Bochnak, M. R., & Matthewson, L. (Eds.) - 2015. Methodologies in Semantic Fieldwork.

Introduction
M. Ryan Bochnak and Lisa Matthewson
1 Context
With as many as half of the world’s languages facing extinction within the next
century (Harrison 2010), research on endangered languages is critical for the ad-
vancement of linguistic theory. However, the vast majority of these endangered
languages remain understudied by linguists, and in particular there has been a
general lack of systematic investigation of their semantic properties. Even for
languages for which there is a tradition of documentation, most of this work his-
torically has focused on phonology and morphology, and to some extent syntax,
while detailed semantic information is typically largely absent (see Van Valin
2006 for similar comments). In claiming that semantics is the most understud-
ied subfield when it comes to endangered languages, we in no way wish to min-
imize the excellent descriptive groundwork done on many languages, on which
formal semanticists gratefully rely whenever possible. However, fieldwork spe-
cifically targeting semantics, and informed by a semantic theory with predictive
power, is required to accurately establish the meaning of elements in a fieldwork
situation.
Part of the reason for the relative paucity of semantic work on minority lan-
guages is that modern formal semantics is a relatively young subfield within lin-
guistics, and many of the early advancements in this area were originally based on
English and German (Partee 2005; von Fintel and Matthewson 2008).
Thankfully, over the past couple of decades or so, there has been an increased
interest in documenting semantic structures in understudied languages, and
linguists have become eager to test semantic theories against a broader range of
languages. A couple of early examples include Maria Bittner’s work on Kalaal-
lisut (Eskimo-Aleut; e.g., Bittner 1987), and an edited volume on cross-linguistic
quantification (Bach et al. 1995). A line of research investigating semantic uni-
versals and variation has greatly benefited from this work, which has introduced
new empirical findings that must now be accounted for by our semantic theories.
1
Methodologies in Semantic Fieldwork. M. Ryan Bochnak and Lisa Matthewson.

© Oxford University Press 2015. Published 2015 by Oxford University Press.
2 Methodologies in Semantic Fieldwork
Conferences such as Semantics of Underrepresented Languages in the Americas

(SULA, which started in 2001) have helped elevate the status of semantic research
based on primary fieldwork, particularly on indigenous languages of the Ameri-
cas, where the rates of language endangerment are unfortunately dire.
Semantic fieldwork1 poses a distinct methodological challenge in its investiga-
tion of the meaning of utterances, or parts of utterances, in the language of study.
The fieldworker attempts to establish semantic facts that are often quite subtle and
highly context-dependent. The semantic properties under investigation are often
not consciously accessible to speakers, and are not necessarily reflected in transla-
tions into another language. Unlike with other areas of linguistics, language data
collected usually do not contain direct information about semantics. For exam-
ple, most utterances by native speakers provide some positive information about
phonetics, phonology, morphology, and syntax—simply by providing exemplars
of grammatical, well-pronounced constructions. But the same is not true of se-
mantics. A semantically well-formed utterance, even if paired with a translation or
a discourse context, provides the researcher with very incomplete clues about what
that utterance really means.
Despite these challenges, there are vanishingly few resources available on
methodological issues for obtaining semantic judgments in fieldwork situations.
For instance, Chelliah and de Reuse’s (2011) fieldwork textbook devotes one chap-
ter (chapter 13) to “Semantics, Pragmatics, and Text Collection,” with a special
focus on lexical semantics, discourse structure, and pragmatic phenomena (in-
cluding deixis, implicature, presupposition, and speech acts). Only one subsec-
tion is devoted to briefly outlining the notions of entailment and acceptability
judgments on the well-formedness of complete linguistic utterances. Likewise,
Bowern (2008) contains one chapter (chapter 8) describing elicitation techniques
for gathering lexical semantic data, and one chapter (chapter 9) that focuses on
the elicitation of texts and discourses. While these two resources serve as excellent
references on general fieldwork techniques and practices, neither contain any sus-
tained discussion on eliciting data on functional semantic categories, which form
the “bread and butter of working semanticists,” as von Fintel and Matthewson
(2008) put it.
Recognizing the special challenges of semantic fieldworkers, Matthewson
(2004) describes a methodology for collecting semantic data from linguistically
untrained consultants. She argues that the fieldworker must take advantage of sev-
eral modes of data collection to arrive at the most complete empirical picture pos-
sible. The methodology argued for includes direct elicitation techniques, including
translations and judgments on felicity and truth conditions, and a system for how
1
We use the term “fieldwork” as it is most commonly used in linguistics, referring to research
conducted on a language of which the linguist is not a native speaker, typically involving one-on-one
interviews with native speaker consultants.
Introduction 3
to interpret the results of such tasks to arrive at the semantic facts. Importantly,
the methodology must allow the researcher to probe for negative data: contexts
where a well-formed utterance is not acceptable. Matthewson furthermore claims
that using a contact language (such as English) for presenting discourse contexts
is unlikely to affect the results of a given task. While certain linguists have recently
argued against gathering elicited data from fieldwork, or asking native speakers
to judge “invented” data (e.g., Mithun 2001; Himmelmann 2006), we follow Mat-
thewson (2004) in maintaining that these forms of data-gathering are indispensa-
ble for the semantic fieldworker.
Many semantic fieldworkers are using techniques similar to those advocated
by Matthewson (2004) in their work on understudied languages, and the growing
number of semantic fieldworkers means that techniques are being refined and new
techniques are being added. (See for example Tonhauser et al. (2013) for excellent
recent methodological discussion.) The advantage of using commonly accepted
methodologies is that the tasks can be systematically replicated by researchers
working on diverse languages, in order to allow for fruitful cross-linguistic com-
parison. Nevertheless, each field situation is unique and presents its own set of
challenges, and so the fieldworker must adapt methodological tools to meet the
challenges encountered in the field. Furthermore, different semantic domains
(e.g., definiteness, modality, comparison, etc.) demand nuanced elicitation tech-
niques to gather the range of data and contrasts that the theoretical literature has
identified as crucial for cross-linguistic comparison.
The goal of this volume is therefore to expand the discussion of methodol-
ogy for semantic fieldwork with contributions from researchers who have inves-
tigated a variety of topics, representing a diverse array of languages. The chapters
in part one explore general elicitation techniques, while those in part two discuss
techniques for investigating particular semantic topics. The chapters in part three
consist of case studies in using language-internal evidence to guide semantic
elicitation. In each case, the methodological discussions are supplemented with
examples from the authors’ own fieldwork, showcasing the successful applica-
tion of the techniques. In all, 11 language families are represented, spanning four
continents.
In the remainder of this introductory chapter, we first provide a brief defense
of fieldwork in general, in response to some recent attacks on the premise that one-
on-one work with a small number of speakers is a valid way to conduct empirical
research. We will then give an overview of the topics that are and are not covered
in the volume, and will end by briefly previewing each chapter.
2 Experiments Large and Small
In recent years there has been a welcome increase in the frequency with which
experimental methods are applied to semantic questions. Recent years have also
seen lively discussion in the literature about the methodologies linguists should
be using in their research. One of the issues of debate is whether it is better, or
even necessary, to obtain large-scale experimental evidence for empirical propos-
als. For the claim that large-scale experiments are the way we must go, see for
example Edelman and Christiansen (2003), Ferreira (2005), Wasow and Arnold
(2005), Featherston (2007), Gibson and Fedorenko (2010a, 2010b) and Gibson
et al. (2013). For defenses of methodologies involving small numbers of speakers,
on the other hand, see den Dikken et al. (2007), Fanselow (2007), Grewendorf
(2007), Haider (2007), Weskott and Fanselow (2008), Phillips (2010), Sprouse and
Almeida (2012a, b, 2013) and Sprouse et al. (2012); see also Featherston (2009).
This debate is of critical importance for any linguist working on an endan-
gered language, or indeed for any linguist who cares about endangered language
data and believes that it should be collected. If large-scale experimental results are
the only reliable results, what is the status of data collected from a language for
which such experiments are impossible for logistical reasons (such as only having
10 speakers left, all over the age of 80)? Gibson and Fedorenko’s answer (2010b:7)
is that “the conclusions that can be drawn from [data from endangered languages]
will be weaker and more speculative in nature than the conclusions based on
quantitative data.” Does this mean that we should not even bother collecting data
from minority languages? Should we give up on linguistic diversity now, and only
work on Indo-European and Sino-Tibetan?2
Our answer to these questions is “no.” The current collection of papers is not
intended as a direct contribution to the debate on whether large-scale experimen-
tal methods are a necessary part of empirical semantic research. The volume is pri-
marily targeted at readers who are already convinced that one-on-one fieldwork
is a viable linguistic methodology, and are interested in the theory and practice
of such work as applied to semantic questions. Nevertheless, we believe that the
volume as a whole serves as a counter-argument to those who believe that seman-
tic questions cannot be fruitfully investigated via fieldwork. By providing explicit
methodologies and examples of their use, we hope to demonstrate the rigor and
success of the relevant techniques. Every paper in this volume discusses and pro-
vides data from a language that is either officially endangered or in decline: Ba-
diaranke, Blackfoot, Cheyenne, Gitksan, Innu-aimun, Inuttut, Kaqchikel, Kiowa,
Navajo, Nez Perce, St’át’imcets, Skwxwú7mesh, Washo, and Yucatec Maya. If the
researchers in this volume (and many others working in the same way) had opted
to work only on languages for which conditions allow large-scale experiments, the
field would be vastly empirically poorer.
2
To see what we are dealing with in terms of the bias against fieldwork on minority languages,
consider these extracts from a recent anonymous review of a paper based on fieldwork: “This reviewer
suspects that fieldwork done through another language is indeed pretty unreliable. The suspicion in-
creases when the native speakers belong to a moribund language . . . How much should one, can one,
believe an informant’s metalinguistic judgments about a language no longer frequently used?”
Introduction 5
It is important to emphasize that our defense of the small-scale experiments

we know as “fieldwork” is not only motivated by practical considerations. On the
contrary, we believe that fieldwork with small numbers of speakers is a robust,
scientifically valid research methodology. As pointed out by Sprouse and Almeida
(2012b), among others, the informal experiments carried out by fieldworkers are
in many crucial respects identical to the experiments carried out in laboratories.
Both methodologies involve the designing of a set of conditions to test the rele-
vant minimal contrasts, and both attempt to rule out nuisance variables. Targeted,
hypothesis-driven elicitation is designed to test the predictions of falsifiable hy-
potheses about language, and as such it meets the primary criterion for scientific
research.
The results of small-scale fieldwork experiments are also reproducible, both
within and across speakers. In fact, even the entirely non-experimental data-
gathering technique of accessing the researcher’s own intuitions can give results
that overwhelmingly correspond to those given by large-scale experiments. Phil-
lips (2010:53) gives examples showing that “carefully constructed tests of well-
known grammatical generalizations overwhelmingly corroborate the results of
‘armchair linguistics’. ” Similarly, Sprouse et al. (2012) randomly selected 146 two-
condition phenomena from articles in Linguistic Inquiry, which were originally
gathered using non-experimental methods. They tested each of these data points
experimentally, and found a replication rate of 95% (with a margin of error of just
over 5%).
Even if fieldwork data were not replicable across large numbers of speakers
(and for minority languages this may never be testable), fieldworkers can still con-
firm intra-speaker reproducibility. That is, for any single speaker, the results of a
variety of grammatical tests over a number of different stimuli should converge on
the same results. Intra-speaker reproducibility is sufficient, given that our object
of investigation is the grammatical competence of individual speakers. Given that
different speakers of the same language have similar but not necessarily identical
grammars, using large numbers of speakers doesn’t necessarily lead to clearer re-
sults, since averaging results over 200 different grammars can be more misleading
than investigating one or two different grammars in depth (see den Dikken et al.
2007, Fanselow 2007, Grewendorf 2007, Phillips 2010, among others).
Finally, there are advantages to one-on-one fieldwork as opposed to large-
scale experiments; these derive from the time spent with each speaker, and from
the fact that the fieldworker–consultant relationship is not fully parallel to the
investigator–subject relationship. For example, we are better able to avoid prob-
lems such as those mentioned by Schütze (2005), having to do with whether
speakers fully understand the tasks we are asking them to do. Speaker confusion
about the task is likely to be noticed much earlier, and mitigated much more easily,
in a context where one is sitting face-to-face with one speaker and assessing the
success of the task in real time, and moreover when one has a long-term, collabo-
rative relationship with one’s speakers.
3 Coverage of the Volume
This volume grew out of a panel that took place at the 2012 Annual Meeting of the
Linguistic Society of America in Portland, Oregon. The chapters in this volume
are largely based on the talks and posters presented as part of this panel.3 One of
the goals of the panel was to bring together both younger scholars making new
discoveries in their semantic research on understudied languages and more sea-
soned fieldworkers who have come to be seen as leaders in this small but growing
field. The panel successfully generated much constructive discussion, the results of
which we are proud to present in this volume.
Before previewing the individual chapters, we would first like to flag a few
issues relevant to semantic fieldwork that are not covered in this book. First, this
book is not meant to be a handbook on general fieldwork methodologies; our
focus is only on the investigation of semantic phenomena. We refer readers to
fieldwork manuals or collections such as Newman and Ratliff (2001), Bowern
(2008) and Chelliah and de Reuse (2011) on linguistic fieldwork in other areas of
linguistics, and also for discussion of practical issues such as fieldwork ethics, data
organization and archiving, grant writing, and use of technologies in the field.
Second, while texts can serve as an important source of data for semantic in-
vestigations, we do not cover general methodologies for eliciting or working with
texts or corpora. However, a couple of chapters do include discussion of the use of
texts in semantic fieldwork: the chapter by Burton and Matthewson discusses the
use of a targeted storyboard methodology to generate texts containing sentences
with modals, and the chapter by Cover discusses the role of texts in the investi-
gation of tense, aspect, and modality markers in Badiaranke. We refer interested
readers to Chelliah and de Reuse (2011, chapter 13) for more detailed discussion on
collecting texts in the field.
Finally, a note on the comprehensiveness of topics covered in this volume.
While we made every attempt to include chapters on as wide a range of topics
within semantics as possible, there are a few such topics not represented here. For
instance, there are no chapters that specifically cover quantification or presupposi-
tion. In the case of quantification, we point readers to Benjamin Bruening’s Scope
Fieldwork Project (http://udel.edu/~bruening/scopeproject/scopeproject.html),
which includes a visual stimulus kit and methodological discussion for collecting
data on quantifiers and scope in the field. For presupposition and projective con-
tent, we recommend Tonhauser et al.’s (2013) article in Language, which discusses a
methodology for setting up contexts to test varieties of projective content, with ex-
amples from Paraguayan Guaraní. We would also like to point out that every effort
3
One exception is Judith Tonhauser’s contribution to the panel, which discussed methodologies
for investigating presupposition and projective content and has been published in Language (Ton-
hauser et al. 2013). We have replaced this contribution with the chapter by Cover and Tonhauser on
theoretically informed fieldwork on temporal and aspectual reference.
Introduction 7
was made to ensure that each chapter will be useful and interesting for fieldwork-
ers who are not necessarily fluent in current formal semantic theories; however,
in places where a certain amount of semantic sophistication is required, we have
asked authors to flag these sections when they arise.
4 Overview of Contributions
Bohnemeyer’s chapter discusses methods of linguistic data collection for reveal-

ing semantic information, which is often much more elusive than information on
phonology, morphology, and syntax. He outlines several types of semantic elicita-
tion techniques, where elicitation is defined as a data-gathering activity involving a
stimulus, a task, and a response. His concerns revolve around the important ques-
tions of what types of evidence can be gained from elicitation activities with native
speakers, and how researchers can gather such evidence in the field and interpret
the data for semantic analysis. He illustrates the various techniques with examples
from his own fieldwork on Yucatec Maya.
Louie’s chapter addresses a seemingly paradoxical challenge for semantic
fieldworkers: on the one hand, semanticists often need to collect data of a par-
adigmatic and systematic nature; on the other hand, fieldwork consultants often
become bored with paradigmatic and systematically planned elicitation sessions.
She proposes a method for alleviating this problem, which involves embedding
test sentences within an over-arching storyline. She furthermore addresses the
question of how to deal with almost-minimal pairs of context-utterance pairings.
Her discussion is accompanied by several entertaining examples from her field-
work on Blackfoot (Algonquian).
Bar-el’s contribution sets forth a proposal for a fieldworker’s toolkit for exam-
ining aspectual classes in the field. She critiques existing questionnaires designed
for eliciting aspectual contrasts, as well as existing classifications of predicates,
which have been based mostly on better-studied languages. Using examples from
Dëne Sųłiné (Athabaskan) and from her own fieldwork on Skwxwú7mesh (Salish),
Bar-el outlines the essentials of a toolkit for documenting a wide range of aspec-
tual features in the field.
Bochnak and Bogal-Albritten present methodologies for investigating degree
constructions, comparison, and gradability. Using examples drawn from their
fieldwork on Washo (Hokan/isolate) and Navajo (Athabaskan), they demonstrate
how notions such as norm-relatedness and crisp judgments can be tested using
visual, tactile, and verbal stimuli. The Washo and Navajo data they present reveal
significant departures from what one might expect based on existing analyses of
familiar languages.
Burton and Matthewson motivate the use of storyboards in semantic elicita-
tion, which are targeted at investigating a particular construction or topic. They
illustrate the technique by showing how storyboards have been successfully used
to make and test hypotheses about modals in Gitksan (Tsimshianic), St’át’imcets

(Salish), and Blackfoot. They argue that data obtained by the storyboard technique
are as natural-sounding as spontaneous speech, obviating a potential drawback of
data elicited verbally.
Deal’s chapter addresses the status of translation and judgment tasks in se-
mantic fieldwork, also taking the semantics of modals as a case study, this time in
Nez Perce (Sahaptian). She makes explicit two hypotheses—the Equivalent Trans-
lations Hypothesis and the Equivalent Judgments Hypothesis—which form the
keystones for reasoning about the outcome of translation and judgment tasks with
native speakers. Deal discusses ways these hypotheses may be fruitfully exploited
in field research, noting along the way certain flaws that may arise.
Gillon’s chapter presents methodologies for testing the semantics of deter-
miners, drawing on examples from her fieldwork on Skwxwú7mesh, Lithuanian
(Indo-European), Innu-aimun (Algonquian), and Inuttut (Eskimo-Aleut). Some
of these languages lack overt determiners, but Gillon shows how the presence (or
absence) of null determiners can be detected through fieldwork. She argues that
the semantics shared by all determiners is domain restriction, and that the differ-
ent semantic possibilities for determiners are instantiated both in overt and null
varieties.
AnderBois and Henderson address the question of which language to use
when presenting discourse contexts to consultants. On the basis of two case
studies—attitude reports and parentheticality in Yucatec Maya, and distributive
pluractionality in Kaqchikel (Mayan)—they argue that both are viable, but that a
complex array of linguistic factors determines whether the object language or the
language of wider communication is a better choice. They present a best practices
guide, according to which researchers should disclose what language was used to
establish the discourse context, and the reasons for this choice.
Cover’s chapter addresses the significant challenges involved in establishing
the semantics of tense, aspect, and modality in a fieldwork situation. She argues
that direct elicitation is a crucial part of such investigations, along with text col-
lection and participant observation. She provides examples of all three techniques
from her fieldwork on the imperfective and discontinuous past tense in Badi-
aranke (Niger-Congo).
McKenzie presents techniques for eliciting information about the discourse
status of noun phrases, drawing on his fieldwork on this issue in Kiowa (Tanoan).
He argues that the pragmatic factors that license noun phrase dislocation in Kiowa
can be investigated without recourse to vague notions such as “aboutness.” He
demonstrates how the subtle discourse properties of these dislocated phrases can
be detected using standard elicitation techniques such as the gathering of accept-
ability judgments in contexts.
Murray’s chapter offers a detailed illustration of the bi-directional interplay
between empirical findings obtained through fieldwork and formal theories of
semantics. Her case study on the reflexive/reciprocal construction in Cheyenne
Introduction 9
(Algonquian) shows that while novel data gained through fieldwork clearly impact
theoretical analyses, theoretical proposals also guide further fieldwork, leading in
turn to empirical discoveries that would likely not have otherwise been made.
Cover and Tonhauser similarly address the issue of the interplay between field-
work and theory. They argue that linguistic theory can guide fieldwork in a way
which leads to new empirical discoveries, and moreover that theoretically informed
research has a greater potential to reveal the language-specific “genius” of the lan-
guage under investigation than does research whose sole aim is description. They
support these proposals by means of examples from their fieldwork on temporal
and aspectual reference in Badiaranke and Paraguayan Guaraní (Tupí-Guaraní).
References
Bach, Emmon, Eloise Jelinek, Angelika Kratzer, and Barbara Partee (eds.). 1995. Quantifica-
tion in Natural Languages. Dordrecht: Kluwer.
Bittner, Maria. 1987. On the semantics of the Greenlandic antipassive and related construc-
tions. International Journal of American Linguistics 53: 194–231.
Bowern, Claire. 2008. Linguistic Fieldwork: A Practical Guide. New York: Palgrave Macmillan.
Chelliah, Shobhana L., and Willem J. de Reuse. 2011. Handbook of Descriptive Linguistic
Fieldwork. Dordrecht: Springer.
den Dikken, Marcel, Judy Bernstein, Claire Tortora, and Raffaella Zanuttini. 2007. Data and
grammar: Means and individuals. Theoretical Linguistics 33: 335–352.
Edelman, Shimon, and Morten Christiansen. 2003. How seriously should we take Minimal-
ist syntax? A comment on Lasnik. Trends in Cognitive Sciences 7: 59–60.
Fanselow, Gisbert 2007. Carrots—perfect as vegetables, but please not as a main dish. The-
oretical Linguistics 33: 353–367.
Featherston, Samuel 2007. Data in generative grammar: The stick and the carrot. Theoretical
Linguistics 33: 269–318.
Featherston, Samuel 2009. Relax, lean back, and be a linguist. Zeitschrift für Sprachwissen-
schaft 28(1): 127–132.
Ferreira, Fernanda 2005. Psycholinguistics, formal grammars, and cognitive science. The
Linguistic Review 22: 365–380.
von Fintel, Kai, and Lisa Matthewson. 2008. Universals in semantics. The Linguistic Review
25: 139–201.
Gibson, Edward, and Evelina Fedorenko. 2010a. Weak quantitative standards in linguistics
research. Trends in Cognitive Sciences 14: 233–234.
Gibson, Edward, and Evelina Fedorenko. 2010b. The need for quantitative methods in syntax
and semantics research. Language and Cognitive Processes. doi:10.1080/01690965.2010
.515080.
Gibson, Edward, Steven Piantadosi, and Evelina Fedorenko. 2013. Quantitative methods in
syntax/semantics research: A response to Sprouse and Almeida. Language and Cogni-
tive Processes 28: 229–240.
Grewendorf, Günther 2007. Empirical evidence and theoretical reasoning in generative
grammar. Theoretical Linguistics 33: 369–380.
Haider, Hubert 2007. As a matter of facts—comments on Featherston’s sticks and carrots.

Theoretical Linguistics 33: 381–394.
Harrison, K. David 2010. The Last Speakers: The Quest to Save the World’s Most Endangered
Languages. Washington, D.C.: National Geographic Society.
Himmelmann, Nokolaus. 2006. Language documentation: What is it and what is it good
for? In Essentials of Language Documentation, Jost Gippert, Nikolaus Himmelman,
and Ulrike Mosel (eds.), 1–30. Berlin: Mouton.
Matthewson, Lisa. 2004. On the methodology of semantic fieldwork. International Journal
of American Linguistics 70(4): 369–415.
Mithun, Marianne. 2001. Who shapes the record: the speaker and the linguist. In Linguis-
tic Fieldwork, Paul Newman and Martha Ratliff (eds.), 34–54. Cambridge: Cambridge
University Press.
Partee, Barbara 2005. Reflections of a formal semanticist as of Feb 2005. http://semanticsarchive.
net/Archive/2YxOWNmY/Reflections_Partee_Feb05.pdf.
Phillips, Colin 2010. Should we impeach armchair linguists? In Japanese-Korean Linguistics
17, S. Iwasaki, H. Hoji, P. Clancy, and S.-O. Sohn (eds.), 49–64. Stanford, CA: CSLI
Publications.
Schütze, Carson 2005. Thinking about what we are asking speakers to do. In Linguistic Ev-
idence: Empirical, Theoretical, and Computational Perspectives, S. Kepser and M. Reis
(eds.), 457–485. Berlin: Mouton de Gruyter.
Sprouse, Jon, and Diogo Almeida. 2012. Assessing the reliability of textbook data in syntax:
Adger’s Core Syntax. Journal of Linguistics, FirstView Articles. doi: http://dx.doi.
org/10.1017/S0022226712000011.
Sprouse, Jon, and Diogo Almeida. 2012b. Power in acceptability judgment experiments and
the reliability of data in syntax. Unpublished manuscript, University of California,
Irvine and New York University, Abu Dhabi.
Sprouse, Jon, and Diogo Almeida. 2013. The empirical status of data in syntax: A reply to
Gibson and Fedorenko. Language and Cognitive Processes 28: 222–228.
Sprouse, Jon, Carson Schütze, and Diogo Almeida. 2012. Assessing the reliability of journal
data in syntax: Linguistic Inquiry 2001–2010. Manuscript under review, available at
http://www.linguistics.ucla.edu/people/cschutze/Assessing.pdf.
Tonhauser, Judith, David Beaver, Craige Roberts, and Mandy Simons. 2013. Toward a tax-
onomy of projective content. Language 89(1): 66–109.
Van Valin, Robert D. Jr. 2006. Some universals of verb semantics. In Linguistic Universals,
Ricardo Mairal and Juana Gil (eds.), 155–178. Cambridge: Cambridge University Press.
Wasow, Thomas, and Jennifer Arnold. 2005. Intuitions in linguistic argumentation. Lingua
115: 1481–1496.
Weskott Thomas, and Gisbert Fanselow. 2008. Scaling issues in the measurement of linguis-
tic acceptability. In The Fruits of Empirical Linguistics I: Process, Featherston, S., and S.
Winkler (eds.), 229–246. Berlin: Mouton de Gruyter.
1
A Practical Epistemology for Semantic Elicitation

in the Field and Elsewhere
Jürgen Bohnemeyer
1 Introduction
The aim of this chapter is to sketch a classification of elicitation methods in seman-

tics based on an analysis of the sources of evidence semanticists can draw on and
the principal components of any elicitation.*
As part of a general empiricist turn that has been slowly but inexorably chang-
ing the cognitive sciences for the last couple of decades, linguists working in the
classical core areas of morphosyntax, semantics, and phonology have begun to feel
the need for a more empirical footing of their research. The current concern with
variation and the growing importance of statistical methods throughout these
subfields seem to be testimony of this.
Furthermore, I believe that the tools for a more empirically grounded linguis-
tics are by and large already in our hands. Linguists have developed a host of pro-
cedures for data gathering and analysis since the days of the Structuralists. What
is called for now is an integration of these methodologies with the epistemological
standards of the social and behavioral sciences such as those that sociolinguists
and psycholinguists have been adhering to in their research. This chapter attempts
to make a modest contribution toward this general goal in the area of semantic
research.
The temptation of thinking of semantics as necessarily hermeneutically based
is perhaps greater than in other subfields of linguistics. After all, meaning is not
*Different parts of the discussion in this chapter are based upon work supported by the National
Science Foundation under Grant No. BCS-0723694 “Spatial language and cognition in Mesoamerica”
(PI J. Bohnemeyer) and by the Max Planck Society, respectively. I owe a wealth of gratitude to the Yu-
catec speakers who participated in the studies discussed here. I would like to thank Rui Chaves, Florian
Jaeger, Jean-Pierre Koenig, and Gail Mauner for helpful suggestions. I am indebted to the editors and
Judith Tonhauser for detailed, engaging, and constructive comments. I alone am to blame for the posi-
tions argued for and for any shortcomings. 13

directly observable—so how else would it be accessible if not through interpreta-

tion? This fallacy is directly responsible for the belief Matthewson (2004) dubbed
“relativist agnosticism”: the widespread assumption that it is impossible to study
the semantics of languages the researcher does not speak, or at least does not speak
native-like. If meaning is accessible through interpretation only, then how could
one study semantics in a language the utterances of which one does not know how
to adequately interpret? For this reason, research methods for semantics in field
contexts are of primary concern in this chapter—which is, of course, why this
chapter is a part of the present volume. But the same apparent conundrum pres-
ents itself in child language research: to the extent that the linguistic competence
and practice of children is not adult-like, how could adult researchers understand
children’s utterances well enough to adequately interpret their semantics?
This apparent conundrum is nothing more than the everyday reality of every
psychologist: mental states and processes are not directly observable, they must
be inferred from the behaviors they are assumed to underlie, in particular, par-
ticipants’ responses to particular stimuli in controlled research settings. I contend
that this is a fairly close analogy to how we should think of semantic research. It
is from this perspective that the survey of semantic elicitation techniques in this
chapter is offered.
The primary guiding maxim is that an empirical linguist cannot rely on their
own native speaker intuitions as their sole source of evidence. This principle places
the semantic field researcher in the same epistemological boat as any other empiri-
cally oriented semanticist. Like the child learning the meanings of the expressions
of her native language, as modeled in Brown’s (1958) “Original Word Game” (see
Figure 1.1), they must start out observing utterances produced by competent speak-
ers and their apparent referential correlates in the observable world. This holds
true even for semanticists who view meaning primarily as a relation between utter-
ances and the mind rather than between utterances and the world, as mental states
are not directly observable. The utterance–world correlations can be augmented
by utterance–utterance correlations, including metalinguistic utterances—for ex-
ample, judgment statements—and contact language utterances. These must not be
naively taken to simply represent the meanings of the target language utterances;
they can, however, serve as further data points in a correlation matrix.
Again like the child in the Original Word Game, the semanticist must now
formulate, based on the observed correlations, hypotheses concerning the mean-
ings of the utterances and their constituents and then proceed to test these. The
central questions of an empirical epistemology and methodology for semantics
are the following:
(1) Which observable properties of communicative behavior can be exploited

as the data of semantic research?
(2) How and under what constraints can evidence of these observable
properties be gathered in a fashion that permits valid analyses?
A Practical Epistemology for Elicitation 15
(3) Which analytical techniques are appropriate and optimal for bringing a
particular dataset to bear on a particular research question?
This chapter is primarily concerned with the second question. But since any mean-
ingful answer to (2) presupposes the existence of plausible and practicable answers
to (1), I briefly consider it in section 3.
Against the backdrop of the general picture of semantic research rooted in the
observation of communicative behavior, the present chapter develops a classifica-
tion of data gathering techniques that can serve as possible solutions in response
to question (2). The focus will be on methods for semantic elicitation. Elicitation
can be defined as a data collection technique that involves three principal com-
ponents: a stimulus, a task, and a response. In any kind of linguistic elicitation,
the stimulus may be a target language utterance; a contact language utterance; a
linguistic representation of some state of affairs (e.g., a description of some sce-
nario the native speaker consultant is asked to assume); a non-linguistic represen-
tation of some state of affairs; or a combination of any of the above. The response
may consist of a target language utterance produced by the speaker; a judgment
that may form the basis for diagnoses of well-formedness, truth conditions, and
so forth; or again a linguistic (e.g., explication by paraphrase) or a non-linguistic
(e.g., in demonstrations and act-out tasks) representation of some state of affairs.
All possible tasks may then be defined as mappings between possible stimuli and
possible responses. I argue that there are only seven possible types of elicitation
techniques in linguistics. Applications of all of these to semantic research will be
illustrated with examples from my fieldwork.
As the semanticist must observe native speaker intuitions about utterance-
world correlations as their primary source of evidence, linguistic and non-linguistic
representations as stimuli play a powerful role in semantic elicitation since they
allow the researcher to manipulate what the world is assumed to be like for the
study of particular utterances. Finally, I propose a Golden Rule of elicitation: the va-
lidity of any elicitation response as a data point in the reconstruction of the speak-
er’s linguistic competence depends on the speaker’s interpretation of the task and
the stimulus and their intended interpretation of their response.
The chapter is organized from here on as follows: I first elaborate on the
answer to question (1) I offered above. I then introduce the proposed classifica-
tion of data gathering techniques and specifically that of elicitation techniques and
proceed to illustrate the various types with examples from my field research on
Yucatec Maya. A discussion section introduces the Golden Rule and revisits the
dichotomy between empirical and hermeneutic research.
2 The Empirical Basis of (Field) Semantics
Empirical research in all areas of linguistics has been making strides, including in
semantics. Nevertheless, an explicit epistemological and methodological footing
for an empirical approach to the study of meaning has been lacking. Laypeople
and many trained linguists alike continue to assume that interpretation is the only
route to meaning and that semantics is therefore necessarily a hermeneutic enter-
prise. For the study of meaning in contexts where hermeneutic approaches alone
are insufficient—for example, in field research on indigenous languages, in work
with aphasic patients, and in the study of meaning in child language and gesture—
this belief system is presenting a serious obstacle. It is obviously not a belief system
held by the contributors to this volume—but many other colleagues, both field-
workers and non-fieldworkers, do seem to subscribe to it.
A principal challenge for an empirical science of meaning is the fact that the
meanings of linguistic utterances are not directly observable. Instead, they must
be inferred from the observable semiotic behavior of language users, which is to
say from their communicative behavior. In this respect, the task of the empirical
semanticist resembles that of the psychologist, who infers the properties of cogni-
tive representations and processes from the observable behavior they are assumed
to support. The semanticist’s position is also not unlike that of a child acquiring
semantic and pragmatic competence in the language she is exposed to. Figure 1.1
presents a cartoon version of the processes presumably involved in semantic ac-
quisition and empirical research in semantics alike: the child/researcher
¤ Observes apparent correlations between utterances (A) and stimuli (C);
¤ Derives inferences concerning the meaning of the utterance and its
constituents (B);
¤ Formulates hypothesis about the semantic system in the mind of the
observed competent speakers (the arrow in the bottom right corner of
Figure 1.1);
¤ Tests these hypotheses by manipulating the stimulus (or waiting for a
variation of it to occur spontaneously) and observing how this affects the
response (the arrow in the bottom left corner of Figure 1.1).
Figure 1.1 An elaborate update of Brown’s (1958) “Original Word Game”

