
Koopman, H., & Sportiche, D. (1989). Pronouns, logical variables and Logophoricity in Abe.
Linguistic Inquiry, 20, 555–589.
Kratzer, A. (1998). Scope or pseudoscope? Are there wide-scope indefinites? In S. Rothstein (Ed.),
Events and grammar (pp. 163–196). Dordrecht: Kluwer Academic Publishers.
Kuno, S. (1987). Functional syntax: Anaphora, discourse and empathy. Chicago: University of
Chicago Press.
Kuroda, S.-Y. (1973). Where epistemology, style, and grammar meet: A case study from Japanese.
In S. Anderson & P. Kiparsky (Eds.), A Festschrift for Morris Halle (pp. 377–391). New York:
Holt, Rinehart & Winston.
Lewis, D. K. (1973). Causation. Journal of Philosophy, 70, 556–567.
Maling, J. (1984). Non-clause-bounded reflexives in modern Icelandic. Linguistics and Philosophy,
7, 211–241.
Patel-Grosz, P. (2012). (Anti-)locality at the interfaces. Ph.D. Dissertation. MIT.
Pearson, H. (to appear). Attitude verbs. In L. Matthewson, C. Meier, H. Rullmann, & T. E.
Zimmermann (Eds.), Companion to Semantics. Wiley.
Pollard, C., & Sag, I. A. (1992). Anaphors and the scope of binding theory. Linguistic Inquiry, 23,
261–303.
Postal, P. (2006). Remarks on English long-distance anaphora. Style, 40, 7–19.
Potts, C. (2005). The logic of conventional implicatures. Oxford University Press on Demand.
Ross, J. R. (1970). On declarative sentences. Readings in English Transformational Grammar,
222–272.
Rutherford, W. (1970). Some observations concerning subordinate clauses in English. Language,
46, 97–115.
Ruwet, N. (1990). En et y: deux clitiques pronominaux antilogophoriques. Langages, 25(97), 51–
81.
Sæbø, K. J. (1991). Causal and purposive clauses. In A. von Stechow & D. Wunderlich
(Eds.), Semantik – Semantics. Ein internationales Handbuch zeitgenössischer Forschung – An
International Handbook of Contemporary Research (HSK 6) (pp. 623–631). Berlin: de Gruyter.
Sells, P. (1987). Aspects of Logophoricity. Linguistic Inquiry, 18, 445–479.
Solberg, P. E. (2017). The discourse semantic of long-distance reflexives. Ph.D. Dissertation,
University of Oslo.
Solstad, T. (2010). Some new observations on ‘because (of)’. In M. Aloni, H. Bastiaanse, T. de
Jager, & K. Schulz (Eds.), Logic, language and meaning: 17th Amsterdam colloquium (pp. 436–
445). Berlin: Springer.
Speas, M., & Tenny, C. (2003). Configurational Properties of Point of View Roles. In A. M.
DiSciullo (Ed.), Asymmetry in Grammar (pp. 315–344). Amsterdam: John Benjamins.
Stephenson, T. (2007). Judge dependence, epistemic modals, and predicates of personal taste.
Linguistics and Philosophy, 30, 487–525.
Sundaresan, S. (2012). Context and (co)reference in the syntax and its interfaces. Ph.D. Disserta-
tion. University of Tromsø and University of Stuttgart, Tromsø.
Thráinsson, H. (1976). Reflexives and subjunctives in Icelandic. In Sixth annual meeting of the
North Eastern Linguistics Society, pp. 225–239.
von Fintel, K., & Gillies, A. (2007). An opinionated guide to epistemic modality. Oxford Studies
in Epistemology, 2, 32–62.
Williams, E. S. (1974). Rule Ordering in Syntax. Ph.D. Dissertation. MIT.
Zu, V. (2018). Discourse participants and the structural representation of the context. Ph.D.
Dissertation. New York University.
Part V
Philosophical Inquiries on Causation
Chapter 14
Causes as Deviations from the Normal:
Recent Advances in the Philosophy
of Causation
Georgie Statham
Abstract There have recently been a number of important advances in the philos-
ophy of causation, which impact our understanding of both the nature of causation
and of causal reasoning. Two stand out in particular: First, a large body of work
on the way that normative factors can influence causal judgement casts doubt
on the intuitive idea that causation is a purely natural relation, independent of
human interests and values. Second, the so-called ‘causal modelling framework’—
developed by computer scientists and statisticians as a formalism for discovering
causal relations—has turned out to be a powerful and extremely fruitful method
for representing causal systems. It has also been incorporated into the philosophy
of causation as the basis of James Woodward’s influential interventionist (or
manipulability) theory (Woodward 2003). The aim of this paper is to provide an
introduction to these recent developments, to show how they are related, and to
comment on their relevance to linguistics.
Keywords Causation · Interventionism · Manipulability · Woodward · Norms ·

Linguistics
Let’s start by considering one of the earliest statements of the idea that everyday
(token) causal judgements are sensitive to normative considerations. In the passage
cited, H. L. A. Hart and Tony Honoré describe what they call the ‘common-sense
concept of cause’. This is a notion of causation that we frequently make use of in
everyday life; it is also of fundamental importance in disciplines like history and
law.
The notion, that a cause is essentially something which interferes with or intervenes in the
course of events which would normally take place, is central to the common-sense concept
of cause . . . Analogies with the interference by human beings with the natural course of
events in part control, even in cases where there is literally no human intervention, what is
G. Statham ()
Polonsky Academy Fellow, The Van Leer Jerusalem Institute, Jerusalem, Israel

https://doi.org/10.1007/978-3-030-34308-8_14
446 G. Statham
to be identified as the cause of some occurrence; the cause, though not a literal intervention,
is a difference to the normal course which accounts for the difference in the outcome (Hart
& Honoré (1959, 27), italics in the original).
Hart & Honoré contrast this common-sense concept with the scientific concep-
tion of cause, which, they claim, aims to ‘discover connexions between types of
events’ (1959, 9). The idea that there are separate everyday and scientific concepts
of causation—where the former attributes responsibility to particular events and the
latter is concerned with making causal generalisations—is common in the literature,
where the former is referred to as either ‘actual causation’ or ‘token causation’, and
the latter as ‘type causation’. Following Hart & Honoré, the claim that causal claims
are influenced by normative considerations is generally taken to be exclusively
restricted to the former.
Peter Menzies points out that in the passage cited above, Hart & Honoré make
three claims about judgements of actual (or token) causation. (i) since we pick
out causes relative to a kind of system, applying this concept of causation to any
particular situation requires that we think of it as part of a system; (ii) we assume
that this system, if it is not subject to any outside intervention, will follow a normal,
or natural course; and (iii) we identify a cause as something that makes a difference,
in that it corresponds to some kind of intervention to, or difference from, the normal
course (2007, 201–202).
Each of the features of everyday causal reasoning listed above connects to one
of the recent developments in the philosophy of causation that I will discuss. (i) and
(iii) connect the passage from Hart & Honoré to the causal modelling framework,
because first, this is perfectly suited to representing the kinds of causal systems
they refer to; and second, at the heart of this framework is a technical notion of
intervention that has enabled philosophers to formalise, and therefore clarify, the
idea that causes are interventions on the normal course of events. (ii), on the other
hand, lies at the centre of recent work on the influence of the normative on (token)
causal judgements.
In the next two sections, I focus on (iii), the claim that the events we pick out
as actual causes are (typically) deviations from the normal course of evolution
of a system, and how this leads to the implication that token causal judgements
(at least) are affected by normative considerations. In Sect. 14.3, I turn to the
distinction between type and token (or actual) causal claims—that is, between causal
generalisations and statements about particular events. I show, on the one hand, that
token causes are not always deviations from the normal, and on the other, that there
is a class of type causal claim that selectively picks out events that are deviations
from the normal. Thus, the taxonomy of kinds of causal claims is more complicated
than is generally acknowledged—our everyday and scientific concepts can’t be
as neatly separated as Hart & Honoré and others have assumed. In Sect. 14.4, I
introduce the causal modelling framework and the associated interventionist theory
of causation, and show how this can be used to flesh out the idea that causes are
(often) deviations from the normal. Finally, in Sect. 14.5 I discuss the implications
of these recent developments for linguistics.
14 Causes as Deviations from the Normal 447
14.1 Causes as Deviations from the Normal
Accounts of causation can be divided into two kinds: process theories and
difference-making theories.1 According to the former, causation is a physical
process that involves the transfer of some physical quantity: for example,
energy or momentum.2 According to the latter, to be a cause of some event is
to make a difference to whether or not that event occurs. Counterfactual and
probabilistic theories are examples of this kind of account.3 The recently influential
interventionist theory of causation (discussed in Sect. 14.4) is also a difference
making theory. This is cashed out in terms of ‘interventionist counterfactuals’—that
is, counterfactuals about the outcomes of intervening in a system. Thus, I assume a
broadly counterfactual approach to causation throughout this paper.
The most basic counterfactual account states that one event, c, is a cause of a
second event, e, if and only if it is true that if c had not occurred, e would not have
occurred. Combining this basic counterfactual account with the idea that causes
are interventions that represent a deviation from the normal leads to the claim that,
roughly, a cause, c, is an abnormal intervention in a system, such that if c had not
occurred, and the system had been left to follow its normal course, the effect, e,
would not have occurred, either.
Of course, normal events also have causes and can be causes, so the kind of
account just sketched can’t possibly account for all token causal claims (I return
to this point at the end of the next section). Intuitively, however, it does capture an
important part of everyday causal reasoning. Consider this commonplace example:
I get into my car, turn the key, and the engine coughs, splutters, and fails to start. Frustrated,
I immediately start to try to work out what is wrong with my car.
The failure of my car to start is a deviation from the normal course of events,
in which the engine starts when the key is turned. In order to determine the cause
of this problem, it is necessary to understand the car (or the car’s ignition system)
as a mechanical system, consisting of a large number of components, which all
have a particular role to play. We assume that the cause of the car’s failure to
start is an alteration to one of the components of the system, such that it no longer
operates within normal parameters, does not play its usual role within the system,
and therefore results in the car being unable to start.
The kind of causal reasoning described in the above example is common in
everyday life. As we will see in the next section, a number of philosophers endorse
accounts of causation that attempt to capture this kind of inference, in which we
1 For an overview of existing theories of causation from two linguists’ perspective, see Copley &
Wolff (2015).
2 For classic expositions of the process theory, see Salmon (1994) and Dowe (2000).
3 For the counterfactual theory, see Lewis (1986) and the papers in Collins et al. (2004). For
examples of the probabilistic theory, see Eells (1991) and Salmon (1993).
448 G. Statham
attempt to causally explain abnormal events by finding causes that are themselves
deviations from the normal.
In the next section, I show that this recent focus on the idea that causes are
(typically) deviations from the normal has lead many philosophers to think that our
causal judgements are influenced by normative factors in a way that may initially be
surprising.
14.2 Causation and the Normative
According to a traditional, and intuitive, understanding of causation, causation is a

natural relation—that is, a relation that holds between events in the natural world.
On this conception, whether or not event c is a cause of event e is independent
of human interests and values, and in particular, of any normative commitments.
However, an increasingly large body of work questions this assumption. Following
Hart & Honoré, both traditional philosophical analyses4 and empirical studies into
the causal claims that people actually endorse5 suggest that we are more likely to
cite abnormal events as causes, where, importantly, the relevant notion of ‘normal’
includes both descriptive and prescriptive norms. If this is right, then the prescriptive
norms we accept—that is, our normative commitments—can make a difference to
our causal judgements.
The claim that our token causal judgments are (and should be) influenced by our
normative commitments is motivated by examples like the following:
I have a pot plant in my office. I go on holiday, and ask my colleague, Casper, to water it
while I’m gone. He forgets, and it dies.
Here, it seems right to say that Casper’s failure to water the plant caused its
death.6
However, Queen Elizabeth II didn’t water my plant either, and it is also true that
if she had watered it, it would have survived. That is, her omission has the same
counterfactual structure as Casper’s. Nevertheless, we don’t think that the Queen’s
failure to water the plant caused its death. The difference here seems to be that
Casper promised, whereas the Queen didn’t (the Queen was thousands of kilometres
away and didn’t even know of the existence of my plant).
4 For example Hall (2007a), Halpern & Pearl (2005), Hitchcock (2007a), McGrath (2005) and
Menzies (2004, 2007, 2009).
5 These studies have been carried out in the fields of experimental philosophy and cognitive
psychology. For example, see Alicke et al. (2011), Hitchcock & Knobe (2009), Knobe (2010),
and Systma et al. (2012).
6 The phrase ‘Casper’s failure to water my plant’ refers to an omission—that is, the non-occurrence
of an event, rather than the occurrence of an event. One of the advantages of the counterfactual
approach to causation is that it allows that omissions can be causes, as does natural language
(think of negligence, for example).
The relevance of the promise can be brought out even more clearly if we note
that I have another colleague, Leora, who’s office is closer to mine than Casper’s
is, and who was also at work while I was away. She didn’t water my plant while I
was away either. Nevertheless, because she didn’t promise to, it doesn’t seem right
to say that her failure to water my plant caused its death.
As noted above, following Hart & Honoré, a popular response to this kind
of example asserts that we are more willing to cite abnormal events as causes,
where, to reiterate, the relevant notion of ‘normal’ includes both descriptive and
prescriptive norms. As noted above, this idea is supported by empirical evidence
from experimental philosophy and cognitive psychology. It is illustrated by the
following examples:
(1) Almog’s height caused him to hit his head.
(2) The power outage caused the oven to turn off.
(3) The driver’s speeding caused the crash.
In examples (1)–(3), the causes are all deviations from some norm. In (1), we
assume the fact that Almog is taller than average explains why he hit his head. Thus,
his height is a deviation from a statistical norm, a kind of descriptive norm. To see
that we really do selectively pick out causes that are deviations from the normal,
notice that if Almog had been of average height, and had still hit his head (it was
a particularly low doorway, say), we would be far more likely to say that the low
lintel was the cause.
The power doesn’t normally go out (it shouldn’t go out). The power outage in
(2) is therefore a deviation from the norm of proper functioning of the electricity
system. Finally, by speeding, the driver in (3) is certainly breaking the law—a legal
norm—and also a moral norm, if he is recklessly endangering other people’s lives.
The view that I have just outlined has been prominently defended by Christopher
Hitchcock & Joshua Knobe (2009). They argue that the reason we are more likely to
cite abnormal events as causes is importantly linked to our ability to intervene in the
world—that is, to the fact that we are not just passive observers, but can manipulate
the course of events to bring about outcomes we want. To illustrate this point, they
consider the case of a student who has failed a test, and wants to prevent it from
happening again. They point out that the following counterfactuals are all true, and
therefore all correspond to possible strategies (at least in theory).
(4) I would not have gotten an F if the teacher had been eaten by a lion.
(5) I would not have gotten an F if the Earth’s gravitational pull had suddenly
decreased.
(6) I would not have gotten an F if I had less to drink the night before the test.
(Sentences taken from Hitchcock & Knobe (2009, 591)).
Counterfactuals (4)–(6) pick out three events (or omissions) that could be
attributed as causes of the student failing, namely her teacher not being eaten by
a lion, the Earth’s gravitational field remaining constant, and her heavy drinking the
night before the test. However, only (6) involves replacing an event that is deviation
from the normal (drinking too much is a deviation from a prudential norm) with a
450 G. Statham
more normal alternative. Fairly clearly, this is also the only counterfactual that the
student should consider to be relevant, in the sense that it identifies an appropriate
target of intervention. Thus, while it may well be possible for the student to avoid
failing her next test by somehow ensuring that her teacher gets eaten by a lion, we
can see why it might be useful to have a concept that identifies the fact that she
drank a lot the night before the test as the cause of her failing. Generalising from
this example, Hitchcock & Knobe’s idea is that abnormal events are often ‘suitable
targets of intervention’ (2009, 591), and that this explains why we have a concept
(which they refer to as ‘actual causation’) that selectively picks out causes that are
deviations from the normal. Their account of this concept can be further illustrated
by returning to the example of my plant.7
Casper’s failure to water my plant is a deviation from the moral norm that says we
should keep promises. Thus, we automatically consider the counterfactual situation
in which Casper does water it. Since we judge that if he had done so, the plant
would have survived, we say that his omission was the cause of its death. Neither
the Queen nor Leora broke any norms, however (assume that my office door was
kept closed and therefore that Leora didn’t see that my plant was becoming less and
less healthy). Thus, we don’t consider the possibility that they could have watered
my plant to be relevant, and don’t judge them to have caused its death.
In summary, Hart & Honoré’s observation that when identifying the causes of
particular occurrences we tend to selectively pick out events that are deviations
from the normal, has been supported by empirical studies and incorporated into
many accounts of the concept of causation. In Sect. 14.5, I show that these kinds
of accounts have further implications for both philosophy and linguistics, because
they force us to rethink the traditional understanding of the connection between the
metaphysics of causation and the semantics of the verb ‘cause’.
Before moving on, it is important to realise the limitations of the kind of account
just discussed. As briefly mentioned in the introduction, the common-sense concept
of causation is often contrasted with the scientific concept of causation. The idea
is that the common-sense concept generates token causal claims, where these token
causes are deviations from the normal. The scientific concept of causation, on the
other hand, is used to make type causal claims, and there is no requirement that these
type causes are abnormal.8 However, notice that as soon as we conceptualise part
of the world as a system with a certain normal course of evolution, it is possible to
enquire about token instances of causation within the normally functioning system,
as well as causes that are deviations from this system. For example, I can ask what
caused my car to start yesterday, when it was working properly, which is just to
ask about the causal structure of the system consisting in the normal running (or
7 Hitchcock & Knobe frame their argument using the causal modelling framework, and their talk
of ‘intervening’ is naturally associated with the formal notion of an intervention that has been
developed within this framework and the associated interventionist theory. However, this is to some
extent misleading: by ‘suitable targets of intervention’, they just mean those events that make sense
for us to try to manipulate.
8 See for example Hitchcock (2007b) and Woodward (2011).
at least starting) of my car. The kind of account we have just been considering
can’t accommodate this kind of case—that is, it can’t account for instances of
causation that are not deviations from the normal. This suggests that our token
causal judgements need to be divided into (at least) two different categories—the
first in which causes are part of the normal course of the evolution of a system, and
the second in which causes are deviations from the normal course of evolution.
In the next section, I argue that there is also a class of type causal claim that
selectively picks out causes that are deviations from the normal. In other words, I
dispute the accepted taxonomy, and provide an alternative.
14.3 Type and Token Causal Judgements
We have seen that in the philosophy of causation, a distinction is made between type
and token (or actual) causal claims. Type causal claims describe generalisations that
hold between kinds of events (e.g. ‘Shots to the head cause death’), whereas token
causal claims describe particular situations, and assert that one event is causally
responsible for another event (e.g. ‘Simon’s getting shot in the head caused his
death’).
There is clearly some connection between type and token causal claims. How-
ever, this connection is not simple. For example, it is possible for there to be a type
causal relation between two kinds of events, C and E; for tokens of both of these
kinds of events, c and e, to be present in a particular situation; and yet it not be true
that c caused e. For example, it is possible that Simon is shot in the head and dies
an hour later, but that the shot to the head was not the cause of his death—perhaps
he would have survived the shot, but was exposed to a lethal dose of cyanide just
afterwards. In that case, the cause of death was the poisoning, not the shooting.
The above example shows that having a complete knowledge of the event tokens
that are instantiated in a particular scenario, as well as the type causal relationships
that hold between events of these types, is not enough to determine the token
causal structure. To see this, note that in our example, the token events that are
instantiated are Simon being shot in the head, being poisoned with cyanide, and
dying. The relevant type causal relationships are that shots to the head cause death,
and that poisonings also cause death. However, all this information is not enough
to determine what caused Simon’s death. Because type and token causal claims can
come apart in this way, philosophers usually give different accounts of type and
token causation.
Hart & Honoré’s claim that causes are deviations from the normal is only
intended to apply to judgements of token causation. When describing the context
in which the common-sense concept of causation is employed, they claim that in
everyday life (and in the law) our concern is primarily to ‘apply generalizations,
which are already known or accepted as true or even platitudinous, to particular
concrete cases’ (1959, 9). Many recent works also assume that only the concept of
token (or actual) causation specifically picks out events that are deviations from the
452 G. Statham
Table 14.1 A taxonomy of causal claims

Normal type Deviant type
e.g. ‘The moon’s gravitational field e.g. ‘Flat batteries cause cars to fail to start’
causes the tides’
Normal token Deviant token
e.g. ‘The moon’s gravitational field caused e.g. ‘The battery’s being flat caused my
the high tide this mornings’ car to fail to start this morning’
normal (e.g. Menzies 2007; Hitchcock 2007b; Hitchcock & Knobe 2009; Woodward
2011). However, this is a mistake. A point that has generally been overlooked is
that just as we say that the battery’s being flat caused my car’s failure to start
this morning, we also say that flat batteries are a cause of cars failing to start in
general. Thus, it is not just token causes that are often deviations from the normal.
Rather, the distinction between deviant and normal causal judgements is orthogonal
to the distinction between type and token causes. There is therefore (at least) a
fourfold taxonomy of causal judgements: normal type, deviant type, normal token
and deviant token. These are summarised in Table 14.1.
The category ‘deviant token causation’ roughly corresponds to the category
that is often referred to in the literature in the philosophy of causation as ‘actual
causation’, or ‘token causation’ (or even just ‘causation’). Deviant token causal
judgements are often backwards-looking, and are also attributions of responsibility,
in the sense that to make a deviant token causal judgement is to claim that one event
is causally responsible for another event. However, the important point is that this is
merely one of many acceptable kinds of causal judgement.
In the next section, I turn to the second recent development in the philosophy
of causation, namely the causal modelling framework, and its incorporation into
the interventionist theory of causation. Although this is separate from the work on
causation and the normal that I have discussed so far, we will see that these two
recent developments are naturally combined.
14.4 The Causal Modelling Framework
The causal modelling framework is a powerful system for representing the structure
of causal systems that was developed by computer scientists and statisticians as
a method for causal discovery—that is, for extracting the existence of causal
relationships from merely correlational data.9 In this framework, causal structure
is represented using graphs that consist of a set of variables and arrows, or directed
edges, each of which represents the existence of a causal relationship between two
9 See Pearl (2000) and Spirtes et al. (2000).

Variables
RH R: the amount of rainfall in the catchment area
HH
H V : the extent to which slopes are covered in vegetation
V - RL - F S : the steepness of the slopes

RL: the river level
S F : whether or not there is a flood
Fig. 14.1 Causes of flooding
variables. For example, the graph in Fig. 14.1 represents a causal structure that is
instantiated in many river catchments.
According to Fig. 14.1, the amount of rainfall in the catchment area, the extent to
which slopes are vegetated, and the steepness of the slopes, are all causally relevant
to the river level. And the river level determines whether or not there is a flood.
Now, we generally want to know more than just that one variable is causally
relevant to another variable; we also want to quantify this causal relevance. For
example, we want to know how much rainfall in a particular river catchment is likely
to result in a flood. This quantitative information is incorporated into causal models
using structural equations, where the structural equation for each effect variable
gives its value as a function of its (direct) causes. For example, the causal model
represented in Fig. 14.1 would include one structural equation expressing RL as a
function of R, V & S, and a second giving F as a function of RL.
The causal modelling framework provides the basis for the interventionist theory
of causation, which has recently been popularised in philosophy by Woodward
(2003). On this theory, variable X is causally relevant to variable Y if and only if
there is a possible intervention on X that would make a difference to the value of Y.
For example, the amount of rainfall is causally relevant to the river level, because if
we manipulated the amount that it rained in a particular region (assuming this were
possible), this would change the river level.
It is important to note that within the interventionist theory, ‘intervention’ is a
technical term that is characterised using causal models. Roughly, an intervention
on X with respect to Y has to be a cause of X, and has to affect Y (if at all) only via
X.10 This idea is best exemplified by a random controlled trial: the whole point of
this experimental design is to ensure (as best as possible) that confounding factors
are controlled for—that is, that any effect on the dependent variable (Y) is due to the
independent variable (X).11
In the terminology introduced at the start of Sect. 14.1, interventionism is a
difference-making theory; one of the advantages of this account over process
10 ForWoodward’s detailed characterisation of the notion of an intervention, see (2003, 98–99).