The richest and most basic source of observational evidence in semantics is

the extensions of linguistic signs: their sets of possible referents. While exten-
sions cannot be observed exhaustively, since they are not finite sets, language
learners and semanticists can observe individual entities or states of affairs that
speakers use the words to label, can form hypotheses about the intensions or
senses governing membership in the extensions, and can test these hypotheses,
for example, by attempting to label new items and observing the response of
competent speakers. The reference of such expressions can also be studied by
observing their impact on the reference of the phrases and sentences they occur
in. A special case of this is the contribution the various constituents of declara-
tive sentences that serve to assert propositions make to the truth conditions of
such sentences. For a simple illustration of how the connection between refer-
ence and truth conditions can be exploited in semantic research, here is a recipe
for the study of the meaning of the preposition under: Take the sentence tem-
plate in (4):
(4) The x is under the y.
Replace x and y with terms describing various objects. Show speakers spatial con-
figurations of instances of the objects in question and ask them for each test con-
figuration whether the sentence is true or false. Try to describe what all those
configurations that make the sentence true according to the participants have in
common.
Two concepts closely related to that of truth conditions are those of entail-
ments and contradictions. Entailments and contradictions can be fairly straight-
forwardly correlated with observable communicative behaviors. For example, a
sequence of two utterances that are incompatible in that one contradicts an entail-
ment of the other will not only make it impossible for a native speaker consultant
to come up with a coherent scenario that instantiates the sequence of utterances,
but is likely to result more immediately in confusion and rejection on the part
of the speaker. Thus, entailments and contradictions open up further alleys for
empirically probing the truth conditions of given utterances. At the same time,
predicting the truth conditions and entailments of particular utterances on the
basis of their form and the meanings of their constituents is of course a central
goal of semantic research.
Referential data is extensional. To get at the intensions or sense meanings—
the conceptual content—of the elicited expressions, semantic and pragmatic
analyses must be performed to separate entailments of lexical and compositional
semantics from pragmatically generated meaning components. For a simple il-
lustration, consider the responses to the Topological Relations Picture Series (af-
fectionately known as “BowPed” after the authors, Melissa Bowerman and Eric
Pederson (Bowerman and Pederson ms.)) by two Mexican Spanish speakers, rep-
resented by the Venn diagrams in Figure 1.2:
Figure 1.2 Linguistic categorization of a subset of the BowPed scenes by two Spanish speakers
(data elicited by Osamu Ishiyama and Arthur Photidiadis)
It seems possible, but not likely, that these two speakers have different mental
lexicon entries for the preposition en. A more parsimonious explanation is that
the first speaker is restricting en to scenes to which more specific alternatives do
not apply—a preemption effect based on a scalar implicature licensed by Grice’s
first Maxim of Quantity, “Make your contribution as informative as is required.”
This is an illustration of how conversational implicatures can make the extension
of an expression appear narrower than it semantically is. Conversely, semantic
transfer—metaphor and metonymy—may widen the expression’s extension.
Another indirectly observable property of the semantic extension of expres-
sions is the potential variation in terms of prototypical and more marginal refer-
ents in the sense of Prototype Theory (Rosch and Mervis 1975; Rosch 1978). Many
expressions have structured extensions that include both focal and more marginal
instances. This kind of gradation can be the result of vagueness (or underspecifica-
tion), but also of membership in the extension being determined by similarity to
a prototype or by the degree of satisfaction of a set of violable criteria (known as
“preference rule systems” (Jackendoff 1983) or “idealized cognitive models” (Lakoff
1987)). Speakers are able to judge the goodness of a given referent and moreover
often flag marginal instances in their spontaneous productions, using hedges—(e.g.,
when saying that a penguin is “sort of a bird”) (Lakoff 1973:471). Similarly, speakers
are able to detect referents licensed by semantic transfer—metaphor or metonymy.
However, this ability diminishes gradually as transferred uses become convention-
alized. Thus, whereas a statement along the lines of (5) is likely to be considered true
without much reflection, (6) is more likely to trigger a double take, because “hot
dog,” unlike “toy dog,” is a fully conventionalized sense of the English word dog:
(5) A toy dog isn’t actually/really a dog.
(6) A hot dog isn’t actually/really a dog.
Furthermore, speakers have the ability to detect whether particular utterances
are appropriate for their contexts. Inappropriateness can have a variety of sources,
including anomaly, but also the failure to meet cultural norms of interaction, such
as norms of politeness or cultural requirements of particular speech acts. Where
these conditions are not met, the utterance is infelicitous and therefore fails to have
the conventional effect it has in the right circumstances. Another kind of semantic
condition that contexts must meet in order for certain utterances to be appropri-
ately interpretable in them is presuppositions. Lastly, language-processing evidence
can be tapped into in exploring the structure of the mental lexicon. For instance, the
degree of semantic relatedness of two lexical items can be assessed by using word
association tasks (Clark 1970). It is also accessible to priming effects in lexical deci-
sion (“Is the stimulus a word?”) or semantic categorization tasks (e.g., “Does the
stimulus label a kind of animal?”); cf. Meyer, Schvaneveldt, and Ruddy (1972, 1974);
Marslen-Wilson et al. (1994); Perea and Rosa (2002a, b); Ferrand and New (2003);
Bueno and Frenck-Mestre (2008); inter alia. Interference effects in chronometric
sentence processing (Lewis et al. 2006 and the literature cited there) and picture-
naming tasks (Costa et al. 2005 and the literature cited therein) are likewise known
to be sensitive to semantic relations. Even neurological evidence can in principal be
brought to bear on the semantic relatedness of two expressions, including evidence
from the impairment complex known as “deep dyslexia” (Marshall and Newcombe
1973; Plaut and Shallice 1993 and references therein). For further discussion on the
use of psycholinguistic evidence in semantic studies, see Krifka 2011.
Important objections have been advanced against semantic externalism, ref-
erencing the potential problem of referential indeterminacy (Quine 1960) or citing
a philosophical rejection of objectivism (Lakoff 1987; Jackendoff 2002). I take these
criticisms extremely seriously. In my personal view of the relation between science
and reality, I favor Constructive Empiricism (van Fraassen 1980) over Scientific
Realism. That is, I consider the measure of good science to be empirical adequacy,
not objective truth. However, it seems to me that the task of inferring the semantic
system of a language from the observation of the communicative behavior of its
speakers must have a solution that is humanly attainable and that has an outcome
that is sufficiently shared among the members of the speech community to allow
the replication of the vast majority of this semantic system in the generations to
which it is transferred. And if children can infer the semantics of a language from
observing the behavior of competent speakers, I see no principal reason why se-
manticists should be unable to do the same, however different the tasks of the child
and the semanticist are in every other respect.
3 Understanding Elicitation
Linguistic data gathering (in the broad sense, as opposed to just elicitation) in-
volves maximally three components: a stimulus, a task, and a response. The stimu-
lus is a linguistic or non-linguistic representation intended as the input of the task
(cf. also Burton and Matthewson, this volume; Bochnak and Bogal-Allbritten, this
volume; Bar-el, this volume). In comprehension and judgment tasks, the input
(stimulus) is an utterance; in production tasks, the input (stimulus) constrains the
content of the utterance to be produced. The semantic elicitation task (as opposed
to the stimulus) is a speech act directed at the participant(s) by the researcher
intended to trigger a set of computations involving the semantic system. These
computations are intended to ultimately result in a response from which the com-
putations, the representations, and the speaker’s (or the speakers’) knowledge and
practices involved in them can then be recovered. In this way, the response per-
mits inductive generalizations and the testing of hypotheses about the semantic
system. The response is a communicative action in the broadest sense. It may be
a target language utterance, a contact language translation, a metalinguistic judg-
ment, or any non-linguistic action that solves the task, for example by pointing
out a possible referent, demonstrating an action that would instantiate a given de-
scription, and so forth. Figure 1.3 presents a cartoon version of a rather pedestrian
example: the field researcher asks a speaker of Yucatec Maya how to say “I’ve got
Stimulus: contact language utterance

Bix kuya’la’l ‘Tengo que irme’?
(’How is it said ‘I have to go’?)
Task: translation
Pa’tik im bin!
(I’ve got to go!’)
Field researcher Consultant
Response: target language utterance
Figure 1.3 Components of linguistic data gathering
to go” in their native language, formulating the question—i.e., the task—in Maya,
but the stimulus utterance in the contact language, Spanish. The speaker responds
with the idiomatic Yucatec way of saying “I’ve got to go” (as an informal way of
taking leave).
Not all of the three components are necessarily present in every study. There
is an implicational relationship here: studies that employ stimuli require tasks, and
all empirical studies of linguistic behavior examine acts of linguistic behavior—
most commonly, utterances—whether these are responses to tasks and stimuli or
not. We thus arrive at the classification in Table 1.1.
The plus and minus signs represent the presence and absence of the particular
component, respectively. For the distinction between spontaneous and “staged”
speech events, see Himmelmann (1998). I assume this distinction to be continu-
ous. No speech event recorded by an observer is 100% spontaneous or staged. The
greater the influence of the researcher, the more staged the event. Recordings of
folk tales and descriptions of cultural practices—arguably the mainstay of linguis-
tic field work in the Boasian tradition—typically exemplify the staged type, as the
speakers realize the event in response to a request by the researcher (see also Cover
and Tonhauser, this volume).
Linguistic elicitation then can be defined as the collection of responses to lin-
guistic or non-linguistic stimuli designed to study the respondents’ linguistic com-
petence and/or their practices of language use.1 This yields a very broad notion of
elicitation, going well beyond the traditional prototype of one speaker answering
a researcher’s questions and including many techniques that are widely considered
“experimental” rather than instances of elicitation. In my view, however, elicita-
tion and experimentation are not mutually exclusive. Elicitation is an approach
to data gathering. As such, it contrasts with recordings of spontaneous and staged
speech events. Experimentation, on the other hand, is broadly any empirical test
of a hypothesis and in the narrow sense involves observations under controlled
Table 1.1
The families of data gathering techniques in linguistics
Recording of spontaneous Recording of staged speech

speech events events Elicitation
Linguistic behavior (“response”) + + +

Task – + +
Stimulus – – +
Experimental methods of psycholinguistics are by this definition not elicitation techniques, since
1
the data they produce serve to test hypotheses about language processing, rather than to provide direct
evidence of the participants’ linguistic competence and practices of language use. For this reason, psy-
cholinguistic paradigms are not included in the further discussion. Nevertheless, as mentioned above,
psycholinguistic data can be exploited for indirect clues about semantic relations.
conditions. From this perspective, psycholinguistic experiments involve the elici-

tation of communicative behavior, and elicitation in turn may but need not be a
part of an experiment, depending on whether it is conducted as a test of some
hypothesis or merely for exploratory purposes.
A classification of the elicitation techniques at the linguist’s disposal can be
achieved by crosstabulating the possible stimulus and response types and identify-
ing the task types as mappings from the stimulus types into the response types, as
depicted in Table 1.2. This presupposes that the identification of stimulus (“input”)
and response (“output”) types alone is sufficient to define the task types. I can see
no obvious theoretical reason why this should be so, but it seems to work out quite
nicely.
The classification in Table 1.2 distinguishes four stimulus types. The stimu-
lus is either an utterance—in which case it may be a target or a contact language
utterance—or the content of some linguistic or non-linguistic representation. Of
course, a stimulus utterance likewise conveys a particular meaning. The difference
between an utterance used as stimulus and the content of a linguistic representa-
tion used as stimulus is that in the former case, the morphosyntactic and phono-
logical properties, the particular set of lexical items involved, and the register are
all part of the stimulus. The speakers’ response will be observed and analyzed as
a response to all of these properties. In contrast, in the case of the content serv-
ing as stimulus, everything besides the meaning of the utterance is considered
just “wrapping” and assumed as irrelevant to the speakers’ response (an assump-
tion that may of course not always be borne out). An example of the content of
Table 1.2
A classification of elicitation techniques by stimulus and response type
Metalinguistic Metalinguistic
Response Target language Contact language utterance: utterance: Non-linguistic
stimulus utterance utterance judgment description representation
Target Type I — Type II — Type V — Type VI — Type VII —

language completion; translation judgment explication by demonstration
utterance association (well- paraphrase, of referents;
formedness, scenario act-out tasks
truth, felicity)
Contact Type II — (these combinations go
language translation beyond target language
utterance elicitation)
Content of Type III —
a linguistic production in a
representation given contextual
(in the target scenario
or contact
language)
Content of a Type IV —
non-linguistic description
representation
a linguistic representation used as stimulus is a context description employed in

combination with a stimulus utterance to test whether the utterance is considered
true and pragmatically appropriate in the context.
The response in turn can be a target language utterance, a contact language ut-
terance, a metalinguistic utterance, or some non-linguistic communicative action.
Among metalinguistic responses, two types may be further distinguished: judg-
ments and descriptions. The former rank a property of the stimulus (or some part
of it), such as its acceptability, idiomaticity, unusualness, and so forth, on a scale,
whereas the latter paraphrase its meaning or describe a scenario or a setting, and
so forth, in which it might be used.
As Table 1.2 shows, target language utterances are the only stimuli that can
be used to elicit all of the five response types. All other stimulus types yield exclu-
sively one valid response type: a target language utterance. This means that every
form of (target language) linguistic elicitation involves target language utterances
as stimulus, response, or both—which makes sense. As for the empty cells, these
combinations yield responses that are not valid as data for target language studies.
For example, a non-linguistic stimulus might be used to elicit a contact language
response in a study on the participants’ second-language competence in the con-
tact language.
The classification in Table 1.2 covers not only the methods for semantic elici-
tation, but—as far as I can see—those for elicitation in any field of linguistics.
However, all of these methods can also play a role in semantic research, including
in fieldwork. For research aiming to identify possible expressions of a given mean-
ing, completion and association tasks, translation tasks, contextualized produc-
tion tasks, and description tasks are suitable. For studies targeting the meaning of a
given expression, eliciting judgments of (non-)contradiction, felicity, and so forth,
explications by paraphrase or scenario, and demonstration and act-out tasks will
be the methods of choice.
In the following, I provide illustrations of the seven types with examples
from my own field research on Yucatec Maya. As will become clear in the process,
elicitation often involves combinations of the seven types of techniques listed in
Table 1.2.
4 Type I: From Target Language Utterance to Target Language Utterance
Completion and association tasks involve both a target language stimulus and a
target language response. They are powerful tools for studying syntagmatic lexi-
cal relations such as selectional restrictions. I have employed association tasks in
several studies of selectional restrictions. One example is a study of the semantics
and argument structure of Yucatec verbs of cutting and breaking (or “separation in
material integrity” (Hale and Keyser 1987; cf. Bohnemeyer 2007; Majid et al. 2008).
My objective was to determine which verbs impose narrow selectional restrictions
on the theme or patient and which impose such restrictions on the instrument.
The hypothesis I was testing, extrapolated from Guerssel et al. (1985), was that
theme/patient-specific verbs have syntactic properties similar to those of English
break, while instrument-specific verbs have syntactic properties similar to those
of cut. According to this hypothesis, members of the break-type class, but not the
cut-type class, would produce inchoative intransitive variants that express the state
change of the theme, but omit the cause, whereas members of the cut-type class,
but not the break-type class, would produce variants that express pure activity
meanings without a state change component. In English, these patterns are instan-
tiated by the causative-inchoative alternation in the case of the break-class and the
conative alternation in the case of the cut-class.
The procedure I used was as follows: for each verb I wanted to test, I gave the
speakers I ran the study with a typical-theme prompt of the format in (7) and a
typical-instrument prompt of the format in (8) (I administered the task in Yucatec,
using Yucatec prompts):
(7) “I want you to tell me the kinds of objects that can be VERBed. If you hear
that somebody VERBed something, what kind of thing are you going to
think it is that they VERBed?”
(8) “I want you to tell me the kinds of objects that one can VERB with. If you
hear that somebody VERBed something, what kind of thing are you going
to think it is that they VERBed it with?”
I ran the task with five speakers. Tables 1.3–1.4 list the responses for hat “tear” and
xot “cut”.
Cursory inspection suggests that the typical themes of hat “tear” form a fairly
coherent set, involving objects that might be conceptualized as being made of
materials of a fibrous structure. This is not actually the case for the plastic bag;
but one can imagine that the category is extended to such objects as plastic bags
Table 1.3
Responses for hat “tear”
Responses to theme prompt (7) Responses to instrument prompt (8)
Clothes, paper, leather, a plastic bag, a letter, One’s hands, feet, mouth, a stick, a machete,
one’s hand, one’s mouth/lips and shoes knife, axe, a piece of wire, scissors
Table 1.4
Responses for xot “cut”
Responses to theme prompt (7) Responses to instrument prompt (8)
Rope, melons, squash, tomatoes, one’s hand, A handsaw, knife, machete, reaping hook,
one’s clothes, a plank or the table, or another hacksaw, axe, shards of glass, pieces torn off an
person aluminum can
because “separation in material integrity” occurs in them in a manner similar to

that typical of fibrous materials. In contrast, coherence in the responses to the
typical-instrument prompt is fairly loose. On the basis of this observation, it might
be tentatively concluded that hat “tear” is semantically theme-specific, but not
instrument-specific. Conversely, responses to the prompts for xot “cut” show co-
herence in the instrument set (all typical instruments can be applied in the manner
of bladed tools, whether or not they actually have blades), but much less so in the
theme set. So this is a verb that seems more likely to be instrument-specific rather
than theme-specific. For further results and analysis, see Bohnemeyer (2007).
Classic readings on association include Ervin and Landar (1963) and Clark (1970).
A very interesting recent application can be found in Evans and Wilkins (2000).
5 Type II: From Contact Language Utterance to Target Language Utterance

or Vice Versa and Type III: From Linguistic Representation of a Stimulus
Content to Target Language Utterance
A translation task directs a speaker to translate a stimulus utterance in the contact

language into a response in the form of a target language utterance or vice versa.
Translation tasks are potentially fraught with two problems. First, they offer
insufficient control over how the speaker construes the stimulus. For example,
the speaker and the researcher may differ in their competence in the contact lan-
guage or use different varieties of it, or they may differ in the inferences involved
in their understanding of the stimulus utterance as a result of differences in cul-
tural knowledge. The second potential concern is the risk of interference effects:
when a speaker has a choice between two or more translations, all of which are
well-formed in the target language and roughly express the intended meaning,
their choice may be influenced by a desire to mimic structural properties of the
stimulus. For example, in a language without definite articles, a speaker might
be tempted to translate a definite article in the stimulus using a demonstrative—
especially if their own imperfect understanding of the function of definite articles
in the contact language treats them as equivalents of text-deictic uses of demon-
stratives in their native language.
Both of these pitfalls can to some extent be checked by providing the stimuli
with contexts that restrict their interpretation, thereby combining Types II and
III, since the context is a stimulus in its own right and a Type-III stimulus at that,
the content of a linguistic representation serving as a stimulus. The most widely
known and successful example of this hybrid approach is the Tense-Mood-Aspect
Questionnaire of Dahl (1985).
In Dahl’s questionnaire, the translation stimuli are utterances that express
event descriptions from a certain temporal, aspectual, and modal perspective.
To avoid interference from the contact language, expressions of tense, aspect,
and mood are omitted from the stimuli and finite verb forms are replaced with
non-finite ones (set in capital letters to flag them) wherever possible.2 Instead, the
intended perspective is controlled by a context that defines a reference time (or
“topic time” in the sense of Klein 1994) for the translation stimulus. This is illus-
trated in (9). The context precedes the translation stimulus and is set in brackets.
(9) TMA Questionnaire item A1:
[Q: What your brother DO when we arrive, do you think?
(= What activity will he be engaged in?)]
He WRITE letters.
The translation target in (9) is simply a description of somebody (a male refer-
ent) writing multiple letters. The translation stimuli are designed so as to cover
all major lexical-aspectual classes that have been identified across languages in
previous research.3 The context defines a topic time for this description that lies
in the future of the utterance time of the stimulus and is included in the runtime
of the letter-writing event. In English, the future progressive is the canonical way
of expressing this perspective. Example (10) is a Yucatec response to the stimulus
in (9). In my experience, the best way to ensure that the speaker takes the context
fully into account during the translation is to ask the speaker to translate the con-
text as well.
(10) Q: Ba’x a=tukul-ik

what(B3SG) A2=think-INC(B3SG)
“What do you think”
k-u=beet-ik a=suku’n chéen k’uch-uk-o’n?
IMPF-A3=do-INC(B3SG) A2=elder.brother SR.IRR arrive-SUBJ-B1PL
“will your big brother be doing (lit. is he doing) when we arrive?”
2
These infinitives tend to be confusing, however, when the task is administered purely orally, for
example when working with speakers not accustomed to reading.
3
A fundamental problem for any research that starts from a set of semantic or notional cat-
egories and asks how these are expressed in a given language is what is known in the typological
and ethnosemantic literature as the ‘etic grid’ problem: the set of semantic or notional categories
that the study is designed to test—the study’s etic grid—biases the possible observations of semantic
categories in the target language. Nowhere has this issue been discussed more prominently—or more
pointedly—than in the critique advanced by Lucy (1997), Saunders & van Brakel (1997), and others of
the methodology of Berlin & Kay’s (1969) seminal study of the semantics of color terminologies (see
Berlin & Kay (1997) and Kay (2006) for the authors’ response). Studies that aim to examine the expres-
sion of a given semantic domain in particular languages should base their etic grid on a careful review
of the distinctions reported in the available crosslinguistic literature; should always be mindful of the
fact that the sets of categories they encounter in grid-based elicitation are partly a function of the grid
they employ; should strive to compare the results of the grid-based elicitation to data obtained from
other (especially non-elicited) sources (see Bohnemeyer 2012 for an illustration); and should present
their findings as an initial step in a series of studies, to be followed up by further research based on a
revised grid that takes into account any shortcomings of the initial grid that emerged during the first
round or from the comparison with other sources of evidence. See also AnderBois & Henderson, this
volume.
A: Chéen k’uch-uk-o’n wal=e’,

SR.IRR arrive-SUBJ-B1PL UNCERT=D3
“When we arrive, I guess”
ts’íib-t-ah+kàartah k-u=meet-ik wal=e’.
write-APP-ATP+letter(B3SG) IMPF-A3=do-INC(B3SG) UNCERT=D3
“letter writing is what he will be (lit. is) doing, I guess.”
The response features the use of the irrealis marker kéen/chéen, which is restricted
to subordinate clauses and governs subjunctive mood; the imperfective aspect
marker in the matrix clauses; a predicate focus construction; and the epistemic
uncertainty particle wal. None of these expressions encodes relative or absolute
future tense.
6 Type IV: From Non-Linguistic Representation to Target Language Utterance
The elicitation of descriptions of non-linguistic stimuli has become the method

of choice in semantic typology since the landmark study by Berlin and Kay 1969
(with much earlier precursors such as Magnus (1877, 1880) in research on the lin-
guistic categorization of color and Chamberlain 1903 and Myers 1904 on that of
tastes). It also plays a prominent role in speech production research and language
acquisition research. Moreover, non-linguistic stimuli are not only used in pro-
duction tasks, but also in various types of comprehension tasks and in so-called
“referential communication” tasks (see below), which combine production and
comprehension. Either way, the principal function of non-linguistic stimuli is to
constrain the referential content of a target language utterance—the response in
the case of production tasks and a second stimulus, a stimulus utterance, in the
case of comprehension tasks. One important caveat for production tasks with
non-linguistic stimuli is that constrain does not mean the stimulus fully deter-
mines the meaning of the response. The meaning of the speaker’s response will
depend above all on their interpretation of both the stimulus and the task, or their
interpretation of the researcher’s intention behind both. Let me illustrate the role
of the task first. Consider Figure 1.4, which shows the first item in the Topological
Relations Picture Series, a.k.a. BowPed. BowPed consists of 71 line drawings fea-
turing spatial configurations. Most of these involve “topological” relations in the
sense of Piaget and Inhelder (1956), (i.e., relations that can be adequately described
without selection of a perspective or reference frame). In each picture, one or more
objects are designated as “figures” (Talmy 2000) or themes of locative descriptions
by arrows pointing to them. The participants’ task is to use the information in the
picture to answer the question “Where is the [figure]?,” asked preferably in the
target language. This question serves as secondary stimulus, thus making BowPed
strictly speaking a combination of Types III and IV.
Figure 1.4 Item #1 of the “Topological Relations Picture Series” aka BowPed (©Eric Pederson;
reproduced with permission)
However, when I ran BowPed with Yucatec speakers, I had an experience I

have heard several other researchers report: in response to the question “Where
is the cup?,” a speaker would look at me with mild puzzlement and point to the
picture: “Uh, right here?” What this response brought home is that I had not been
specific enough about the task. To fix this problem, I constructed the following
scenario, which I asked the speaker to assume as an elicitation frame, that is, a
more elaborate context within which to respond, a context that put a certain inter-
pretation on the Where-question:
(10) “Imagine you are talking to somebody who is looking for the [figure]. This
person knows where the [ground] is, but does not know where the [figure] is.
You know where the [figure] is; but neither of you can see the [figure] and the
[ground] right now. The person asks you “Where is the [figure]?” Imagine
you want to tell the person where the [figure] is. How do you respond?”
I would repeat this frame with every new picture until I got the impression that the
speaker understood and remembered the point.
In comprehension tasks, the visual stimulus is presented along with a target
language utterance—a typical example of a hybrid technique. For instance, in verifi-
cation tasks, the speaker is to determine whether the utterance can serve as a descrip-
tion of the visual stimulus (or, more generally, whether it instantiates its extension—a
type of judgment elicitation). In matching tasks, another subtype of comprehension
tasks, the speaker is asked to select from among two or more visual representations
the one that is best (most accurately, etc.) described by the utterance, or select among
two or more utterances the one that best describes a given visual representation.
Verification and matching tasks thus combine elements of Type IV and Type V. Ref-
erential communication tasks are a combination of production and comprehension
distributed across two participants (cf. Clark and Wilkes-Gibbs 1990). They involve
at least two speakers per trial: one describes the content of a stimulus and the other
rematches the description to a set of non-linguistic stimuli. There are numerous pos-
sible realizations of this, including picture to picture, picture to toy model, and so
forth. Figure 1.5 illustrates the setup of a picture-to-picture matching task.
Figure 1.5 Setup of the Ball & Chair picture matching task
The “Ball & Chair” referential communication task (Bohnemeyer 2011) was
developed to replace and improve upon a similar task, “Men & Tree,” designed at
the Max Planck Institute for Psycholinguistics in the 1990s (Pederson et al. 1998).
The goal of both tasks is to assess the participants’ use of spatial frames of reference
in discourses referring to small-scale space. To this end, two speakers sitting side
by side are asked to match identical sets of photographs placed in front of them in
different orders, while a screen between them prevents them from sharing a visual
field (I used the suitcase in which I had hauled my field equipment). The screen
forces the participants to produce maximally explicit descriptions in order to solve
Figure 1.6 Ball & Chair picture 2.5

the task. The photos all show a ball and a chair. There are four sets, each compris-
ing 12 pictures, which differ from one another in the orientation of the chair and
the location of the ball vis-à-vis the chair. Example (11) reproduces a description of
one picture, shown in Figure 1.6, in full.
(11) a. Estée, u séegere-e-e . . . chan fòotoa’,

esté u=séegir le=chan fòoto=a’,
HESIT A3=follow DET=DIM photo=D1
“Uh, this next-uh-little photo,”
b. u frèente e sìiyao’, tu tohile don Jorgeo’,

u=frèente le=sìiya=o’ tu=tohil le=don Jorge=o’
A3=front DET=chair=D2 PREP:A3=straight:REL DET=don Jorge=D2
“the front of the chair, in the line of that don Jorge (i.e., JB),”
c. ti’ yàani’. Tu’x ku nakta’ máako’,

ti’=yàan=i’ tu’x k-u=nak-tal máak=o’
PREP=EXIST(B3SG)=D4 where IMPF-A3=lean-INCH.DIS person=D2
“there it is. The back rest (lit. where a person leans (against)),”
d. estée, ta frèente súutu’.

estée ta=frèente súut-ul
HESIT PREP:A2=front turn\MIDDLE-INC(B3SG)
“uh, it’s turned (toward) your front.”
e. Ta xno’hk’abile’
ta=x-no’h+k’ab-il=e’
PREP:A2=F-right+hand-REL=TOP
“On your right,”
f. ti’ yàan ump’ée bòolai’.

ti’=yàan hun-p’éel bòola=i’
PREP=EXIST(B3SG) one-CL.IN ball=D4
“there is a ball.”
g. Ta xts’íi - ta xts’íihk’abil [unintel.], ti’ yàan

ta=x-ts’íik+k’ab-il=e’ ti’=yàan
PREP:A2=F-left+hand-REL=TOP PREP=EXIST(B3SG)
ump’éel bòolai’,
hun-p’éel bòola=i’
one-CL.INball=D4
“On your le—on your left [unintelligible], there is a ball,”
h. kàasi tu tohil u yòok yàan ti’.