11 Notice that ‘intervention’ is itself a causal notion. This entails that interventionism is a non-
reductive theory of causation—that is, it doesn’t attempt to reduce causal facts to facts about some
non-causal phenomenon. Since Woodward doesn’t intend to provide a metaphysics of causation
(see Sect. 14.5), this doesn’t create a problem.
454 G. Statham
theories is that to establish the existence of a causal relation it is enough to

demonstrate the right kind of responsiveness to intervention. We don’t need to
understand the mechanism (or process) linking the cause and effect, or even to show
that there is one.12
The connection between interventionism and causal discourse is blurred because
of the fact that the theory is intended to be an account of causal inference, rather than
causal locution. In other words (as explained in more detail below), it is intended as
an account of the practice of causal reasoning, not just of sentences involving the
verb ‘cause’ (and perhaps other causal verbs). Woodward writes that:
[M]y focus is not just on how people use words, but on larger practices of causal inference
and explanation in scientific and nonscientific contexts, practices that involve substantial
nonverbal components . . . my project focuses on (what I take to be) the purposes or goals
behind our practices involving causal and explanatory claims; it is concerned with the
underlying point of our practices. (2003, 7)
As such, Woodward intends his theory to be normative, rather than purely

descriptive—that is, it is intended to describe how we ought to go about causal
reasoning, not just how we do go about causal reasoning (Woodward 2003, 7; 2015,
3578). I discuss this feature of Woodward’s account in the next section. For now,
what is relevant is that we can’t expect the interventionist theory to align neatly
with linguistic use of the verb ‘cause’.
There are two features of interventionism that are especially relevant to the above
point. First, the theory is primarily intended as an account of type causal claims, not
token (or actual) causal claims. We saw above that according to interventionism,
variable X is causally relevant to variable Y if and only if certain conditions are met.
The kind of causal claim that is the primary focus of Woodward’s account therefore
asserts that one variable is causally relevant to another—for example that mass is
causally relevant to acceleration, or that the amount of rainfall is causally relevant
to whether or not there is a flood. Since most causal claims do not have this form,
the theory doesn’t have anything to say about most causal discourse. In particular,
it doesn’t say anything about token causal claims. This is easily remedied, however:
Woodward gives an interventionist account of token causation (2003, 74–86), to
which I return below. In fact, one advantage of the causal modelling framework is
that it makes explicit the link between type and token causal claims.
The second feature of interventionism that is relevant to its connection to causal
discourse is that it is a very inclusive account of causation. On this theory, for X to
be a cause of Y, there only has to be some conceptually possible intervention on X,
in some background circumstances, that makes a difference to Y. Thus, for example,
the oxygen levels in the earth’s atmosphere count as a cause of my having cereal for
breakfast this morning, because if the oxygen concentration had been significantly
different, I wouldn’t have eaten the cereal (you don’t eat much cereal when you are
12 This is not to say that causes and effects are not (generally) connected by a physical process,
but just that according to interventionism, the existence of a particular kind of physical process is
neither necessary nor sufficient for the existence of a causal relationship.
dead). For this reason, interventionism is better understood as an account of causal

relevance than of the verb ‘cause’.
Again, this means that the theory, as currently formulated, does not neatly align
with utterances of the word ‘cause’ (or even with a broader construal of causal
discourse). However, it would be mistaken to conclude that the causal modelling
framework and the associated interventionist theory of causation are irrelevant to
understanding causal language, and in particular, to the concept of actual causation.
To see this, let’s return to the passage from Hart & Honoré cited in the introduction,
and the three features of causal reasoning that are implied.
Recall that these are (i) that causal reasoning requires that we represent causal
systems; (ii) that these systems are assumed to have a normal course of evolution;
and (iii) that many (token) causes are interventions on this normal course. The
causal modelling framework obviously speaks to (i), since this is designed to be
a sophisticated way of representing causal structure. Furthermore, both the causal
modelling approach and the interventionist theory are connected to (iii), since these
both take causes to be difference-making interventions.13
There is no general requirement that causal models represent systems that have
a normal course of evolution—that is, that (ii) holds, nor that (actual) causes are
specifically deviations from such a normal course. However, Menzies shows that
the interventionist account can be modified to specifically pick out the category that
I have referred to as ‘deviant token causal claims’ (see Sect. 14.3).14
Menzies’ Account: variable X taking the value x causes variable Y to take the value y relative
to the default values15 x and y if and only if:
(i) the actual values of X and Y are x and y respectively; and
(ii) if an intervention were to change the value of the X variable from x to x, the value of
the Y variable would change from y to y.16
Menzies’ account applies to a situation in which the actual values of both X and
Y are different from the default values—that is, they are abnormal. However, if X
had taken its normal value, then Y would also have taken its normal value—that is,
13 Causal models are further connected to (iii) in that each model encodes a set of counterfactuals.
For example, we have seen that Fig. 14.1 asserts that there is a possible intervention on the amount
of rainfall (R) that makes a difference to the river level (RL). This entails that there is a true
counterfactual with the following form: if it were to rain x amount (rather than x amount), the
river level would be y (rather than y ).
14 See also Hitchcock (2007a) and Hall (2007a). Note that the interventionist theory can also be
used to give an account of the other kinds of causal claims listed in Table 14.1. See Statham
(2017).
15 The default values are the values that the variable normally takes (i.e. those that it takes in the
normal course of evolution).

16 Adapted from Menzies (2009, 360).
456 G. Statham
Variables
CW XX CW : whether or not Casper waters my plant
XX
A QW : whether or not the Queen waters my plant

QW A: whether or not the plant is alive when I return
Fig. 14.2 Potential causes of my plant’s death
it was the intervention on X that resulted in the abnormal value of Y.17 Notice that
this is exactly the situation that Hart & Honoré describe.
To see how Menzies’ account is incorporated into the causal modelling frame-
work, let’s return to Fig. 14.1, the causal graph that is intended to represent many
river catchments. This can be used to represent the normal state of a particular
river catchment by setting V (the extent to which slopes are vegetated) and S (the
steepness of slopes) to their actual values; R (the amount of rainfall) to a value
corresponding to the average rainfall within the river catchment18 ; RL (the river
level) to its normal value; and F (whether or not there is a flood), to no. A period
of unusually heavy rain thus corresponds to a deviant value of R, which makes a
difference to the value of RL and (if it is heavy enough) also to F.
In the example just considered, the default values represent descriptive norms.
However, in many other systems, the default states will represent prescriptive norms.
For example, we can represent the situation in which Casper promises to water my
plant while I am away using the causal model in Fig. 14.2.
In our example, the default values of the variables are CW = yes, QW = no and
A = yes. The actual values of the variables (as specified) are CW = no, QW = no
and A = no. Thus, relative to the default model, Casper’s failure to water my plants
(unlike the Queen’s) was a difference-making deviation, and therefore counts as the
(actual) cause of the plant’s death. Here, the default value of CW is an instantiation
of a prescriptive norm, namely that we should keep promises.
To conclude, although there is no essential connection between the causal
modelling framework/interventionist theory of causation and recent work on the
ways that normative factors influence our causal judgements, these two recent
developments in the philosophy of causation are naturally combined. In the final
section, I discuss some implications of these recent developments for linguistics.
17 Menzies expresses his account more formally as follows:

A value of a variable X makes a difference to the value of another variable Y in a default
causal model if and only if plugging in the default values of the variables in the structural
equations yields X = x and Y = y and there exist actual values x = x and y = y such that
the result of replacing the equation for X with X = x yields Y = y (2007, 208, italics in the
original).
18 Perhaps over a particular period of time: the average rainfall for June, say.
14.5 Relevance to Linguistics
The philosophy of causation has traditionally been seen as a branch of metaphysics.

Because of the traditional understanding of the relationship between metaphysics
and semantics, this means that when giving an account of causation, philosophers
are generally aiming to answer two questions. First: what is the semantics of
‘cause’? The accepted way of answering this question is to provide a set of necessary
and sufficient conditions, such that any particular causal claim19 is true if and only
if those conditions are satisfied. Second: which feature of the world makes causal
claims true (or grounds these claims)? In other words, what is the metaphysics of
causation? These two projects are generally taken to be related, due to a commitment
to a certain picture of the link between language and the world, which I will call
‘representationalism’. According to representationalism, the function of statements
is to represent the world; true statements are therefore those that succeed in doing
so. On this picture, to give a correct semantics of causation is to identify the parts of
the world to which causal statements refer—that is, to identify the truth-makers of
causal claims. If this is right, then to provide a semantics of ‘cause’ is also to provide
a metaphysics of causation; the two questions introduced above are answered by a
single enquiry.20
Woodward, however, takes himself to be engaged with the semantic question,
but not the metaphysical question. He says that his ‘aim is to give an account of
the content or meaning of various [causal] locutions’ (2003, 38). Thus, he is clearly
engaged in a project that can be understood as semantic.21 However, he describes his
project as ‘methodological’, rather than metaphysical. It is focused on elucidating
how we think about, learn about, and reason with various causal notions’ (2008,
194), rather than determining their metaphysical foundations.22
To be clear, Woodward is committed to there being something mind-independent
that underpins causal relationships, and that we can latch on to when we manipulate
the world (Woodward 2003, 118–123). It is just that he is not committed to any
particular understanding of the nature of these relations. Because his project is
methodological rather than metaphysical, Woodward takes himself to be addressing
19 The philosophers’ term ‘causal claim’ is ambiguous between an actual piece of causal
discourse—that is, a causal locution—and an abstract causal statement, independent of any actual
utterance. In this section I have disambiguated by referring to the former as a causal locution, and
only the latter as a causal claim.
20 Philosophers of causation working within the counterfactual (Lewis 1986; Collins et al. 2004),
agency (Menzies & Price 1993), regularity (Paul & Hall 2013), and process theories (Salmon 1994;
Dowe 2000) all take themselves to be doing metaphysics.
21 Woodward himself describes his project as ‘semantic or interpretive’ (2003, 38).
22 For a defence of the claim that interventionism should be seen a methodological project, see
Woodward (2014, 2015). Roughly, he argues that the methodological questions he wants to answer
are largely independent of metaphysical considerations, and that interventionism is consistent with
a range of different positions in the metaphysics of causation (2008, 194), between which he has
no interest in adjudicating.
458 G. Statham
different questions to those that have traditionally been seen as the province of
the philosophy of causation; questions including: How should we go about causal
reasoning? What is the purpose of the concept of causation? And are there important
differences between different (kinds of) causal locutions?
At first glance, linguistic approaches to causation may appear to be closely
connected to the tradition of conceptual analysis, and to the questions asked by
these metaphysically-orientated philosophers. Perhaps linguists also take these to
be the relevant questions. However, I want to suggest, first, that moving away from
metaphysics might be a good move, for both philosophers and linguists; and second,
that the expertise of linguists could help assess some of the claims being made by
less metaphysically-oriented philosophers of causation.
The first point to make is that there are good reasons to question the assumption
that we can read the metaphysics of causation off our causal discourse. Certainly the
claim that judgements of actual causation are influenced by normative commitments
suggests that this concept isn’t directly connected to fundamental metaphysics—
prescriptive norms like the requirement that we keep promises don’t seem to be
the kinds of things that ‘carve nature at the joints’ as metaphysicians like to put
it. Additionally, causal models—and thus our most sophisticated representations of
causal structure—are understood to be context sensitive, and the choice of which
variables to include is acknowledged to be interest relative.23 Again, this casts doubt
on the claim that causal claims succeed in referring to fundamental joints in nature.
An important advantage of giving up the idea that there is a tight connection
between everyday causal claims and the metaphysics of causation is that it allows
us to answer many questions about causal discourse and causal reasoning without
having to first adjudicate on the metaphysics of causation. For example, causation
is most commonly held to be a relation between events. This makes intuitive
sense in many cases—for example, the sentences ‘Nea’s flipping the switch caused
the light to turn on’ and ‘The lightning strike caused the fire’ both appear to
describe a relationship between two events. However, many causal claims cite
causes and/or effects that are not obviously events. Consider: ‘The state of high
unemployment caused the socio-economic instability’, ‘The driver’s negligence
caused the accident’, and ‘The tightness of her trousers caused her stomach ache’.
In the first sentence, the cause looks like a state of affairs, in the second an omission,
and in the third a property.
There are ways of getting round problematic examples like the three listed above,
for example, by defining an expansive notion of ‘event’,24 and translating seeming
counterexamples into event terminology.25 However, it is far more natural to take
the interventionist approach, according to which causes and effects are variables (or
values of variables) and there are no restrictions on which metaphysical category
23 See for example Hitchcock (2007a) and Woodward (2016).