kàasi tu=tohil uy=òok yàan ti’=i’
almost PREP:A3=straight:REL A3=leg/foot EXIST(B3SG) PREP(B3)=D4
“it’s almost in the line of its leg with respect to it.”
Line g is a correction of line e. The individual propositions of this description can

be analyzed under the assumption that they are true of the described stimulus
item, that is, the picture in Figure 1.5. X-ts’íik “left” in line g could be ambiguous
with respect to Figure 1.5, permitting both a relative interpretation projected from
the body of the addressee and an intrinsic one projected from the chair itself as
reference entity or “ground” of the locative description. However, the morphologi-
cally bound second-person possessor pronoun rules the second interpretation out,
making it clear that line g involves a relative frame of reference.
In discussions of the topic of using referential communication designs in field
research, concerns about ecological validity are regularly voiced. There are two
aspects to this problem: the artificiality or unfamiliarity of the stimulus and that of
the task. The former problem pertains to any research with non-linguistic stimuli
that are alien to the culture of the speech community, whereas the latter is more
or less a unique property of referential communication tasks. Let me address the
more specific issue first.
Members of traditional cultures may not be very accustomed to using speech
in contexts where gaze and gesture cannot serve to disambiguate referents. More
importantly, few members of any speech community are accustomed to commu-
nicating detailed spatial information in such contexts. Except for visually impaired
speakers and highly technical genres of communication, the conveyance of rich
small-scale spatial information naturally relies heavily on gaze and gesture. What
this means is that responses to a task such as Ball & Chair can tell us something
about the cognitive and communicative resources that the members of a given
community tap into when faced with an unfamiliar task of certain specifications,
but they do not permit a direct assessment of the actual practices of language use
in the community. To make this more concrete: the description in (11) involves
three spatial reference frames, a frame anchored to my body standing near the
camera that recorded Figure 1.4 (“in the line of that don Jorge”) and two frames
anchored to the body of the addressee or a generic observer, one that does not
involve projection of the body’s axes onto the chair (“turned toward your front”)
and one that does (“on your left”, i.e., on the observer’s left of the chair). What this
shows is that this particular speaker is capable of using these kinds of frames in
reference to small-scale space. Furthermore, if the description results in a success-
ful match, this suggests that the addressee is capable of using the same frames in
the comprehension of the speaker’s descriptions. Next, by analyzing the total set
of descriptions by a particular speaker, we can assess that speaker’s preferences
among the strategies available to them for solving this artificial task. By comparing
preferences across participants, one can assess the preferences of a generic or aver-
age or typical Maya speaker in rural central Quintana Roo. This in turn permits
comparisons across speech communities both among speakers of the same lan-
guage and among speakers of different languages. It allows us to conclude, for
example, that rural Yucatec speakers make more frequent use of relative frames
in solving this artificial task than speakers of many other Mesoamerican lan-
guages, but do so much less frequently than speakers of European languages (see
the descriptions in O’Meara and Pérez Báez (eds.) 2011, including Bohnemeyer
2011). This is an interesting and important finding: it suggests that relative refer-
ence frames cannot play the same role in reference to small-scale space that they
play in Euro-American speech communities (and, e.g., among Japanese speakers
(Kita 2006)), where they are the default for this domain. But it does not tell us
much about what Yucatec speakers habitually do to communicate about space.
Assuming that natural referential practice relies heavily on gaze and gesture, it
is a foregone conclusion that designs such as Ball & Chair necessarily produce
rather distorted representations of it. The standard response in the semantic typol-
ogy community to this problem has long been that elicitation results—especially,
but not restricted to, results obtained with referential communication designs—
should always be complemented by other sources of evidence, both elicited—for
example, in the case of spatial reference, route descriptions—and non-elicited, that
is, staged discourses (firsthand witness accounts of natural disasters and local his-
tory narratives may prove useful in spatial studies) and the observation of sponta-
neously occurring interactions. More on this below.
When it comes to elicitation stimuli, considerations of ecological validity
must take into account three factors:
¤ Are there conventional descriptors for the stimuli in the target language?
¤ Are the stimuli culturally appropriate?
¤ How do speakers of the target language interpret the stimuli?
The first issue is usually the most trivial in my experience, as it tends to be confined
to lexical expressions. At the lexical level, the problem of missing descriptors is
readily addressed by the researcher negotiating with the speakers either the use
of a contact language loan or a reinterpretation of the stimulus item in question
that makes it describable in the target language. For example, if a stimulus pic-
ture or video shows a plant or animal of a species that does not occur in the local
environment, it may be possible to ask the speakers to treat it as an instance of a
similar plant or animal that does occur. Lexicalization gaps can create problems
that spill over into the grammar in cases in which the grammatical classification
of lexical items is concerned, for example when noun class or aktionsart markers
are involved.
As for the second issue, both linguistic and non-linguistic stimuli can be
offensive to members of particular cultures for a variety of reasons: exposure of
body parts that is considered indecent, characters hunting animals or eating foods
considered taboo, and so forth. There is no other solution to this type of problem
than to avoid it during the design of the stimuli.
Lastly, the interpretation of visual stimuli is subject to non-trivial cultural
conventions. Consider Figure 1.7, a line drawing created by David Wilkins as
part of a series of stimuli designed for the elicitation of expressions of manner of
motion by children learning the Pama-Nyungan language Arrernte spoken in and
around Alice Springs in central Australia (see Wilkins 1997). The intended inter-
pretation of the picture was that of a horse in full gallop. However, the Arrernte
children instead understood it as showing a dead horse lying in the dirt. What was
intended to be seen as clouds of dust thrown up into the air by the horse’s legs was
instead understood as the traces the onset of rigor mortis had left behind in the
sand. These different interpretations are the result of different cultural conventions
for visual representations: whereas the default perspective for such representations
in Asian and European cultures is horizontal, it is the bird’s-eye view in Aboriginal
cultures. The different conventions in turn may be linked to the most widespread
traditional media for visual representations in each culture: paper and canvas in
Eurasian cultures vs. campground dirt in cultures of Aboriginal Australia.4 (Now,
of course, all of these materials are increasingly being replaced by digital media,
with the inevitable result of a globalization of the horizontal perspective.)
The dependence of stimuli on culture-specific interpretations only increases
with the semiotic complexity of the stimuli. Consider, for example, the representa-
tion of events by single snapshot images vs. cartoon-strip sequences vs. video clips.
This, too, is subject to changing cultural conventions—e.g., medieval and non-
western artists often represent temporal as spatial relations, as in the case of the
Figure 1.7 Galloping horse or dead horse? (Wilkins 1997: 157; ©David P. Wilkins; reproduced
with permission)
4
Sand drawings on the campground are of course exclusively viewed from above—hence the
naturalness of the bird’s-eye perspective.
Bayeux Tapestry, which shows the events of the Norman conquest of England as
if they all happened simultaneously but in adjacent places. In contrast, contempo-
rary Western imagery strictly follows a convention according to which everything
that is represented within the same drawing is understood to (have) happen(ed)
simultaneously. Consequently, representing a sequence of events requires a se-
quence of images, for example in separate panels, as in a comic strip.
Guarding against the effects of culture-specific interpretations of visual stim-
uli is but one example of a much more general principle: a stimulus only impacts
the response via the speaker’s interpretation of it. This is captured by the Golden
Rule of Elicitation proposed below.
Another often-commented-on limitation of non-linguistic stimuli is their
restriction to perceivable and thus concrete information. This limitation can be
overcome by combining non-linguistic and linguistic stimuli or through complex
task designs. An example is the TEMPEST design for the elicitation of temporal
relations described in Bohnemeyer (1998a, b; 2000), a referential communication
task in which speakers match videos that show the same events in contrasting
orders. Cf. also Burton and Matthewson (this volume) and Bar-el (this volume).
7 Type V: From Target Language Utterance to Judgment
Judgments are metalinguistic utterances that may comment on a variety of prop-

erties of linguistic stimuli: their grammaticality, interpretability, idiomaticity, ste-
reotypicality, pragmatic appropriateness (which covers a large variety of different
properties; see above), and—arguably most importantly for the purposes of em-
pirical semantics—whether particular individuals or states of affairs are elements
of their extension. These metalinguistic utterances are typically prompted by ques-
tions or requests. The cognitive basis of such judgments is not entirely understood,
but is presumably related to the speaker’s ability to detect ill-formed, semantically
false, or contextually inappropriate (constituents of) utterances both in processing
the speech of their interlocutors and in their own production.
As pointed out by Matthewson (2004) and Tonhauser et al (2013), it makes
little sense to ask a linguistically untrained speaker whether an utterance has a
given entailment, carries a certain presupposition, and so forth. This is trivially
true for the simple reason that untrained speakers will not have the relevant tech-
nical notions of “entailment,” “presupposition,” and so on. This means that it is
up to the skill of the researcher to construct an elicitation stimulus and task that
allow the speaker to express a judgment from which the researcher can then infer
whether or not the utterance has the relevant property in the speaker’s judgment.
The successful elicitation design must circumvent the problematic technical no-
tions by instead tapping into the definition of the relevant semantic property.
For illustration, consider entailment. An utterance entails another if any pos-
sible world that makes the former true also makes the latter true. To test whether
this is the case for a given pair of utterances, the researcher may ask the speaker
whether they can think of a situation in which the first utterance is true but the
second is not. This question is likely still too complex and abstract for most un-
trained speakers to readily answer it on first trial. To simplify matters further,
the researcher can construct a few scenarios on a trial basis and ask the speaker
whether both utterances are true in them. This of course means that the researcher
is not asking for a direct judgment of entailment, but rather for a series of judg-
ments about the truth of a pair of utterances in a series of scenarios. Another
important avenue for eliciting entailment data are judgments of contradiction.
Speakers appear to be able to tell relatively immediately whether two statements
are logically consistent or not. Consequently, one method for testing whether an
utterance has a given entailment is by combining it with a second utterance, which
negates the hypothetical entailment. If in the speaker’s judgment the conjunc-
tion of the two utterances may be true in the same scenario, this suggests that the
proposition negated by the second utterance is not an entailment of the first. But if
the speaker judges the utterances to be inconsistent, this supports the entailment
analysis.
Judgments are almost always of a graded nature. That is, even if a speaker
gives a categorical response to a simple polar question, this response can be ranked
in relative strength with respect to the same speaker’s responses to other stimuli.
There are a number of principal obstacles that may beset the elicitation of
judgments:
¤ Judgments may not reflect the speaker’s own production well.
¤ Judgments may be influenced by normative beliefs.
¤ The same stimulus utterance may be judged differently by the same
speaker in different contexts.
¤ A speaker’s judgment in response to a particular utterance will depend
on which aspect of the utterance the speaker understands they are
asked to judge, or in other words, which type of judgment they are
asked to make—a judgment of well-formedness, idiomaticity, and so
forth. However, linguistically untrained speakers may not find it easy to
distinguish between these different types of properties and judgments.
¤ Similarly, a linguistically untrained speaker cannot always be expected to
be able to locate the source of a violation of well-formedness, idiomaticity,
interpretability, or the like. In general, a speaker can tell that an utterance
“sounds funny” (in a given context), and may even associate the anomaly
with non-native speakers of a particular background. But they are less
likely to be clear on why the utterance “sounds funny.”
I have encountered the phenomenon alluded to in the first point on numerous

occasions: speakers will reject a certain construction or the use of a certain term
in reference to a particular state of affairs and later produce that very construction
during a different task or use that very term in reference to the state of affairs.
There can be a variety of reasons for why a decontextualized expression appears to

us differently than when we come upon a context in which that same expression is
used by others or in which we might use the expression ourselves. Moreover, judg-
ments are always susceptible to normative beliefs—the second point above—and
such beliefs may cause speakers to reject particular expressions even though they
themselves use them. Such beliefs may not always be the result of standardization,
but can also be influenced by folk theories of language use. As an example, my
work on spatial reference frames mentioned in the previous section has taught me
that many speakers of Yucatec and other Mayan languages operate on a belief that
tasks such as the one shown in Figure 1.4 have correct and incorrect solutions and
that the correct ones employ cardinal direction terms. The origin of this belief is
at present unclear to me.
An instance of the context-dependence of judgments that many linguists are
familiar with from their own practice and that is also well documented in the psy-
cholinguistic literature is that of satiation: the phenomenon that the acceptability
of utterances that appear initially anomalous sometimes seems to improve with
time as the same utterance is repeated again and again (Snyder 2000; Hiramatsu
2000; Goodall 2005; Francom 2009; inter alia). Some types of anomaly are known
to satiate much more easily than others. Why this is the case is unknown, and the
causes of satiation itself are poorly understood.
The ability to distinguish between different types of anomaly—as induced by
syntactic vs. semantic vs. pragmatic clashes, and so forth.—depends on a con-
sultant’s declarative, metalinguistic understanding of linguistic phenomena and
thus grows with the consultant’s experience and training (see also Cover and Ton-
hauser, this volume). An independent potential challenge may be the terminology
available to the researcher and the native speaker consultants—in either the target
language or a contact language—to distinguish the relevant sources. My prefer-
ence has always been to ask general acceptability questions and try to construct the
stimuli so as to minimize the risk of ambiguity in the speaker’s response. Research-
ers should of course always aim to make sure that their stimuli do not feature any
anomalies other than the one to be tested. But they cannot possibly always succeed
at this unless they are omniscient about the target language except perhaps for the
anomaly under investigation. Typical query formats I would use are listed in (12):
(12) a. “What about this one, how does it sound to you: [stimulus utterance]”
b. “[stimulus utterance] Is it said well like that?”
c. “[stimulus utterance] Is it possible to be said like that?”
d. “[stimulus utterance] Are there people, you think, who talk (lit. say it)
like this?”
e. “In the photo/picture/video here, can it be said that [stimulus
utterance]?”
f. “In the photo/picture/video here, if a person says that [stimulus
utterance], would that be true?”
g. “Let’s say [verbal description of scenario]. In that case, can it be said that
[stimulus utterance]?”
h. “Let’s say [verbal description of scenario]. In that case, if a person says
that [stimulus utterance], would that be (lit. is that) true?”
The templates in (12a–d) might be used to test the well-formedness of an ut-

terance as per its morphosyntactic and morphophonological structure and the
selectional restrictions of its lexical items. In contrast, (12e–h) can be used to test
whether a given description is accurate and pragmatically appropriate in reference
to a particular nonverbal stimulus (12e–f) or a verbally described scenario (12g–h).
Both of these options are illustrated below.
To test whether a particular entity or state of affairs falls into the semantic
extension of a given descriptor, the researcher can ask speakers, in the simplest
case, “Can X be called Y?,” where X is a verbal or nonverbal representation of the
referent or simply an instance of it and Y is the descriptor to be tested. However,
unless the descriptor is a non-relational common noun—and often even then—it
is generally preferable to insert it into a declarative sentence and study its impact
on the truth conditions of assertions of such a sentence. This can be understood
as an application of the Context Principle often attributed to Frege 1884 and Witt-
genstein 1921.
The referent X to be tested for inclusion in the extension of the expression
can be an actual instance of a particular kind of entity or state of affairs or a repre-
sentation of it. The representation in turn can be linguistic or non-linguistic. Let
me illustrate elicitation of truth judgments against both verbally and nonverbally
represented scenarios, beginning with the former type. My example for this comes
from testing Yucatec verb phrases for telicity. Telicity has no syntactic reflexes in
this language (Bohnemeyer 2002: 172–192). There is, for example, no distinction
between duration (i.e., for-type) and time-span (i.e., in-type) adverbials. “Spend
X time VERBing” and “take X time to VERB” are expressed the same way. The
aspectual verbs translating “finish”/”complete” are compatible with telic and atelic
verb phrases alike.
The only way to test for telicity is by tapping into the entailment patterns col-
lectively known as the “imperfective paradox” (Dowty 1979; cf. also Bohnemeyer
and Swift 2004), which are used in Vendler 1957 to distinguish accomplishments
from activities and thus effectively to define telicity (even though Vendler did
not use that notion). As illustrated in (13), activity descriptions in the progressive
entail their simple-tense counterparts (13a), whereas the same does not hold for
accomplishment descriptions in the progressive (13b):
(13) a. Floyd was pushing a cart

∴ Floyd pushed a cart.
b. Floyd was drawing a circle.
not ∴ Floyd drew a circle.
To study the behavior of particular Yucatec verbs vis-à-vis these entailment pat-
terns, I consulted with five native speaker consultants to find scenarios in which the
event described by the verb phrase is plausibly interrupted at a time at which the
VP marked for progressive aspect applies. I then asked whether a perfective or per-
fect form of the same VP can be truthfully asserted at the time of the interruption.
(14) Pedro=e’ táan u=k’àay,

Pedro=TOP PROG A3=sing\ATP
“Pedro, he was singing,”
káa=t-u=k’at-ah u=báah Pablo.

CON=PRV-A3=cross-CMP(B3SG) A3=self Pablo
“(when/and then) Pablo interfered.”
Pedro=e’ t-u=p’at-ah u=k’àay.

Pedro=TOP PRV-A3=leave-CMP(B3SG) A3=sing\ATP
“Pedro,” he stopped singing.”
Be’òora=a’ ts’o’k=wáah u=k’àay Pedro?

now=D2 TERM=ALT A3=sing\ATP Pedro
“Now, has Pedro sung?”
The researcher should be prepared for surprises: for example, most consul-
tants answer negatively in response to (14) since kàay “sing,” the antipassive stem
of the transitive root k’ay “sing,” is normally interpreted as “sing a song” (cf. Bohne-
meyer 2002:172–199 for details).
If possible, a visual stimulus should be used to clarify the scenario against
which one wishes to test entailments. This is the verification method mentioned
above. As an example, in a study reported in Bohnemeyer 2010, I examined
whether Yucatec verbs of “inherently directed motion” (Levin 1993) entail transla-
tional motion of the figure or merely change of location, as described by Kita 1999
for Japanese hairu “enter” and deru “exit.” To test this, I employed the Motion verb
(MoVerbs; Levinson 2001). MoVerbs comprises 96 computer-animated video clips
featuring a variety of location change scenarios varied according to the spatial
relation between the “figure” or theme and some reference entity or “ground” in
the source or target state or in between, the involvement of figure motion, and the
perspective (toward/away from observer vs. lateral to the observer’s viewing axis).
I would, for example, test whether Yucatec speakers find (15) acceptable in refer-
ence to the clip whose first and last frame are depicted in Figure 1.8, in which a
plank slides under a ball and cylinder:
(15) H-na’k le=chan kanìika y=óok’ol le=tàabla=o’

PRV-ascend(B3SG) DET=DIM marble A3=on DET=plank=D2
“The marble, it went up the plank”
Figure 1.8 First and last frame of “FIGURE_GROUND 14” (Levinson 2001; ©Stephen C.
Levinson; reproduced with permission)
Out of context, Yucatec speakers will reject this description in reference to the sce-
nario shown in the clip. However, as discussed in the next section, this is not be-
cause the scene violates an entailment of (15), but rather because it is incompatible
with a stereotypical interpretation of it, which a hearer will assume by implicature
unless it is canceled or blocked.
8 Type VI: From Target Language Utterance to Linguistic Representation
A very powerful strategy for elucidating the meanings of linguistic expressions

involves a reversal of sorts of the method described in the previous section: ask
speakers to come up with and describe to you a scenario in which a given utter-
ance might be used to make a truthful and pragmatically appropriate statement.
For example, I asked speakers to modify (15) above to turn it into an acceptable
description of the scenario depicted in Figure 1.8. One response to this procedure
is shown in (16):
(16) Le=chan tàabla=o’ h=péek-nah-ih, káa=h-na’k

DET=DIM plank=D2 PRV=move-CMP-B3SG CON=PRV-ascend(B3SG)
le=chan kanìika y=éetel che’ te’l y=óokol=o’.

DET=DIM marble A.3=with wood there A3=on=D2
“The little plank, it moved, and the little marble and the tree ascended there
on top.”
As soon as it is made explicit that it was the ground—the plank—that moved,

this speaker has no problem with asserting the location change description of the
stationary figure. This strongly suggests that the inference to figure motion trig-
gered by (15), which clashes with the scenario shown in Figure 1.8, is merely an
implicature, not an entailment. The most likely type of implicature is a stereotype
implicature—a generalized conversational implicature licensed by Grice’s second
Quantity maxim: for Yucatec speakers, just like for English speakers, the stereo-
typical way for someone or something to change location is for them/it to move.
“Reverse-engineering” scenarios or contexts in which a given expression
might be used can provide powerful insights into the semantics and pragmatics
of the expression. However, not every speaker will find the task of coming up with
such an instantiation equally easy to solve. It requires imagination, a gift appar-
ently not evenly distributed among people. In my experience, of all the skills that
may qualify a good native speaker consultant, the ability to envision scenarios and
contexts is the rarest and most precious for the purposes of empirical semantic
research.
9 Type VII: From Target Language Utterance to Non-Linguistic

Representation
The final type of elicitation task has speakers produce a non-linguistic representa-
tion of the meaning of a target language expression. Demonstration seems to me an
appropriate general label for this type of task. A special subtype of demonstrations
are act-out tasks, in which a speaker instantiates a described action or event liter-
ally or by playacting it (or through a combination of both).
The example I would like to offer to illustrate demonstrations as a type comes
from a study I did a few years ago on the semantics of “dispositional” roots. Such
roots lexicalize non-inherent spatial properties such as postures. Mayan languages
have hundreds of roots of this kind, and the majority of these select for inanimate
figures. For this reason, I prefer “dispositional” to the traditional Mayanist term
“positional,” which suggests postures. In Yucatec and many other Mayan languages,
dispositionals represent a root class sui generis with unique privileges of producing
stems of various lexical categories, among which verbs do not necessarily stand out.
Yucatec dispositional roots produce transitive and intransitive verb stems, derived
stative predicates, numeral classifiers, and more, depending on the derivational
morphology used (Bohnemeyer and Brown 2007). Distinctions that enter the con-
ceptualization of dispositions include support, suspension, blockage of motion,
orientation (mainly in the gravitational field), shape, and configuration of parts
of the figure with respect to one another. Location is not a dispositional concept;
rather, dispositions can be thought of as “manners of location” (Belloro et al. 2008).
The greatest challenge in analyzing dispositional semantics is that the dimen-
sions of contrast are poorly understood, since dispositions are not lexicalized in
Indo-European languages at the level of specificity at which they are lexicalized in
Mayan languages. To overcome this challenge, I applied a two-step process inspired
by Berlin’s 1968 classic study of Tseltal numeral classifiers. In a first step, I elicited
typical themes or figures for each previously identified dispositional root with
six speakers, applying an association task very similar to the typical-theme and
typical-instrument prompts described above. I then consolidated the responses by
identifying the 20 most frequently mentioned types of themes—humans; various

species of animals, including horses, dogs, birds, and snakes; ropes; clothes and
pieces of fabric; and so forth—and the total set of roots in association with which
each type of figure had been mentioned by at least one speaker. Then, in a second
elicitation phase, I asked the same six speakers to demonstrate all the dispositions
associated with a given type of figure contrastively, by showing me how it would
have to be manipulated to get it from an instantiation of the last demonstrated
disposition to one of the disposition described by the root I was prompting the
speaker with now. For some of the figure types, actual exemplars were used; others
were represented by toys. I videotaped these sessions, resulting in a total of about
26 hours of videotape. Since then, several students in the University at Buffalo
Semantic Typology Lab have been working on the coding of these video files, at-
tempting to identify the properties shared across demonstrations of the same root-
figure pair by different speakers and those that distinguish dispositions expressed
by different roots. Figure 1.9 illustrates four types of suspension configurations
described by different roots. This work is generating hypotheses regarding the se-
mantics of the roots, which are being tested in follow-up fieldwork.
The Dialectical Pivot: Empirical and Hermeneutic Approaches Revisited
This chapter started from the premise that semantic research within the social and
behavioral sciences must be an empirical endeavor based on the observation of
Figure 1.9 Suspension dispositions described by (clockwise from top left) choh, ch’uy, lech, and t’oy
the communicative behavior of interlocutors. Yet, somewhat paradoxically, having

taken the reader several steps along the way toward an answer to the question how
empirical semantic research is possible, I am now about to argue that a mature em-
pirical semantics must in fact avail itself of the techniques of hermeneutic analysis
in order to achieve its goal.
We have seen above that a speaker’s response to an elicitation stimulus de-
pends on the speaker’s interpretation of the stimulus. This of course holds not
just for linguistic elicitation, but for any type of empirical research with human
or animal participants. It is a valid question—and one routinely asked—in the
analysis of experimental results in psychology or interview responses in sociology
and political science how the presentation of the task and/or the stimuli may have
influenced the observed responses. Think, for example, of how easily the findings
of a marketing research study or an opinion poll can be influenced by the way the
questions are asked, the orders in which they are asked, and of course the sampling
procedures used to recruit participants. These are all questions that go to the valid-
ity of study designs in empirical research. But linguistic elicitation adds a potential
further layer to this problem complex. In any quantitative research design, the goal
is to determine whether there are significant correlations between predictor and
response variables. As long as the research design is valid, any such correlation is
a reportable outcome, and so is the absence of a predicted correlation. Linguistic
elicitation, however, including semantic elicitation, may produce data for quanti-
tative or qualitative analysis or both. All of the studies discussed above produced
primarily data for qualitative semantic analyses, meaning analyses that draw direct
conclusions concerning the meanings of particular utterances and expressions. In
such analyses, the speaker’s interpretation of the stimuli and task and the intended
interpretation of their response—and the researcher’s assumptions about all of
these—are not merely a validity concern, but have direct bearing on the content of
the analyses. By way of illustration, the Yucatec speakers who rejected (15) as a de-
scription of the animation in Figure 1.8 did not apparently intend this judgment to
be understood to the effect that (15) would be false as a representation of the scene,
but rather to the effect that it would be misleading. This, however, did not become
apparent until I asked them to think about how the utterance might be amended in
order to make it acceptable in reference to Figure 1.8 with the result shown in (16).
The Golden Rule of elicitation states that an elicitation response only becomes
a data point in formulating generalizations about the linguistic competence and
practices of language use of the members of a speech community once the speak-
ers’ interpretation of the task and stimulus and the intended interpretation of the
response have been ascertained. A “raw” elicitation response does not document
much of anything about the speaker’s knowledge except for the fact that they are able
to produce it, which does not even tell us whether the responses are well-formed.5
5
A similar view is stated in Matthewson (2004).
10 Summary
Researchers who study semantics in the field working with speakers of understud-
ied languages or in the lab working with small children have to proceed without
being able to rely on their own native speaker intuitions or on those of expert
speakers with linguistic training. This chapter has argued that this is not only
possible, but that in fact all semantic research conceived of as part of the social
and behavioral sciences should not content itself with the researcher’s own native
speaker intuitions as the sole source of evidence. Such introspective approaches
presuppose a hermeneutic view of semantic research with interpretation as the
fundamental source of evidence. In contrast, the present chapter has advocated for
an empirical semantics based on the observation of communicative behavior as it
reveals the referential extension of linguistic expressions, their selectional restric-
tions, the structure of their sense spectra, the pragmatic conditions of using them,
and their processing properties. The empirical semanticist infers these properties
from observations of how competent speakers use the expressions under study,
not unlike a child acquiring the semantic systems of the languages she is exposed
to by observing competent speakers in the act of using them.6
References
Belloro, Valeria, Jürgen Bohnemeyer, Dedre Gentner, and Kathleen Braun. 2008. Thinking-
for-Speaking: Evidencia a Partir de la Codificación de Disposiciones Espaciales en Es-
pañol y Yucateco [Thinking for Speaking: Evidence from the Encoding of Spatial
Dispositions in Spanish and Yucatec]. Memoria del IX Encuentro Internacional De
Lingüística En El Noroeste. Vol. 2. Hermosillo: Editorial UniSon. 175–190.
Berlin, Brent. 1968. Tzeltal Numeral Classifiers: A Study in Ethnographic Semantics. Janua
Linguarum (All Series), Bloomington, IN; 70. The Hague: Mouton de Gruyter.
Berlin, Brent, and Paul Kay. 1969. Basic Color Terms. Berkeley: University of California Press.
Bohnemeyer, Jürgen. 1998a. Temporal reference from a Radical Pragmatics perspective: Why
Yucatec does not need to express ‘after’ and ‘before.’ Cognitive Linguistics 9(3): 239–282.
Bohnemeyer, Jürgen. 1998b. Time relations in discourse: Evidence from Yukatek Maya.
PhD diss., Tilburg University.
Bohnemeyer, Jürgen. 2000. Event order in language and cognition. In Linguistics in the
Netherlands 17, H. de Hoop and T. van der Wouden (eds.), 1–16. Amsterdam: John
Benjamins.
Bohnemeyer, Jürgen. 2002. The Grammar of Time Reference in Yukatek Maya. Munich:
LINCOM.
6
Of course, the researcher’s goal is fundamentally different from the child’s: one is aiming for
(primarily) declarative knowledge, the other for procedural knowledge. The approaches the two take
are tailored toward these different goals and are consequently not interchangeable. What they have in
common, however, is that they both rely on the same types of observational evidence.
Bohnemeyer, Jürgen. 2007. Morpholexical transparency and the argument structure of

verbs of cutting and breaking. Cognitive Linguistics 18 (2): 153–177.
Bohnemeyer, Jürgen. 2010. The language-specificity of Conceptual Structure: Path, Fictive
Motion, and time relations. In Words and the World, Barbara Malt and Philip Wolff
(eds.), 111–137. Oxford: Oxford University Press.
Bohnemeyer, Jürgen. 2011. Spatial frames of reference in Yucatec: Referential promiscuity
and task-specificity. Language Sciences 33: 892–914.
Bohnemeyer, Jürgen. 2012. Yucatec demonstratives in interaction: spontaneous vs. elicited
data. In Practical theories and empirical practice, Andrea C. Schalley (ed.), 99–128. Am-
sterdam: John Benjamins.
Bohnemeyer, Jürgen, and Penelope Brown. 2007. Standing divided: Dispositional verbs and
locative predications in two Mayan languages. Linguistics 45 (5–6): 1105–1151.
Bohnemeyer, Jürgen, and Mary D. Swift. 2004. Event realization and default aspect. Linguis-
tics and Philosophy 27 (3): 263–296.
Bowerman, Melissa, and Eric Pedersen no date. Cross-linguistic perspectives on topo-
logical spatial relationships. Unpublished manuscript, Max Planck Institute for
Psycholinguistics.
Brown, Roger. 1958. Words and Things. New York: Free Press.
Bueno, Steve, and Cheryl Frenck-Mestre. 2008. The activation of semantic memory: Effects
of prime exposure, prime-target relationship, and task demands. Memory & Cognition
36(4): 882–898.
Chamberlain, Alexander Francis. 1903. Primitive taste words. The American Journal of Psy-
chology 14: 146–153.
Clark, Herbert H. 1970. Word associations and linguistic theory. In New Horizons in Lin-
guistics, John Lyons (ed.), 271–286. Baltimore, MD: Penguin.
Clark, Herbert H., and Deanna Wilkes-Gibbs. 1990. Referring as a collaborative process. In
Intentions in communication, Philip Cohen, Jerry Morgan, and Martha Pollack (eds.),
463–493. Cambridge, MA: MIT Press.
Costa, Albert, F.-Xavier Alario, and Alfonso Caramazza. 2005. On the categorical nature
of the semantic interference effect in the picture-word interference paradigm. Psycho-
nomic Bulletin & Review 12(1): 125–131.
Dahl, Östen. 1985. Tense and Aspect Systems. Oxford: Blackwell.
Dowty, David R. 1979. Word Meaning and Montague Grammar: The Semantics of Verbs and
Times in Generative Semantics and Montague’s PTQ. Dordrecht: Reidel.
Ervin, Susan M., and Herbert Landar. 1963. Navajo Word Association. American Journal of
Psychology 76: 49–57.
Evans, Nicolas, and David P. Wilkins. 2000. In the mind’s ear: The semantic extensions of
perception verbs in Australian languages. Language 76 (3): 546–592.
Ferrand, Ludovic, and Boris New. 2003. Semantic and associative priming in the mental lexi-
con. In The mental lexicon, Patrick Bonin (ed.), 25–43. New York: Nova Science Publishers.
van Fraassen, Bas. 1980. The Scientific Image. Oxford: Oxford University Press.
Francom, Jerrid. 2009. Experimental Syntax: Exploring the effect of repeated exposure to
anomalous syntactic structure—evidence from rating and reading tasks. PhD diss.,
University of Arizona.
Frege, Gottlob. 1980 [1884]. The Foundations of Arithmetic. Trans. J. L. Austin. Second Re-
vised Edition. Evanston, IL: Northwestern University Press.
Goodall, Grant. 2005. On the syntax and processing of wh-questions in Spanish. In WCCFL
23 Proceedings, B. Schmeiser, V. Chand, A. Kelleher, and A. Rodriguez (eds.), 101–114.
Somerville, MA: Cascadilla Press.
Guerssel, Mohamed, Kenneth Hale, Mary Laughren, Beth Levin, and Josie White Eagle.
1985. A cross-linguistic study of transitivity alternations. In Papers from the Para-
session on Causatives and Agentivity at the Twenty-First Regional Meeting, W. H.
Eilfort, P. D. Kroeber, and K. L. Peterson (eds.), 48–63. Chicago: Chicago Linguistic
Society.
Hale, Kenneth L., and Samuel J. Keyser. 1987. A view from the middle. Lexicon Project Work-
ing Papers, 10. Cambridge, MA: Center for Cognitive Science, MIT.
Himmelmann, Nikolaus P. 1998. Documentary and descriptive linguistics. Linguistics 36:
161–195.
Hiramatsu, Kazuko. 2000. Accessing linguistic competence: Evidence from children’s and
adults’ acceptability judgments. PhD diss., University of Connecticut.
Jackendoff, R. 2002. Foundations of Language: Brain, Meaning,Grammar, Evolution. Oxford:
Oxford University Press.
Jackendoff, R. S. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.
Kay, Paul. 2006. Methodological issues in cross-language color naming. In Language, Cul-
ture, and Society, Christine Jourdan and Kevin Tuite (eds.), 115–134. Cambridge: Cam-
bridge University Press.
Kay, Paul, and Brent Berlin. 1997. Science ≠ imperialism. Behavioral and Brain Sciences 20:
196–201.
Kita, Sotaro. 1999. Japanese ENTER/EXIT verbs without motion semantics. Studies in Lan-
guage 23: 307–330.
Kita, Sotaro. 2006. A grammar of space in Japanese. In Grammars of Space, S. C. Levinson
and D. P. Wilkins (eds.), 437–474. Cambridge: Cambridge University Press.
Klein, Wolfgang. 1994. Time in Language. London: Routledge.
Krifka, Manfred. 2011. Varieties of semantic evidence. In Semantics. An International Hand-
book of Natural Language Meaning, vol. I, Claudia Maienborn, Klaus von Heusinger,
and Paul Portner (eds.), 242–267. Berlin: Mouton de Gruyter.
Lakoff, George. 1973. Hedges: A study in meaning criteria and the logic of fuzzy concepts.
Journal of Philosophical Logic 2: 458–508.
Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the
Mind. Chicago: University of Chicago Press.
Levin, Beth. 1993. English Verb Classes and Alternations. Chicago: University of Chicago
Press.
Levinson, Stephen C. 2001. Motion verb stimulus, version 2. In ‘Manual’ for the Field Season
2001, S. C. Levinson and N. J. Enfield (eds.), 9–11. Nijmegen: Max Planck Institute for
Psycholinguistics.
Lewis, Richard L., Shravan Vasishth, and Julie A. Van Dyke. 2006. Computational prin-
ciples of working memory in sentence comprehension. Trends in Cognitive Science 10
(10): 447–454.
Lucy, John A. 1997. The linguistics of “color.” In Color Categories in Thought and Language,
C. L. Hardin and Luisa Maffi (eds.), 320–346. Cambridge: Cambridge University Press.
Magnus, Hugo. 1877. Die geschichtliche Entwicklung des Farbensinnes [The Historic Devel-
opment of the Color Sense]. Leipzig: Viet.
Magnus, Hugo. 1880. Untersuchungen ueber den Farbensinn der Naturvoelker [Investiga-
tions on the Color Sense of the Primitive Peoples]. Jena: Fraher.
Majid, Asifa, James S. Boster, and Melissa Bowerman. 2008. The cross-linguistic categoriza-
tion of everyday events: A study of cutting and breaking. Cognition 109 (2): 235–250.
Marshall, John C., and Freda Newcornbe. 1973. Patterns of paralexia: A psycholinguistic
approach. Journal of Psycholinguistic Research 2: 175–199.
Marslen-Wilson, William, Lorraine Komisarjevsky Tyler, and Rachelle Waksler. 1994. Mor-
phology and meaning in the English mental lexicon. Psychological Review 215 (1): 3–33.
of American Linguistics 70: 369–415.
Meyer, David E., Roger W. Schvaneveldt, and Margaret G. Ruddy. 1974. Functions of graphe-
mic and phonemic codes in visual word-recognition. Memory & Cognition 2: 309–321.
Myers, Charles S. 1904. The taste-names of primitive peoples. British Journal of Psychology
1: 117–126.
O’Meara, Carolyn, and Gabriela Pérez Báez. 2011. Spatial frames of reference in Mesoameri-
can languages. Language Sciences 33: 837–852.
Pederson, Eric, Eve Danziger, David P. Wilkins, Stephen C. Levinson, Sotaro Kita, and Gunter
Senft. 1998. Semantic typology and spatial conceptualization. Language 74: 557–589.
Perea, Manuel, and Eva Rosa. 2002a. Does the proportion of associatively related pairs
modulate the associative priming effect at very brief stimulus-onset asynchronies?
Acta Psychologica 110: 103–124.
Perea, Manuel, and Eva Rosa. 2002b. The effects of associative and semantic priming in the
lexical decision task. Psychological Research 66: 180–194.
Piaget, Jean, and Bärbel Inhelder. 1956. The Child’s Conception of Space. London: Routledge.
Plaut, David C., and Tim Shallice. 1993. Deep dyslexia: A case study of connectionist neuro-
psychology. Cognitive Neuropsychology 10(5): 377–500.
Rosch, Eleanor. 1978. Principles of Categorization. In Cognition and Categorization, Eleanor
Rosch and Barbara B. Lloyd (eds.), 27–48. Hillsdale, NJ: Lawrence Erlbaum Associates.
Rosch, Eleanor, and Carolyn B. Mervis. 1975. Family resemblances: Studies in the internal
structure of categories. Cognitive Psychology 7: 573–605.
Saunders, Barbara A. C., and Jaap van Brakel. 1997. Are there nontrivial constraints on
colour categorization? Behavioral and Brain Sciences 20: 167–228.
Snyder, William. 2000. An experimental investigation of syntactic satiation effects. Linguis-
tic Inquiry 31(3): 575–582.
Quine, W. V. O. 1960. Word and Object. MIT Press.
Talmy, Leonard. 2000. Toward a Cognitive Semantics. Vol. I. Cambridge, MA; London, Eng-
land: MIT Press.
onomy of projective content. Language 89(1): 66–109.
Vendler, Zeno. 1957. Verbs and times. The Philosophical Review LXVI: 143–160.
Wilkins, David P. 1997. Alternative representations of space: Arrernte narratives in sand. In
Proceedings of the CLS Opening Academic Year ’97/’98, Monique Biemans and Joost van
der Weijer (eds.), 133–162.
Wittgenstein, Ludwig. 1921/1922. Tractatus Logico-Philosophicus. Trans. C. K. Ogden. London:
Routledge and Kegan Paul.
2
The Problem with No-Nonsense Elicitation Plans