24 The notion of ‘event’ that is used in the philosophy of causation is generally accepted to include
states of affairs.
25 For a good introduction to this problem, see Hall (2007b).
these must fall into.26 Using the interventionist theory, philosophers of causation
have been able to analyse causal reasoning and discourse without assuming that
causal claims are closely connected to metaphysical facts. Perhaps linguists, too,
could benefit from being able to focus on language, without being waylaid by
questionable metaphysical distinctions.
As soon as we give up the idea that there is a direct connection between the
semantics of causal claims and the (fundamental) metaphysics of causation, we
are faced with a set of questions about what we are doing when we make causal
claims—that is, we are faced with Woodward’s questions.
Let’s consider one such question: what is the purpose of causal reasoning?
Woodward’s answer is that the purpose is to identify correlations that are exploitable
for the purpose of manipulation and control (2003, 9–12)—that is, we care about
causal relations, as opposed to mere correlations, because the former tend to be
stable under intervention; they are therefore handles that we can exploit in order to
manipulate the world. As it stands, this assertion from Woodward has to be taken
as a plausible sounding hypothesis: he doesn’t back it up with empirical evidence.
However, linguistics could conceivably acquire evidence for or against this claim,
by asking questions like: ‘Is it true that the purpose of most causal discourse is
to enable us to better manipulate the world?’ and ‘What other purposes does causal
discourse serve?’ Comprehensive answers to these questions would also help answer
the normative question of how closely the concept of causation should be linked to
the notion of an intervention.
We can think of Woodward’s claim about the purpose of causal reasoning as
a theoretically derived hypothesis about the role played by causal discourse and
causal reasoning more generally. Since this hypothesis could plausibly be tested
empirically, it can be thought of as a starting point for research by more empirically
oriented disciplines, including linguistics.
14.6 Conclusion
In this paper, I have introduced recent work in the philosophy of causation on the
interaction between normative commitments and causal reasoning, and argued for
the need to distinguish between (at least) four different kinds of causal claims:
normal type, deviant type, normal token and deviant token. This distinction has
significance for both philosophy and linguistics: with respect to the former, it
suggests the need for a rethinking of the traditional classification of types of
causal claims; for the latter, it can be thought of as a hypothesis ripe for empirical
confirmation (or disconfirmation).
26 There are, however, non-metaphysical restrictions on the choice of variables. See Hitchcock
(2007a, 520–503).
460 G. Statham
In the second half of the paper, I discussed the incorporation of developments

from computer science and statistics in the form of the causal modelling framework.
Finally, I showed that both these recent developments have implications for how
linguists approach the analysis of causal discourse, and could also benefit from
linguists’ input. Thus, I conclude that both linguistics and philosophy would benefit
from further collaboration.
I end by considering two further issues that arise from within the interventionist
approach to causation that also provide potential starting points for empirical
research, and thus could (and perhaps should) also contribute to the discussion of
causation beyond philosophy.
First, Woodward argues that, just as important as determining the difference
between causal and non-causal relationships is ‘elucidating and understanding the
basis for various distinctions that we (both ordinary folk and scientists) make
among causal relationships’ (2010, 287).27 For example, we have seen that the
interventionist theory defines a notion of causation that is extremely broad, and is
therefore best thought of as an account of causal relevance. One consequence is that
the set of causal statements that the interventionist theory adjudicates as being true is
much larger than the set of causal sentences that we would be willing to either assert
or assent to. Thus, there must be linguistic distinctions that the interventionist theory
of causation is blind to. We can therefore expect to find distinctions between kinds of
causal locutions—and perhaps even between different concepts of causation—that
can each be thought of as picking out some subset of causal relationships. In this
paper, for example, I have argued there are important distinctions between causal
claims that selectively pick out causes that are deviations from the normal, and those
that don’t.28 If Woodward and I are right, there should be linguistic evidence for
different kinds of causal locutions—that is, this work in the philosophy of causation
can be thought of as providing suggestions for further empirical investigation.
Second, there is a question about how realistically we have to take the causal
models that are associated with the interventionist theory. Consider again the model
representing various causes of flooding (Fig. 14.1) and the associated structural
equations. Interventionism seems to imply (and, in fact, require) that we have
the ability to cognitively construct and manipulate causal structures like this.29
However, it seems safe to assume that we don’t generally represent the detail
of these structures linguistically (or explicitly). For example, in a conversation
about the causes of flooding, it would be very unusual to present a causal model,
or even to convey all the information needed to construct one. Nevertheless, we
are able to make and understand quite sophisticated claims about this kind of
27 In the paper cited above, Woodward considers the interrelated notions of stability, level of
description, and specificity, which are used to distinguish different (kinds of) causal relationships.
28 Others, for example Hall (2004) and Hitchcock (2007b,c), have also argued that there are
important distinctions between different causal concepts.

29 Much recent work in the cognitive psychology of causal inference also assumes that causal
inference requires that we are able to represent a network of directed relations between variables.
For an overview, see Lagnado (2011).
causal system. Thus, interventionism suggests that there is a gap between the
linguistic representation of causal structure and the cognitive representation of
these structures.30 This raises the question, what cognitive apparatus is necessary
to explain our causal judgements, if the interventionist theory is correct? And is
there evidence that we possess this? These, too, are questions that linguists (and
psychologists) are better equipped to answer than philosophers.
References
Alicke, M. D., Rose, D., & Bloom, D. (2011). Causation, norm violation, and culpable control.
Journal of Philosophy, 108, 670–696.
Collins, J., Hall, N., & Paul, L. A. (Eds.) (2004). Causation and counterfactuals. Cambridge: MIT
Press.
versa. In B. Copley & F. Martin (Eds.), Causation in grammatical structures (pp. 11–57).
Dowe, P. (2000). Physical causation. Cambridge: Cambridge University Press.
Eells, E. (1991). Probabilistic causality. Cambridge: Cambridge University Press.
Hall, N. (2004). Two concepts of causation. In J. Collins, N. Hall, & L. A. Paul (Eds.), Causation
and counterfactuals (pp. 225–276). Cambridge: MIT Press.
Hall, N. (2007a). Structural equations and causation. Philosophical Studies, 132, 109–136.
Hall, N. (2007b). Causation. In F. Jackson & M. Smith (Eds.), The Oxford handbook of
contemporary philosophy (pp. 507–533). Oxford: Oxford University Press.
Halpern, J. Y., & Pearl, J. (2005). Causes and explanations: A structural-model approach. Part 1:
Causes. British Journal for the Philosophy of Science, 56: 843–887.
Hart, H. L. A., & Honoré, T. (1959). Causation in the law. Oxford: Clarendon Press.
Hitchcock, C. (2007a). Prevention, preemption, and the principle of sufficient reason. Philosophi-
cal Review, 116, 495–532.
Hitchcock, C. (2007b). Three concepts of causation. Philosophy Compass, 2, 508–516.
Hitchcock, C. (2007c). On the importance of causal taxonomy. In A. Gopnik & L. Schulz (Eds.),
Causal learning: Psychology, philosophy, and computation (pp. 101–114). Oxford: Oxford
University Press.
Knobe, J. (2010). Person as scientist, person as moralist. Behavioral and Brain Sciences, 33, 315–
329.
Lagnado, D. (2011). Causal thinking. In P. M. Illari, F. Russo, & J. Williamson (Eds.), Causality
in the sciences (pp. 129–149). Oxford: Oxford University Press.
Lewis, D. (1986). Causation. In Philosophical papers (Vol. 2, pp. 159–213). Oxford: Oxford
University Press.
McGrath, S. (2005). Causation by omission: A dilemma. Philosophical Studies, 123, 125–148.
Menzies, P. (2004). Difference-making in context. In J. Collins, N. Hall, & L. A. Paul (Eds.),
Causation and counterfactuals (pp. 139–180). Cambridge: MIT Press.
Menzies, P. (2007). Causation in context. In H. Price & R. Corry (Eds.), Causation, physics, and
the constitution of reality (pp. 191–223). Oxford: Clarendon Press.
Menzies, P. (2009). Platitudes and counterexamples. In H. Beebee, C. Hitchcock, & P. Menzies
(Eds.), The Oxford handbook of causation (pp. 341–367). Oxford: Oxford University Press.
30 Copley & Wolff discuss this issue; see (2015).

462 G. Statham
Menzies, P., & Price, H. (1993). Causation as a secondary quality. British Journal for the
Philosophy of Science, 44, 187–203.
Paul, L. A., & Hall, N. (2013). Causation: A user’s guide. Oxford: Oxford University Press.
Pearl, J. (2000). Causality: Models, reasoning, and inference. Cambridge: Cambridge University
Press.
Salmon, W. C. (1993). Probabilistic causality. In E. Sosa & M. Tooley (Eds.), Causation (pp. 137–
Salmon, W. C. (1994). Causation without counterfactuals. Philosophy of Science, 61, 297–312.
Spirtes, P., Glymour, C., & Scheines, R. (2000). Causation, prediction, and search. Cambridge:
MIT Press.
Statham, G. (2017). Contrastive causal claims: A case study. British Journal for the Philosophy of
Science, 68, 663–688.
Systma, J., Livengood, J., & Rose, D. (2012). Two types of typicality: Rethinking the role
of statistical typicality in ordinary causal attributions. Studies in History and Philosophy of
Science Part C, 43, 814–820.
Woodward, J. (2003). Making things happen: A theory of causal explanation. Oxford: Oxford
University Press.
Woodward, J. (2008). Response to Strevens. Philosophy and Phenomenological Research, 77,
193–212.
Woodward, J. (2010). Causation in biology: Stability, specificity, and the choice of levels of
explanation. Biology and Philosophy, 25, 287–318.
Woodward, J. (2011). Psychological studies of causal and counterfactual reasoning. In C. Hoerl,
T. McCormack, & S. R. Beck (Eds.), Understanding counterfactuals, understanding causation
Woodward, J. (2014). A functional account of causation. Philosophy of Science, 81, 691–713.
Woodward, J. (2015). Methodology, ontology, and interventionism. Synthese, 192, 3577–3599.
Woodward, J. (2016). The problem of variable choice. Synthese, 193, 1047–1072.
Chapter 15
Counterfactuals and Causal Reasoning
Boris Kment
Abstract Counterfactual conditionals are used extensively in causal reasoning.

This observation has motivated a philosophical tradition that aims to provide a
counterfactual analysis of causation. However, such analyses have come under
pressure from a proliferation of counterexamples and from evidence that suggests
that the truth-conditions of counterfactuals are themselves causal. I offer an
alternative account of the role of counterfactuals in causal thought that is consistent
with these data: counterfactuals are used in a common method of causal reasoning
related to John Stuart Mill’s method of difference. The method uses background
beliefs about causal relationships, history, and the natural laws to establish a
new causal claim. Counterfactuals serve as a convenient tool for stating certain
intermediate conclusions in this reasoning procedure, and that is part of what makes
counterfactuals useful. This account yields a functional explanation of why our
language contains a construction with the truth-conditions of counterfactuals.
Keywords Counterfactuals · Causation · Causal reasoning · Counterfactual

thinking · Laws of nature · Probability · John Stuart Mill · David Lewis
15.1 Causation and Counterfactuals: The Order

of Explanation
Counterfactual thought is an important element of our cognitive lives. In making

practical decision, we are often led to ask what would happen if we were to carry
out a certain action, and we frequently support causal claims by showing that
the putative effect depends counterfactually on the supposed cause. It, therefore,
B. Kment ()
Department of Philosophy, Princeton University, Princeton, NJ, USA
e-mail: bkment@princeton.edu

https://doi.org/10.1007/978-3-030-34308-8_15
464 B. Kment
does not come as a surprise that scholars in many disciplines—from philosophy to

cognitive and social psychology to computer science to linguistics—have shown a
keen interest in understanding counterfactuals.1
One point of contention is whether causal notions should figure in a semantic
account of counterfactuals. A number of philosophers, motivated by examples like
those described in Sect. 15.2 below, have favored such causal theories of counterfac-
tuals. However, this approach stands opposed to a prominent philosophical tradition,
going back at least to David Hume and most prominently defended by David
Lewis,2 that aims to give a reductive analysis of causation in counterfactual terms.
The two views advocate for opposite directions of analysis and are consequently
mutually exclusive—combining them would lead to circularity.
The goal of this paper is to provide further support for a causal account of
counterfactuals, by showing that it can explain the phenomenon that motivates
counterfactual analyses of causation as convincingly as these analyses themselves
can. The phenomenon under consideration is our pervasive tendency to use coun-
terfactuals to support causal claims. A counterfactual account of causation provides
a straightforward explanation of this datum: causal relationships consist (at least
partly) in certain patterns of counterfactual dependencies, and to ask whether X
is a cause of Y is therefore (at least in part) to ask whether certain counterfactuals
hold. Nevertheless, counterfactual analyses face considerable obstacles, which I will
review in Sect. 15.2. That will provide initial motivation to look for an alternative
explanation of the role of counterfactuals in causal reasoning and will set the stage
for my own account.3 I will offer a somewhat idealized and simplified rational
reconstruction of our use of counterfactuals in ordinary-life causal reasoning,
focusing on deterministic contexts in Sect. 15.3 and on indeterministic ones in Sect.
15.4. On my account, the procedure uses certain background beliefs about causal
relationships, natural laws, and history, to establish a new claim about relationships
of (actual token) causation. Counterfactuals serve as a convenient tool for stating
certain intermediate conclusions in this reasoning procedure, and that is one of
the reasons why we have a counterfactual construction. This account yields a
good functional explanation of why our language contains a construction with the
truth-conditions of counterfactual conditionals: they are exactly the truth-conditions
that a construction needs to have to adequately serve the purpose of stating the
1 See Roese & Olson (2014) and Hoerl et al. (2011) for contributions by psychologists, Pearl (2009:
ch. 7) for discussion by a computer scientist and philosopher, and the references in the rest of this
paper for further literature on counterfactuals.
2 Hume (1995: 87), Lewis (1986a, 2004a).
3 For an interesting alternative explanation of the connection between causation and counterfactu-
als, see Maudlin (2004).

15 Counterfactuals and Causal Reasoning 465
intermediate conclusion in the reasoning practice I have described. As we will see,

these truth-conditions involve the concept of causation.4,5
15.2 The Counterfactual Analysis of Causation

and the Causal Account of Counterfactuals
Y counterfactually depends on X just in case Y would not have existed (obtained,

occurred) if X had not existed (obtained, occurred). Counterfactual accounts of
causation aim to analyze causation in terms of counterfactual dependence. I
will discuss three of the main challenges confronting this approach. Firstly, for
the counterfactual account to be tenable, there must be necessary and sufficient
condition for causation that can be stated in counterfactual terms. However, it is very
hard to find such conditions, even for causation under determinism. Secondly, even if
this problem could be solved for deterministic causation, it would be hard to extend
the account to probabilistic causation (which presents its own set of challenges).
Thirdly, one may not find the main motivation for pursuing a counterfactual analysis
of causation particularly compelling.
Start with the first point: the difficulty of formulating necessary and sufficient
conditions for causation. The main problem is that simple counterfactual depen-
dence between distinct matters of particular fact is not a necessary condition for
causation.6 There are two types of cases that are commonly used to show this.
Over-Determination. Fred’s rock and Susie’s rock simultaneously hit the window,
each causing sufficient damage to shatter it. Fred’s throw and Susie’s throw are
both causes of the shattering (the shattering is causally overdetermined), but the
shattering does not counterfactually depend on either throw. If one of the throws
had not occurred, the other would still have broken the window.
Preemption. Susie and Fred are both getting ready to throw a rock at the window.
Susie throws hers first and shatters the window, thereby preempting Fred’s plan.
Susie’s throw is a cause of the breaking of the window, but the breaking does not
counterfactually depend on her throw. If she had not thrown her rock, Fred would
have thrown his, which would have shattered the window.
4I do not claim that that is the only function of counterfactual conditionals. They clearly also serve
other purposes, e.g. in making practical decisions. See, e.g., Stalnaker (1981), Gibbard & Harper
(1981), Lewis (1981), and Joyce (1999). For arguments against counterfactual decision theory,
see Ahmed (2014). See Edgington (2003) and Bennett (2003) for discussions of further uses of
counterfactuals.
5 For a much fuller development of the view proposed in this paper, see Kment (2010), and in
particular Kment (2014: Chs. 10–12). Also see Kment (2015: Sect. 5).
6 By “matters of particular fact”, I mean, roughly speaking, facts about the goings-on in specific
space-time locations.
466 B. Kment
As these examples show, effects need not counterfactually depend on their

causes. Consequently, if causation is to be analyzed in counterfactual terms at all,
it cannot be analyzed as simple counterfactual dependence, but at best as some
complex pattern of counterfactual dependencies (possibly combined with other
conditions). There have been numerous attempts to provide such an analysis in a
way that gets the extension of causation right, but in my opinion they met with only
limited success.7,8
We face additional obstacles when trying to give a counterfactual analysis of
probabilistic causation. If indeterminism is pervasive, so that it is almost always a
matter of chance what happens, then it is almost never true that the effect would
not have happened if the cause had not happened. The effect still might have
happened (though it might or would have been less likely). For that reason alone, it
seems hopeless to try to define probabilistic causation in terms of the counterfactual
dependence of the effect on the cause. Instead, philosophers have tried to define
probabilistic causation in terms of the counterfactual dependence of the chance of
the effect on the cause. The most popular version of this account starts from the idea
that causes are probability raisers, a thought that can be refined in various ways. For
a classic statement of this view, see Lewis (1986a: postscript B), also see Menzies
(1989).
Unfortunately for this approach, whether C is a cause of E cannot be read off
the way in which E’s chance counterfactually depends on C. Consider an example
due to Jonathan Schaffer (Schaffer 2000). Merlin casts a spell to turn the prince
and the king into frogs at midnight, and Morgana casts a spell to turn the prince
and the queen into frogs at midnight. Once one of these spells has been cast, its
chance of success remains constant at 50% until midnight. Since the results of
the two spells are stochastically independent, the two spells result in a chance of
75% that the prince will become a frog at midnight. At midnight the prince is
transformed along with king, while the queen is not. The result proves that Merlin’s
spell worked, while Morgana’s was ineffective. So, Merlin’s spell is a cause of the
prince’s transmutation while Morgana’s is not. Note, however, that each spell raised
the probability of the prince’s transmutation by the same amount, from 50% to 75%.
The example shows that the way in which a factor influences the chance of X need
not determine whether it makes a causal contribution to the occurrence of X.
7 For some of the strategies for dealing with overdetermination and preemption problems within the
framework of the counterfactual account, see Lewis (1986a, 2004a), Menzies (1989), McDermott
(1995), Ramachandran (1997), Yablo (2004). For an overview, see Paul & Hall (2013). More
recently, philosophers using the framework of causal models have proposed a number of other
treatments of overdetermination and preemption problems. See, e.g., Hitchcock (2001), Woodward
(2003), Halpern & Pearl (2005), and Hall (2007). Other philosophers have tried to address the
problems for counterfactual accounts in part by arguing that there are several concepts of causation,
and that the problems arise from choosing the wrong notion as the target for a counterfactual
analysis (Hall 2004; for a reply, see Kment 2014: Sect. 9.1.2, in particular p. 225 n.1, Sect. 10.4.1).
8 There are also cases that seem to show that counterfactual dependence between distinct matters
of particular fact is not sufficient for causation (see Bennett 1984; also Kment 2010: 84–5, 2014:
248–9). These examples have been discussed less extensively.
It seems plausible enough that we can use counterfactuals about chances to

support claims about the causes of these chances. If we know that E would not have
had chance p at time t if A had not obtained, then we have reasons for concluding
that A is a cause of the fact that E had chance p at t. But, given that indeterministic
causation of other effects (i.e., effects that are not facts about chances) is not merely
a matter of influencing chances, it is not obvious how to extend the account to such
instances of causation. However, such an extension would be needed to obtain a
unified counterfactual account that covers all cases of causation.
Not only do counterfactual accounts of causation face these obstacles, but the
considerations motivating the analysis seem very resistible. The analysis receives
its greatest support from its ability to explain our tendency to infer causal claims
from counterfactual connections. However, we have an equally strong tendency
to draw the contrapositive inference from the absence of a causal connection to
counterfactual independence. Moreover, just as beliefs about causal relationships
are often guided by counterfactual judgments, so counterfactual judgments are fre-
quently based on prior causal beliefs. We could appeal to the former phenomenon to
support an analysis of causation in counterfactual terms, but we could equally well
appeal to the latter phenomenon to motivate a causal account of counterfactuals.
Before considering a case in which our counterfactual judgments are guided
by causal beliefs, we need a simplified working account of the truth-conditions
of counterfactual conditionals. I will endorse the standard account of their truth-
conditions in terms of comparative closeness or overall similarity between possible
worlds. To simplify somewhat, If it had been that P, then it would have been that Q
is true just in case Q is true at the P-worlds closest to the actual world.9 It is well-
known that the operative standards of overall similarity between worlds must differ
from those underlying our offhand similarity judgments. It sounds true to say “If
Nixon had pressed the button, then there would have been a devastating nuclear
war.” That means that the closest worlds where Nixon presses the button are those
where the earth is devastated, not those where the signal dies in the wire on its way
to the launch pad, even though offhand we would judge the latter worlds to be more
similar to actuality than the former.10
What are the standards of inter-world similarity that matter to the truth-conditions
of counterfactuals? Suppose that Susie throws a rock at the window and the window
shatters, and ask yourself whether the window would still have shattered if Susie
had not thrown the rock. Roughly speaking, the closest possible worlds where Susie
doesn’t throw the rock are those that are just like the actual world at the time of the
9 Throughout this paper, I will make the simplifying assumption that the “limit assumption” is true,
i.e. that for any antecedent, there is a set of antecedent-worlds that are equally close to actuality and
closer than any antecedent-worlds not in the set. Although this assumption is likely to be false, the
simplification is harmless. For, there are well-known ways of doing without the limit assumption
(Lewis 1973) and they could easily be applied (with some loss of simplicity) to the discussion in
this paper.
10 See Bennett (1974), Fine (1975), Lewis (1973: 76, 1986b). The Nixon case is a variant of Fine’s
example.
468 B. Kment
rock throw, except that Susie does not throw her rock. After that, the world evolves
in accordance with the natural laws of the actual world. If the window breaks in that
world, then we can express this by saying that the window would still have shattered
if Susie had not thrown her rock. Otherwise, it is true to say that the window would
not have shattered if Susie had not thrown her rock, i.e. that the shattering depends
counterfactually on Susie’s throw. We can generalize from this example. Let A and
E be matters of particular fact that actually obtain, with A obtaining at time tA and
E obtaining at some later time:
(1) Under determinism, E counterfactually depends on A just in case, at every pos-
sible world that is like actuality at tA except that A does not obtain and that
conforms to the actual laws of nature after tA , E fails to obtain.11,12
Under determinism, the state of an antecedent-world at tA and the actual laws
of nature together determine the entire rest of history. The same is not generally
true under indeterminism. Two antecedent-worlds might both be like actuality at
tA except for the fact that A does not obtain, and they might both conform to the
actual laws thereafter, and yet they may differ in the outcomes of some post-tA
random processes. That raises the question whether similarity to the actual world in
the outcomes of post-tA chance processes is an additional criterion of closeness to
actuality.
The answer is a qualified ‘yes.’ Some post-antecedent similarities matter to the
closeness ordering, others do not. Consider a variant of an example due to Dorothy
Edgington (2003, 2011). You are about to watch an indeterministic lottery draw on
television when someone offers to sell you ticket number 17. You decline. As luck
would have it, ticket number 17 wins. It seems true to say, “If you had bought the
ticket, you would have won.” But that presupposes that the following is true:
11 This is simplified in a number of ways. For example, there is, strictly speaking, no possible world
where A fails to obtain but which matches actuality in all facts about tA other than A. (For some
of the facts about tA other than A necessitate A, e.g. the fact that A and B both obtain, where B is
some other fact about tA ). A more precise description of the closest worlds where A fails to obtain
would say that these worlds maximize match in facts about tA other than A. That is to say, of all
the worlds where A fails to obtain, the closest ones (other things being equal) are those that come
closest to matching actuality in all the facts about tA other than A. Of course, this account is still
simplified. For a fuller and more precise account, see Kment (2006, 2014: Chs. 8–9).
12 Counterfactuals are notoriously context-dependent (Quine (1950), Lewis (1973, 1986b); differ-
ent standards of similarity are relevant to their truth-conditions in different contexts. However,
like David Lewis (1986b), I believe (Kment 2006: 262–3, 2014: 44–46) that there is a specific
standard of closeness that serves as our default—we use this standard in interpreting and evaluating
counterfactuals unless our presumption in its favor is canceled by distinctive features of the context.
Lewis’s account of causation analyzes causation in terms of this default standard of closeness.
(1) and (2) describe the conditions for counterfactual dependence under the default standard.
Moreover, the method of evaluating causal claims in the light of counterfactual dependencies that
I will discuss in this paper employs the default standard as well.
If you had bought ticket number 17, that ticket would still have won.
Now suppose that the lottery company has two qualitatively indistinguishable lottery
machines that give the same chance to every possible outcome. They used machine
A in the draw but could have used machine B instead. Consider:
If a different machine had been used, 17 would still have won.
That seems false. If a different machine had been used, then 17 might still have
won, or some other number might have won. It is not true that 17 would still have
won. In the first case, we hold the outcome of the lottery draw fixed, in the second
we do not. It seems very plausible that this difference is due to underlying causal
judgments. Your decision about whether to buy the ticket is not causally connected
to the outcome of the draw (or so we believe). That is why the outcome can be
held fixed when we are thinking about what would have happened if you had made
a different decision. By contrast, the use of a particular lottery machine is part of
the causal history of the outcome. That is why the outcome of the draw cannot be
held fixed in the second case. In these examples, we are drawing on prior causal
judgments to decide whether certain facts can be held fixed—i.e., whether they
would still have obtained if the antecedent had been true, or in other words, whether
they are counterfactually independent of the antecedent.13
The upshot is that, when we think about what the world would be like if A had
not obtained, we are holding fixed just those post-antecedent matters that are not
causally connected to A in the actual world. For the indeterministic case, therefore,
we can give the following, somewhat simplified account of the conditions for
counterfactual dependence. Suppose that A and E are actual matters of particular
fact, with A actually obtaining at tA and E actually obtaining at some later time.
(2) Under indeterminism, E counterfactually depends on A just in case E does not

obtain at any world w that meets the following conditions:
(a) A fails to obtain at w,
(b) w is otherwise like actuality at tA ,
(c) w matches actuality after tA in all matters of particular fact that are not
actually caused by A, and
(d) w conforms to the actual laws after tA .
13 Examples like this are sometimes called “Morgenbesser cases,” in honor of Sydney Mor-
genbesser, who was among the philosophers who discovered them (although Morgenbesser did
not publish the result). Examples similar to the one described are discussed in Adams (1975: ch.
IV, Sect. 8, in particular pp. 132–3.), Tichý (1976), Slote (1978), Bennett (2003), Edgington (2003,
2011), Schaffer (2004), and Kment 2006: Sects. 3–4, 2014: Chs. 8–9, in particular Sects. 8.3–8.4.
470 B. Kment
It seems that the conditions for counterfactual dependence are themselves causal.14
If that is true, then causation cannot be analyzed in terms of counterfactual
dependence without circularity.
15.3 Counterfactual Dependence and Deterministic

Causation
15.3.1 The Determination Idea
I will argue that our practice of using counterfactuals to evaluate causal claims rests
on an assumption I will call the “determination idea.” Separate versions of this idea
apply to deterministic and to indeterministic contexts.15 The deterministic version
will be considered in this section and the indeterministic version in Sect. 15.4.
The deterministic version of the determination idea (“D/d,” for short) runs as
follows:
(D/d) Under determinism, the causes of E together nomically determine E.