(for Semantic Fieldwork)
Meagan Louie
1 Introduction
This paper begins with two observations. The first observation is that semanti-
cists, being linguists, love systematicity and paradigms. Unsurprisingly, this love
of systematicity and paradigms often results in elicitation plans that are systematic
and paradigmatic. Thus a budding semantic fieldworker, inspired by the semantic
fieldwork methodology presented in Matthewson (2004), may be tempted to create
elicitation plans structured as follows: the fieldworker systematically presents their
language consultant with context-utterance pairs that differ minimally, back-to-
back, in order to check for variation in truth-value and/or felicity judgments.
This tendency toward systematic/paradigmatic elicitation plans is problematic,
however, when one considers the second observation: unlike linguists, language
consultants do not love systematic and paradigmatic elicitation plans. In this paper,
I discuss the problems associated with systematic/paradigmatic elicitation plans.
I then propose an approach that avoids these problems: I propose an approach to
elicitation plans where the minimal pairs required for linguistic analysis are em-
bedded in an overarching storyline. I then consider problems associated with over-
arching storylines, but conclude that it is nonetheless preferable to the alternative.
1.1 Background on Semantic Fieldwork: Matthewson (2004)
Prior to Matthewson (2004), the general consensus was that semantic fieldwork
on something other than the investigator’s native language was impossible—that
is, that one had to be a native speaker of the target language in order to access the
sort of intuitions required for formal semantic analysis. Matthewson (2004), how-
ever, laid out a methodology for eliciting, from a non-linguistically trained native
speaker (i.e., a consultant), the sorts of judgments that a researcher could indirectly
use for semantic analysis. The basic set up is as follows: the investigator/elicitor 47

(who must already have a solid understanding of the basic phonology, morphol-
ogy, and syntax of the target language) presents the consultant with a context, via
verbal description, visual cues, or a combination of the two, and then presents the
consultant with a constructed sentence in the target language. The elicitor then
may ask the consultant for a translation of the constructed sentence in the pro-
vided context. Or the elicitor may ask the consultant for a judgment regarding (i)
whether the constructed sentence is something that the consultant (or other native
speaker of the language) would say in the given context and/or (ii) whether the
constructed sentence is true or false in the given context. The former kind of judg-
ment is termed a felicity judgment, and the latter is a truth–value judgment.
One important point that Matthewson (2004) argues is that the semantic field-
worker should only ask for translations of and judgments for full sentences. She
points out that not only (i) is it difficult for a speaker to consciously be aware of and
pick out a sentence’s subconstituents, but also (ii) the meaning of the targeted sub-
constituent may depend on its surrounding environment. The example she gives
is the interpretation of Chinese (both Mandarin and Cantonese) NPs, which bear
no particular definite/specificity morphology, but nonetheless differ in the range
of interpretations they allow, depending on their syntactic positions. One of the
observations I make in this paper is that this reasoning can be extended beyond the
sentence level. Many lexical items or syntactic constructions may have interpreta-
tions that depend on their larger (i.e., extra-sentential) environment. Judgments
associated with isolated context-utterance pairs may thus miss out on significant
interpretational aspects of the lexical item or syntactic constructions under inves-
tigation. An undisguised paradigmatic elicitation plan (which will be described in
more detail in section 2), I argue, does not provide the natural extra-sentential en-
vironment required to identify these interpretational aspects. Before moving onto
the main observations about paradigmatic elicitation plans and their problems,
however, I will describe the fieldwork situation from which my observations arise.
1.2 The Fieldwork Context
The examples used in this paper are drawn from the author’s experiences work-
ing on Blackfoot,1 a member of the Algonquian language family. Blackfoot is an
1
Guide to glosses: 1 = first person, 2 = second person, 3 = third person (proximate), 3’= third
person (obviative), 0 = inanimate, 21 = “inclusive plural” or “unspecified person,” X:Y = X acts on Y
theme marker (where {X,Y} correlate to 1st, 2nd, 3rd, or 0 (inanimate) persons), 1/2 = “local person”
(i.e., 1st or 2nd person), vta = transitive animate (object) verb, vti = transitive inanimate (object) verb,
vai = animate (subject) intransitive verb, vii = inanimate (subject) intransitive verb, cl.an = animate
classifier/verb, pl = animate plural, 0pl = inanimate plural, ic = “initial change” (a morphophonological
process), rl = relative root (indicates a contextually-salient time or place), rel = not-visible-to-speaker,
fut = future, DIST = distributive, ints = intensifier, neg = negation, sbj = subjunctive clause type mor-
phology, unr =“unreal” clause type morphology, 3pl:DTP = “distinct third person pronoun plural,”
3:nonaff.sg = “third person non-affirmative singular,” bism = “be in specified manner” verb.
The Problem with No-Nonsense Elicitation Plans 49
understudied and endangered aboriginal language spoken on three reserves in

Alberta, Canada. As of the 2011 Canadian census, there were approximately 3,250
speakers in Canada, with another 100 speakers in the reservation in Montana
(Ethnologue (2013)). Blackfoot is classified as EGIDS level 7—shifting—which
means that although there are still native speakers, there is currently no transmis-
sion of the language to children (Ethnologue (2013)). The majority of speakers is
over the age of 60 and bilingual, also speaking English. There is a grammar and
dictionary (Frantz 1991; 2009, Frantz and Russell 1995); and collections of older
texts (Uhlenbeck 1911; Uhlenbeck (1912); Josselin de Jong (1914)); as well as early-
education materials (Tailfeathers 1993).
The speaker that I work with is female, in her mid-to-late 60s, and from the
Blood/Kainaa Reserve outside of Lethbridge. She was a monolingual Blackfoot
speaker until age seven, when she was forced to attend St. Paul’s residential school
in Cardston. She was allowed to return home during the summers, however, where
she would begin speaking Blackfoot again and learned more about the language
from her grandfather. She has been working with linguists at the University of
British Columbia for 15 years (since 1998), where she served as the language con-
sultant for three field methods classes (1998–1999, 2004–2005, 2011–2012). She is
thus an experienced language consultant. And although she has had no systematic,
formal training in linguistics, she is a “natural linguist” in the sense that she is in-
terested in examining her own language, often spontaneously volunteering opin-
ions about various parts of the language and spontaneously providing her own
contexts to replace the contexts in infelicitous context-utterance pairs constructed
by the elicitor. She is thus particularly suited for semantic fieldwork,2 where, in my
experience, not all language consultants have the necessary interest/inclination/
curiosity for participating in the sort of context-utterance-judgment task required.
Nonetheless, not even the most enthusiastic and insightful of consultants, I argue,
should be subjected to paradigmatic elicitation plans.
2 The Problem with Paradigmatic Elicitation Plans
The main claim of this paper is that the linguist’s love of the systematic, paradig-
matic presentation-style is something that the semantic fieldworker must actively
disguise when forming their elicitation plans. In other words, paradigmatic elici-
tation plans are problematic for three reasons, which are listed, as follows, in the
order that I address them:
1. They are too boring for the language consultant
2. They are too mentally straining for the language consultant, and
3. They are too transparent for the language consultant
2
Second only, perhaps, to having an enthusiastic semanticist as your language consultant.
Before proceeding to discuss these problems, I give an example of a paradig-

matic elicitation plan.
2.1 A Paradigmatic Elicitation Plan
The target phenomenon was the availability of distributive readings in Black-

foot. I was testing to see whether the availability of distributive readings varied
according to the morphological form of the predicate, looking at four different
morphological forms: (i) a bare predicate, (ii) a predicate marked with the uni-
versal quantifier ohkan(a), (iii) a predicate marked with what appears to be a
distributive marker (n)i’t, and (iv) a predicate marked with both ohkan(a) and
(n)i’t. I was also testing to see whether the availability of distributive readings
varied with respect to the grammatical role of the pairs of entities 〈x, y〉 , such that
y was being interpreted as referentially dependent on x. In other words, whether
the grammatical roles of the Distributive Key (x), and the Distributive Share (y)
affect the availability of distributive readings (cf. Szabolcsi 2010). The relevant
grammatical roles I considered were: (i) Actor; (ii) Theme, where both the “Actor”
and “Theme” nominals have morphological reflexes on the verbal complex and
are generally considered to be syntactic “arguments”; (iii) iiht-NPs, which are
nominals that are licensed by the “means/instrument/source/reason” preverbal
element iiht; (iv) locational NPs, which are linked by a preverbal element it; (v)
events; and (vi) non-grammatical themes. A “non-grammatical theme” refers to
things that are part of the meaning of a predicate but not considered arguments.
For instance, the verb sinaa “draw a picture of ” takes two arguments: the drawer
and the person that is being drawn. There is no overt nominal correlating to the
“picture” that is produced, nor any agreement morphology on the predicate. I
refer to these “picture”-type things as “non-grammatical themes.” The elicitation
plan was structured as follows: each example correlated with a particular 〈Dist-
Key, DistShare〉 pair, and for each example, the consultant was presented with
two pictorial contexts: one where a distributive reading is true (Context A), and
one where a distributive reading is false (Context B). Each example consisted of
Table 2.1
Summary Table for the Boring, Paradigmatic Elicitation Plan
DistKey–DistShare Bare P (n)i’t-P ohka(a)n-P ohkana-(n)i’t-P
Actor – Theme 1a (A/B?) 1b (A/B?) 1c (A/B?) 1d (A/B?)

Actor–iiht NP 2a ((A/B?)) 2b (A/B?) 2c (A/B?) 2d (A/B?)
Actor–Locational NP 3a (A/B?) 3b (A/B?) 3c (A/B?) 3d (A/B?)
Theme–Event 4a (A/B?) 4b (A/B?) 4c (A/B?) 4d (A/B?)
Theme–iiht NP 5a (A/B?) 5b (A/B?) 5c (A/B?) 5d (A/B?)
iiht NP–non-grammatical theme 6a (A/B?) 6b (A/B?) 6c (A/B?) 6d (A/B?)
iiht NP–Actor 7a (A/B?) 7b (A/B?) 7c (A/B?) 7d (A/B?)
four subexamples (a), (b), (c), (d), which correlated with the four morphologi-
cal forms I was testing. For each of these subexamples, I presented my language
consultant with the relevant morphological form and asked whether they were
true in (i) Context A, and then whether they were true in (ii) Context B. This is
summarized in Table 2.1, and the elicitation plan is presented in (1)–(8) below. I
only provide the actual Blackfoot forms tested for (1); for (2)–(8) I only provide
the contexts.3
(1) Distributive Key: Actor; Distributive Share: Theme
Context A: Context B:
Three dogs chased two cats

a. niyookskamiiksi imitaiksi iiksisaisskoyiiyaa
niyookska-waami-iksi imitaa-iksi ii-oksisaissko-yii-yaawa
three-cl.an-pl dog-pl ic-chase.vta-3:3’-3pl:DTP
naatokaami poosiksi
naatok-waami poos-iksi
two-cl.an cat-pl
“Three dogs chased two cats.” BARE FORM
Q: True in Context A? Y/N Q: True in Context B? Y/N

b. niyookskamiiksi imitaiksi i’toksisaisskoyiiyaa

niyookska-waami-iksi imitaa-iksi ni’t-oksisaissko-yii-yaawa
three-cl.an-pl dog-pl DIST-chase.vta-3:3’-3pl:DTP
naatokaami poosiksi
two-cl.an cat-pl
“Three dogs each chased two cats.” ni’t/DIST-marked

The reader may notice that the numerals vary in terms of whether they take plural morphology.
3
The morphosyntactic and semantic properties of of plural marking on numerals in Blackfoot remain
mysterious to me.
c. niyookskamiiksi imitaiksi iihkanaoksisaisskoyiiyaa

niyookska-waami-iksi imitaa-iksi ii-ohkana-oksisaissko-yii-yaawa
three-cl.an-pl dog-pl ic-all-chase.vta-3:3’-3pl:DTP
naatokaami poosiksi
two-cl.an cat-pl
“Three dogs all chased two cats.” ohkan/∀-marked

d. niyookskamiiksi imitaiksi iihkanai’toksisaisskoyiiyaa

niyookska-waami-iksi imitaa-iksi ii-ohkana-ni’t-oksisaissko-yii-yaawa
three-cl.an-pl dog-pl ic-all-DIST-chase.vta-3:3’-3pl:DTP
naatokaami poosiksi
two-cl.an cat-pl
“Three dogs all chased two cats.”

(2) Distributive Key: Actor; Distributive Share: iiht-NP (NP linked with a
“means/source” preposition)
Context A: Joel, Mario, and Donald are playing tricks on Amelia. Each boy
gathered up three snakes and scared her with them.
J M D
Context B: Joel, Mario, and Donald are playing tricks on Amelia. Each boy
found a snake, and they all scared her with the snakes.
J M D
Those boys scared Amelia with three snakes.
(3) Distributive Key: Actor; Distributive Share: Locational NP

Context A: Donald, Mario, and Joel at a food fair. Donald goes to eat at
three Asian food tents: the Chinese tent, Korean tent, and Japanese tent.
Mario goes to eat at three European food tents: the German tent, French
tent, and Swedish tent. Joel goes to eat at three African food tents: the
Ethiopian tent, Nigerian tent, and Moroccan tent.
D M J
Context B: Donald, Mario, and Joel at a food-fair. Donald goes to eat at the
Chinese tent. Mario goes to eat at the Swedish tent. Joel goes to eat at the
Moroccan tent.
D M J
(4) Distributive Key: Theme; Distributive Share: Events

Overall Context: Heather’s taking a photography course, and her
assignment is to take portraits. At the park she sees a bunch of eight girls.
(5) Distributive Key: Theme; Distributive Share: iiht-NP (NP linked with a
“means/source” preposition)
Context A: My relatives are all getting their hair cut. The hairdresser
considers herself an artist, so each cousin gets a unique haircut style, and
each style is extremely elaborate and requires three different pairs of scissors
in order to get done.
Context B: My relatives are all getting their hair cut. They have such thick
hair that the hairdresser’s scissors get dull, having to cut all that hair. The
hairdresser has to discard two pairs of scissors because they get dull in the
process, and a total of three pairs of scissors are used by the time she’s done.
(6) Distributive Key: iiht-NP (NP linked with a “means/source” preposition);

Distributive Share: non-grammatical theme (i.e., not morphologically
marked)
Context A: I have three pens. I use each of these pens to produce a separate
picture of Amelia.
Context B: I have three pens. I use all of these pens to produce a single
picture of Amelia (e.g., one is for outlines, one is for crosshatching, etc.)
(7) Distributive Key: iiht-NP (NP linked with a “means/source” preposition);

Distributive Share: Theme
Context A: I’m considering buying a knife. There are three options, and I’m
testing them all to see if i) they can handle cutting large fruit, and ii) they
can handle the intricacies of cutting small fruit. So I use each knife to cut a
large apple and a small apple.
Context B: The knives are all dull! In the end, I had to use three knives to
cut a measly two apples.
(8) Distributive Key: iiht-NP; Distributive Share: Actor

Context A: We’re at a craft fair. At the collage table, there are three pairs of
scissors, a big pair, a medium-sized pair, and a little pair. Collage turned
out to be a fairly popular attraction. Each of the scissors was used by two
different men.
Context B: We’re at a craft fair. There are three pairs of scissors, a big pair,
a medium-sized pair, and a little pair. But no one was interested in collage.
Only a total of two men used the scissors.
The planned elicitation actually consisted of 10 examples. The examples in

(1)–(8) represent how far I got before my consultant revolted, and began to give
me frustrated comments to the tune of “It just means ‘two men used three pairs of
scissors!’ ” Attempts to continue this elicitation plan in further elicitation sessions,
as well as attempts to confirm the data judgments in further elicitation sessions,
were met with similar expressions of frustration.
I now move to a discussion of the problems inherent in paradigmatic elicita-
tion plans. The three issues I have identified are as follows:
1. They are too boring for the language consultant
2. They are too mentally straining for the language consultant
3. They are too transparent for the language consultant
2.2 Paradigmatic Elicitation Plans are too boring
The biggest problem associated with paradigmatic elicitation plans is that they
are boring. And while a linguist, as an academic, may be trained to endure situ-
ations with advanced levels of boringness, this is rarely the case for their lan-
guage consultants. Thus, a paradigmatic elicitation plan often results in a bored
consultant.
There are three possible consequences associated with having a bored lan-
guage consultant. The first is that the consultant may be too bored to pay attention
to the minute differences in context, or the minute differences in morphological
form, that you are presenting. Because these minute differences in context and
morphological form are the crucial conditioning factors that a semantic field-
worker aims to manipulate, if there is a possibility that the language consultant is
too bored to pay attention to these, a fieldworker cannot be confident of the judg-
ments they elicit.
Consider the following example, where the goal was to determine whether or
not the presence/absence of plural morphology on numerals affects the availability
of distributive readings. Having already elicited distributivity judgements for the
form in (1a), which has plural morphology (-iksi) on the numeral modifying the
actor/agent (niyookska “three”), I attempted to elicit judgements for a form that
lacked this plural morphology on the numeral.
(9) Morphological Minimal Pair:
1. Elicitor: If there were three dogs, and each of them chased two different
cats, like in the picture in B, would it be true to say:
niyookskamiiksi imitaiksi iiksisaisskoyiiyaa naatokaami poosiks?

2. Consultant: No. It just means “Three dogs chased two cats.”

3. Elicitor: What if I say:
niyookskamii imitaiksi iiksisaisskoyiiyaa naatokaami poosiks?
4. Consultant: Yes, that’s what you said last time.
niyookskamiiksi imitaiksi iiksisaisskoyiiyaa naatokaami poosiks.
5. Elicitor: Right, so last time I said niyookskamiiksi imitaiksi
iiksisaisskoyiiyaa naatokaami poosiks—but what about if I just say
niyookskamii imitaiksi iiksisaisskoyiiyaa naatokaami poosiks? Is that true
in the picture in A?
6. Consultant: niyookskamii imitaiksi iiksisaisskoyiiyaa naatokaami
poosiks? That means “three dogs chased two cats”—niyookskamiiksi
imitaiksi iiksisaisskoyiiyaa naatokaami poosiks.
The second possible situation that may arise from having a bored language
consultant is as follows: the language consultant may be too bored to even pretend
that they are paying attention to the contexts you are providing them and may at-
tempt to commandeer the elicitation plan. The commandeering may be achieved
by various methods. For example, once asked about a given form, a language con-
sultant may begin to rapidly volunteer loosely related forms that are not relevant
to the fieldworker’s experimental goal. As a fieldworker, I don’t want to suggest

to my language consultant that there are parts of the target language (or related
culture) that I don’t care about. Fieldworkers with a similar mindset will likely feel
obligated to express interest in these loosely related forms, and may have trouble
resteering the elicitation back toward the type of material they originally aimed to
elicit.
For example, my research interests generally focus on functional elements
in Blackfoot, and thus I gear my elicitations toward getting semantic/pragmatic
minimal pairs that contrast different functional elements. My language consultant,
however, is quite interested in teaching me the slightly different meanings associ-
ated with related lexical items. Thus I may be trying to gauge the semantic effect of
an aspectual morpheme with the various kinds of aspectual classes, choosing the
lexical item, ihpiyi “dance,” as an instance of the aspectual class “activity.” This has
previously resulted, however, in my consultant turning the elicitation into an op-
portunity to teach me the difference between the verbs ihpiyi “dance” and ipaskaa
“dance (at a dance),” and then continuing to inform me about the various (mor-
phologically related) kinds of dances and dancing like ssikotoyiipasskaa “do the
black-tailed deer dance,” o’taksipasskaan(istsi) “round/circle dance(s)” (originally
done solely by women) and sipisttoipasskaan(istsi) “owl dance(s).” My consultant
can become very focused on these lexically related items, and often can continue
to think about them after I try to steer the elicitation back toward the original elic-
itation. This is evidenced by the fact that she often pauses in the middle of giving
me an utterance/translation/judgment, in order to offer me another lexically re-
lated item.
Another common method that a language consultant can use to commandeer
an elicitation is storytelling. For instance, a bored language consultant may start
telling you about the time that they were six years old and stole their father’s boots,
and then got stuck on the side of the hill after it rained since they were only six and
not strong enough to pull the large gumboots out of the mud.4 Fieldworkers in this
situation, who want to maintain friendly relations with their language consultant,
cannot easily interrupt their language consultant’s story, even as they see their am-
bitious plans for the elicitation session slipping away.5
The third possible consequence is that a bored language consultant may be
predisposed against coming back for the following week’s elicitation. This is the
worst possible consequence for a semantic fieldworker of an endangered language
and a major reason why paradigmatic elicitation plans should be avoided.
4
This is a real-life example. The story “Going to a Medicine Pipe Dance” can be accessed (both
in English and in Blackfoot with morpheme breakdowns and glosses) at https://www.academia.
edu/8516822/Going_to_a_Medicine_Pipe_Dance_Blackfoot_Text_.
5
If the fieldworker is lucky, the language consultant might tell the story in the target language, so
that the elicitation session is not a loss for collecting data. If the language consultant tells the story in
the meta language, there isn’t much a fieldworker can do.
2.3 Paradigmatic Elicitation Plans are too mentally straining
The second major problem associated with paradigmatic elicitation plans is that
they are often quite mentally straining for a language consultant. The observation
underlying this claim is as follows: it’s very easy for a language consultant to still be
thinking of a previous context when he or she give you a judgment. The observa-
tion can be illustrated with the following exchange between elicitor and language
consultant:
(10) 1. Elicitor: OK, so here’s the context. Amelia and I bought lunch and mine
cost five dollars, but Amelia got a better deal and only paid three dollars.
How can I say: “My lunch cost two dollars more than hers”?
2. Consultant: omistsi aoówaahsinistsi nitohpommatoo’pistsi

om-istsi aoówaahsin-istsi nit-ohpommatoo-’p-istsi
dem-0pl food-0pl 1-buy.vti-1:0-0pl
náátokisopoksi o’tsitskahkohtoo’p
náátoki-sopoksi o’tsitsk-ahkohtoo-’p
two-dollar over-cost.vti-21
“The food that I bought was two dollars more.”
3. Elicitor: All right, now let’s say that Amelia and I rented movies. Hers
was one hour long, and mine was three hours long. How could I say: “My
movie lasted two hours longer than hers”?
4. Consultant: omi saiksistohsini nitohpommatoo’pi

om-yi saiksistohsini-yi nit-ohpommatoo-’p-yi
dem-0l movie-0 1-buy.vti-1:0-0
náátokisopoksi o’tsitskahkohtoo’p
náátoki-sopoksi o’tsitsk-ahkohtoo-’p
two-dollar over-cost.vti-21
“The movie that I bought was two dollars more.”
(ML/BB Blackfoot Elicitation 2011-10-11)
The consultant’s answer to the question of which movie was longer (“The
movie that I bought was two dollars more,”) makes it obvious that the consultant is
still half-thinking of the previous context, where the relevant contextual difference
was price, as opposed to the targeted contextual difference of (time) length.
The observation isn’t surprising when one considers the task that paradig-
matic elicitation plans impose on the consultant: the elicitor first asks the language
consultant to imagine a described context and to hold that context in their head so
that he or she can give us a truth-judgment or felicity judgment for an utterance
within that context. The elicitor then asks the consultant to completely forget that
context and instead imagine a new, but only very minimally different context, and
hold this new context in their head so that they can give a (possibly different) judg-
ment. This is an extremely difficult mental task. Given an exchange as in (10), the
elicitor can easily tell that such confusion took place, and can adjust their elicita-
tion plan accordingly.
But if the relevant back-to-back contexts are ones that differ only minimally,
then such confusion might take place without overt warning signs like (10). Given
this possibility, how can a semantic fieldworker be confident that the language
consultant has the targeted context in mind? Although the consultant has only
been asked to provide a translation in (10), one could extrapolate from this and
assume the consultant could similarly still be half-thinking of a previously pre-
sented context when giving truth or felicity judgments.
2.4 Paradigmatic Elicitation Plans are too transparent
The third major problem associated with paradigmatic elicitation plans is that they
may be too transparent to the language consultant. If this is the case, and the lan-
guage consultant starts to figure out what the goal of the elicitation session is, the
fieldworker runs the risk that the language consultant will come up with their
own hypothesis regarding the distribution/meaning of the target morphemes/
constructions. If this happens, then the consultant may give judgments according
to their hypothesis as opposed to their natural intuitions. Consider, for example,
the following exchange:6
(11) 1. Elicitor: Quick survey for native English speakers: compare and contrast
the following two sentences:
Sentence 1: If they would come tomorrow, they would see him.
Sentence 2: If they came tomorrow, they would see him.
2. CONSULTANT 1: If they would come tomorrow they would see him.
Came is past tense and how can tomorrow be past tense?
3. CONSULTANT 2: Correct. Agree with CONSULTANT 1 about tense.
Although the subjunctive conditional given as “Sentence 2” is not actually

objectionable in the given consultants’ dialects,7 the consultants’ hypothesis that
past-tense morphology can only be used to indicate past temporal reference inter-
feres with their felicity judgments.
6
Thanks to Carmela Penner Toews, whose Facebook status update triggered this exchange [Oct.
28th, 2011] (Penner Toews 2011).
7
After given a detailed context, CONSULTANT 2 accepted the sentence, commenting that “It
would be easier to overlook the grammar if I was hearing it.” This suggests that the speaker does
find subjunctive conditionals like “Sentence 2” acceptable in use, but considers them prescriptively
“ungrammatical.”
3 A Solution: Overarching Storylines

3.1 Avoiding the Problems with Paradigmatic Elicitation Plans
I suggest that the problems associated with paradigmatic elicitation plans can be
solved by taking the target utterances and relevant contexts, and linking them to-
gether into an overarching storyline.
This extra structure will (i) (it is hoped) be less boring for your consultant
and (ii) be less straining for the consultant. The reasoning behind this claim is
that a storyline is a natural flow of changing contexts; a language consultant won’t
have to completely wipe their mind of a previous context (something atypical to
normal discourse), they only have to update a context, as the new information
and circumstances within a coherent story line will usually build on, or modify,
the previous context (as is standard in typical discourse).8 For example, with an
overarching storyline, the consultant should only need to incorporate the passing
of time, or the occurrence of a few events, as happens in everyday life. They won’t
have to rethink everything they know (i.e., what has been set up in a previous
context) with an entirely new set of assumptions. There can be real-life scenarios
where one has to rethink through everything they thought they knew, and these
real-life situations are mentally straining. The simulation of such scenarios in elic-
itation contexts is similarly mentally straining. The extra required structure will
also (iii) help the fieldworker camouflage what they are targeting. This is because
having to fit the target utterance-context pairs into an overarching, coherent struc-
ture forces the fieldworker to (a) come up with contexts that aren’t quite so obvi-
ous in how they are minimally different, or (b) separate the minimally different
pairs chronologically. These consequences follow from the imposition of a natural
storyline—it’s a rare story where two only minimally different contexts fit chron-
ologically back-to-back.
The easiest way of incorporating an overarching storyline into your elicita-
tions is to find a ready-made overarching storyline, such as the storyboard elici-
tations provided by Burton and Matthewson (this volume). However, because a
fieldworker may often be curious about a mysterious morpheme/morphology
construction particular to the language that they work on, and their hypothesis
about the mysterious morpheme may not fit standard theoretical approaches, it
may be the case that there will not be a ready-made storyboard for their purposes.
8
At least, I haven’t tried to incorporate Quentin Tarantino–type non-linear narratives into my
elicitation plans. Or Waynes World–style “multiple ending” narratives. Using such narratives in elicita-
tion seems like it would be contrary to solving the discussed problems, although I will admit to being
mostly naive about the schemata/narrative conventions of different cultures. It is certainly the case
that I have trouble following the narrative structure of older Blackfoot texts (for example, the stories in
Josselin de Jong 1914); however, the stories that my consultant tells me follow a more familiar narrative
structure. Given this, I don’t think a standard western-style narrative structure is problematic for my
elicitation plans.
The fieldworker will have to create their own elicitation plan with an overarching
storyline. In the following section, I discuss my basic approach to creating elicita-
tion plans with a narrative structure.
3.2 Creating an Overarching Storyline: Process and Comments
I begin with a caveat: this is just the method that comes most easily to me, and
works for my consultant. Different elicitors and different consultants will likely
have different preferences for creating and being subjected to elicitation plans re-
spectively. That being said, my particular basic approach to creating an elicitation
plan within an overarching storyline is as follows:9
1. List your goals for the elicitation.
(a) If possible, it’s useful to have a couple of different goals related to
different phenomena. This way you can interleave the examples that
target different phenomena, and the examples you elicit won’t seem
too repetitive.10
(b) It’s useful to have these goals in a place you can refer to while
eliciting, so that during the whirl of elicitation you can see if you’re
getting the right kinds of examples.
For instance, sometimes the examples you plan for won’t work for a
particular reason, and you and your consultant will compromise and
arrive at a different way of conveying the targeted meanings. Having
the explicit goals to refer to during elicitation is useful, so you can
check to see if the modified target examples can still achieve your
specific goals.11
(c) Examples of specific goals:
i. “See how comparatives conditionals are formed.”
ii. “Determine whether conditional antecedents with unreal clause-
type morphology can be interpreted with a present or future
orientation.”
iii. “Determine whether morpheme X can take non-agentive
complements.”
9
Thanks to a reviewer for suggesting this part of the chapter. He also provided a very useful com-
ment about “interleaving” of different elicitation topics, which I briefly touch on below.
10
I don’t have a systematic way of incorporating two topics into an elicitation. The interweaving
of examples is usually driven by what seems to me to make a better narrative, as opposed to worries
about priming, and so forth.
11
Although some fieldworkers may be able to remember exactly what they were getting at with
particular examples, and which morphological bits must not change, this is not the case for me. Writing
down the goals has resulted in far more productive elicitation sessions. I have also found that identify-
ing these goals is useful for writing up elicitation notes afterward; while I do the detailed morpheme
breakdown and glossing, I can look at what the goal was, and determine what the “result” was,
2. Make a list of the sorts of minimal pairs you need to resolve the
aforementioned elicitation goals.
(a) By “sorts of minimal pairs” I mean a general structural/functional
frame—for example, “I need a minimal pair, one where this utterance
is presented in (i) a context where the conditional antecedent
we’re describing could happen tomorrow, and (ii) another where it
happened the day before.”
(b) I try to include at least two or three examples of each targeted
minimal pair (with the same structural/functional frame, but with
different lexical items) into the elicitation narrative. This is to gauge
for the possible effects that the particular lexical items may have on
the judgments.
3. Build up the storyline by coming up with a sequence of gradually
unfolding contexts that allow for the aforementioned minimal pairs,
allowing the narrative to determine the lexical material for your
functional “minimal pair” frames.
(a) In the interest of making it as easy as possible for my consultant to
“see (the story/context) in her mind’s eye” I tend use common friends
and acquaintances to “play the role” of the characters in the narrative
context. I have observed a marked increase in my consultant’s ease
with imagining contexts and giving me judgments when doing this,
as opposed to having imaginary characters.
(b) I have also observed that my consultant reacts better to somewhat
humourous12 narratives with bombastic/ridiculous characters—for
example, if our common friends/acquaintances play roles like “the
vain art/toy/pie thief ” or “the evil mastermind.” Unnatural contexts
are more easily accepted in these sorts of narratives.
(c) My consultant also reacts better to elicitation narratives that have
a clear story structure. Lazy elicitation plans that involve gradually
unfolding contexts with no (or a poor) sense of plot/story-structure
are as subject to my consultant’s indifferent boredom as much as
paradigmatic elicitation plans.
3.3 Emerging Contrasts with Overarching Narratives
The introduction of an overarching narrative structure to an elicitation plan tends

to lessen the problems identified with paradigmatic elicitation plans.13 But there
12
Or my best aim at “humorous,” anyway.
13
The introduction of an overarching narrative structure, of course, isn’t infallible. A consult-
ant can still get bored with an ovearching storyline (this may have to do with the fact that I am not a
bestselling novelist). And I think a fieldworker and their consultant just have to resign themselves to
sometimes having rough elicitation days.
is also another benefit to situating one’s elicitation plan into an ovearching sto-
ryline: contrasts unelicitable with paradigmatic elicitation plans can emerge with
overarching storyline elicitations. This can be illustrated by considering how the
two approaches fared at eliciting the difference between two types of conditionals
in Blackfoot. The relevant types of conditionals illustrated in (12). (12i) is an ex-
ample of a subjunctive conditional, characterized by the presence of a preverbal
morpheme kam- (glossed as “if ”), and subjunctive morphology on the anteced-
ent clause. (12ii) is an example of an unreal conditional, characterized by the
presence of unreal morphology on the antecedent clause. The goal of both elici-
tation plans (P1) and (P2) was to discover whether there are differences in terms
of what sorts of contexts unreal and subjunctive conditionals were felicitous.
(12) i. kamááwahkaawsaa apinákosi áakomo’tsaakiyaa

kam-ááwahkaa-si-yi-aawa apinákosi áak-omo’tsaaki-yaawa
if-play.vai-sbj:3-3pl-dtp tomorrow fut-win.vai-3pl:DTP
“If they play tomorrow, they would win.”
ii. ááwahkaawohtopi(yaa) apinákosi áakomo’tsaakiyaa
ááwahkaa-ohtopi(yaawa) apinákosi áak-omo’tsaaki-yaawa
play.vai-unr-(3pl:DTP) tomorrow fut-win.vai-3pl:DTP
“If they had played tomorrow, they would have won.”
P1 is a paradigmatic elicitation plan, structured such that the elicitor pres-

ents the consultant with one of the contexts in (13), and then asks the consultant
whether (12i) is felicitous. Once that judgment is received, the elicitor then asks
whether (12ii) is felicitous in that same context. Once that judgment is received,
the elicitor moves onto the next context, which is temporally and pragmatically
disconnected from the previous. The elicitor then asks first whether (12i), and then
(12ii), is felicitous. The goal for the paradigmatic elicitation plan was a very broad
goal, as I created this plan very near the beginning of my research into Blackfoot
conditionals: I was aiming to see whether the clause-type morphology on condi-
tional antecedents (subjunctive versus unreal clause-type morphology) has a
discernable semantic effect. The particular variables I was using were contextual
ones: whether the hypothetical situation described by the antecedent clause was
(i) past, present, or future-oriented, (ii) whether it was possible or counterfactual,
and (iii) whether the hypothetical event had a past, present, or future counterpart.
(13) P1: A Paradigmatic Elicitation Plan (ML/BB Elicitation 10-25-2011)

a. Antecedent Situation: (i) Future, (ii) counterfactual, (iii) with a future
real-world counterpart situation
Context: {My brother’s hockey team has their final game scheduled for
tomorrow/Monday, but their unbeatable goalie doesn’t get back until
the day after tomorrow. Depressed I say:
→ Both (12i) and (12ii), modified so that the temporal adverbial is
mistapotonni “the day after tomorrow,” accepted as felicitous.
Consultant Comment: “Means the same.”
b. Antecedent Situation: (i) Future, (ii) counterfactual, (iii) with a no
Context: My brother’s hockey team had their final game scheduled for
two days from now (Tuesday), but since an earthquake was predicted
for tomorrow (Monday), all games were canceled. I say:
→ Both (12i) and (12ii) accepted as felicitous
Consultant Comment: “Means the same.”
c. Antecedent Situation: (i) Future, (ii) counterfactual, (iii) with a past
Context: Our final game was yesterday (Saturday), although our
unbeatable goalie got back from vacation today (Sunday). I say:
→ Both (12i) and (12ii) accepted as felicitous.
Consultant Comment: “I can say that too. Means the same.”
d. Antecedent Situation: (i) Future, (ii) possible, (iii) with a (possibly)
future real-world counterpart situation
Context: The team has one more game to play. They haven’t finalized
the schedule yet. But we’re hoping they get scheduled for tomorrow
(Monday), because their unbeatable goalie is leaving on vacation the
day after that (Tuesday):
→ Both (12i) and (12ii) accepted as felicitous.
Consultant Comment: “Both are good, but (ii) pertains more to the
context.”
The comments and judgments yielded by the consultant in such an elicita-

tion session are unenlightening—both types of conditionals are accepted, along
with the comment that the two types of conditionals “mean the same.” Contrast
this with P2, an elicitation plan where the context-utterance pairs have been
linked into an overarching storyline. Note that the elicitation is still paradigmatic
in that, for each example context, I asked my consultant for two different felicity
judgments: a felicity judgment for a conditional with subjunctive clause-type
morphology, and a felicity judgment for a conditional with unreal clause type
morphology. The non-paradigmatic part of the elicitation refers to the gradually
unfolding sequence of example contexts, which require different lexical items for
each subsequent example.14
14
I have used this elicitation plan twice, the first time presenting it in the paradigmatic way described
above. The second time I presented it (a couple of weeks later), I tried to present it by going through the nar-
rative and asking judgments for only one form of conditionals (the “unreal” conditional forms). Although
I got the same results, my consultant quickly became frustrated when I continually presented her with the
infelicitous unreal conditional forms. And she would always offer the felicitous subjunctive form as the
(14) P2: An Elicitation Plan with an Overarching Storyline