That is to say: E obtains at every possible world where all the actual causes of E
obtain and which conform to the actual laws of nature.
Some clarification is in order concerning the notion of cause used in (D/d). We
can distinguish two ways in which x can cause y.16 On the one hand, x might be
part of what produced y (the stone throw produced the breaking of the window,
the poisoning caused the patient’s death, etc.). On the other hand, x might cause
y without being among the producers of y. These non-producing factors include
omissions, such as the absence of various kinds of possible interference with the
14 See Mårtensson (1999), Edgington (2003, 2011), Bennett (2003: ch. 15), Hiddleston (2005), and
Wasserman (2006) for causal analyses of counterfactuals motivated by examples like the above
lottery case, and see Kment 2006, 2014: Chs. 8–9 for an analysis in terms of (causal and non-
causal) explanation. Veltman (2005) and Schulz (2011) take a similar line. Also see Pearl (2009,
ch. 7), who uses the framework of causal models to give a causal account of counterfactuals and
of what is held fixed in counterfactual reasoning. For an early causal theory of counterfactuals, see
Jackson (1977).
15 By “determinism” I mean the thesis that the state of the universe at any given moment and the
laws of nature together determine all of history: any possible world that matches actuality at one
time and that conforms to the actual laws of nature matches actuality at all times.
16 For more on this distinction, see Ned Hall’s discussion in his (2004) and David Lewis’s in his
(2004b). I don’t agree with their thesis that we need a counterfactual account of causation (or
of one notion of causation) to accommodate the thought that omissions are causes. I think that
our belief in omissions as causes is closely connected to the idea of causes as nomic determiners
of their effects (Kment 2014: Sect. 10.4.1, and in particular Sect. 10.4.2), and that this idea can
also explain the close connection in ordinary causal thinking between causation by omissions and
counterfactual dependence (Kment 2014: Ch. 10).
causal processes that produced y.17 They also include factors (so-called “double
preventers”) that prevent such interferences and thereby cause their absence, as well
as the producing and non-producing causes of such double preventers. For example,
the fact that the would-be assassin failed to kill the president on the eve of her speech
forms part of the causal history of the speech, as does the action of the police agent
who arrested the assassin before he could strike. (It is partly because of the action
of the police agent and the absence of assassins that the president holds the speech
the next day.) But neither the police agent’s action nor the absence of assassins is
among the factors that produced the speech. The notion of cause used in (D/d) is to
be understood in a broad sense, as including not only the producers of E, but also
E’s various non-productive causes.18
For the purpose of illustration, assume that determinism is true and suppose that
Susie throws a rock at a window and breaks it. Consider all the causes of the window
shattering, including omissions. These causes include Susie’s throw, the position and
molecular structure of the window, etc. They also include the absence of any factors
that could interfere with the shattering, such as obstacles in the path of the flying
rock, strong winds that could blow the rock off its path, bystanders trying to catch
the rock, and so forth. (D/d) tells us that, if you complete this list of causes in the
right way, then you get a set of factors that nomically determines the breaking of the
window.
Note that the determination idea merely states a necessary condition for causa-
tion; it does not state a sufficient condition. That is to say, it is true of a set of factors
that it contains all and only the causes of E only if the set nomically determines E.
But clearly, it is not true of every set of factors that nomically determines E that it
contains all and only the causes of E. (Moreover, there is no reason for thinking that
it is possible to formulate non-trivial necessary and sufficient condition for causation
in terms of nomic determination. Philosophers who have tried to do so, typically
with reductionist ulterior motives, have been in for a disappointment.)
An assumption that is slightly stronger than (D/d) seems plausible as well:
(D/d*) Under determinism, the causes of E that obtain at t nomically determine E
(where t is earlier than the time at which E obtains).
The causes of E that obtain at t—I will call them the ‘t-causes’ of E—make up a
complete temporal cross-section of E’s causal history. They nomically determine all
17 Admittedly, not all philosophers are happy with the idea that omissions can be causes. For
example, Beebee (2004) denies that any omissions are causes, while others hold that they are
causes only in a secondary sense, or that they are not causes but stand in some other, closely related
relation to effects (Dowe 2000, 2001; Armstrong 2001). Others think that they can be causes in one
sense but not in another (Hall 2004). I cannot jump into the fray on this occasion, but see Kment
2014: Sects. 10.4.1–10.4.2, and also Sects. 9.1.2–9.1.3).
18 Philosophers sometimes distinguish between causes and causally relevant background condi-
tions, or between causes and enablers. However, the term “cause” as used in (D/d) is to be
understood in a broader way, as covering all factors that are causally relevant to E, including
background conditions or enablers. The same is true for the principles (D/d*), (D/i), and (D/i*)
below.
472 B. Kment
later causes of E and they screen off any previous causes. (Earlier causes do not act
at a temporal distance. They contribute causally to E only by causing t-causes of E.)
Hence, if all the causes of E together nomically determine E, then so do the t-causes
of E. Here is another way of looking at it. Under determinism, the state of the world
at t contains a set of factors that nomically determines E. (D/d*) tells us that, if you
remove from the state of the universe at t all the factors that are not causally relevant
to E, then the remaining factors still nomically determine E.
It is not of critical importance for my purposes whether the determination idea
should be regarded as true in light of our best philosophical and scientific theories.19
My reconstruction of everyday causal and counterfactual reasoning requires only
the premise that the determination idea is commonly used in ordinary explanatory
thinking, at least as a working assumption. And that much seems very plausible.
Suppose that you made a certain type of cake on two different occasions. The first
time it was delicious, the second time it was chalky and unappealing. Then it seems
very tempting to say: you must have done something the second time that you
did not do the first time and which made the second cake taste chalky. In other
words, we can conclude from the fact that the two cakes taste different that the
factors that are causally responsible for the taste of the first cake are somewhat
different from those responsible for the taste of the second cake. Different effects,
therefore different causes. That is the contrapositive of: same causes, same effect.
And the latter principle, in turn, is most likely motivated by an application of the
determination idea.
15.3.2 The Method of Difference and the Counterfactual

Method Under Determinism
I think that the use of counterfactuals to evaluate causal claims is an extension of

another method of causal reasoning, which I will consider first: John Stuart Mill’s
method of difference (Mill 1956, bk. III, ch. VIII, sect. 2). Let me start with an
admittedly highly simplified and idealized description of this procedure.
Scenario 1
tA ABCD
tA +1 E
Scenario 2
t Ā B C D
t+1 Ē
19 However, in Kment 2014: Sect. 10.4, in particular Sect. 10.4.2, I argue that some of the criticisms
that have been leveled at the determination idea are misguided, and that some of them rely on
controversial views (that I reject) about what the relata of causation are for example on the view
that they are events (or entities similar to events) rather than facts.
You observe a scenario (Scenario 1) in which the causal factors A, B, C, and D are
present at time tA . A little later, E obtains. You want to know what caused E. Now
suppose that you also observe Scenario 2. In Scenario 2, B, C, and D obtain but A
does not, and E does not obtain a moment later. You infer from these observations
that A is a cause of E in Scenario 1.
In order for this line of reasoning to be justified, you need to assume that the
initial states of the two scenarios match each other with respect to all the factors
that are causally relevant to whether E obtains at the later time, with the possible
exception of A. There must not be any other causally relevant differences between
the initial states of the two scenarios. (If there is another such difference, then that
difference might be what is responsible for the fact that E occurs at the later time in
Scenario 1 but not in Scenario 2. Then you cannot blame A for the E’s occurrence in
Scenario 1.) In other words, the causal factors with respect to which the initial states
of the two scenarios match each other—B, C, and D—include all factors obtaining
at tA in Scenario 1 that are causally relevant to E, with the possible exception of A.
Equivalently:
(3) A, B, C, and D include all the factors that are tA -causes of E in Scenario 1.
I propose that we reconstruct the method of difference as follows. By assumption
(3), the set of E’s tA -causes must be a subset of {A, B, C, D}. However, you are not
sure whether all members of this set are causes of E or only some of them. In
particular, you do not know whether A is a cause of E. Now you observe Scenario 2.
In this scenario, B, C and D occur but E does not occur a moment later. That shows
that
B, C and D do not nomically determine E.
However, according to (D/d*), the facts that are tA -causes of E in Scenario 1 taken
together must nomically determine E. Hence:
(4) B, C and D do not include all the tA -causes of E in Scenario 1.
From (3) to (4) you can infer that A is a cause of E in Scenario 1.
For an illustration of this form of reasoning, consider once more the example of
the previous section in which you tried to bake the same cake on two occasions. The
first time it tasted good but the second time it did not. Given the difference in taste,
you conclude that you must have prepared the two cakes in somewhat different
ways. That is an application of the determination idea: you infer a difference in
causes from a difference in effects. You look more closely and discover that you
used somewhat different ingredients on the two occasions. The first time you used A,
B and C, while the second time you used only B and C. That was the only difference
between the two cases. You conclude that your use of ingredient A on the first
occasion must have been a cause of the pleasant taste. More sophisticated versions
of this procedure are applied in scientific experiments. (In these cases, Scenario 1 is
the “experimental condition,” Scenario 2 is the “control condition,” and B, C, and
D are the background factors that the experimenters are controlling for.) However,
my discussion will focus on everyday uses of the method.
474 B. Kment
The method of difference has two characteristic limitations. Firstly, it requires

you to either find or create a control scenario that matches Scenario 1 in all those
tA -factors that are causally relevant to E in Scenario 1. However, you might not be
lucky enough to find such a scenario and it may be beyond your powers to create
one. Secondly, you may not even know what scenario to look for or be able to tell
whether you have found what you need. For, you may not know very much about
which factors at tA were causally relevant to E in Scenario 1 and therefore may not
know in what respects the control scenario needs to match Scenario 1.
If my reconstruction of the method of difference is on the right track, however,
then it is easy to solve these two problems. Start with the first problem. Suppose that
you can narrow down the range of factors that might be tA -causes of E in Scenario
1 to a fairly small set X. However, you are unable to find or create an actualized
control scenario that matches Scenario 1 with respect to all the factors in X–{A}. On
my account, this is no serious obstacle. For, I claim that you need a control scenario
only in order to show that X–{A} does not nomically determine E. An actualized
control scenario is not needed to show this. You can instead consider a possible
scenario where the factors in X–{A} obtain and which conforms to the actual laws. If
you can show that E does not obtain in that possible scenario, then you can conclude
that X–{A} does not nomically determine E. The rest of the argument proceeds in
the way described before. This shift from looking for an actualized control scenario
to merely looking for a possible one allows us to solve the second problem as well.
Suppose you know very little about what caused E in Scenario 1, and are therefore
unable to narrow down the range of factors that might (for all you know) have been
tA -causes of E to a small set. Then you may have little hope of finding an actualized
control scenario that matches Scenario 1 in all tA -causes of E other than A. However,
this problem disappears once we recognize that a merely possible scenario can serve
as control scenario. We can simply use as our control scenario a possible world that
matches actuality at tA in all factors other than A and that conforms to the actual
laws after tA . Suppose that we can show that E fails to obtain at worlds like that, i.e.
that the following is true:
(5) E fails to obtain at those possible worlds where (i) A fails to obtain, (ii) the
state of the universe at tA is otherwise just like in actuality, and (iii) events
conform to the actual laws of nature after tA .
We can infer from (5) that the factors other than A that actually obtain at tA do not
nomically determine E. Given (D/d*), it follows that these factors do not include all
the actual tA -causes of E and that A must therefore be one of E’s causes.20
20 Again, this is a little simplified. Let S be the set of facts about tA other than A. As mentioned
in fn. 8, the closest worlds where A fails to obtain do not match actuality with respect to all fact
in S. Other things being equal, they match actuality as closely in S-facts as is compatible with A’s
failure to obtain, but there might be a small range of S-facts that fail to obtain at these worlds.
Consequently, the inference from the premise that
E fails to obtain at the closest worlds where A fails to obtain
to the conclusion that
(5) is what we express by saying that E would not have obtained if A had
not obtained. Counterfactuals allow us to state this intermediate conclusion of
the reasoning process concisely, and that is likely one of the purposes for which
counterfactuals exist. I will call the method of causal reasoning described in this
section the “counterfactual method” of supporting causal claims.
15.3.3 Comparison with John Mackie’s Account
The account sketched in Sect. 15.3.2 is similar in some respects to John Mackie’s
view. Let me briefly compare the two proposals.
In The Cement of the Universe, his seminal study of causation, Mackie (1974:
chs. 1–3) aims to answer two questions: “What do causal claims mean?”, and “What
is causation ‘as it exists in the objects’?” (Mackie 1974: 60). On his account of
the meaning of causal claims, the content of “A is a cause of E” includes certain
counterfactual conditionals, such as the claim that E would not have occurred if A
had not occurred. Among the “grounds” (ibid.) of these counterfactuals is a certain
fact about the actual world, namely the fact that A is a member of a set of actual
conditions that are minimally sufficient for E (Mackie 1974: ch. 3).21 Mackie holds
that this fact is part of what constitutes causation as it exists in the objects. He goes
on to discuss how the method of difference can be used to show that A is part of a
minimal sufficient condition for E (Mackie 1965, 1974: ch. 3). We need to start from
some assumptions about Scenarios 1 and 2, including the premise that E has a cause
(and that there is therefore a minimal sufficient condition for E) in Scenario 1, and
that the two scenarios are alike in all relevant factors except A. Given E’s absence
in Scenario 2, there can be no sufficient conditions for E in Scenario 2. It follows
that in Scenario 1, any sufficient condition for E includes A, and that A is therefore
part of a minimal sufficient condition for E. That in turn supports the counterfactual
component of what is asserted by the claim that A is a cause of E in scenarios similar
to Scenario 1. In this way, the observation of Scenarios 1 and 2 can provide support
for this causal claim.
the facts in S do not nomically determine E (and therefore do not include all the tA -causes of E)
is defeasible. The premise might be true and the conclusion false if S includes facts that nomically
determine E but some of these facts fail to obtain at the closest worlds where A fails to obtain.
A fuller version of my account therefore predicts that the inference from E’s counterfactual
dependence on A to the claim that A is a cause of E is defeasible, or in other worlds, that
counterfactual dependence is not quite a sufficient condition for causation. I think that this
prediction is borne out (see footnote 7). A fully developed version of the view propounded in
this paper can explain why the inference from counterfactual dependence to causation fails in just
those cases where it does (see Kment 2014: Sect. 12.1).
21 I am simplifying by ignoring the fact that Mackie is relativizing such causal claims to a “causal
field,” which is essentially a set of background factors.