(ML/BB Elicitation 12-13-2011)
Context 1: My sister has been spending a lot of time caring for an apple
tree, fertilizing it, pruning it, and so forth. And this year it finally bears
fruit. I see an apple that looks really good, but I know my sister hasn’t had a
chance to eat any of the apples yet.
a. kamikamo’saatá’i niki oma apasstaaminaam annahk
kam-i-ikamo’saata-iniki om-wa apasstaaminaam ann-wa-hk
if-steal.vta-sbj:1/2 dem-3 apple dem-3-rel
ninstahk áaksiikokiitaki
n-inst-wa-hka áak-iik-okiitaki-wa
1-sister-3-rel fut-ints-get.mad.vai-3
“If I steal that apple, my sister will be really mad.” subjunctive
b. nits’i’i kamo’saata’ohtopi oma apasstaaminaam annahk
nit-ii-ikamo’saata-ohtopi om-wa apasstaaminaam ann-wa-hk
1-ic-steal.vta-unr dem-3 apple dem-3-rel
ninstahk áaksiikokiitaki
n-inst-wa-hka áak-iik-okiitaki-wa
1-sister-3-rel fut-ints-get.mad.vai-3
“If I had stolen that apple, my sister would have been really mad.” unreal
Antecedent Situation: (i) Future, (ii) possible, (iii) with a (possibly) future

Context 2: I’m feeling really hungry though . . . I think to myself:
a. kamsaykamo’saatá’i niki oma apasstaaminaam ókowaan
kam-sa-ikamo’saata-iniki om-wa apasstaaminaam n-okowaan
if-neg-steal.vta-sbj:1/2 dem-3 apple 1-stomach
áakitomatapohtako
áak-it-omatap-ohtako-wa
fut-rl-start-sound.vai-3
“If I don’t steal that apple, my stomach will start sounding.” subjunctive
proper way to express the targeted meaning, so I failed at avoiding this aspect of paradigmaticity. This is
often the case; my consultant will feel frustrated with her efforts at teaching me Blackfoot if I continually
present her with “bad” examples. So I usually try to (use my non-native speaker intuitions to) interleave
what I expect might be positive and negative judgments. Note that this is not an issue of boredom—my
consultant is elated when I present her with a streak of well-formed and felicitous Blackfoot! Rather it is an
issue to do with the elicitor-consultant relationship being both (i) a student-teacher relationship, and (ii) a
(sort of) experimenter-participant relationship. Resolving the conflicts that arise from this two-faced nature
of the elicitor-consultant relationship is interesting, but I don’t have anything insightful to say on the subject.
b. #nits’i’i saykamo’saata’ohtopi oma apasstaaminaam ókowaan

nit-ii-sa-ikamo’saata-ohtopi om-wa apasstaaminaam n-okowaan
1-ic-neg-steal.vta-unr dem-3 apple 1-stomach
áakitomatapohtako
áak-it-omatap-ohtako-wa
fut-rl-start-sound.vai-3
“If I hadn’t stolen that apple, my stomach would have started sounding”.
unreal
Comment: That means you already took the apple.

Antecedent Situation: (i) Future (but negative), (ii) possible, (iii) with a
(possibly) future real-world counterpart situation

Context 3: So I steal it and eat it, and hope that my sister doesn’t notice.
I even go to the trouble of eating most of the core, so that there is no
evidence. I throw the seeds into the compost box and figure.
a. kamsaykowai’piksiisii omi átakssakssin

kam-sa-ikowai’piksiisi om-yi atakssakssin
if-neg-open.vti-sbj:3 dem-0 box
nimáátáakokimokaatsiksi
ni-maat-áak-okimo-ok-waatsiksi
1-neg-fut-get.mad.at.vta-3:1/2-3:nonaff.sg
“If she doesn’t open that box, she won’t get mad at me.” subjunctive
b. #iisáykowai’piksimohtopi omi átakssakssin

ii-sa-ikowai’piksi-m-ohtopi om-yi atakssakssin
ic-neg-open.vti-3>0-unr dem-0 box
nimáátáakokimokaatsiksi
ni-maat-áak-okimo-ok-waatsiksi
1-neg-fut-get.mad.at.vta-3:1/2-3:nonaff.sg
“If she hadn’t opened that box, she wouldn’t have gotten mad at me.”
unreal
Comment: That means that she opened the box.

Antecedent Situation: (i) Future (but negative), (ii) possible, (iii) with a
(possibly) future real-world counterpart situation

Context 4: My sister throws her banana peel in the compost box, and then
pauses, looking in the box. Then she whirls around and accuses me:
a. omistsi kamáánissta’pisa kiníínoko’s kitsííkamo’tsaaki

om-istsi kam-aanist-a’pii-si-yaa kiniinoko’s kit-ikamo’saa-ki
dem-0pl if-manner-bism.vii-sbj:3 seed 2-steal.vta-2:1
apasstaaminaam
apasstaaminaam
apple
“If these are seeds, then you stole an apple from me” subjunctive
b. #omistsi áánista’pi’ihtopi kiníínoko’s kitsííkamo’tsaaki

om-istsi aanist-a’pii-ohtopi kiniinoko’s kit-ikamo’saa-ki
dem-0pl manner-bism.vii-unr seed 2-steal.vta-2:1
apasstaaminaam
apasstaaminaam
apple
“If these had been seeds, then you stole an apple from me” unreal
Antecedent Situation: (i) Present, (ii) possible, (iii) with a (possibly)

present real-world counterpart situation

Context 5: I weakly try and deny that they are apple seeds. She throws up
her hands and asks:
a. kamsaykiníínokokisaa, tsá níístapííwaiksaa

kam-say-kiniinokoS-ki-si-yaawa, tsa niit-a’pii-waiksaa
if-neg-seed-be?-sbj:3-3pl:DTP, what manner-bism.vii-3:nonaff.pl
“If they aren’t seeds, what are they?” subjunctive
b. isaykinííniki’ihtopi, tsá níístapííwaiksaa

ii-sa-kiniinokoS-ki-ohtopi, tsa niit-a’pii-waiksaa
ic-neg-seed-be?-unr, what manner-bism.vii-3:nonaff.pl
“If these hadn’t been seeds, what are they?” unreal
Antecedent Situation: (i) Present (negative), (ii) counterfactual, (iii) with

a present (different polarity) real-world counterpart situation

Context 6: I’m totally caught and totally feeling bad. But mostly because I
got caught.
a. #kamowatooiniki omistsi kiniinokiistsi nimáátohkoisskssinioko

kam-oowatoo-iniki om-istsi kiniinokii-istsi ni-máát-ohkoisskssini-ok
if-eat.vti-sbj:1 dem-0pl seed-0pl 1-neg-find.out.about.vta-3:1
“If I eat the seeds, I won’t be found out.” subjunctive
b.nitsoowatoo’ohtopi omistsi kiniinokiistsi nimáátohkoisskssinioko

nit-oowatoo-ohtopi om-istsi kiniinokii-istsi ni-máát-ohkoisskssini-ok
1-eat.vti-unr dem-0pl seed-0pl 1-neg-find.out.about.vta-3:1
“If I had eaten the seeds, I wouldn’t have been found out.” unreal
Antecedent Situation: (i) Past (negative), (ii) counterfactual, (iii) with a
past (different polarity) real-world counterpart situation
Unlike elicitation plan P1, where the unreal and subjunctive conditionals
were uniformly accepted in all contexts and judged as having the same meaning,
the unreal and subjunctive are only both acceptable in the first given context
in elicitation plan P2.15 Once the overarching storyline begins to develop, felicity
differences between the two types of conditionals begin to emerge. Notice that
there is a correlation between (i) whether the context is a natural development of
a previous context, and (ii) whether the different types of conditionals yield differ-
ent felicity judgments: the first context in P2, like all of the contexts in P1, lacks a
connection to a previously developed context. This context, like the contexts in P1,
was unable to elicit any contrast between the two types of conditionals. The non-
initial contexts in P2, on the other hand, which all developed from a previously
given context, do elicit a contrast between the two types of conditionals. From
this generalization, we can extrapolate that certain types of felicity judgments only
arise in contexts with complex narrative structures, which paradigmatic elicitation
plans lack. I suggest that this can be seen as evidence for an extension of Matthew-
son (2004)’s proposal that consultants should only be asked for translations of, and
judgments for, full sentences. Recall that her arguments for restricting judgment
tasks to full sentences are (i) that it is difficult for a consultant to consciously pick
out a sentence’s subconstituents, and (ii) that a lexical/functional item’s meaning
may depend on its surrounding environment. I suggest that the reasoning in (i) and
(ii) can extend beyond the sentence level. The truth-conditions and use-conditions
of a sentence may depend on its extra-sentential context, both linguistic and non-
linguistic, such as previous utterances and previous contexts. Presenting a consul-
tant with a judgment task that isolates a single utterance-context pair might thus
hamper their ability to identify subtle context-dependent interpretational aspects
of the targeted minimal pairs.
15
An exception is (18a) and (18b). Note, however that this doesn’t appear to be an artifact of elici-
tation; it is a robust generalization that both subjunctive and unreal conditionals are felicitous when
the antecedent describes a present-oriented counterfactual situation.
4 Conclusion
In this paper, I have presented problems associated with paradigmatic elicitation

plans for semantic fieldwork. I further presented an alternative type of elicitation,
in which the context-utterance miminal pairs that a semanticist wants to elicit
should be embedded within an overarching storyline. To conclude, I will discuss
some of the problems associated with an elicitation plan structured with an over-
arching storyline. I suggest that despite these problems, an overall assessment still
balances in favor of the overarching storyline–style methodology.
The first problem is that the adaptability potential of your elicitation plan de-
creases. When some unknown factor derails an in-progress elicitation (e.g., your
consultant can’t think of the word for “contest,” or the lexical items you looked
up in preparation have connotations that are independently infelicitous in the
planned contexts), it is not so difficult to adapt an utterance-context pair that is
independent of the preceding and following utterance-context pairs. Adapting an
utterance-context pair that is intertwined with an overarching story, however, is
more difficult. If this sort of eventuality arises, a fieldworker has (at least) two op-
tions. The first option is to try and salvage the elicitation session so that it’s still
useful for your original goal. Because you still have the prepared examples, you
can reframe the elicitation so that the goal is not semantic fieldwork, but prepara-
tion for semantic fieldwork. That is, one can use the elicitation session to find out
which lexical/syntactic forms work, and which ones don’t, and then embed these
forms into next week’s elicitation plan with an overarching storyline. If this doesn’t
seem possible, then the second option is to give up salvaging the elicitation plan’s
original goal. In this sort of situation, a fieldworker can always ask their language
consultant to tell the story that may have derailed an elicitation weeks prior (this
time in the target language!)
The second problem associated with an overarching storyline is a practical
one—this sort of approach requires more time on the part of the fieldworker.
Much more preparation time and brainwork is required on the part of the field-
worker to transform paradigmatic elicitation plans into elicitation plans with
overarching storylines. And more time is required to process the data afterward:
depending on how well the elicitor disguised the paradigmatic minimal pairs in
the first place, it will take a correlating amount of time to pull and rearrange the
context-utterance pairs out of the obscuring overarching storyline, into a orga-
nized data set that allows the semanticist to find interesting generalizations. Near
the end of a research project, this is not so difficult, because the elicitation plans
generally have a more targeted purpose, and it’s easy to remember which example
was testing for what. You can often tell during elicitation how your hypotheses are
faring. However, near the beginning of a research project, when the fieldworker is
fishing for possible generalizations and hypotheses, extracting the generalizations
from a narrative elicitation plan is more difficult. And it is very difficult to tell
how your multiple hypotheses are faring during elicitation. I suggest that taking
the time to do this, however, is preferable to the alternative, which may result in
elicitation sessions where the consultant is bored, and may risk all of the possible
consequences discussed in section 2. The general narrative structure of the elici-
tation plan can also then be reworked into a less language-specific template for
storyboards á la Burton and Matthewson (this volume).
The third problem associated with this sort of elicitation structure is that the
extra structure provided by an overarching storyline tends to increase the prob-
ability of pragmatic/stylistic embellishment by one’s consultant. The pragmatic/
stylistic embellishment usually takes the form of additional mystery morphemes
that the fieldworker has never seen before, and which the field worker is often
forced to gloss with question marks.16 This increase in pragmatic/stylistic embel-
lishment is inversely correlated with the percentage of true minimal pairs, which
is problematic for a linguist who wants to take a scientific approach to language.
An example is given below in (20)—the overarching storyline was about an un-
wanted houseguest. For reasons of space, I haven’t included the targeted/elicited
utterances for each stage of the elicitation narrative. I have only provided the actual
elicited example for the fourth context, where the targeted utterance has been sup-
plemented with the interjection yáá .
(20) Context: 1: My second uncle is the worst houseguest. He takes advantage
of everything in your house, while simultaneously criticizing everything
about your house.
Context: 2: He’s visiting Vancouver this week, and he called us up, asking if
he could stay. My mom and dad really didn’t want him to stay.
Context: 3: But my mom is terrible at saying no, so she said yes, thinking
that it would only be for a couple of days.
Context: 4: The day before he was supposed to leave, however, he decides
to extend his stay. I exclaim:
yáá, nóóhkáyaakomatapoo apinákosi

yáá, nóóhk-áyaak-omatap-oo apinákosi
intj???, c-exp-pfut-start-go.vai tomorrow
“He was supposed to leave tomorrow” OR
“He WAS going to leave tomorrow.”
Consultant: If you’re really upset, you say yáá !
Because I was targeting the semantic/pragmatic effects of the preverbal

morpheme noohk-, which has similar properties to a discourse particle, the un-
prompted addition of an interjection like yáá may be non-trivial. As I see it, there
are two options that a fieldworker can take in dealing with this issue. The first
16
Or even worse, the fieldworker may be forced into labeling the mystery morphemes as “particles.”
option is that a fieldworker can persist in asking their consultant for judgments
on the unembellished minimal pairs. I think this option is problematic, however,
because the fieldworker is asking their consultant to give nuanced judgments for
pragmatically unnatural utterances. In these sorts of situations, the consultant
usually lacks confidence in their judgments. A linguist who takes these sorts of
judgments as their data thus has to (and should!) worry about basing their analy-
ses on possibly compromised data—that is, data that may merely be artifacts of
their consultant being presented with a forced-choice task. The second option is
that the fieldworker bites the bullet and works with the more natural-sounding,
albeit non-minimal pairs. Although the linguist who loves systematicity and para-
digms may cry out at this option, I think this route is preferable. Because in the
end, these embellished near-minimal pairs can always be re-analyzed in light of
further research on the mysterious particles. Possibly compromised data consist-
ing of judgments that a consultant feels uncertain about, on the other hand, isn’t
much good for linguistic analysis either now or later.
References
Burton, Strang, and Lisa Matthewson. 2014. Targeted construction storyboards in semantic
fieldwork. This volume.
Frantz, Donald G. 1991. Blackfoot Grammar. Toronto, Canada: University of Toronto Press.
Frantz, Donald G. 2009. Blackfoot Grammar, 2nd edition. Toronto, Canada: University of
Toronto Press.
Frantz, Donald G., and Norma J. Russell. 1995. Blackfoot Dictionary of Stems, Roots and Af-
fixes. Toronto, Canada: University of Toronto Press.
Josselin de Jong, JPB de. 1914. Blackfoot Texts from the Southern Piegans Blackfoot Reserva-
tion Teton County, Montana. Amsterdam: Johannes Müller.
Lewis, M. Paul, Gary F. Simons, and Charles D. Fennig (eds.). Ethnologue: Languages of the
World, Seventeenth edition, Dallas, Texas: SIL International.
Penner Toews, Carmela. 2011. Facebook status update: Quick survey for native english speakers.
http://www.facebook.com/CarmelaToews/posts/10150432832736335. Accessed: 10/28/2011.
Szabolcsi, Anna. 2010. Quantification. New York: Cambridge University Press.
Tailfeathers, Esther. 1993. Blackfoot for Beginners. Books 1, 2, and 3. Edmonton, Alberta:
ERIC.
Uhlenbeck, Christianus Cornelius. 1911. Original Blackfoot Texts. Amsterdam: Johannes
Müller.
Uhlenbeck, Christianus Cornelius. 1912. A New Series of Blackfoot Texts from the Southern
Peigans Blackfoot Reservation, Teton County, Montana: With the Help of Joseph Tatsey,
Collected and Pub. with an English Translation, Vol. 131. Amsterdam: Johannes Müller.
3
Documenting and Classifying Aspectual Classes

Across Languages
Leora Bar-el
1 Overview
Descriptions of aspectual categories cross-linguistically tend to focus on variation

in grammatical aspect (e.g., perfective/imperfective contrasts), as opposed to lex-
ical aspect (henceforth, aspectual classes).* Descriptions of individual languages
that address aspectual contrasts also tend to focus on grammatical aspect distinc-
tions, leaving aspectual classes unexamined. The lack of attention to aspectual
classes in the language documentation literature can likely be attributed to a com-
monly held assumption that the inventory of aspectual classes is universal (see Van
Valin 2006; Chelliah and de Reuse 2011 for this view). Recent research, however, re-
veals that the semantics of aspectual classes is subject to cross-linguistic variation.
As a result, fieldworkers who do not investigate aspectual classes are likely to miss
the potential variation in the aspectual classes of those languages. Fieldworkers
who do investigate aspectual classes tend to rely on standard tests for classification
(e.g., Dowty 1979). These tests, however, are language-specific, rely on metalinguis-
tic intuitions, and have also been claimed to give false results (see Walková 2013).
The goal of this chapter is to examine the commonly held assumptions about
aspectual classes and their reported cross-linguistic variation. The proposal put
forth is that fieldworkers need a toolkit that enables the researcher to document
and classify the full range of variation in the semantics of aspectual classes. In this
chapter I compile a range of documented variation from the typical Vendlerian
classes, not only for English, but cross-linguistically, in order to illustrate the types
of contrasts that need to be examined when conducting cross-linguistic research
*My sincere thanks to Ryan Bochnak and Lisa Matthewson for organizing the LSA Method-
ologies in Semantic Fieldwork panel and poster sessions, and for putting together this volume. The
number of people who have contributed to the thoughts represented in this chapter are too numerous
to mention here, but my sincere thanks to all of them. Any errors or misrepresentations are my own. 75

of this type. The fieldworker’s toolkit should include these documented contrasts
and suggestions of how to investigate them.
This chapter is structured as follows: section 2 presents the problem by exam-
ining the proposed universality of aspectual classes and the standard diagnostics
for classification, and demonstrating some of the problems with these tests. Section
3 provides a brief overview of how aspectual classes are (and are not) discussed in
the descriptive literature. Section 4 describes the two major questionnaires avail-
able for fieldworkers embarking on a study of aspect in a target language, and
the advantages and disadvantages of each. Examining the documented variations
in the behaviors of four aspectual classes cross-linguistically (activities, achieve-
ments, accomplishments and states), section 5 presents a sample of the features
that should be targeted in the fieldworker’s toolkit. Section 6 concludes this dis-
cussion by pointing to some unanswered questions and issues for further research.
2 The Problem
The claims of universality of aspectual classifications and the diagnostics used

to arrive at these categories suggest that documenting aspectual classes cross-
linguistically should be a relatively straightforward task. However, an examina-
tion of the problems with the standard tests for class membership, as well as the
limitations of the tests across predicates of the same class, suggests that the tools
that researchers rely on to investigate aspectual classes are problematic.
2.1 why ASPECTUAL CLASSES SEEM EASY TO CATEGORIZE ACROSS

LANGUAGES
Although the classification of aspectual classes dates back to Aristotle (see Rosen
1999 for discussion), perhaps the most influential inventory is attributed to Vendler
(1967), who proposes a four-way classification: activities (e.g., run, push a cart, stroll
in the park), achievements (e.g., win the race, find/lose something, die), accomplish-
ments (e.g., write a letter, cross the street, eat an apple), and states (e.g., be tired,
be dead, know the answer). Since this seminal work, Vendler’s inventory has been
modified and extended to other languages. Smith (1997) adds a fifth class, semelfac-
tives (e.g., knock on the door, sneeze, hiccup, flap a wing). Van Valin (2006) claims
that Vendler’s classification, with the addition of semelfactives, is universal.
The literature on aspect applies several standard tests for aspectual classes
(e.g., Vendler 1967; Dowty 1979; Smith 1997; Van Valin 2006, etc.), suggesting that
the classes should be easy to categorize cross-linguistically. Some example criteria
are given below:
(1) a. Occurs with the progressive

b. Occurs as the complement of stop/finish
Documenting and Classifying Aspectual Classes 77
c. Occurs with adverbs such as deliberately, carefully, vigorously, gently

d. Has habitual interpretation in the simple present
e. Occurs with ϕ for an hour; spend an hour ϕ-ing
f. Occurs with ϕ in an hour; take an hour to ϕ
The central problem with these tests is their lack of cross-linguistic application
(discussed in further detail in section 2.2).
The fieldwork literature also promotes the universality of aspectual classes.
For example, Chelliah and de Reuse (2011) point the fieldworker to a few refer-
ences on predicate classification and suggest that “the fieldworker should remain
cautious in distinguishing Aktionsart from derivational or inflectional aspect”
(p. 323). However, they also state that Vendler’s classification has “stood the test of
time” (p. 292). These directions suggest that data collection strategies to determine
aspectual classes in a given language should be straightforward.
There are at least two major tools available for fieldworkers examining as-
pectual contrasts as either a focus of a study, or as part of a larger documentation
project: Dahl’s (1985) tense and aspect questionnaire, and §2.1.3.3. of Comrie and
Smith’s (1977) questionnaire.1 The availability of such tools, which aim to present
all the possible contrasts that the fieldworker should look for, again suggests that
the categorization of aspectual classes across languages should be an easy task.
Unfortunately, this is not the case.
2.2 WHY ASPECTUAL CLASSES ARE NOT EASY TO CATEGORIZE

ACROSS LANGUAGES
Despite the proposed universality of aspectual classification, variation in the in-

ventory of the classes is widely observed. Two main problems that illustrate that
aspectual classes are not so easy to categorize are: (i) standard tests are not appli-
cable in all languages, and (ii) even within a language, the tests do not necessarily
behave the same across all predicates in a class.2
2.2.1 Limitations of Standard Tests Cross-Linguistically

While standard tests are used to classify predicates by their aspectual features,
these are not always cross-linguistically applicable. As Van Valin (2006) suggests,
it is necessary to adapt the tests to the features of the language under investi-
gation . . . one of the tests Dowty (1979) gives to differentiate states from non-
states is “has habitual interpretation in simple present tense,” which is clearly
an English-specific test.
1
These two questionnaires are discussed in further detail in section 4.
2
See also Walková (2013) for a discussion of how Dowty’s (1979) tests can give false results. She
suggests that some diagnostics test for features other than aspectual contrasts, and that classes are sub-
ject to aspectual shift and coercion.
The application of tests even within a language is subject to disagreement. For

example, the in x time/for x time contrast is a typical test used to distinguish activ-
ities (atelic predicates) from accomplishments (telic predicates). Activity predicates
(such as drink beer in (2a) below) are said to be compatible with the prepositional
phrase for x time, but not in x time. Accomplishment predicates (such as drank the
glass of beer in (2b) below), on the other hand, are said to be compatible with the
prepositional phrase in x time, but not for x time:
(2) a. Tom drank beer for/*in an hour.
b. Tom drank the glass of beer in/*for an hour. (Van Valin 2006:161)
Smith (1997) suggests that the starred readings above are “odd, or require a marked
interpretation.” (p. 41) While this may seem to be a difference in labels, the varying
descriptions of the results of the same test in the same language suggest that standard
tests are problematic. When examining these contrasts in other languages with lin-
guistically untrained speakers, the fieldworker is faced with the task of determining
which judgment the speaker has provided. Without a better understanding of how
to evaluate the results of these tests, cross-linguistic comparison may be flawed.3
Filip (2012) suggests that the in/for x time test is one of the three “most reli-
able and commonly used” diagnostics to distinguish telic from atelic predicates
in English. However, languages such as Skwxwú7mesh (a.k.a. Squamish; Central
Salish), as well as other Salish languages, and Dëne Sųłiné (Athapaskan) do not
exhibit the in x time/for x time contrast. In Dëne Sųłiné “there is only one general
time adverbial which covers both the “in an hour” and the “for an hour” meaning.
Therefore, the telicity test based on the ‘in/for an hour’ contrast also is not avail-
able” (Wilhelm 2003:73). Activities (3a) and accomplishments (3b) in Dëne Sųłiné
are both compatible with the postpositional phrase, and in some cases, the same
predicate yields both “in x time” and “for x time” interpretations:4
(3) a. ʔįɬá dziné ø dighitɬ’ís

one day P perf.1s.write
“I wrote for one day/in one day.” (adapted from Wilhelm 2003:74)
b. ʔįɬághe ts’udzáhi k’e bekatheɬt’e
one hour P perf.cook
“She cooked it in/for an hour.” (adapted from Wilhelm 2003:75)
3
Xiao and McEnery (2006) question the reliability and validity of the in/for x time telicity test in
English and Chinese corpus studies. They show, for example, that not all predicates that take the for x
time adverbial are atelic (i) and not all predicates that take the in x time adverbial are telic (ii):
(i) I stood and read the menu for a while, discovering it served mainly hamburgers.
(ii) He hadn’t worked in six months.
See also Cover and Tonhauser (this volume) for further discussion of the cross-linguistic application
of this test.
4
The interlinear glosses of the Dëne Sųłiné data have been simplified for expository purposes. The
reader is referred to Wilhelm (2003) for details. The following abbreviations are used in the Dëne Sųłiné data
presented here: Ø=omission of adposition, 1s=first person singular, O=object, P=adposition, perf=perfective.
Similarly, the contrast between “finish” and “stop” distinguishes telic predi-
cates (e.g., accomplishments) from atelic predicates (e.g., activities). In English,
telic predicates are compatible with both “finish” and “stop” (4a) whereas atelic
predicates are compatible with “stop” only (4b):
(4) a. Mary stopped/finished building the sandcastle.
(adapted from Smith 1997:41)
b. Mary stopped/?finished walking in the park.
(adapted from Smith 1997:43)
In Dëne Sųłiné, however, ʔanast’e, the closest equivalent to “stop”/ “finish”, is com-
patible with both activities (5a) and accomplishments (5b):5
(5) a. yasti ʔanast’e

1s.speak/talk/pray 1s.stop/finish
“I finished praying/speaking, I stopped praying/quit talking.”
b. yoh hoɬtsį ʔanat’e
house make.O stop/finish
“S/he made the house and finished it, s/he quit making the house.”
(adapted from Wilhelm 2003:77)
2.2.2 Limitations of the Tests Across All Predicates of a Class

In the literature on aspect, typical examples of predicates from Vendler’s classes are
shown to follow the patterns predicted by the aspectual tests. Little time is spent
accounting for predicates that either do not fall into the expected classes, or the
typical predicates that sometimes deviate from the behavior predicted by standard
tests.
Achievement predicates are notoriously difficult to classify. Typical examples
of achievements include win the race, leave the house, reach the summit. A com-
monly used diagnostic for achievement predicates is their incompatibility with
the progressive (due to their punctual/non-durative nature), as (6a) is meant to
illustrate. However, as Smith (1997) shows, some achievement predicates are com-
patible with the progressive (6b):
(6) a. *The firecracker is popping. (Van Valin 2006:159)6
b. He was winning the race. (Smith 1997:46)
5
Despite the fact that two of the standard tests for telicity do not apply in the language, Wilhelm
proposes that telicity is a relevant feature in Dëne Sųłiné. However, one needs to look elsewhere in the
language to find the contrast between telic and atelic predicates (e.g., k’ájëne “almost” induces contrast-
ing interpretations with activities and accomplishments). See Wilhelm (2003) for further discussion.
6
The ungrammaticality here does not take into account the acceptability of progressive achieve-
ments with plural subjects (e.g., The firecrackers are popping) (as discussed in Van Valin 2006), or with
a slow-motion interpretation (see Rothstein 2004 for discussion).
The fact that even in English there is disagreement regarding the compatibility of
some predicates in a given class with certain constructions suggests that the field-
worker cannot rely on translations of typical predicates in order to classify aspec-
tual classes in the language under investigation.
Despite the assumption that predicates in all languages can be categorized ac-
cording to Vendler’s classes,7 alternate classifications are reported. Tatevosov (2002)
compiles 16 reported classifications, many of which contain similar classes labeled
with different terminology. Although some inventories may vary from the tradi-
tional classes, Tatevosov suggests that they all rely on the assumption that these
classes are universal, and thus they have little to say about cross-linguistic variation.
Tatevosov motivates a parametric view of aspectual classes. He proposes a
10-way classification of “cross-linguistic actional types”, which he defines as a “ty-
pologically stable clustering of properties” (p. 326). Tatevosov’s classification, sum-
marized in (7) below, is based on an examination of four unrelated languages and
“can only be regarded as tentative, and must be tested against a wider-ranging
typological sample” (p. 376):
(7) a. Stative
b. Atelic
c. Strong telic
d. Weak telic
e. Punctual
f. Strong inceptive-stative
g. Weak inceptive-stative
h. Strong ingressive-atelic
i. Weak ingressive-atelic
j. Multiplicative (Tatevosov 2002:376)
Tatevosov suggests that classifying predicates in a given language are

linguistic facts that have to be recognized via investigation, and not to be
taken as a pre-theoretical assumption . . . Language-specific actional classes
are more numerous and their relations to each other are more complicated
than appears in light of the traditional classes of Vendler’s (p. 393–394)
While fieldworkers have some tools available to them for the investigation of
aspectual classes cross-linguistically, these tools often assume a universal set of
classes—they do not test for them (see section 4).
Finally, although aspectual class and grammatical aspect are independent
(Smith 1997), the circular nature of their interaction is problematic. Many diag-
nostics of aspectual classes depend on the identification of grammatical aspect
7
See Tatevosov (2002) for a list of language-specific studies.
contrasts (see also Tatevosov 2002 and references therein for discussion). For ex-
ample, compatibility of a predicate with the progressive construction is a diag-
nostic used to classify a predicate as an activity or accomplishment versus state
or achievement. However, determining compatibility, and thus categorization,
requires that the progressive construction has been identified in the language.
Dahl (1985) suggests that progressives are “normally used only of dynamic—that
is, non-stative—situations” (p. 93). Thus, using a test for non-stativity requires al-
ready having categorized a predicate as non-stative. A related issue is that while
(8a) is a typical example showing that states are incompatible with the progressive,
even in English, states can occur in the progressive (e.g., in non-standard registers
(8b), under slow-motion readings (8c), and adjectival predicates with volitional
readings (8d)):8
(8) a. *Kim was knowing the answer. (Smith 1997:40)
b. I’m loving the hot hue, the sweet, off-the-shoulder neckline. (glamour.
com)
c. Mary is spotting her archenemy at the party at the moment. (Rothstein
2004:56)
d. John is being silly.9
Examples such as these suggest that studies of aspectual classes that are based on a
limited number of predicates are at risk of missing a more comprehensive under-
standing of aspectual classification cross-linguistically.
3 Aspectual Classes in the Descriptive Literature
Reference grammars, common outcomes of language documentation projects,

rarely include discussion of aspectual classes. The reasons for this absence, how-
ever, are unclear. First, it may be the case that fieldworkers simply do not test for
aspectual classes; this would be unsurprising since the investigation of seman-
tic universals, as von Fintel and Matthewson (2008) point out, “is a particularly
young and understaffed discipline, which has only quite recently started to seri-
ously look at cross-linguistic variation and uniformity” (p. 140). Second, if aspec-
tual class distinctions are morphologically unmarked in the target language, these
contrasts may go unexamined in the language documentation literature, which
8
Dahl (1985) suggests that progressive statives have “special interpretations” (p. 28). That gener-
alization however, does not explain the extensive number of exceptions. See also Jóhannsdottír (2011)
for detailed discussion on English and Icelandic statives in the progressive. She argues that progressive
states are shifted states that have prototypical features of events. Fieldworkers must take into account
the possibility of a shifted state when examining stative predicates cross-linguistically.
9
Although Walkova (2013) suggests that sentences like this are not felicitous (e.g., ?? You are being
tall), the availability of examples like (8d) suggest that this is not true of all states.
often focuses on morphologically marked contrasts. Third, some fieldworkers may

simply assume a Vendlerian classification based on English translations. This pos-
sibility is confirmed by Ebert (1995), who suggests in her cross-linguistic study of
perfect-progressive ambiguity that “most often it is assumed that a verb or verb
phrase has the same actional character as its closest English counterpart” (p. 186).
Fourth, it may be the case that the results of diagnostics for predicate classes show
unclear categories; given the complications with standard tests outlined in section
2.2, and the fact that standard tests may be difficult to adapt, this too would be an
unsurprising result.
A cursory look at recent work in the University of California Publications Lin-
guistics series10 reveals that reference grammars often focus on grammatical aspect
alone, describing, for example, perfective/imperfective contrasts. When aspectual
classes are categorized, they are often based on morphological (rather than semantic)
tests. For example, Thompson et al. (2006) propose what they call a “semantically-
motivated” (p. 53) set of three verb classes in Wappo: the durative class, the impera-
tive class, and the infinitival class. Class membership is based on which form of the
durative, imperative, and infinitival suffixes can attach to the root. Each class has
multiple sub-types that, in some cases, relate to transitivity. However, this classifica-
tion seems to be based on morphological compatibility, as opposed to semantic tests.
Perhaps most surprisingly, research focused on aspect may lack analysis of as-
pectual classes. For example, Hintz (2011) is devoted to examining aspectual con-
trasts in South Conchucos Quechua. Yet, from the very onset he asserts that “[s]
ince the focus of this study is the grammatical expression of aspect rather than its
lexical expression, Aktionsarten in that sense will not be discussed further” (p. 4,
fn 1). This may be problematic given that the analysis of distinctions of grammat-
ical aspect is often based on the interplay of grammatical and lexical aspect. The
point here is that while reference grammars are invaluable resources for linguists
studying several different areas of the grammar, they cannot necessarily be relied
on to investigate aspectual classes in a target language. This in turn means that the
typological literature is also disadvantaged.
There are cases where aspectual classes are addressed overtly. For example,
Nichols (2011) discusses a five-way Aktionsart distinction in Ingush, though not
the typical Vendlerian classes: state, progressive, ingressive, punctual, and telic
(p. 319). Although she does outline the properties of each class, she suggests also
that “[m]uch work remains to be done on aktionsart” in the language (p. 319).
Thus, when fieldworkers do examine these contrasts in understudied languages,
they do not always result in a Vendlerian system. This is further evidence for the
need to investigate aspectual classes cross-linguistically so that we can more com-
prehensively understand these classifications.
10
This series focuses on descriptive research on lesser-known languages, especially Indigenous
languages of the Americas, and includes several reference grammars. It is included here as a sample of
the descriptive literature on understudied languages that results from fieldwork.
4 Tools for the Fieldworkers: Questionnaires on Aspectual Categories
In some cases, reference grammars that contain descriptions of aspectual catego-

ries are based on data collected by questionnaires designed to identify all pos-
sible distinctions in the target language. Two tools available that have at least a
main or partial focus on aspect are Comrie and Smith’s (1977) Lingua Descrip-
tive Studies Questionnaire and Dahl’s (1985) Tense and Aspect Systems Question-
naire. In this section we examine each and identify their respective advantages
and disadvantages.
4.1 THE LINGUA DESCRIPTIVE STUDIES QUESTIONNAIRE
Comrie and Smith’s (1977) Lingua Descriptive Questionnaire is “primarily de-

signed for grammar-writing, but with useful structural questions that should be
addressed in the field . . .[to assist] a field worker . . . in gathering the full range of
data in that area.” Its aim is to
provide descriptions of individual languages and language-groups, rang-
ing from descriptions of individual phenomena to overall descriptions,
arranged in such a way that the information on the language or language-
group concerned should be readily accessible to linguists working on lan-
guage universals, language typology, or comparative syntax, morphology, or
phonology (p. 1).
The Lingua questionnaire forms the basis for reference grammars published as
part of the North Holland, Croom Helm, and Routledge descriptive grammar
series. The advantage of this questionnaire is that it outlines a set of aspectual
features that are common cross-linguistically. Sample features are illustrated in (9)
below, from Comrie and Smith (1977:49-50):11
(9) 2.1.3.3.2.1. Which of the following, if any, are marked formally, either (a)
regularly for all verbs where applicable, (b) only for certain lexical items? . . .
2.1.3.3.2.1.9. semelfactive aspect (a single occurrence of a situation)
2.1.3.3.2.1.10. punctual aspect (a situation that is viewed as not being able to
be analyzed temporally, e.g. he coughed, referring to a single cough; contrast
perfective aspect, where the situation is not analyzed, although there is no
specification that it could not be)
2.1.3.3.2.1.11. durative aspect (a situation that is viewed as necessarily lasting
in time) . . .
11
The questionnaire’s section on aspect (§2.1.3.3.) also includes (i) a sub-section that asks “What
possibilities are there for combining different aspectual values?” (§2.1.3.3.2.2.1.), and (ii) a sub-section
that asks “Are there any restrictions on the combination of different aspectual values with the various
voices, tenses, moods, and finite/non-finite forms?” (§2.1.3.3.2.2.2.).
2.1.3.3.2.1.14. Is there any way of indicating overtly a situation that leads to a

logical conclusion (telic, accomplishment), as opposed to one that does not?
Thus English drink a gallon of water is telic (the action must come to an end
when the gallon of water is consumed, and may not come to an end before
then), and drink some water atelic (it can continue indefinitely, or stop at any
time), although English makes no overt distinction here. Is there a way of
indicating that the logical conclusion of a telic situation has been reached?
The disadvantages of this questionnaire are that (i) the questions do not distinguish
between lexical and grammatical contrasts, (ii) the focus is on overtly marked dis-
tinctions, and (iii) the fieldworker must develop his/her own diagnostics to target
these distinctions.
For some features in the questionnaire, grammatical aspect and lexical aspect
seem to overlap. For example, semelfactive aspect is described as “a single oc-
currence of a situation,” thus treating it as a grammatical category rather than a
possible inherent property associated with a predicate. The questionnaire refers
to punctual aspect as “a situation that is viewed as not being able to be analyzed
temporally,” and uses the typical semelfactive predicate in he coughed “referring to
a single cough” as an illustration. The challenge for the fieldworker is how to distin-
guish between punctual aspect and semelfactive aspect. Without further informa-
tion regarding how these features are distinguished cross-linguistically and thus,
the possible nature of the variation in the target language, the fieldworker may not
know what to test for in the target language.
For some features in the questionnaire, there is an implicit assumption about
aspectual classes. For example, progressive aspect is described as “continuous aspect
of a nonstative (dynamic) verb,” yet no definition of stativity or dynamicity is given.12
This may result in the fieldworker making assumptions about these classes based on
classifications of the metalanguage or his/her native language.
The questionnaire targets the telic/atelic distinction, a contrast crucial to an un-
derstanding of aspectual classification. However, the questionnaire asks only if this
contrast is indicated overtly.13 While English does not make a morphologically overt
distinction between telic drink and atelic drink, the telic/atelic distinction manifests
in other constructions in English (see section 5 of this chapter for examples). For a
comprehensive understanding of the telic/atelic distinction in the language under
investigation, the fieldworker needs to look further than overt contrasts.
A disadvantage of the Lingua questionnaire is its lack of suggestions of how
to examine these distinctions in the target language. Asking the untrained speaker
about meanings directly may provide clues, but those elicited meanings should not
12
See section 5.1 and section 5.4 of this chapter for some discussion of the features dynamic and
stative, respectively.
13
Given their discussion of drink a gallon of water versus drink some water below, I take “overtly”
here to mean morphologically marked.
themselves form the basis of an analysis (see Matthewson 2004 for further discus-
sion). For example, perfective aspect is described in the Lingua questionnaire as “a
situation viewed in its totality, without distinguishing beginning, middle, and end.”
The question remains, what sort of data does the fieldworker look for to determine
how perfective aspect works in the target language? When asking a linguistically
untrained speaker of English whether the sentence John ate an apple describes a
situation that is, or is not viewed in its totality, one might be faced with several
different responses. A brief survey resulted in the following four: (i) “I guess. He
finished eating it, but it doesn’t give any details about how he ate it,” (ii) “My first
response was that it does . . . but then I got curious about where he ate it . . . I think
it does tell the bare bones of the event,” (iii) “I understand that he ate a whole apple,
but you get an impression that something else happened next,” and (iv) “If I under-
stand what you mean by ‘viewed in its totality’ I would say yes, it does. I know who
was involved, what he was doing, and in this case, the object he was consuming.”
These responses are enlightening, but they do not necessarily reveal whether or
not a speaker’s understanding of “viewed in its totality” is the same as a linguist’s.
While the Lingua questionnaire is an invaluable resource for investigating
several features of a language’s grammar, it alone cannot be relied upon to com-
prehensively investigate aspectual classes in the target language. This suggests that
reference grammars based on this questionnaire likely do not describe the lan-
guage’s inventory of aspectual classes. Fieldworkers would benefit from a toolkit
that takes into account the range of cross-linguistic variation in aspectual classes
and contains a set of cross-linguistically viable diagnostics.
4.2 THE TENSE MOOD ASPECT QUESTIONNAIRE
Dahl’s (1985) cross-linguistic study of tense and aspect systems is based on the
results of the Tense Mood Aspect (TMA) questionnaire completed by native
speakers of 64 languages. Although Dahl’s TMA questionnaire is not designed or
promoted as a field tool, but instead forms the database for a typological study, its
structure lends itself to fieldwork. Dahl’s questionnaire has, in fact, been adopted
by fieldworkers exploring tense and aspect systems of other languages.
The main part of the questionnaire consists of 156 English sentences and rel-
evant contexts. Speakers are asked to translate the sentences into their native lan-
guage. The verbs (or auxiliary+adjective) are capitalized and uninflected so as to
minimize priming the translation in the target language (i.e., “we wanted to avoid
literal translations of English TMA categories” (p. 45)). Some sample sentences are
given in (10) below (contexts, not to be translated, are given in square brackets):
(10) a. [Talking about a house which the speaker saw for the first time yesterday
and doesn’t see now:] The house BE BIG
b. [Q: What your brother usually DO after breakfast? A:] He WRITE letters
c. Whatever you TELL him, he not ANSWER (Dahl 1985:198, 199, 201)
In some cases, subjects were linguists who were native speakers of the target lan-
guage. In other cases, subjects were untrained speakers working with a linguist.
Subjects were told to write the most natural translation of the sentence in their
native language. They were also told that if there were several equally natural
translations they should include all of them. If a translation was more syntactically
complex/periphrastic than the others, they were told to omit it.14
The main advantage of the TMA questionnaire is that it yields cross-linguistic
comparative data. Unlike the Lingua questionnaire, which primarily lists contrasts
to examine, Dahl’s questionnaire consists of sentences and contexts designed to
target linguistic features/contrasts. In addition to the problems with the TMA
questionnaire that Dahl describes (e.g., the use of culturally specific nouns, issues
with relying on translations, instructions were misunderstood), for the field-
worker investigating aspectual classes and not simply grammatical aspect, the
TMA questionnaire is not a comprehensive resource. The questionnaire assumes
a stative versus dynamic contrast, and notes that the border between the two fea-
tures varies, but importantly, the TMA questionnaire does not examine aspectual
classes themselves. Thus, while Dahl’s questionnaire may reveal some facts about
aspectual classes in a given language, it is not designed to examine those contrasts
and thus cannot be relied upon for that purpose.
5 Aspectual Class Variations and the Diagnostics Used to Test Them
A fieldworker’s toolkit for the investigation of aspectual classes should focus

on testing for the range of semantic variation that has been documented cross-
linguistically. The toolkit should build on other tools available to the fieldworker,
but also address the problems mentioned above (e.g., include specific contrasts to
target as well as the diagnostics to do so). The diagnostics included in the toolkit
should be informed by Matthewson’s (2004) semantic methodology framework
and include nonverbal stimuli (e.g., stills/images, video) which avoid translation
problems, yield comparable data, target subtle contrasts, and can be fun to use (see
Lüpke 2011 and Bar-el 2007 for discussion; see also examples from the Language
and Cognition Field Manuals and Stimulus Materials published by the Language
and Cognition Department at the Max Planck Institute for Psycholinguistics
http://fieldmanuals.mpi.nl). Video clips can be realistic and adaptable; editing can
be used to create a context and to focus on different parts of the event (see Bar-el
2009 for discussion). Crucially, the toolkit should reflect newly reported aspectual
14
Note that Matthewson (2004) cautions against asking speakers to translate sentences that are
ungrammatical in the contact language. While most subjects of the TMA questionnaire were linguists,
it is unclear how untrained subjects were asked to interpret these seemingly ungrammatical sentences
or how instructions were distributed to them.
contrasts to examine. In that sense, the toolkit is a work-in-progress and relies on

contributions from researchers.
In this section, I examine the typical descriptions of the Vendlerian four-way
classification. For each class I present some documented cross-linguistic varia-
tions in the behavior of predicates from these classes and the diagnostics used
to uncover them. These are the contrasts that fieldworkers should examine when
investigating aspectual classifications. In cases where nonverbal stimuli have been
used to target the contrast, or are proposed as a useful tool for examining the con-
trast, a description of the methodology is given. The results of informed fieldwork
on aspectual classes will move us closer to a comprehensive documentation of as-
pectual classes cross-linguistically.
5.1 ACTIVITIES
Activities are typically described as atelic dynamic events. Van Valin suggests
that dynamicity “relates to whether the state of affairs involves action or not”
(p. 157).15 Smith (1997) calls activities “processes that involve physical or mental
activity and consist entirely in the process” (p. 23). Different properties of activ-
ity predicates have been identified in the literature (e.g., lack of homogeneity,
variation in termination entailments/implicatures, semelfactive behaviour). In
some cases, predicates exhibiting these properties have been analyzed as activ-
ities, and in other cases, these predicates have been analyzed as a sub-class of
activities.
5.1.1 Homogeneity
Activities are often described as having the feature homogeneity (Vendler 1957,
among others), which gives rise to their entailment pattern (Smith 1997:23):
(11) If an Activity event A holds at interval I, then the process associated with
that event holds at all intervals of I, down to the intervals too small to
count as A.16
However, Smith herself notes that activities are not entirely homogeneous since
their “endpoints involve change to and from a state of rest” (p. 23). She also cites
Parsons (1990) who suggests that an entailment pattern similar to that expressed
in (11) above is questionable for activity events that have just started.
15
Reference to “action” in this description is problematic as we find predicates such as think about
in English which behaves as an activity (dynamic), and contrasts with think, which behaves as a state
(Smith 1997).
16
As Smith notes, “[t]he qualification of interval size is necessary because Activities cannot be
said to take place at vanishingly small intervals. Waltzing, chuckling, running, for instance, require
certain motions; at a small enough interval, a person may be lifting a foot but not running” (p. 23). See
Rothstein (2004) for a similar proposal.
That activity predicates may not, in fact, be homogenous is proposed by Bar-el

(2005) to account for inceptive interpretations of activity predicates. This proposal
is based on activities in Skwxwú7mesh, however, similar facts also hold for English
activity predicates. The inceptive interpretation of Skwxwú7mesh activities can be
seen by comparing the perfective and imperfective construction of the same pred-
icate when modified by a punctual clause. In (12a), the perfective activity induces
an inceptive (sequential) reading whereas the imperfective activity induces an on-
going (overlapping) reading (12b):17
(12) a. chen xaa-m kwi s-es tl’ik kwa John

1s.sg cry-intr det nom-3poss arrive det John
“I cried when John got here.”
b. chen wa xaa-m kwi s-es tl’ik kwa John
1s.sg imperf cry-intr det nom-3poss arrive det John
“I was crying when John got here.”
Speaker’s comments: “You say this if you had been crying for a while before.”
Bar-el (2005) argues that the inceptive reading is a result of the fact that the incep-
tion is part of the inherent meaning of the activity predicate. The punctual clause
modifies, but does not introduce, the inception of the event. Tatevosov (2002) sug-
gests that these predicates constitute a category separate from the atelic (activity)
class, namely the ingressive-atelic class.
Smith (1997) and Rothstein (2004) both acknowledge that activities in English
have inceptive interpretations when modified by punctually locating expressions/
clauses. In (13a) the running event begins at 9 p.m.; in (13b), the swimming event
and bell ringing event are sequential, not overlapping:
(13) a. John ran at 9 p.m. (Rothstein 2004:25)

b. Mary swam when the bell rang. (Smith 1997: 64)
Rothstein suggests that the inception is the “only privileged instantaneous event
available” for the activity (p. 25). Smith proposes that the predicate is interpreted
as inceptive due to the durative nature of the activity predicate. Neither account
explains why other durative predicates (e.g., accomplishments) are incompatible
with punctual clauses; i.e., why (14a) cannot be used to convey that John started
to write his dissertation at 5 o’clock last Tuesday, and why (14b) cannot be used to
convey that Mary started to paint a picture at midnight:
(14) a. ??John wrote his dissertation at 5 o’clock last Tuesday. (Bertinetto 2001)
b. #Mary painted a picture at midnight. (Rothstein 2004:25)
17
Past tense is not obligatorily marked in Skwxwú7mesh (see Bar-el et al. 2005). Some translations
of perfective/bare activity predicates in this context include “started to” or “began.”
The inconsistency in these data and the lack of an explanation for the generaliza-
tions may be responsible for the absence of punctual clauses as a test for dura-
tivity or (a)telicity in the literature. The fact that, according to Tatevosov (2002),
languages may differ with respect to whether the perfective form of an activity
predicate can describe the inception (e.g., Russian) or is ambiguous between an
inception and resultant process reading (e.g., Bagwalal, Tartar, and Mari), further
suggests that this feature must be explored cross-linguistically.
5.1.2 Termination Entailments/Implicatures

Smith (1997, 1999) argues that English activities have a termination entailment.
Perfective activities are interpreted as terminated events when expressed in isola-
tion, as in (15) below:
(15) a. Lily worked
b. The dancers rehearsed.
Evidence that the termination is an entailment and not an implicature is demon-
strated in examples such as (16) below in which activity sentences conjoined with
assertions of simple continuation are infelicitous:18
(16) a. #Lily worked and she may still be working. (Smith 1999:488)
b. #The dancers rehearsed and they may still be rehearsing.
(Smith 1999:488)
Wilhelm (2003) argues that activities in Dëne Sųłiné have a completion en-
tailment. As (17) illustrates, the event completion cannot be canceled without in-
ducing infelicity. Wilhelm suggests that speakers regularly judge these types of
sentences as “nonsense”:19
(17) yɑghiɬti #kúlú ʔɑnɑst’e-íle

perf.1s.talk/pray but 1s.finish-not
(“I talked/prayed [but I didn’t finish].”) (adapted from Wilhem 2003:61)
Bar-el (2005) proposes that, unlike English and Dëne Sųłiné, Skwxwú7mesh
activities have a termination implicature. This is illustrated by (18) below in which
a clause containing a bare activity predicate can be conjoined with a clause assert-
ing continuation of the event:
(18) na kw’eyilsh lha Mary i na7-xw wa kw’eyilsh

rl dance det Mary conj rl-still imperf dance
“Mary danced and she’s still dancing.” (Bar-el 2005:77)
18
According to Smith, if a new unit of activity is asserted, the conjunctions are acceptable:
(i) Lily worked and continued working after that without a break.
(ii) The dancers rehearsed and kept on rehearsing.
19
The translation is Wilhem’s suggestion of what the sentence might mean if it were felicitous.
The sentence in (19) below demonstrates that a clause containing a bare activity
predicate can be conjoined with a clause explicitly canceling the inference of the
event termination:
(19) na t’ich-im lha Mary welh haw k-as i huy

rl swim-intr det Mary conj neg irr-3cnj part finish
“Mary swam but she didn’t finish swimming.” (Bar-el 2005:76)
Bar-el argues that the termination implicature is evidence that activities in

Skwxwu7mesh lack final points. Wilhelm argues that the culmination entail-
ment of activities is a property of the perfective in Dëne Sųłiné (the perfective
has a post-time that induces a culmination entailment), not a property of the
activities themselves. Regardless of the preferred analysis of this variation, the
fact remains that the termination/culmination entailments of activity predi-
cates can vary cross-linguistically, thus there is a need to test for these proper-
ties overtly.
5.1.3 Semelfactives
A subset of activity predicates can involve either single events or iterated events.
The sentence in (20) below could be interpreted as a single knock or a series of
knocks:
(20) John knocked on the door
= one tap
= a series of taps
These predicates are categorized as semelfactives in some frameworks (e.g., Smith
1997). Rothstein (2004) suggests that these predicates are “used to denote single
instances of events usually considered to be activities” (p. 184).
Targeting single versus iterative readings of this sub-class of activities in field-
work can be aided by the use of performance or nonverbal stimuli, such as video
clips. Short videos can present, for example, a single tap on a door or a series of
taps in appropriate contexts. Speakers can be asked for descriptions of the scenes,
or for felicity judgments of sentences in the target language without having to rely
on translation from the metalanguage.
One could argue that it would be unusual for the sentence in (20) to be in-
terpreted as a single knock. Levin (1999) suggests that some verbs are necessarily
“durative” (e.g., batter, beat), whereas others are ambiguous between the two pos-
sible readings (e.g., hit, kick, slap). Levin also suggests that this behavior of activity
predicates may not be enough to warrant a separate aspectual class. Regardless of
whether these types of predicates are treated as part of the class of semelfactives
or as a sub-type of activity predicates, examining these types of predicates must be
part any study of aspectual classes.
5.2 ACHIEVEMENTS
Achievement predicates are typically described as (near-)instantaneous punctual

events that result in a change of state. Van Valin (2006) suggests that the feature punc-
tual “differentiates events with internal duration from those which lack it” (p. 157).
Examining “die” verbs in 18 languages, Botne (2003) argues that the descrip-
tion of achievements as constituting culminations with no duration is incorrect.
He proposes a typology of four types of achievement predicates, given in (21)
below:
(21) a. Acute: nucleus alone

b. Inceptive: nucleus plus onset
c. Resultative: nucleus plus coda
d. Transitional: nucleus plus onset and coda
Acute achievements have no perceived temporal duration; inceptive achievements

include a potential durative onset phrase; resultative achievements encode a post-
liminary stage or ensuing state; transitional achievements encode all three phases.
Botne proposes that this range of verb types for “die” predicates can be extended
to all achievement predicates and that a language may exhibit achievements of
different classes. For example, he argues that English has acute achievements (e.g.,
notify, find), inceptive achievements (e.g., die, reach), and resultative achievements
(recognize), but no transitional achievements.
According to Botne, t’a “die” in Assiniboine (Siouan) is of the acute class in
that it encodes only the punctual nucleus of the event, as in (22a), where it can only
refer to a point in time. T’a is incompatible with the progressive suffix -hã (22b), or
the continuative suffix -ga (22c), which Botne suggests is evidence for an inceptive
phase. Events with a durative nucleus in Assiniboine (such as mani “walk”) are
compatible with these suffixes. For t’a to get the onset reading, the auxiliary aya
“become” must be added (22d), which is incompatible with mani “walk.” The sta-
tive meaning (being dead) is unavailable for t’a; this suggests that the verb is not of
the resultative type. To induce the stative meaning, the event must be described as
having occurred relative to another time (22e):
(22) a. T’a
“He dies/(has) died” (Botne 2003:269)
b. *T’a-hã
die-prog (Botne 2003:270)
c. *T’aa-ga
die-cont (Botne 2003:270)
d. T’a aya
die become
“He is becoming dead.” (Botne 2003:269)
e. Wanãgash t’a
long.ago die
“He died long ago.” (Botne 2003:270)
The verb muut in Egyptian Arabic is considered a member of the inceptive

class since it can readily occur with the progressive (23a), as do non-achievement
verbs such as qra “read.” The onset phase of muut can occur in the past without
leading to completion (23b):
(23) a. Bi-y-muut
prog-3sg.m.pres-die
“He is dying.” (Botne 2003:243)
b. Kaan bi-y-muut bas il-amaliyya ʔanaqaz-it-u
3sg.m.was prog-3sg.m.pres-die but def-operation save-3sg.f-3sg.m
“He was dying, but the operation saved him.” (Botne 2003:244)
The verb ta:y in Thai is considered a member of the resultative class. ta:y does
not have an onset phrase, and thus cannot co-occur with the progressive, as illus-
trated in (24):20
(24) *Ta: bun kamlaŋ ta:y

Grandfather Boon prog die (Botne 2003:254)
Crucially, ta:y does encode a stative coda such that the predicate can be interpreted
as referring to either the moment of dying or the state of death, as in (25):
(25) Khaw ta:y ma: sɔ:ŋ dɯan lœ:w

3.person die come two month already
i. “He died two months ago.”
ii. “He has been dead for two months.” (Botne 2003:254)
Akan (a Kwa language) is a member of the transitional class in that achieve-

ments can encode all potential phases: onset, nucleus, and coda. This is illustrated
in (26) below where wu can occur in the progressive to mark the onset phase (26a),
and can also refer to the nucleus or the stative reading (26b):
(26) a. ɔ-re-wu
3sg-prog-die
“S/he is dying (imminent).” (Botne 2003:267)
20
Tones have been omitted in these examples.
b. W-a-wu
3sg-pf-die
“S/he has died” or “s/he is dead.” (Botne 2003:267)
Examining only a few predicates in a study of achievements in a given lan-

guage may not provide a fieldworker with a comprehensive view of this aspectual
class. Although (in)compatibility with the progressive is a standard test for as-
pectual classification, an examination of the range of possible interpretations of
a given predicate must be conducted as well. Botne’s cross-linguistic study of the
verb “die” provides further evidence that translations of predicates may not behave
the same way across languages, and thus, may not be categorized in the same class
cross-linguistically. As Botne suggests “the same ‘concept’ will not necessarily
be encoded with the same phrases in every language. Consequently, appropriate
cross-linguistic comparison and analysis of these kinds of verbs will perforce re-
quire a close analysis of a particular verb in each language” (p. 276).
5.3 ACCOMPLISHMENTS
Accomplishment predicates are typically described as dynamic telic events that

have a natural endpoint. Cross-linguistic research on accomplishments reveals
that the culmination of accomplishment predicates is not an entailment across
languages, and that the status of culmination entailments/implicatures is, in some
languages, due to the nature of the object of the predicate. As well, variation in
the presence of an initial point in the meaning of accomplishments motivates
the need for a more detailed understanding of the nature of these predicates.
This understanding can only arise if more comprehensive cross-linguistic data
are available.
5.3.1 Culmination Entailment versus Implicature

Smith (1997) argues that perfective accomplishments in English (27) entail
culmination:
(27) Mrs Ramsey wrote a letter. (Smith 1997:67)
The culmination of the accomplishment event is argued to be an entailment rather

that an implicature. A clause containing a perfective accomplishment conjoined
with a clause canceling the culmination (28a) or with a clause asserting the con-
tinuation of the event (28b) is infelicitous:
(28) a. # Mrs Ramsey wrote a letter, but she didn’t finish writing it.
(Smith 1997:68)
b. # Mrs Ramsey wrote a letter and she may still be writing it.
(Smith 1997:67)
Bar-el et al. (2005) propose that accomplishments in Salish have culmination

implicatures, not entailments. In out-of-the-blue contexts, and in the absence of
overt tense and aspect morphology, accomplishment predicates are interpreted as
having culminated. This is illustrated for Skwxwú7mesh (Central Salish) in (29a)
and St’át’imcets (Interior Salish) in (29b):
(29) a. na xel’-t-as ta sxwexwiy’am’ lha Mary

rl write-tr-3erg det story det Mary
“Mary wrote a story.”
Speaker’s comments: “She wrote it . . . she finished.” (Bar-el et al. 2005)
b. ts’áqw-an’-lhkan ta n-kíks-a
eat-tr-1sg.su det 1sg.poss-cake-det
“I ate my cake.”
Speaker’s comments: “Sounds like you ate all of it.” (Bar-el et al. 2005)
As the data in (30) below show, these culminations are implicatures, not en-
tailments. When the culmination is explicitly canceled, it does not result in a
contradiction:21
(30) a. na p’ayak-ant-as ta John ta snexwilh-s welh haw k-as

rl heal-tr-3erg det John det canoe-3poss conj neg irr-3cnj
7i huy-nexw-as
part finish-lc-3erg
“He fixed his canoe but he didn’t finish (fixing) it.” (Skwxwú7mesh)
b. ts’áqw-an’-lhkan ti n-kíks-a lhkúnsa ku sq’it,

eat-tr-1sg.su det 1sg.poss-cake-det now det day
t’u7 qelh-cál-lhkan ku k’wík’wena7 t’u natcw
but save-act-1sg.su det few until tomorrow
“I ate my cake today, but I saved a little for tomorrow.” (St’át’imcets)
Culmination implicatures are associated with control transitives in Salish

languages. (In)Transitive morphology in Salish encodes not only valency but the
degree of control that the agent has over the event: control transitives are predi-
cates in which the agent has full control over the event, while non-control or
limited-control transitives are predicates in which the agent does not have full
21
Although the English translations are contradictory (thus, infelicitous), the Salish sentences are
felicitous. Note further that the first clause in each sentence is not interpreted as an imperfective event.
Salish predicates are perfective if not overtly marked as imperfective; a progressive or imperfective
morpheme is required for imperfective readings of accomplishment predicates (see Bar-el 2005 and
Bar-el et al. 2005 for further discussion).
control over the event.22 The latter induce interpretations such as “accidentally”
and “managed to.” Both control and non-control/limited control transitives are
classified as accomplishments. Kiyota (2008) proposes that while control tran-
sitives in Sənčáθən (Central Salish) have culmination implicatures, non-control
transitives in Sənčáθən have culmination entailments. This is demonstrated in (31)
below where canceling the culmination is infelicitous:
(31) # ləʔə sən kwəʔ lé-nəxw tsə latem ʔiʔ ʔawa sən šəq-naxw
aux 1.sg inf get.fixed-nctr d table acc neg 1.sg complete-nctr
“I (managed to) fix the table, but I didn’t finish it.”
Jacobs (2011) argues that limited control accomplishment predicates in

Skwxwú7mesh entail culmination since the culmination of predicates marked
with the limited-control transitivizer cannot be denied without inducing a contra-
diction, as illustrated in (32) below:
(32) chen p’ayak-nexw-Ø ta tetxwem #welh haw k’-an i