476 B. Kment
Mackie’s account of how the method of difference works rests on an elaborate

theory that aims to provide necessary and sufficient conditions for causation. By
contrast, my own explanation of the method merely assumes that there is a certain
necessary condition for causation: certain factors are the t-causes of E only if they
nomically determine E. Moreover—and this is crucial for the topic of this paper—
Mackie’s account of the connection between counterfactuals and causal claims is
completely different from mine. He does not use his account of the method of
difference to explain the connection between counterfactuals and causation. Instead,
he thinks that the connection simply consists in the fact that certain counterfactuals
are part of the content of a causal claim. In my view, by contrast, counterfactuals are
not part of what is said by a causal claim. They merely express the intermediate
conclusions of a common way of supporting causal claims that is justified in
essentially the same way as the method of difference.
15.3.4 Limitations of the Counterfactual Method
We saw in Sect. 15.2 that there are well-known cases (such as those of over-
determination and preemption) in which effects do not counterfactually depend
on their causes. As mentioned above, that creates a challenge for any attempt to
formulate necessary and sufficient conditions for causation in counterfactual terms
and therefore for the counterfactual analysis of causation. But it presents no serious
difficulty for the view outlined in this paper. What it shows is simply that the
counterfactual method (or at least the version of it discussed in this paper22 ) is more
useful for supporting causal claims than for refuting them. If we can show that E
counterfactually depends on A, then that supports the claim that A is a cause of E.
But if E is counterfactually independent of A, then that does not provide similarly
strong evidence for the claim that A is not a cause of E, since the case at hand may
involve over-determination or preemption.
That is just what we would expect on my account. In fact, as some authors
have noted (Mackie 1965, in particular sect. 5; Strevens 2007), the datum can be
explained by the earlier observation that the determination idea states merely a
necessary but not a sufficient condition for a set to contain all the tA -causes of
22 Ihave only described the simplest way of using counterfactuals to evaluate causal claims. More
sophisticated methods may proceed by determining not only whether E counterfactually depends
on A, but also whether A and E are linked by certain more complex patterns of counterfactual
dependencies. (See Pearl 2009, in particular chs. 7–8, Woodward 2003, and the papers cited in
footnote 7 as propounding sophisticated forms of the counterfactual analysis.) While it is open
to doubt whether any complex pattern of counterfactual dependencies is necessary and sufficient
for causation, some such patterns might come much closer to being necessary and sufficient than
simple counterfactual dependence. The fact that the relevant patterns fail to hold between E and A
might then lend (strong but defeasible) support to the claim that A is not a cause of E, even if it
does not entail the latter claim.
E. If a set does not nomically determine E, then it does not contain all of E’s tA -
causes. But if the set does nomically determine E, nothing interesting follows. In
particular, it does not follow that the set contains all tA -causes of E. Apply this to
the counterfactual method. Let S be the set of all factors that obtain at tA other than
A. If E counterfactually depends on A, then S does not nomically determine E. Given
the determination idea, it follows that S does not contain all the tA -causes of E, so
that A must be a cause of E. But if E is counterfactually independent of A, then the
most we can conclude is that S does nomically determine E. However, that does not
entail that S contains all the tA -causes of E or that A is not a cause of E.
We would expect, therefore, that there is causation without counterfactual
dependence whenever there are factors at tA that don’t include all of E’s tA -causes
but that nevertheless nomically determine E. That is the case in over-determination
and preemption scenarios. Consider first an over-determination case. Fred’s rock
and Susie’s rock simultaneous hit the window, each causing sufficient damage to
break it. Consider the set of all matters of particular fact that obtain at the time
t of Susie’s throw, except for her throw itself. This set does not contain all t-
causes of the window shattering, since it does not contain Susie’s throw. But the set
nomically determines the window shattering. For, it contains Fred’s throw, as well
as background facts that nomically determine that his rock will hit the window with
sufficient force to break it. Since all these factors obtain at the closest worlds where
Susie does not throw her rock, the window breaks at these worlds. The shattering
does not counterfactually depend on Susie’s throw.
Similarly in cases of preemption. Suppose that Susie throws her rock first and
shatters the window. Fred, who intended to break the window, sees that the job has
already been done and walks away. Consider the set of all matters of particular fact
obtaining at the time t of Susie’s throw, except for her throw itself. This set does not
contain all the causes of the window’s shattering, since it does not contain Susie’s
throw. But it does nomically determine the window’s shattering. For the set contains
Fred’s intention to shatter the window, as well as background facts that nomically
guarantee that nothing will prevent him from carrying out his intention except for
something else’s shattering the window first. Given that these factors obtain at the
closest worlds where Susie does not throw her rock, the window breaks at these
worlds. The window shattering does not counterfactually depend on Susie’s throw.
15.4 The Counterfactual Method Under Indeterminism
Under pervasive indeterminism, we can almost never establish that A is a cause of

E by showing that E counterfactually depends on A, since effects rarely or never
counterfactually depend on their causes. Perhaps we can show that E’s chance
counterfactually depends on A, but the inference from this observation to the claim
that A is a cause of E is problematic, as is shown by Schaffer’s example of Sect.
15.2. (The chance of the prince’s turning into a frog depends counterfactually on
Morgana’s spell, despite the fact that her spell is not a cause.) However, from the
478 B. Kment
fact that E’s chance counterfactually depends on A, we can infer that A is a cause of
the fact that E had a certain chance. That is to say, if E had chance p at t (“cht (E) =
p,” for short) and we can show that
E would not have had chance p at t if A had not obtained,
then we can conclude that
A is a cause of the fact that cht (E) = p.23
This version of the counterfactual method, just like the deterministic variant,
rests on a certain version of the determination idea. This version is restricted to the
causes of one special kind of fact, namely facts about chances. I will call it (D/i),
for “determination idea/indeterministic version.” It says that
(D/i) The causes of the fact that cht (E) = p jointly nomically determine that
cht (E) = p.
This principle seems very plausible. If E has a certain chance at t, then there must
be some matters of particular fact that causally determine that E has that chance at
t. A slightly stronger version of this principle seems plausible as well:
(D/i*) Those causes of the fact that cht (E) = p that obtain after t* nomically
determine that cht (E) = p (for any time t* before t).
The causes of the fact that cht (E) = p that obtain after t* screen off earlier causes.
(Earlier causes of the fact that cht (E) = p do not act at a temporal distance. They
influence E’s chance at t only by way of influencing what happens between t* and
t.) It follows that, if (D/i) is true, then (D/i*) is true as well.
Now suppose that we can show the following:
(6) cht (E) = p at every possible world w that meets the following conditions:
(a) A fails to obtain at w,
(b) w is otherwise like actuality at tA ,
(c) w matches actuality after tA in all matters of particular fact that are not
actually caused by A, and
(d) w conforms to the actual laws after tA .
23 This indeterministic version of the counterfactual method of evaluating causal claims is subject
to the same limitations as the deterministic version: in cases of over-determination and preemption,
the fact that cht (E) = p may fail to depend counterfactually on A despite the fact that A is a cause
of the fact that cht (E) = p. This limitation can be explained in the way discussed in the previous
section.
From (6) we can infer the following:
(7) Those post-tA factors that are not caused by A do not nomically determine that
cht (E) = p.
(D/i*) entails this:
(8) The causes of the fact that cht (E) = p that obtain after tA nomically determine
that cht (E) = p.
From (7) to (8) we can infer the following:
(9) The causes of the fact that cht (E) = p include some factors that were caused
by A.
Finally, given the assumption that causation is transitive (if A is a cause of B and B
is a cause of C, then A is a cause of C),24 (9) entails the conclusion:
A is a cause of the fact that cht (E) = p.
Again, (6) is what we express by saying that E would not have had chance p at t if
A had not obtained. Counterfactuals mediate the inference to the causal conclusion
and that is one of the ways in which counterfactuals are of use to us.
15.5 Conclusion
Counterfactuals play a central role in causal reasoning. Counterfactual analyses of

causation give one explanation of this fact, my account presents another. Let me
conclude by briefly comparing the two approaches in light of the results of the earlier
sections.
24 It
is somewhat controversial whether causation is transitive. For discussion of this question, see
McDermott (1995), Paul (2004), Hitchcock (2001), Hall (2004, 2007), Lewis (2004a), Paul & Hall
(2013: ch. 5), and (Kment 2014: Sect. 12.4). If you believe that causation fails to be transitive,
you can easily adjust the account I gave of counterfactuals and the counterfactual method to this
background belief of yours. You just need to replace all talk about causation with talk about the
ancestral relation of causation. For illustration, consider how the counterfactual method under
indeterminism would need to be revised. (The deterministic version of the method could be revised
in an analogous way.) To begin with, (2)(c) and (6)(c) need to be replaced with the claim that w
matches actuality after tA in all matters of particular fact to which A does not actually stand in the
ancestral relation of causation. The revised counterfactual method is a procedure for showing that
(i) A stands in the ancestral relation of causation to the fact that cht (E) = p.
The method starts by showing that the reformulated version of (6) is true. From that result, one
can infer that those post-tA factors to which A does not stand in the ancestral relation of causation
do not nomically determine that cht (E) = p. Given (D/i*), it follows that the post-tA causes of E
include some factors to which A stands in the ancestral relation of causation. That in turn entails
(i). Counterfactuals can be used to express the reformulated version of (6) in a concise manner and
are therefore useful to us in applying the revised version of the counterfactual method.
480 B. Kment
Over-determination and preemption cases show that counterfactual dependence

between distinct matters of particular fact is not a necessary condition for causation.
That presents a serious problem for the counterfactual analysis of causation, given
that the viability of that account depends on the possibility of finding necessary and
sufficient conditions for causation cast in counterfactual terms. The same examples
present no difficulty for the view outlined in this paper, according to which coun-
terfactual dependence merely provides evidence for causal connections but does not
constitute them. All we need to conclude from the data is that the counterfactual
method is useful mostly for establishing causal claims, not for refuting them. What
is more, the account predicts this limitation of the counterfactual method, and
explains it by appealing to the fact that the determination idea provides merely a
necessary but not a sufficient condition for certain factors to include all the causes
of E. Hence, far from presenting a difficulty for the theory, the examples of over-
determination and preemption confirm the account.
The need for causal notions in the theory of counterfactuals threatens the
counterfactual analysis of causation with circularity. But it does not constitute a
problem for the view that counterfactual reasoning is useful for supporting causal
claims. All we need to conclude is that counterfactual thinking cannot generally
create causal knowledge from scratch. Given that causal notions figure in the truth-
conditions of counterfactuals, we cannot in general acquire causal knowledge by
counterfactual reasoning unless we have some causal knowledge to begin with. But
there is no circularity and no regress. For, the causal knowledge required for our
counterfactual reasoning is different from that which we acquire as a result of it
(see Kment 2014: Sects. 10.6.2, 11.5 for more detail). We use one item of causal
knowledge to gain another.25 In that way, counterfactual reasoning extends our stock
of causal knowledge. And that is what makes it useful.
Finally, by portraying the use of counterfactuals in establishing causal claims as
an extension of Mill’s method, my view provides a unified account of the workings
of the counterfactual method and of the method of difference (including the method
of controlled experiments). This theoretical unification is a further virtue of the
account.
Acknowledgements This paper is based on a talk I gave at the Linguistic Perspectives on

Causation workshop at the Hebrew University. I am grateful to the organizers of this workshop,
Nora Boneh and Elitzur Bar-Asher Siegal, to the other workshop participants, and to two referees
for this volume. I am also indebted to Oxford University Press for permission to use passages from
my book Modality and Explanatory Reasoning (Oxford University Press, 2014), and to Wiley for
permission to use passages from my paper “Causation: Determination and Difference-Making,”
Noûs 44: 80–111 (© 2010 Wiley Periodicals, Inc.).
25 More specifically, in establishing that the fact that cht (E) = p counterfactually depends on A, we
need to draw on knowledge about which facts about specific post-tA events were actually caused by
A (only those that were not actually caused by A can be held fixed). Once we have established the
counterfactual dependence, we can infer a new causal claim: A is a cause of the fact that cht (E) = p.
References
Adams, E. (1975). The logic of conditionals. Dordrecht: Reidel.