1s.sg fix-lctr-3obj det car conj neg sbj-1s.conj pres
húy-nexw-Ø
finish-lctr-3obj
“I fixed the car #but I didn’t finish it.” (Jacobs 2011:124)
Singh (1998) shows that in Hindi, perfective accomplishments denoted by

simple verbs need not culminate in certain contexts;23 for example, if the object
is not totally affected by the event denoted by the predicate (33a), or if the object
comes into existence as a result of the event denoted by the predicate (33b):
(33) a. us ne ciTThii paRhii par puurii nahii kii

he erg letter read-perf but complete neg de-perf
“He read a letter but did not complete it.” (Singh 1998:193)
b. miiraa ne sleT par cakkar banaayaa par puuraa nahii
Mira erg slate on circle make-perf but completely neg
banaayaa
make-perf
“Mira drew a circle on the slate but did not complete it.” (Singh 1998:193)
22
See Jacobs (2011) for detailed discussion of control in Salish.
23
Simple verbs are contrasted with compound verbs, such as khaa liye ‘eat take.perf’, which add
subtle meanings such as intentions of the actor, positive or negative attitude, and so forth. Compound
verbs entail culmination in all contexts (see Singh 1998 for further discussion).
Hindi provides further evidence that languages may not categorize all accomplish-
ment predicates in the same way. Thus, the properties that may potentially affect
the culmination entailments of accomplishment predicates must be explored
cross-linguistically.24
With respect to elicitation strategies, culmination implicature/entailment is
another feature of aspectual classes where nonverbal stimuli such as video can
aid in exploring contrasts. The advantage of video here is that events that take
place over long stretches of time can be shown with time lapses. Furthermore, the
contexts in which events take place can also be shown. For example, to determine
whether an accomplishment entails culmination, judgments of simple accomplish-
ment clauses according to scenes in which an agent crosses a bridge (see Clip 3.1a
at www.oup.com/us/methodologiesinsemanticfieldwork), and scenes in which an
agent only gets half-way across the bridge (see Clip 3.1b), can target the culmina-
tion contrast without relying on translation from the metalanguage.25 The clips can
be used by the fieldworker in different ways. For example, the fieldworker could
play Clips 3.1a and 3.1b and ask the speaker whether a sentence containing the per-
fective form of the predicate cross the bridge (the equivalent of the English sentence
John crossed the bridge) is felicitous in the context provided in the clip (i.e., whether
it is an appropriate description of the event in the clip). For English, we expect that
the sentence John crossed the bridge would be compatible with Clip 3.1a but not
Clip 3.1b. For Skwxwú7mesh, we would expect that the perfective accomplishment
marked with a control transitivizer would be compatible with both clips, but the
perfective accomplishment marked with a limited control transitivizer would only
be compatible with Clip 3.1a, as in English.26
Non-culminating accomplishments are attested in several languages, in-
cluding Japanese, Mandarin, Malagasy, Tagalog (see Travis 2010, and references
therein).27 As in Salish languages, the culmination of an accomplishment predicate
is strongly implied even though it is not entailed (see also Tatevosov 2002 for a dis-
cussion of “weak-telic” predicates and the languages that exhibit or lack this class
of predicates). The growing body of research that reveals that non-culminating
accomplishments are not rare suggests that this is an area where more cross-
linguistic work is necessary. Furthermore, any study of aspectual classes in a given
language should explore the status of culmination implicatures/entailments.
24
Arunachalam and Kothari (2010) examine simple and complex verbs in Hindi via experimenta-
tion. They conclude that the contrast between the two verb types cannot be reduced to the status of
their objects alone, but that pragmatics plays a greater role in determining culmination.
25
See Arunachalam and Kothari (2011) for some discussion of possible limitations arising from
using video stimuli to explore culmination.
26
Since events marked by Salish limited control/non-control transitivizers are also interpreted as
being accidental or requiring difficulty, fieldworkers might need other video clips that target the addi-
tional meanings.
27
See also Hay et al. (1999) for a discussion of non-culminating verbs of consumption, creation,
and motion in English.
5.3.2 Role of the Object

Other studies that focus on culmination patterns of accomplishments examine the
role of objects in determining implicatures versus culminations.
An examination of accomplishments in Mandarin, Soh and Kuo (2005) show
that different classes of objects of creation induce culmination requirements. Some
predicates allow for partial objects (e.g., build a house, bake a cake, draw a circle,
write a word) and others do not (e.g., write a letter, draw a picture). Moreover,
Mandarin predicates with numeral objects must culminate (34a) while predicates
with demonstrative objects need not culminate (34b):
(34) a. # Ta chi-le liang-ge dangao, keshi mei chi-wan.

he eat-LE two-Cl cake but not eat-finish
(“He ate two cakes, but he did not finish eating them.”)
b. Ta chi-le na-ge dangao, keshi mei chi-wan.
he eat-LE that-Cl cake but not eat-finish
“He ate that cake, but he did not finish eating it.”
(adapted from Soh and Kuo 2005:204)
As suggested in section 5.3.1, the use of nonverbal stimuli such as video can aid in
testing for culmination entailments versus implicatures. Determining whether the
object plays a role in the culmination requirement is no exception. For example,
to establish whether the object in predicates such as peel a banana is considered
totally affected by the event, judgments of clauses according to a scene in which
a banana is only partially peeled (see Clip 3.2a), and a scene in which a banana is
completely peeled (see Clip 3.2b), can once again target the contrast without rely-
ing on translation from the metalanguage. As discussed for Clips 3.1a-b above,
the fieldworker could show Clips 3.2a-b and ask the speaker whether a sentence
containing the perfective form of the predicate peel a banana is felicitous in the
context provided in the clip (i.e., whether it is an appropriate description of the
event in the clip). For English, we would expect that the sentence John peeled a
banana is only compatible with Clip 3.2b. For Mandarin, we might expect that
the perfective form of the predicate with a definite object, as in John peeled that
banana, would be compatible with the contexts provided by Clips 3.2a and 3.2b.
However, we might also expect that the perfective form of the predicate with a
numeral object, as in John peeled one banana, would only be compatible with the
context provided in Clip 3.2b.28
28
Note that video clips can be used in a variety of ways. For example, a clip can be played in its
entirely to reveal to the speaker the entire event from beginning to end. Then the clip could be played
again, stopping the clip at certain points to ask the speaker for felicity judgments. The researcher could
also ask questions while the clip is playing in order to target different types of interpretations (e.g.,
events in progress).
5.3.3 Inclusion of the Inception

While activity predicates can be modified by punctual clauses to induce an in-
ceptive reading (see section 5.1.1), Rothstein (2004) suggests that accomplishment
predicates are incompatible with punctual clauses, and are not interpreted as in-
ceptive. This is repeated in (35) below:
(35) a. ??John wrote his dissertation at 5 o’clock last Tuesday. (Bertinetto 2001)
b. #Mary painted a picture at midnight. (Rothstein 2004:25)
This generalization is inconsistently reported in the literature. For example, Smith
(1997) suggests that durative verbs have inceptive interpretations with momentary
adverbials, as in (36):
(36) They ate dinner at noon. (Smith 1997:42)
However, since activities and accomplishments are both durative in Smith’s frame-
work, it is unclear whether (36) is considered an activity (with a mass noun object)
or an accomplishment.
Terry (2004) suggests that only events that can “reasonably occur during a
very short period of time” are possible answers to a question containing an event
such as enter a room, as in (37) below:
(37) What happened while Esther was entering the room?
a. ? Eugene ate the cake.
b. ?? Eugene wrote his dissertation. (Terry 2004:77)
Terry suggests that an inception or culmination interpretation is not available for
these sentences: “[t]he entire cake-eating event must be contained within the topic
time.” (p. 77).
Bar-el (2005) examines accomplishment predicates modified by punctual
adverbs/clauses in both English and Skwxwú7mesh. Native English speakers were
asked to judge the felicity of sentences in (38) below:
(38) a. Mary ate an apple at 1:00 p.m.
b. Mary fixed the car at 1:00 p.m.
c. Mary wove a blanket at 1:00 p.m.
For most speakers, the “entire event” reading is the preferred reading: “sounds
like she wove the entire blanket at 1.” Felicity judgments provided by the speakers
varied; for example, (38a) was sometimes judged felicitous because it “doesn’t take
long to eat an apple.” A similar explanation was provided for (38b) where a speaker
suggested that it “could simply be a loose bolt to tighten.” But this explanation was
not offered for (38c), since “weaving cannot be done in that minute.”29
29
A generalization which, to my knowledge, is as of yet unexplored, is the availability of an inceptive
reading of perfective accomplishment predicates modified by a punctual clause/adverbial in a list context:
(i) Q: What did you do yesterday?
(ii) A: I wrote a letter at 1, washed the car at 3, cooked dinner at 5 . . .
Examining similar data in Skwxwú7mesh, Bar-el (2005) shows that

accomplishments modified by punctual clauses/adverbials are compatible
with inceptive, medial, and culminating contexts (variation is observed across
speakers and predicates). It is as of yet unclear whether there is a preferred
interpretation:
(39) na xel’-t-as ta sxwexwiy’am’ kwa John na7 t-kwi an’us-k

rl write-tr-3erg det story det John loc obl-det two-o’clock
“John wrote the story at 2 o’clock.”
✓Context: inceptive (Figure 3.1a)
✓Context: medial (Figure 3.1b)
✓Context: final (Figure 3.1c)
Figure 3.1a Inceptive context

Figure 3.1b Medial context
Figure 3.1c Final context

I showed the pictures to each speaker and asked whether I could use the
Skwxwú7mesh sentence in (39) to describe them. This sentence was compatible
with each picture. However, this was not the case for other accomplishment predi-
cates and corresponding picture sets, such as climb a mountain and wash the floor,
once again, pointing to the fact that several predicates need to be tested.30 That
Skwxwú7mesh and English differ in this respect, and that there is disagreement
on the status of these types of sentences within each language suggests that these
properties of accomplishment predicates need to be explored in greater detail not
only cross-linguistically, but in English as well.
5.4 STATES
States are typically described as static unbounded situations. Van Valin (2006)
suggests that the feature ±static distinguishes between verbs that denote a “hap-
pening” from verbs that denote a “non-happening.” Several variations of the typi-
cal stative predicate have been proposed. The variations are dependent on whether
or not the entry into the state is part of the meaning of the state. Here we examine
different proposals of this variation across several languages.
5.4.1 A Typology of Stative Predicates

Tatevosov (2002) argues for three different types of stative-like predicates cross-
linguistically: (i) stative (predicates that can only refer to a state, not a change of
state (40a)), (ii) strong inceptive-stative (predicates that have change of state inter-
pretations (40b)), and (iii) weak inceptive-stative (predicates that are ambiguous
between the two (40c)). This is illustrated for Tatar (Turkic) below:31
(40) a. daut gɤlnara kit-ü-e-nä šatlan-dɤ

Daut Gulnara come-nmn-3-dat be.glad-pst
“Daut was glad / *came to be glad that Gulnara has come.”
(Tatevosov 2002:376)
b. daut rašit-nɤ kür-de
Daut Rashid-acc see-pst
“Daut caught sight of Rashid.” (Tatevosov 2002:383)
c. daut bu burɤč-nɤ aŋna-dɤ
Daut this task-acc understand-pst
(i) “Daut came to understand this task.”
(ii) “Daut understood this task (for some time).” (Tatevosov 2002:383)
30
Note that the video clips described in section 5.3.1 and section 5.3.2 above could easily be used
to examine the different parts of the event here.
31
Translations are the same as those given in Tatevosov (2002).
5.4.2 Individual-Level/Stage-Level Distinction
A well-known distinction among predicates is the individual-level (I-L) and stage-
level (S-L) contrast (Carlson 1977). Traditional aspectual classifications do not take
this contrast into account. Bar-el (2005) argues that S-L states form an aspectual
class in Skwxwú7mesh that is distinct from I-L states. S-L states in Skwxwú7mesh
are inchoative states. In the perfective, they yield both inchoative readings and
stative readings, as illustrated in (41)) below:
(41) chen kw’ay’

1s.sg hungry
(i) “I got hungry.”
(ii) “I am hungry.” (Bar-el 2005:171)
When modified by a punctual clause, S-L inchoative states in Skwxwú7mesh have

inchoative readings:
(42) chen kw’ay’ kwi-n-s na kw’ach-nexw-an ta sch’exwk

1s.sg hungry det-1poss-nom rl see-tr(lc)-1s.sg det fried.food
“I got hungry when I saw the bannock.” (Bar-el 2005:171)
Skwxwú7mesh inchoative states would be categorized as weak inceptive-statives

in Tatevosov’s (2002) classification since they have “entry into a state” and “state”
interpretations in their perfective forms and “state” interpretations in their imper-
fective forms.
Kiyota (2008) shows similar facts for Sənčáθən. Inchoative states yield past
inchoative or current state readings, as in (43)) below:
(43) k’wey sən

get.hungry 1sg
“I’m hungry / I got hungry.” (Kiyota 2008:248)
Kiyota argues that Sənčáθən also has a class of homogenous (I-L) states, which have
neither an inherent initial nor final point. When modified by a punctual clause,
Sənčáθən inchoative states only allow the inchoative reading (44a)), whereas ho-
mogenous states are infelicitous when modified by a punctual clause (44b)):
(44) a. t’ečəq’ tə Jack kws kwɬ təl-nəxw-s tə sqwəl’qwəl’

mad d Jack sub perf hear-nctr-3.sg d news
“Jack was mad when he heard the news.”
√Jack was not mad before, but he became mad because of the news
*Jack was already mad when he heard the news (Kiyota 2008:42)
b. #k’wámk’wəm tə Jack kws kwɬ kw�n-ət-əŋ-s tə stél’ŋəxws

strong d Jack sub perf take-ctr-pass-3.sg d medicine
“Jack felt strong when he took the medicine.” (Kiyota 2008:43)
Kiyota extends his analysis to Japanese and proposes that its inventory of aspec-
tual classes consists of both homogenous states and inchoative states. These ex-
amples show that the S-L/I-L distinction may be relevant for the class of states
cross-linguistically.
5.4.3 Inchoative Statives versus Inchoative Achievements

Examining Spanish reflexive psychological verbs, Marín and McNally (2011) argue
that verbs of this class (such as aburrirse “to be/become bored” and enfadarse
“to become angry”) are inchoative, but do not denote a change of state, nor are
they dynamic. Instead, they suggest that these verbs constitute two classes: stative
predicates that refer to the left boundary (aburrirse class) and atelic achievement
predicates (enfadarse class), which, contrary to telic achievements, do not denote a
right boundary. Aburrirse verbs in the progressive are interpreted as stative (45a),
while enfadarse verbs in the progressive are not yet in the state (45b):
(45) a. Juan se está aburriendo

Juan SE is boring
“Juan is (already) bored.”
b. El perro se está asustando
the dog SE is frightening
“The dog is getting (but is not yet) frightened.”
(Marín and McNally 2011:475)
Aburrirse predicates behave like typical statives in that they have nonhabitual in-
terpretations in the non-progressive present (46a), while enfadarse predicates only
allow generic or habitual interpretations (46b):
(46) a. Con su hermana pequeña se aburre

with his sister little SE bores
“With his little sister he is bored (now).” (Marín and McNally 2011:485)
b. Se asusta con los fuegos artificiales
SE frighten with the fires artificial
“He is (generally) frightened by fireworks.”
Not: “He is frightened (now) by the fireworks.”
(Marín and McNally 2011:485)
Although they use some standard tests of aspectual classification, Marín and Mc-
Nally demonstrate that a more fine-grained analysis of inchoative predicates is
needed in order to fully comprehend the inventory of aspectual classes in Spanish.

Their study reveals that the behavior of these types of predicates need to be exam-
ined cross-linguistically.
5.4.4 Property Concepts versus Result States

Koontz-Garboden (2005) examines the contrast between states that denote prop-
erty concepts (“states related to speed, age, dimension, color, value, etc. and that
presuppose no prior event” p. 86) and those that denote result states (“the result
of some action” p. 86). These two types of states have different entailments: prop-
erty concepts do not entail a prior event while result states do entail a prior event.
These entailment patterns are illustrated in (47–48) below. A clause containing
a property concept can be conjoined with a clause that explicitly denies a prior
event leading up to the state (47). However, a clause containing a resultant state
conjoined with a clause explicitly denying the prior event leading to the resultant
state is infelicitous (48):
(47) a. The dirt is red, but nobody reddened it.

b. Mount Chimborazo is tall and has always been so.
(48) a. # The glass is broken, but it never broke.

b. # Sandy is dressed, but neither she nor anyone else dressed her.
(Koontz-Garboden 2005:86)
Without considering their entailment patterns, these predicates could both be

considered Vendlerian states. However, taking into account their entailment pat-
terns, these predicates could be categorized in separate classes: property concepts
are “true” states, while result states pattern with achievements.
6 Conclusions, Implications, and Further Issues
While the view of the universality of aspectual classes persists in the literature, a
growing body of research reveals that the semantics of aspectual classes is sub-
ject to cross-linguistic variation. Several standard tests used to distinguish among
the classes are language-specific, and thus not applicable cross-linguistically. This
chapter has proposed that fieldworkers need a comprehensive resource for the in-
vestigation of aspectual classes. This toolkit would enable the researcher to docu-
ment and classify the full range of variation in the semantics of aspectual classes
in a target language. Taking the Vendlerian classes as a starting point, this chapter
presented a range of documented variations among these classes, as well as the di-
agnostics used to reveal these variations, in order to illustrate the types of contrasts
that need to be examined when conducting cross-linguistic research on aspectual
classes. A summary of the classes and some sample variations are given in Table 3.1.
Table 3.1
Summary: Aspectual classes and their documented variations
Vender’s classes Sample definitions Typical examples Sample variations
Activities Atelic dynamic events Run, push a cart, • Lack of homogeneity

stroll in the park • Termination entailment vs.
implicature
• Semelfactives as a part of
the same/separate class
Achievements Instantaneous/ Win the race, find/ • Botne’s (2003) typology:
near-instantaneous lose something, die - Acute
punctual events - Inceptive
- Resultative
- Transitional
Accomplishments Dynamic telic events Write a letter, cross •C ulmination entailment vs.
that have a natural the street, eat an implicature, based on:
endpoint apple - Predicate features
- Object features
• Inclusion of the inception
States Static unbounded Be tired, be dead, • Tatevosov’s (2002) typology:
situations know the answer - Stative
- Strong inceptive-stative
- Weak inceptive-stative
• I-L/S-L distinction
• Inchoative statives vs.
inchoative achievements
• Entailment of a prior event
The remainder of this section outlines some implications of this discussion and
some issues for further research.
6.1 UNIVERSALITY AND VARIATION
The Vendlerian four-way aspectual class contrast is subject to variation not only
in English, upon which the system is based, but cross-linguistically. The challenge
facing proponents of the universality of aspectual classes is accounting for this
range of variation. As von Fintel and Matthewson (2008) and Tatevosov (2002)
both suggest, what may be universal is an inventory of building blocks that lan-
guages use to construct aspectual classes.
A question that researchers are starting to ask and answer is where the var-
iation lies. It may be the case that it is the inventory of classes itself which is
subject to variation (as in Tatevosov’s 2002 catalog of 10 cross-linguistic actional
types), or it may be the case that the meanings of the aspectual classes are subject
to variation (as in Botne’s 2003 description of achievements that he claims can
be of different types, but are still achievements, based on their punctual, cul-
minative nature (p. 276)). This is a theoretically rather than empirically driven
question and will require further exploration of what minimally constitutes an
aspectual class.
6.2 NEGATIVE EVIDENCE
The importance of negative evidence in documenting aspectual classification

cannot be overestimated. Matthewson (2004) argues that textual materials lack
negative evidence (and also have a dearth of positive evidence). In an opposing
view, Himmelman (2006) suggests that the absence of a construction in a given
corpus is no less useful than collecting negative evidence (p. 23):
with regard to the usual way of obtaining negative evidence (i.e., asking one or
two speakers whether examples x, y, z are “okay”), it is doubtful whether this
really makes a difference in quality compared to evidence provided by the fact
that the structure in question is not attested in a large corpus.
Himmelman acknowledges that elicitation can be useful if it yields data that are
carefully elicited. He also suggests that negative evidence is already included in the
collection of metalinguistic knowledge that makes up the language documenta-
tion format. It remains the case that a common result of a language documentation
project is a descriptive grammar. Matthewson (2013) notes that descriptive gram-
mars rarely provide negative data; thus while negative evidence may be collected,
it is not systematically disseminated. The fact remains that the examination of as-
pectual classes via data collection of any type (corpus-based or elicitation-based)
will be impoverished if the extent of possible language variation in this part of the
grammar is not explored.
6.3 Range of Data
The value of comprehensive documentation of aspectual classes also cannot be

overstated. Both the fact that classification is not determined on the basis of a
single test, but that each aspectual class is defined by a constellation of features,
and the fact that not all tests are cross-linguistically applicable, point to the need
for a more comprehensive investigation in, not just across, languages.
That predicates from a given class may behave differently from the typical
examples suggests that we need to test a range of predicates from each class, and
not simply the few repeated examples that appear in many studies. That aspectual
classes across languages can vary with respect to entailments/implicatures, and the
fact that individual predicates can have both telic and atelic readings, point to the
necessity of exploring ambiguity, as well as available and preferred interpretations.
While some studies may focus on a specific contrast in a given language (e.g., ac-
tivities vs. accomplishments), studies of aspectual classes should ideally examine
all classes in the target language as important parallels between the classes may
be revealed (e.g., the relationship between inchoative states and inceptive events).
What constitutes a comprehensive documentation is itself an unanswered
question. Variations among the Vendlerian classes that are discussed in this
chapter suggest that the commonly used tools do not do the trick. In the same
vein as Matthewson (2013), who suggests that “[t]ypological work can only be as
good as its primary sources,” devising an all-inclusive toolkit that reflects the full
range of variation in aspectual classes can only be as good as our cross-linguistic
reporting. The upside is that a toolkit can easily be updated to reflect new find-
ings. The more comprehensive our documentation and classification of aspectual
classes cross-linguistically, the more we can uncover about their universality and
diversity.
References
Arunachalam, Sudha, and Anubha Kothari. 2010. Telicity and event culmination in Hindi
perfectives. In Proceedings of Verb 2010 Interdisciplinary Workshop on Verbs: The Iden-
tification and Representation of Verb Features, P. M. Bertinetto et al. (eds.), 16–19.
Arunachalam, Sudha, and Anubha Kothari. 2011. An experimental study of Hindi and Eng-
lish perfective interpretation. Journal of South Asian Linguistics 4(1): 27–42.
Bar-el, Leora. 2005. Aspectual Distinctions in Skwxwú7mesh. PhD diss., University of Brit-
ish Columbia.
Bar-el, Leora. 2007. Video as a tool for eliciting semantic distinctions. In Proceedings of Se-
mantics of Under-Represented Languages in the Americas 4. University of Massachusetts
Occasional Papers in Linguistics 35, Amy Rose Deal (ed.)., 1–16. Amherst, MA: GLSA
Publications.
Bar-el, Leora. 2009. Not just for techies: Using video for fieldwork and beyond. Paper pre-
sented at the 5th Conference on Endangered Languages and Cultures of Native Amer-
ica, University of Utah.
Bar-el, Leora, Henry Davis, and Lisa Matthewson. 2005. On non-culminating accomplish-
ments. In Proceedings of the 35th North East Linguistics Conference, L. Bateman and
C. Ussery (eds.), 87–102. Amherst, MA: GLSA Publications.
Botne, Robert. 2003. To die across languages: Toward a typology of achievement verbs. Lin-
guistic Typology 7: 233–278.
Carlson, Gregory N. 1977. Reference to kinds in English. PhD diss., University of Massa-
chusetts, Amherst.
Comrie, Bernard, and Norval Smith. 1977. The Lingua Descriptive Studies Questionnaire.
Lingua 42(1): 1–71.
Cover, Rebecca, and Judith Tonhauser. This volume. Theories of meaning in the field: Tem-
poral and aspectual reference. In Methodologies in Semantic Fieldwork, M. Ryan Boch-
nak and Lisa Matthewson (eds.). New York: Oxford University Press.
Dowty, David R. 1979. Word meaning and Montague grammar: The Semantics of Verbs and
Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel.
Ebert, Karen. 1995. Ambiguous perfect-progressive forms across languages. In Temporal ref-
erence, Aspect and Actionality, vol. II, Typological Perspectives, Pier Marco Bertinetto,
Valentina Bianchi, Östen Dahl, and Mario Squartini (eds.), 185–203. Torino: Rosen-
berg and Sellier.
Filip, Hana. 2011. Lexical aspect. In The Oxford Handbook of Tense and Aspect, Robert Bin-
nick (ed.), 721–751. Oxford: Oxford University Press.
25: 139–201.
Hay, Jennifer, Christopher Kennedy, and Beth Levin. 1999. Scalar structure underlies telic-
ity in “Degree Achievements.” In Proceedings of Semantics and Linguistic Theory IX,
T. Mathews and D. Strolovitch (eds.), 127–144. Ithaca, NY: CLC Publications.
Himmelman, Nikolaus P. 2006. Language documentation: What is it and what is it good
for? In Essentials of Language Documentation, Jost Gippert, Nikolaus Himmelman and
Ulrike Mosel (eds.), 1–30. Berlin: Mouton.
Jacobs, Peter. 2011. Control in Skwxwu7mesh. PhD diss., University of British Columbia.
Jóhannsdóttir, Kristín. 2011. Aspects of the progressive in English and Icelandic. PhD diss.,
University of British Columbia.
Kiyota, Masaru. 2008. Situation aspect and viewpoint aspect: From Salish to Japanese. PhD
diss., University of British Columbia.
Koontz-Garboden, Andrew. 2005. On the typology of state/change of state alternations.
In Yearbook of Morphology 2005, Geert Booij and Jaap van Marle (eds.), 83–117. Dor-
drecht: Springer.
Levin, Beth. 1999. Objecthood: An Event Structure Perspective. In CLS 35 Part 1: Papers
from the Main Session, Sabrina J. Billings, John P. Boyle and Aaron M. Griffith (eds.),
223–247. Chicago: Chicago Linguistic Society.
Lüpke, Friederike. 2009. Data collection methods for field-based language documentation.
Language Documentation and Description 6: 53–100.
Marín, Rafael, and Louise McNally. 2011. Inchoativity, change of state, and telicity: Evidence
from Spanish reflexive psychological verbs. Natural Language and Linguistic Theory
29: 467–502.
Matthewson, Lisa 2013. On how (not) to uncover cross-linguistic variation. Proceedings of
the North East Linguistic Society 42. Amherst, MA: GLSA, 323–342.
Nichols, Johanna. 2011. Ingush Grammar. UC Publications in Linguistics 143. University of
California Press.
Parsons, Terence. 1990. Events in the Semantics of English: a Study in Subatomic Semantics.
Cambridge, MA: MIT Press.
Rosen, Sara Thomas. 1999. The syntactic representation of linguistic events. Glot Interna-
tional 4: 3–11.
Rothstein, Susan. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect.
Oxford: Blackwell.
Soh, Hooi Ling, and Jenny Yi-Chun Kuo. 2005. Perfective aspect and accomplishment situ-
ations in Mandarin Chinese. In Perspectives on Aspect, H. J. Verkuyl, H. de Swart, and
A. van Hout (eds.), 199–216. Dordrecht: Springer.
Singh, Mona. 1998. On the semantics of the perfective aspect. Natural Language Semantics
6: 171–199.
Smith, Carlota S. 1999. Activities: Events or states? Linguistics and Philosophy 22: 479–508.
Smith, Carlota S. 1997. The Parameter of Aspect. Dordrecht: Kluwer.
Tatevosov, Sergej. 2002. The parameter of actionality. Linguistic Typology 6: 317–401.
Terry, J. Michael. 2004. On the articulation of aspectual meaning in African-American

English. PhD diss., University of Massachusetts, Amherst.
Thompson, Sandra A., Joseph Sung-Yul Park, and Charles N. Li. 2006. A Reference Gram-
mar of Wappo. UC Publications in Linguistics 138. Berkeley and Los Angeles: Univer-
sity of California Press.
Travis, Lisa deMena. 2010. Inner Aspect. Dordrecht: Springer.
Van Valin, Robert D. 2006. Some universals of verb semantics. In Linguistic Universals,
Ricardo Mairal and Juana Gil (eds.), 155–178. Cambridge: CUP.
Vendler, Zeno. 1967. Philosophy in Linguistics. Ithaca, NY: Cornell University Press.
Walková, Milada. 2013. The aspectual function of particles in phrasal verbs. Groningen Dis-
sertations in Linguistics 118.
Wilhelm, Andrea. 2003. The grammatization of telicity and durativity in Dëne Sųłiné
(Chipewyan) and German. PhD diss., University of Calgary. Published as 2007 Telicity
and Durativity: A Study of Aspect in Dëne Sųłiné (Chipewyan) and German. New York:
Routledge.
Xiao, Richard, and Tony McEnery. 2006. Can completive and durative adverbials function
as tests for telicity? Evidence from English and Chinese. Corpus Linguistics and Lin-
guistic Theory 2(1): 1–21.
4
Investigating Gradable Predicates, Comparison,

and Degree Constructions in Underrepresented
Languages
M. Ryan Bochnak and Elizabeth Bogal-Allbritten
1 Introduction
The meaning of gradable predicates, comparison, and related constructions has

been a topic of interest for linguists for decades.* Many key insights into this
domain were noticed as early as Sapir (1944), and researchers in formal semantic
frameworks have since made great progress in our understanding of the relevant
facts in this area (e.g., Kamp 1975, Cresswell 1976, Klein 1980, von Stechow 1984,
Heim 1985, Bierwisch 1989, Kennedy 1999, Kennedy and McNally 2005, among
many others). Additionally, typological studies including Ultan (1972), Stassen
(1985), and Bobaljik (2012) have explored the morphosyntactic landscape of com-
parison and gradability in the world’s languages.
Only recently, however, have the theoretical proposals, based largely on data
from English and German, been tested by formal semanticists researching under-
studied and typologically disparate languages in detail (e.g., Beck et al. 2009). First
steps toward integrating data from under-represented languages into theories of
comparison and gradability can be found in studies such as Galant (1998), Bogal-
Allbritten (2008, 2013), Beck et al. (2009), Eckardt (2009), Pearson (2010), Boch-
nak (2012a, 2013a,b), Hohaus (2012), and Francez & Koontz-Garboden (2013). This
semantic domain is thus relatively underrepresented within the small but growing
field of formal semantic fieldwork on understudied languages.
*We thank Washo consultants Ramona Dick and Steven James, and Navajo consultants Ellavina
Perkins and Louise Kerley. We would also like to thank Robert Henderson and Lisa Matthewson for
comments on a previous version that helped us improve this chapter. M. Ryan Bochnak’s work was
supported by grants from the Jacobs Fund of the Whatcom Museum, the American Philosophical Soci-
ety, and the National Science Foundation under Grant #1155196. Elizabeth Bogal-Allbritten’s work was
110 supported by a National Science Foundation Graduate Research Fellowship.