Ahmed, A. (2014). Evidence, decision and causality. Cambridge: Cambridge University Press.
Armstrong, D. M. (2001). Going through the open door again: Counterfactual versus singularist
theories of causation. In G. Preyer (Ed.), Reality and humean supervenience: Essays on the
philosophy of David Lewis (pp. 163–176). New York: Rowman and Littlefield.
Beebee, H. (2004). Causing and nothingness. In J. Collins, N. Hall, & L. Paul (Eds.), Causation
Bennett, J. (1974). Review of counterfactuals. The Canadian Journal of Philosophy, 4, 381–402.
Bennett, J. (1984). Counterfactuals and temporal direction. Philosophical Review, 93, 57–91.
Bennett, J. (2003). A philosophical guide to conditionals. Oxford: Clarendon Press.
Dowe, P. (2000). Physical causation. New York: Cambridge University Press.
Dowe, P. (2001). A counterfactual theory of prevention and ‘causation’ by omission. Australasian
Journal of Philosophy, 79, 216–226.
Edgington, D. (2003). Counterfactuals and the benefit of hindsight. In P. Dowe & P. Noordhof
(Eds.), Causation and counterfactuals. London: Routledge.
Edgington, D. (2011). Causation first: Why causation is prior to counterfactuals. In C. Hoerl
et al. (Eds.), Understanding counterfactuals, understanding causation: Issues in philosophy
and psychology (pp. 230–241). Oxford: Oxford University Press.
Fine, K. (1975). Review of counterfactuals. Mind, 84, 451–458.
Gibbard, A., & Harper, W. (1981). Counterfactuals and two kinds of expected utility. In W. Harper,
R. Stalnaker, & G. Pearce (Eds.), Ifs: Conditionals, belief, decision, chance, and time (pp. 125–
162). Dordrecht: Reidel.
Hall, N. (2004). Two concepts of causation. In J. Collins, N. Hall, & L. Paul (Eds.), Causation and
counterfactuals (pp. 225–277). Cambridge, MA: Bradford Books, MIT Press.
Hall, N. (2007). Structural equations and causation. Philosophical Studies, 132, 109–136.
Halpern, J. Y., & Pearl, J. (2005). Causes and explanations: A structural-model approach. Part 1:
Causes. British Journal for the Philosophy of Science, 56, 843–887.
Hiddleston, E. (2005). A causal theory of counterfactuals. Noûs, 39, 632–657.
Hitchcock, C. (2001). The intransitivity of causation revealed in equations and graphs. Journal of
Philosophy, 98, 273–299.
Hoerl, C., McCormack, T., & Beck, S. (Eds.). (2011). Understanding counterfactuals, understand-
ing causation: Issues in philosophy and psychology. Oxford: Oxford University Press.
Hume, D. (1995). An inquiry concerning human understanding. Upper Saddle River: Prentice Hall.
Jackson, F. (1977). A causal theory of counterfactuals. Australasian Journal of Philosophy, 55,
3–21.
Joyce, J. (1999). The foundations of causal decision theory. Cambridge: Cambridge University
Press.
Kment, B. (2006). Counterfactuals and explanation. Mind, 115, 261–310.
Kment, B. (2010). Causation: Determination and difference-making. Noûs, 44, 80–111.
Kment, B. (2014). Modality and explanatory reasoning. Oxford: Oxford University Press.
Kment, B. (2015). Modality, metaphysics, and method. In C. Daly (Ed.), The Palgrave handbook
of philosophical methods (pp. 179–207). New York: Palgrave Macmillan.
Lewis, D. (1973). Counterfactuals. Cambridge, MA: Harvard University Press.
Lewis, D. (1981). Causal decision theory. Australasian Journal of Philosophy, 59, 5–30.
Lewis, D. (1986a). Causation. In Philosophical papers (Vol. 2, pp. 159–213). New York/Oxford:
Lewis, D. (1986b). Counterfactual dependence and time’s arrow. In Philosophical papers (Vol. 2,
pp. 32–66). New York/Oxford: Oxford University Press.
Lewis, D. (2004a). Causation as influence. In J. Collins, N. Hall, & L. Paul (Eds.), Causation and
482 B. Kment
Lewis, D. (2004b). Void and object. In J. Collins, N. Hall, & L. Paul (Eds.), Causation and
Mackie, J. L. (1965). Causes and conditions. American Philosophical Quarterly, 2, 245–264.
Mackie, J. L. (1974). The cement of the universe. Oxford: Oxford University Press.
Mårtensson, J. (1999). Subjunctive conditionals and time (Acta Universitatis Gothoburgensis).
Gothenburg: University of Gothenburg.
Maudlin, T. (2004). Causation, counterfactuals, and the third factor. In J. Collins, E. J. Hall, & L.
A. Paul (Eds.), Causation and counterfactuals. Cambridge, MA: MIT Press.
McDermott, M. (1995). Redundant causation. British Journal for the Philosophy of Science, 46,
523–544.
Menzies, P. (1989). Probabilistic causation and causal processes: A critique of Lewis. Philosophy
of Science, 56, 642–663.
Mill, J. S. (1956). A system of logic, ratiocinative and inductive. London/New York: Longmans,
Green.
Paul, L. A. (2004). Aspect causation. In J. Collins, N. Hall, & L. Paul (Eds.), Causation and
Paul, L. A., & Hall, N. (2013). Causation: A user’s guide. Oxford: Oxford University Press.
Pearl, J. (2009). Causality: Models, reasoning, and inference. Cambridge, MA: Cambridge
University Press.
Quine, W. V. O. (1950). Methods of logic. New York: Holt, Rinehart, and Winston.
Ramachandran, M. (1997). A counterfactual analysis of causation. Mind, 151, 263–277.
Roese, N. J., & Olson, J. M. (Eds.). (2014). What might have been: The social psychology of
counterfactual thinking. New York: Psychology Press.
Schaffer, J. (2000). Overlappings: Probability-raising without causation. Australasian Journal of
Philosophy, 78, 40–46.
Schaffer, J. (2004). Counterfactuals, causal independence and conceptual circularity. Analysis, 64,
299–309.
Schulz, K. (2011). If you’d wiggled A, then B would’ve changed. Synthese, 179, 239–251.
Slote, M. (1978). Time in counterfactuals. Philosophical Review, 87, 3–27.
Stalnaker, R. (1981). Letter to David Lewis. In W. Harper, R. Stalnaker, & G. Pearce (Eds.), Ifs:
Conditionals, belief, decision, chance, and time (pp. 151–152). Dordrecht: Reidel.
Strevens, M. (2007). Mackie remixed. In J. K. Campbell, M. O’Rourke, & H. S. Silverstein (Eds.),
Causation and explanation. Cambridge, MA: MIT Press.
Tichý, P. (1976). A counterexample to the Stalnaker-Lewis analysis of counterfactuals. Philosoph-
ical Studies, 29, 271–273.
Veltman, F. (2005). Making counterfactual assumption. Journal of Semantics, 22, 159–180.
Wasserman, R. (2006). The future similarity objection revisited. Synthese, 150, 57–67.
Woodward, J. (2003). Making things happen: A theory of causal explanation. New York: Oxford
University Press.
Yablo, S. (2004). Advertisement for a sketch of an outline of a proto-theory of causation. In J.
Collins, N. Hall, & L. Paul (Eds.), Causation and counterfactuals (pp. 119–138). Cambridge,
MA: Bradford Books, MIT Press.

Dokumen - Pub - Perspectives On Causation Selected Papers From The Jerusalem 2017 Workshop 1st Ed 9783030343071 9783030343088

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dokumen - Pub - Perspectives On Causation Selected Papers From The Jerusalem 2017 Workshop 1st Ed 9783030343071 9783030343088

Uploaded by

Copyright:

Available Formats

Jerusalem Studies in Philosophy and History of Science

Elitzur A. Bar-Asher Siegal

More information about this series at http://www.springer.com/series/16087

ISSN 2524-4248 ISSN 2524-4256 (electronic)

© Springer Nature Switzerland AG 2020

We would like to conclude by noting that this book represents a community

Jerusalem, Israel Elitzur A. Bar-Asher Siegal

Part I Perspectives on Causation

Part II Methodology: Uncovering the Representation of Causation

Part III Meaning Components of Causation

8 Aspectual Differences Between Agentive and Non-agentive

Part IV Syntactic and Semantic Aspects of Causation

Part V Philosophical Inquiries on Causation

Odelia Ahdout Humboldt Universität zu Berlin, Berlin, Germany

Christopher Hitchcock Division of Humanities and Social Sciences, California

Elitzur A. Bar-Asher Siegal and Nora Boneh

Abstract This paper examines reciprocal connections between the discussions

Keywords Cause · Effect · Dependency · Counterfactuality · Causal Selection ·

© Springer Nature Switzerland AG 2020 3

1.1 Introduction: Philosophical and Linguistic Discussions

26–27), and Hitchcock’s contribution to this volume.

(i) a cause (c);

propose, however, a different approach as to what should be considered as a causative construction,

1.2 Setting the Scene

1.2.1 Theoretical Questions and Their Background

C. As for the philosophical discussions on causation, we inquire whether they are

analysis in terms of closeness between possible worlds (cf. Anscombe 1981).

This understanding of causation, to a large extent, follows Lewis’ (1973a, b)

(2) a. A kangaroo is a marsupial because it has a pouch. (Dowty 1979: 132b)

1.2.2 Causative Constructions

We center on three central types of causative constructions in English and Hebrew.

1.2.2.1 Overt Causative Verbs

(3) a. [c The neighbor/the music] caused / made / enabled [e the kids

(4) a. [e The kids danced] because of [c the music].

(5) a. [e ha-delet niftexa] biglal / me- [c ha-ruax].

1.2.2.3 Lexical Causatives

(16) a. The recession caused Jerry to lose his home. ⇒

(17) She is not functioning from stress.

(20) ha-zamar hirkid et ha-yeladim.

Similarly, in English, (21) can describe a situation in which counterfactuality

(21) John marched the prisoners. (Cruse 1972: ex. 7)

Table 1.1 Counterfactuality Dc : pcause cDe ⇒ (∼c → ∼e)

1.4 Singling out the Cause

(22) pcDe ⇒ “C is the cause of E”

1.4.1 Overt Causative Verbs

(24) The traffic jam caused Pat’s cooking spaghetti late.

33 We thank Arnon Levy for bringing up this issue.

(27) a. Ann ate broccoli today because it’s Monday.

1.4.3 Lexical Causatives

(32) a. The baby opened the door. ⇒

(34) a. The key opened the door.

Table 1.2 Causal Selection Dc : pcause cDe (22)

1.5 Causation Under Negation

1.5.1 Negating the Dependency: D Entailed or Not?

(38) a. [e ha-yeladim lo paxadu] biglal / me- [c ha-ruax].

(40) ha-zamar lo hirkid et ha-yeladim.

Table 1.3 D is asserted (one pcause cDe ⇒ c&e&[cDe]

1.5.2 Negating the Dependents

(41) a. [c NEG taking the medicine] D [e her death]

In comparison, to a certain extent, the application of constituent negation is

(47) She didn’t die from the medicine.

(48) hi lo meta me-ha-trufot.

(50) hi lo metafkedet me-ha-laxac