Investigating Gradable Predicates 111
Probably the biggest challenge facing the fieldworker investigating compar-

ison and gradability is the fact that the meaning of gradable predicates is highly
context-dependent. A single individual may count as tall in some contexts, but not
in others, all the while maintaining the same absolute height. A further compli-
cation is the fact that gradable predicates are not a uniform class. Variables such
as scale structure, ordering polarity, and whether a predicate is associated with a
measurable dimension have important consequences for entailment patterns and
the distribution of modifiers, among other things. These factors therefore need to
be taken into consideration when investigating constructions involving gradable
predicates.
In this chapter, we discuss these issues and offer methods for eliciting certain
key data in the field. The discussion is based on our own experiences investigating
the semantics of gradability and comparison in Washo and Navajo.1 Using case
studies from these two languages, we propose methodologies for obtaining data
on semantic distinctions that have been highlighted in the theoretical literature
and briefly outline how the data obtained using those methodologies fit into the
formal semantic frameworks developed for English and other languages.
The chapter is organized as follows. In section 2, we review Kennedy and Mc-
Nally ’s (2005) theory of scale structure and scalar modifiers, and discuss data from
Washo and Navajo, showing that scalar modifiers are not necessarily sensitive to
all or only those aspects of scale structure identified as linguistically relevant by
Kennedy and McNally. In section 3, we discuss the notions of norm-relatedness
and crisp judgments, and propose that certain visual stimuli can be effective in
eliciting crucial contrasts in the target language. We suggest that use of visual
stimuli can help obviate some problems posed by the inherent context sensitivity
of gradable predicates. Section 4 also addresses norm-relatedness and its interac-
tion with lexical competition, where we argue for certain non-visual elicitation
techniques for eliciting key data. We summarize and conclude in section 5.
Before continuing, we would like to flag from the outset that this chapter
does not contain a questionnaire of sentence types that must be collected by a
fieldworker investigating gradable predicates and related constructions in their re-
search language. Rather, the chapter is largely focused on methods for digging
deeper into the semantics of these constructions, once a range of sentence types
has been collected. For a questionnaire that outlines the basic sentence types to
collect, see Beck et al. (2009), especially p. 13 and the appendices.
1
Washo is highly endangered, spoken by about 10 elderly native speakers in the Lake Tahoe area
of California and Nevada. It is typically considered a language isolate, though it has been linked to the
proposed Hokan group of languages (Mithun 1999). English was used as the contact language for this
work. Navajo is a member of the Southern (or Apachean) subfamily of the Athabaskan language family.
Navajo is spoken by between 120,000 and 170,000 individuals based in the southwestern United States.
All Navajo consultants are fully bilingual in English, and English was used as the contact language. For
discussion of contact language choice, see AnderBois and Henderson (this volume) and Matthewson
(2004).
2 Adjective Classes and Modifier Licensing
In this section, we review adjective classes that have been argued in the theoretical
literature to be linguistically relevant, for example in their ability to license scalar
modifiers. We then present case studies from modifiers in Washo and Navajo that
are translated into English as very, but which have different distributions from
their English counterpart.
2.1 Methods of classifying adjectives
Much of the literature on the semantics of gradable adjectives has focused on the
fact that a subset of them, and indeed the most prototypical of them, are vague
in the bare form.2 Sentences with vague predicates are subject to contextual var-
iability in their truth conditions. That is, it is not a matter of fact whether Joe at
a height of 5’8” counts as tall or not. If Joe is a fifth-grader and we are comparing
him with other boys in his class, then the sentence Joe is tall is intuitively true.
However, if Joe is an adult male and we are comparing him with other members of
his professional basketball team, then the same sentence is intuitively false. Not all
context-sensitive predicates are vague, but all vague predicates seem to be context-
sensitive. Gradable adjectives such as tall, short, deep, and expensive show proper-
ties of vagueness.
While vague gradable predicates have historically received the most attention
in the analysis of gradable adjectives and comparison, there is a subset of grada-
ble adjectives that do not have vague interpretations. An example is wet: an object
counts as wet if it has a non-zero amount of moisture. In this case, there is a fact of
the matter for when an object counts as wet. In this sense, such predicates are not
subject to the same kind of contextual variability as predicates like tall.
Thus, the first major distinction that we observe between different types of
adjectives is the type of standard they are associated with, namely whether the
standard is relative or absolute. Vague gradable adjectives like tall have a rel-
ative standard that is context-dependent. Non-vague gradable predicates like wet
have endpoint-oriented standards, which do not vary across contexts in the same
way. Specifically, we can say that wet has a minimum (lower endpoint) standard.
Conversely closed has a maximum (upper endpoint) standard; a door counts as
closed only if it is completely closed. The term endpoint-oriented already makes
reference to the types of scales these predicates encode. In the theoretical literature,
Kennedy and McNally (2005) and Kennedy (2007b) argue that a gradable predi-
cate whose scale contains a maximum or minimum element will have its standard
fixed to one of those values, and will have an absolute interpretation. Meanwhile,
2
The ‘bare form’ of scalar predicates is much more frequently referred to as the “positive” form.
We use “bare” here to avoid confusion with positive-polar adjectives, such as tall and heavy, in contrast
to negative-polar adjectives, such as short and lightweight.
a scale that does not contain endpoints, i.e., an “open” scale, cannot possibly have
an endpoint-oriented standard, and therefore must have a relative interpretation.3
The scale structure associated with a predicate is correlated with certain en-
tailment patterns observed with the comparative form. As shown in (1), the com-
parative forms of relative-standard predicates like tall and pretty do not entail that
the bare form holds of the subject. By contrast, the comparative form of predicates
with lower-bounded scales like wet entails that the bare form of that predicate
holds of the subject. In order for the bare form to hold, the predicate must hold of
the subject to a degree exceeding some standard.
(1) a. Mary is taller than Ellen. ⇒
/ Mary is tall.
b. Mary is prettier than Ellen. ⇒
/ Mary is pretty.
c. The table is more wet than the counter. ⇒ The table is wet.
Kennedy and McNally (2005) argue that the acceptability of certain com-
binations of degree modifiers with scalar predicates can diagnose whether the
corresponding scale contains a maximum or minimum element. That is, the dis-
tribution of certain modifiers is tied to the scale structure of the predicate. Intu-
itively, the maximality modifier completely targets a maximal element on a scale.
It is therefore only compatible with predicates that contain a maximal element.
Meanwhile, the intensifier very is only compatible with relative-standard predi-
cates, and infelicitous with maximum-standard predicates. Thus we observe the
contrasts in acceptability in (2):
(2) a. The bottle is very tall/pretty/#closed.
b. The bottle is completely closed/#tall/#pretty.
The generality of these scale-structural distinctions and the corresponding
contrasts in (2) have yet to receive much cross-linguistic scrutiny. While the dis-
tribution of modifiers can be an informative locus of study, the fieldworker should
not take the availability (or, unavailability) of a particular modifier-predicate pair
as necessarily indicative of a particular scale structure. In the next subsection, we
discuss modifiers in Washo and Navajo that are often translated into English as
very, but which have different distributions from their putative English counter-
part. In both languages, the modifiers appear to track semantic distinctions other
than scale structure.
A second major division in gradable adjectives is Bierwisch ’s (1989) classi-
fication of adjectives into dimensional and evaluative classes. Dimensional
properties are quantitative and are usually associated with physically real systems
of measurement (e.g., tall, short, heavy). Evaluative properties are qualitative and
include properties describing personality (e.g., hardworking, lazy), color, and tex-
ture. Such properties are not typically associated with systems of measurement.
3
See Kennedy (2007b) and Bochnak (2013a) for more details on the distinction between relative
and absolute gradable adjectives.
Bierwisch’s dimensional properties generally pattern as open-scale adjectives with

relative standards. Descriptive properties are more heterogeneous: while some
pattern as open-scale adjectives (e.g., pretty) others pattern more like closed-scale
adjectives.4 While modifier selection in English does not seem to track this dis-
tinction, there are other areas of the grammar of gradability where this distinction
is linguistically relevant, for example in licensing norm-related interpretations in
how questions and other degree constructions (see Bierwisch 1989; Rett 2007). It
remains to be explored how this distinction may be exploited in other languages.
Another class of adjectives that has recently received some attention in the
theoretical literature is the class of so-called extreme adjectives like gigantic and
gorgeous (Paradis 2001; Morzycki 2012). While we will not discuss this class of ad-
jectives in this paper, we highlight it here for two reasons. First, this represents an-
other class of adjectives that shows special restrictions on modifier c o-occurrence
in English. Specifically, as shown in (3), this class licenses a distinct class of intensi-
fiers, but rejects modification by very, even when they arguably share a scale with
a non-extreme counterpart.
(3) a. Your shoes are downright/positively gigantic/gorgeous/#big/#pretty.
b. very big/pretty/#gigantic/#gorgeous (Morzycki 2012)
Second, extreme adjectives represent a class that has not yet received any attention
in the cross-linguistic literature, as far as we know. This therefore represents an
area within degree semantics that deserves further investigation from semantic
fieldworkers.
Finally, we point out the distinction between positive- and negative-polar
adjectives. Most gradable adjectives come in antonymic pairs such as tall/short,
wet/dry, open/closed. Under many theories, it is claimed that both adjectives share
the same scale, but have the opposite perspective. To put it another way, tall makes
reference to the positive degrees on the scale of height, while short makes reference
to negative degrees on the same scale. This distinction is linguistically relevant
in English for licensing a norm-related interpretation in certain degree construc-
tions, as shown in (4).5 While the positive-negative distinction does not appear
to have a reflex in modifier selection in English, we will see in the next section a
modifier in Navajo that does track this distinction.
(4) a. Charlie is as tall as David. ⇒
/ Charlie and David are tall.
b. Mary is as short as Ellen. ⇒ Mary and Ellen are short.
Before moving on to our examples from Washo and Navajo, we would also
like to point out the typological work of Dixon (1982) in distinguishing several
4
We refer the interested reader to Bogal-Allbritten (to appear) for more discussion of how evalu-
ative properties fit into Kennedy and McNally ’s (2005) typology of gradable adjectives.
5
We discuss norm-relatedness in more detail in sections 3–4; also see Rett 2007 for overview and
discussion.
lexical classes of adjectives. His work focuses on establishing a hierarchy of the

types of scalar attributes that tend to be lexicalized as a distinct syntactic category
(i.e., adjectives) in the world’s languages.
The distinctions between adjective classes discussed in this section have so far
remained relatively unexplored in semantic fieldwork on understudied languages.
This is definitely an area of research on the semantics of gradable predicates and
degree constructions that deserves more cross-linguistic scrutiny. We submit that
these distinctions are important for the fieldworker to be aware of, even if not all
of them turn out to be relevant for modifier selection or norm-related entailments
in the language being studied.
2.2 Scalar modifiers cross-linguistically
As shown above, scale structure distinctions are linguistically relevant for the dis-
tribution of degree modifiers. However, as seen in both case studies discussed
below, English translations of modifiers may make incorrect predictions about the
distribution of a modifier in another language. We therefore want to test modifiers
with predicates with different scale structures, and indeed across the other distinc-
tions described above, to arrive at the correct modifier and predicate meanings in
the language under investigation.
Our first example from the field comes from Washo. Before showing the data
on modification, we first introduce the basic morphosyntax of gradable predicates
in this language. Washo lacks a distinct morphosyntactic category of adjectives;
scalar concepts are typically lexicalized as verbs, as in (5).6 These verbal predicates
take the regular morphology found on verbs in this language, for example the im-
perfective suffix -i (Jacobsen 1964). Alternatively, a verbal gradable predicate may
appear with the nominalizing prefix de-, in which case a copula hosts verbal mor-
phology, as shown in (6).7
(5) verbal predicate:

méːhu ʔilkáykayiʔ-i
boy tall-ipfv
“The boy is tall.”
6
Baker (2003) argues that all languages make the distinction between the syntactic categories V,
N, and A. In making the above claim that scalar concepts are encoded as verbs in Washo, we are simply
referring to the fact that they are verb-y in that that occur with verbal morphology. Future research may
reveal a category of adjectives that has a distinct syntactic distribution from verbs in this language, in
line with Baker’s cross-linguistic claim. For now, we simply note that the question of whether or not a
language has degree-denoting expressions is likely separate from whether it has expressions of category
A. Thanks to Robert Henderson for discussion on this point.
7
We use the following short forms for glosses: 3 = 3rd person; aor = aorist; ca = comparative
aspect; cop = copula; ipfv = imperfective; nca = non-comparative aspect; neg = negation; nmlz =
nominalizer.
(6) nominalization + copula:

méːhu de-ʔilkáykayiʔ k’eʔ-i
boy nmlz-tall 3.cop-ipfv
“The boy is tall.”
The modifier šemu is often translated into English by speakers as very when it
modifies relative-standard predicates. However, as we see below, šemu has a much
wider distribution than very in English. In fact, šemu is compatible with predicates
that have either relative or absolute standards. Correspondingly, the translation
“really” is more appropriate when šemu modifies absolute standard predicates.
(7) a. ʔilkáykayiʔ šémuyi

tall šemu-ipfv
“He is very tall.”
(relative standard predicate) washo
b. ʔilší:šibiʔ šémuyi
straight šemu-ipfv
“It’s really straight.”
(maximum standard predicate) washo
c. ʔilk’únk’uniʔ šémuyi
bent šemu-ipfv
“It’s really bent.”
(minimum standard predicate) washo
Based on such data, it appears then that šemu is not a modifier that tracks scale
structure or the relative/absolute distinction. The fact that šemu can apply to pred-
icates of all scale types may indicate that the semantics of šemu is different from
English very (despite translations offered by speakers), or that scalar predicates
in Washo don’t lexicalize scale structure in the same way that scalar predicates in
English do, or both.8 In any case, a translation task that only focused on predicates
that license very in English would have missed the distributional differences be-
tween very and šemu, despite initial translational equivalence.
Our second example comes from Navajo. Once again, we consider a modifier
(yee,) that is translated by speakers and Young and Morgan (1987) as very. Like in
Washo, the modifier’s distribution does not track the distinction between adjec-
tives with relative and absolute standards. However, in contrast with Washo, the
Navajo modifier has a narrower distribution than English very. The modifier yee,
8
See Bochnak (2012a) for an analysis.
is felicitous with open scale adjectival predicates denoting negative dimensions

(8a,b) but not with predicates denoting positive dimensions (8c).9
(8) a. shideezhí ’áɬts’óózí yee’

my.little.sister 3-slender yee’
“My little sister is very slender.” navajo
b. díí dibé yázhí ’áɬtsísí yee’

this sheep small 3-little,small yee’
“This lamb is very small.” navajo
c. #shideezhí nineez yee’
my.little.sister 3-tall yee’
(Intended: “My little sister is very tall.”) navajo
The distribution of yee’ suggests that it is sensitive to the distinction between

positive- and negative-polar adjectives. Given an antonymic pair of dimensional
adjectives, yee’ is only felicitous with the negative member. Although this distinc-
tion does not appear to be relevant for licensing modifiers in English, we noted
above that it is relevant for licensing norm-related interpretations in some degree
constructions (section 2.1) (Rett 2007).
In this section, we considered modifiers in Washo and Navajo that are both
translated into English as very but which do not have the same distribution as very.
The availability of English very has been argued to indicate that an adjective has a
relative standard and a scale lacking endpoints. However, it would be a mistake for
a fieldworker to conclude based on the grammaticality of Washo ʔilšíːšibiʔ šémuyi
(“straight šemu”) that ʔilšíːšibiʔ “straight” has an open scale and a relative stan-
dard in Washo. Other diagnostics point away from this conclusion. For instance,
absolute-standard predicates are pragmatically marked in certain comparison
constructions in Washo (Bochnak 2013a,b). Similarly, it would be a mistake for a
fieldworker to conclude from the infelicity of Navajo #nineez yee’ (“tall yee’ ”) that
nineez “tall” does not have an open scale and relative standard in Navajo. We will
see in section 4.2 that nineez has a relative standard.
It is crucial to first establish the distribution of the modifier in the language
of study before taking the modifier as diagnostic of a particular feature of scale
structure. This is particularly important for work on underrepresented languages,
where not only might English translations be misleading but also where existing
linguistic materials may not contain complete descriptions of the distribution of
degree modifiers.
9
To the second author’s knowledge, the only positive adjective that appears to be felicitous by
yee’ is nizhóní ‘pretty.’ This also seems to be the only evaluative predicate with which yee’ is felicitous.
3 Norm-Relatedness and Crisp Judgments
In the remainder of the paper, we turn our attention to comparative construc-

tions. We address the study of comparative constructions themselves, as well as
their usefulness in diagnosing the scale structures of gradable predicates. In this
section, we examine the distribution of norm-relatedness and crisp judgments
in Washo comparative constructions, and consider how they are related to each
other. We show that using the right kind of visual stimuli in the field can be ex-
ploited to bring out the relevant semantic distinctions, and illustrate the success of
such techniques with data from Washo.
A construction is norm-related if it entails that the bare form of the adjec-
tive holds (Bierwisch 1989).10 For instance, the comparative form of the dimen-
sional, relative-standard predicate tall in (9a) is not norm-related, since it does not
entail that the bare form in (9b) holds.
(9) a. Alice is taller than Brian.
b. Alice is tall.
In English, the comparative, superlative, and excessive constructions of dimensional
predicates lack the property of norm-relatedness (Bierwisch 1989).11 One environ-
ment in English where the bare form is not norm-related occurs in the compared
to construction, as in (10). In fact, as argued by Sawada (2009), there is actually an
implicature that the bare form is not true of either individual being compared.
(10) Alice is tall compared to Brian.
What counts as tall varies both with the sorts of objects under consideration
(e.g., adult humans versus skyscrapers) and also with the particular context of ut-
terance. Correspondingly, during the investigation of degree constructions in the
field, it is crucial that the fieldworker and consultant share a conversational back-
ground that allows the fieldworker to evaluate whether a certain construction is
norm-related. We use examples from the first author’s investigations of Washo to
illustrate. In this language, the primary comparative construction is a conjoined
comparison (Bochnak 2013a,b), whereby two clauses containing antonymous
predicates are juxtaposed, and a comparison between two objects is inferred (Stas-
sen 1985). An example is given in (11):
(11) t’éːliwhu delkáykayiʔ k’éʔi šáwlamhu delkáykayiʔéːs k’áʔaš

man nmlz-tall 3.cop-ipfv girl nmlz-tall-neg 3.cop-aor
“The man is taller than the girl.”
(literally: “The man is tall, the girl is not tall.”) washo
10
Other terms for norm-relatedness found in the literature includes “orientedness” (Seuren 1978)
or “evaluativity” (Neeleman et al. 2004; Rett 2007).
11
As already mentioned in section 2, certain other constructions may vary depending on the polar
variant of the scalar predicate. See Rett (2007) for an analysis of such facts in English.
Since the comparison in (11) makes use of the bare form of scalar predicates, we
may wonder whether the scalar predicates receive a norm-related interpreta-
tion or whether this is similar to the English comparative construction in being
non-norm-related.
In some cases, presenting a consultant with a context verbally may suffice if
we are confident that both the fieldworker and the consultant share a common
ground to make the relevant judgments with respect to norm-relatedness. For
instance, the following context was presented verbally in the contact language
(English), whereby the fieldworker and consultant share the common ground that
heights of five feet or four and a half feet do not count as tall for adult humans:
(12) a. Context: comparing a man who is five feet tall and a woman who is four
and a half feet tall (i.e., both are clearly short for adult humans)
b. # t’éːliwhu delkáykayiʔ k’éʔi daʔmóʔmoʔ delkáykayiʔéːs

man nmlz-tall 3.cop-ipfv woman nmlz-tall-neg
k’áʔaš
3.cop-aor
Intended: “The man is taller than the woman.” washo
c. The man is taller than the woman. english
We observe that the conjoined comparison using the predicate delkáykayiʔ “tall”
is unacceptable when neither individual being compared counts as tall. Note that
the English comparative in (12c) is perfectly felicitous in the same context. We thus
have some evidence that conjoined comparisons in Washo are norm-related, in
contrast with English comparative constructions, which are not.
However, many contexts that one would want to use to test norm-relatedness
are not as straightforward to explain verbally to a field consultant. The context
in (12a) works well because the contact language (English) provides a means by
which we can precisely talk about heights (e.g., measure phrases like five feet).
But there are many scalar predicates in English for which there is no standardized
measurement system. Take big for example. Most native speakers probably have
good intuitions about when an object counts as big, though there is no conven-
tional means for precisely measuring bigness. Additionally, big is a scalar pred-
icate that is associated with many dimensions. Indeed, an adult human can be
considered big if she or he is quite tall, or quite wide, or quite heavy. Furthermore,
our intuitions about what precise measurements count as big seem to break down
here: is 50 cm wide considered big for an adult human? We therefore run into diffi-
culty when trying to verbally establish a context like (12a) to test norm-relatedness
in constructions involving the scalar predicate big.12
12
Another potential problem is that you may be working in a contact language that has very few
or even no measure phrases, making it very difficult to talk about precise measurements in the first
place.
We argue that the use of targeted visual stimuli can help overcome this diffi-
culty. The task we have in mind involves the fieldworker collecting a set of objects
that differ along some dimension (e.g., height, width). This set should have two
properties. First, it should consist of objects of the same type. For instance, com-
paring a pencil and a piece of rope with respect to length is undesirable, since
what counts as long for a pencil may not count as long for a piece of rope. Visually
presenting objects of the same type obviates this difficulty. Second, this set should
contain objects that differ along the relevant dimension within a wide enough
range such that it is clear that a certain subset holds the relevant property, while
some other subset does not. In the investigation of degree constructions in Washo,
the first author has used sets of pine cones for this task, since pine cones are plen-
tiful in the part of the world where the fieldwork is conducted. They additionally
come in a variety of shapes and sizes, which allowed us to compare them along
several different dimensions.
To begin, the fieldworker and consultant come to an agreement over which
members of the set count as big, which ones count as small, and which ones are in-
between. This can be done in either the contact or the research language. However,
we recommend that the consultant confirm set membership using the targeted
predicates in the research language, in case the extensions of gradable predicates
do not quite overlap in the two languages.13 After coming to an agreement with the
consultant about which pine cones count as big, which ones count as small, and
which ones are medium-sized, the consultant is presented with two pine cones
from the ‘small’ set, but which still differ in size, and is asked to compare them. In
this context, the consultant utters (13b):
(13) a. Context: comparing two small pine cones; both are clearly not big (for
pine cones)
b. wíːdiʔ behéziŋaš lák’aʔ wíːdiʔ t’íːyeliʔ wéwši
this small-aor one this big slightly-ipfv
“This one is small, that one is a little bit big.” washo
c. That one is bigger than this one. english
We observe that the consultant resists using the unmodified predicate t’íːyeliʔ ‘big’
to describe a pine cone that does not fall under its extension, even though it is
bigger than another pine cone. Indeed, a follow-up elicitation shows that the com-
parison in (13b) is unacceptable without the modifier wéwši “slightly.” Note once
again that the English comparative in (13c) is perfectly acceptable in the same con-
text. We thus have more evidence for our hypothesis from above that conjoined
comparisons in Washo are norm-related.
13
For example, in some languages, one might not use the same predicate for “tall” to describe
humans and non-human objects.
To summarize, using this elicitation technique can help elucidate the proper-
ties of degree constructions with respect to norm-relatedness through the creation
of contexts where the fieldworker and consultant share a common ground of what
objects count as big, for example, in a particular context. The use of real-life visual
stimuli facilitates the investigation since it controls for the fact that such predi-
cates are highly context-sensitive, and ensures that the fieldworker and consultant
have the same context in mind. We now go on to show that a similar task can also
be used to elicit judgments on whether comparative constructions support crisp
judgments.
crisp judgment contexts are those in which the two objects being compared
differ minimally along the relevant dimension (Kennedy 2007a,b). Different com-
parison constructions differ with respect to whether they support crisp judgments
or not. For instance, the English comparative in (14b) using the comparative mor-
pheme -er is felicitous in the crisp judgment context given in (14a), while the com-
pared to construction in (14c) is infelicitous.
(14) a. Context: Charlie is 5’6” tall and John is 5’5” tall.
b. Charlie is taller than John.
c. # Charlie is tall compared to John.
In fact, it is a general property of the bare form of vague predicates that they do
not support crisp judgments. That is, if two objects x and y differ only slightly with
respect to a vague predicate P, then if x is P is true, then we are unable or unwill-
ing to accept y is P as false (the “Similarity Constraint”: Fara 2000). Investigating
whether a comparative construction supports crisp judgments thus requires set-
ting up contexts such as in (14a) where objects differ minimally with respect to
some property.
The visual stimulus task described above can be modified to create the rele-
vant contexts for testing crisp judgments. The fieldworker simply needs to juxta-
pose two objects that differ minimally with respect to the relevant dimension. In
the case of Washo, the following sentence was offered by a speaker comparing two
pine cones that differ in a perceptible but minimal amount with respect to size:
(15) a. Context: comparing two pine cones that differ minimally in size
b. wíːdiʔ dewdíʔiš y’áːk’aʔ wíːdiʔ t’íːyeliʔ šemuyéːs lák’aʔ deltétebiʔ

this tree cone this big very-neg one nmlz-fat
wéwɨš
slightly
“This pine cone, this one is not very big, this one is sort of fat.” washo
Here, we have a first clue that conjoined comparisons in Washo do not support
crisp judgments. The speaker avoids using the canonical x is P, y is not P con-
joined comparison in this context, but rather opts to use two distinct predicates
to compare the pine cones (t’íːyeliʔ “big” and deltétebiʔ “fat”). If this construction
in Washo does not support crisp judgments, then it would be infelicitous for a
speaker to use the canonical x is P, y is not P construction, due to the Similarity
Constraint on the use of vague predicates.
A follow-up elicitation session confirms this hypothesis. In this case, images
of ladders were presented to the speaker on a computer screen. While the use of
real-life visual stimuli allow the fieldworker to create more natural contexts (rather
than having a speaker imagine some hypothetical scenario), it may not always be
possible to find a set of objects that meets the ideal specifications described above.
This is where technology can be useful: electronic images can be manipulated and
scaled by the fieldworker to create the appropriate set of visual stimuli for testing
the relevant distinctions. For instance, the image in Figure 4.1 shows a set of lad-
ders that differ along the dimension of height.14
In the follow-up elicitation, the fieldworker presented the following conjoined
comparison to a Washo speaker in the context of comparing the second and third
ladders from the left in Figure 4.1 (i.e., a crisp judgment context). The sentence is
rejected by the speaker in this context:
(16) a. Context: comparing two ladders, one only slightly taller than the other
b. # wíːdiʔ t’éːweʔ dewgíʔiš k’éʔi wíːdiʔ t’éːweŋa dewgíʔišéːs

this much height 3.cop-ipfv this much-nc height-neg
k’éʔaš
3.cop-aor
Intended: “This one is taller than that one.” washo
As discussed in Bochnak (2013a,b), the findings for norm-relatedness and

crisp judgment effects suggest that Washo gradable predicates should be analyzed
as vague predicates rather than scalar expressions. In other words, Washo gradable
Figure 4.1 “Ladders” visual stimulus for testing norm-relatedness and crisp judgments
14
Of course, we still need to make sure the fieldworker and consultant agree about which objects
count as tall and short in this scenario, since even the tallest ladders in Figure 4.1 could in principle
be miniscule. Thanks to Yaron McNabb for sharing images used in experimental studies on the use of
gradable predicates in English, upon which Figure 4.1 is based.
predicates represent an alternative type of semantic object to adjectives as usu-

ally conceived (cf. Cresswell 1976, von Stechow 1984, Kennedy 1999, Kennedy
and McNally 2005). We refer the interested reader to Bochnak (2013a,b) for more
discussion.
Crucially, the investigation of norm-relatedness and crisp judgment effects
was facilitated by the use of visual stimuli. These visuals can be used both for in-
itial explorations where the speaker is invited to spontaneously offer a sentence
to describe a certain situation, and also for targeted follow-up elicitations where
the fieldworker tests a sentence against the pictured context. The advantage of this
methodology is that the fieldworker and consultant can explicitly negotiate to-
gether over whether objects hold a certain property, obviating some of the difficul-
ties related to the context-sensitivity of the predicates being investigated.
We close this section with a couple of caveats about using this type of visual
stimuli in elicitation. First, since we have shown that the same type of stimulus
can be used to test both norm-relatedness and crisp judgments, the fieldworker
must make sure to test for only one semantic feature at a time. For instance, in
testing norm-relatedness, it is crucial that the context presented is not also a crisp
judgment context, since the fieldworker needs to be able to identify the source
of infelicity if a speaker judges a sentence unacceptable. Second, while we have
shown the applicability of these visual stimuli for testing dimensional predicates, it
can be difficult to find or create appropriate stimuli to test norm-relatedness with
non-dimensional (i.e., evaluative) predicates. We discuss this issue further in the
next section.
4 Norm-Relatedness and Lexical Competition
To this point, we have concentrated on elicitation of judgments for sentences

containing dimensional predicates, namely tall and big. Even when the field-
worker focuses on this relatively small set of gradable predicates, we have seen
that the format of the elicitation materials is key. In this section, we consider
the methodological considerations that arise when the fieldworker expands the
study to include evaluative adjectival properties. The data in this section come
from the second author’s investigation of Navajo. Elicitation techniques for di-
mensional and evaluative predicates are discussed in the context of a fieldwork
study that probes for norm-relatedness within the adjectival domain. We argue
that competition between different morphological forms of a single property
leads to norm-relatedness in comparative constructions. We continue to advo-
cate for the value of visual and tactile stimuli, but also make a case for the use of
verbal stimuli in certain cases. We argue that verbal stimuli are especially useful
for evaluative adjectival predicates, such as pretty, since verbal stimuli allow the
context to be more carefully defined and agreed upon by the fieldworker and the
consultant.
4.1 Background on Navajo Adjectival predicates
The core semantic meaning of Navajo adjectival predicates—like event-denoting

predicates—is carried by the stem, which invariably occurs at the right edge of the
predicate. Stems are necessarily marked with prefixes encoding nominal arguments
and may bear additional prefixes. For event-denoting predicates, the additional
prefixes include markers of situational and viewpoint aspect, iterativity, and dis-
tributivity. Adjectival stems also bear additional prefixes. We will refer collectively
to the prefixes borne by adjectival stems as comparative aspect (ca) and non-
comparative aspect (nca). Choice of ca versus nca morphology also may affect
pronunciation of the stem (e.g., tones). For detailed discussion of ca and nca mor-
phology, please see Young and Morgan (1987) and Bogal-Allbritten (2010, 2013).16
Not all adjectival stems can bear ca morphology: ca morphology is largely
restricted to Bierwisch ’s (1989) dimensional properties (Young and Morgan 1987).
The existence of a ca-marked form of a given stem entails the existence of an nca-
marked form of the same stem. ca- and nca-marked forms of the stem -neez “tall”
are shown in (17).
(17) a. -neez
tall
b. nineez
3-tallNCA
c. ’áníɬnééz
3-tallCA
Many more adjectival stems can only be nca-marked. Adjectival predicates
with only nca-marked forms are more heterogenous in meaning. This set of pred-
icates includes predicates whose English translations have closed scales (e.g., digiz
“crooked”) and predicates whose English translations have open scales, according
to Kennedy and McNally ’s (2005) typology. Among the latter are predicates ex-
pressing Bierwisch ’s (1989) evaluative properties (e.g., nizhóní ‘pretty.’)17
(18) a. -zhóní
pretty
b. nizhóní
3-prettyNCA
c. No ca-marked form in common use
16
The use of the term “aspect” with specifically ca or nca morphology should not be taken as
a theoretical claim that these morphemes are, in some sense, true aspectual morphemes. We use the
term “aspect” here to maintain continuity with previous literature on Athabaskan languages, including
Young and Morgan (1987). The term is used by the authors to capture the linear proximity of ca/nca
prefixes on adjectival predicates to viewpoint and situational markers on eventive predicate.
17
A small number of ca-marked forms of evaluative properties are reported by Young and Morgan
(1987). However, the second author found that these forms, including pretty, were rarely volunteered
by consultants in comparative constructions. Consultants found some of the ca-marked forms to be
unfamiliar or disagreed on their morphological form. This suggests that these forms are becoming
obsolete or irregular in a way that dimensional ca-marked predicates are not.
To summarize, dimensional properties are expressed in Navajo by predi-

cates that reliably have both ca- and nca-marked forms. By contrast, evaluative
properties are only expressed by predicates with nca-marked forms. In the next
section, we show that with respect to norm-relatedness in comparative construc-
tions, predicates with both ca- and nca-marked forms pattern differently from
predicates with only nca-marked forms. In order to draw this conclusion, it was
necessary to test for norm-relatedness in comparative constructions containing
dimensional and evaluative predicates. Different types of elicitation materials
were found to be ideal for each class of property. While visual and tactile contexts
were useful for dimensional predicates, verbal contexts were used with evaluative
predicates.
4.2 Testing for norm-relatedness
We first examine the semantics of bare adjectival predicates in Navajo (i.e., predi-
cates not further modified by comparative morphology). Recall that in English,
bare English adjectives are always norm-related, except when used in the compared
to construction (section 3). We confirm below that when nca-marked Navajo ad-
jectival predicates appear without further modification, they are felicitous in the
same types of norm-related contexts that would license English bare adjectives
with relative standards.
(19) a. Context: You are showing me some rugs. One of the rugs stands out for
the high quality of the weaving, the color of the yarn, and the intricateness
of the pattern. You are telling me about this rug. You say to me:
b. díí diyógí nizhóní
this rug 3-prettyNCA
Comment: “You can say that, you’re saying this rug looks good.”
(20) a. Context: You are describing your family to me. Your little sister is tall for
a young girl. She is only 10 years old but is already 5’ tall.
b. shidéézhí nineez
my.younger.sister 3-tallNCA
Comment: “You can say this.”
If the context was not norm-related, consultants judged sentences with un-
modified adjectival predicates to be infelicitous.
(21) a. Context: You are showing me some rugs. One of the rugs has uneven
weaving and unattractive colors. You are telling me about this rug. You
say to me:
b. # díí diyógí nizhóní
this rug 3-prettyNCA
Comment: “If the rug isn’t pretty, why are you saying it is?”
Only nca-marked adjectival predicates are shown above because ca-marked

predicates can never appear without further modification (22). Reasons for the
ungrammaticality of (22) are discussed at length in Bogal-Allbritten (2013).
(22) *shidéézhí ’áníɬnééz

my.younger.sister 3-tallCA
(Intended: “My younger sister is tall.”)
We now turn our attention to norm-relatedness of adjectives in comparative

constructions. Both ca- and nca-marked adjectival predicates can appear in the
comparative construction.18 Kennedy and McNally (2005) use norm-relatedness
in comparative constructions as a diagnostic of minimum endpoints on adjectival
scales (section 2.1). In Navajo, lexical competition appears to be another source of
norm-relatedness in comparatives:
(23) a. If an nca-marked adjective has a ca-marked counterpart, then the

comparative construction with the nca-marked adjective is necessarily
norm-related.
b. If an nca-marked adjective does not have a ca-marked counterpart,
then the comparative construction with the nca-marked adjective is not
necessarily norm-related.
We first examine comparative constructions containing adjectival predicates

for which only an nca-marked form is available. Consultant comments in (25) in-
dicate that nizhóní ‘pretty’ is not obligatorily norm-related in the comparative con-
struction. Norm-relatedness was tested for by setting up contexts in which both
objects of interest were established to meet, or not meet, the relative standard of
comparison for the relevant property. We discuss the contexts used to elicit the
judgments in (24) and (25) in section 4.3.
(24) a. Context: Verbal context identifying two rugs as attractive (30).
b. díí diyógí ’eii diyógí yilááh ’át’éego nizhóní

this rug that rug 3-beyond 3-being 3-prettyNCA
“This rug is prettier than that rug.”
c. Comment: “You can say this. This rug is pretty but this one is better
looking.”
18
The “greater than” ordering relation is expressed by the standard marker -lááh “beyond.” The
predicate does not bear any special morphology (cf. English more/-er). This is a cross-linguistically
common method of expressing comparative relations (Stassen 1985). For more discussion of the syntax
of Navajo comparative constructions, see Bogal-Allbritten (2013).
(25) a. Context: Verbal context identifying two rugs as unattractive (31).
b. díí diyógí ’eii diyógí yilááh ’át’éego nizhóní

this rug that rug 3-beyond 3-being 3-prettyNCA
“This rug is prettier than that rug.”
c. Comment: “It is relative to what you’re comparing. If you have two ugly
rugs, then you’re just saying that this one is better than the other.”
Navajo nizhóní behaves like its English translation pretty: neither is obligatorily
norm-related in comparative constructions. This suggests that both are open-scale
adjectives according to Kennedy and McNally ’s (2005) diagnostics.
We now turn to Navajo adjectival predicates for which both ca- and nca-
marked forms exist. Example (26) shows that ca-marked adjectives can be used in
comparative constructions in non-norm related contexts.
(26) a. Context: You are describing your family to me. Your mother and your
younger sister are both short women. Your mother is 5’2” and your
younger sister is 4’11”.
b. shimá shidéézhí yilááh ’áníɬnééz

my.mother my.younger.sister 3-beyond 3-tallCA
“My mother is taller than my younger sister.”
c. Comment: “This sounds good. They aren’t tall, but my mom is taller.”
Both the Navajo predicate ’áníɬnééz and its English translation tall behave like open-
scale adjectives: neither is obligatorily norm-related in comparative constructions.
English and Navajo diverge when we look at Navajo nca-marked predicates
that have a ca-marked form. The comparative containing nca-marked nineez
“tall” is judged infelicitous in the context in (27), where the individuals under
comparison do not meet the relative norm or standard for tall. The consultant
suggested that the ca-marked form of the predicate be used for instead.
(27) a. Context: Same as (26)
b. # shimá shidéézhí yilááh ’át’éego nineez

my.mother my.younger.sister 3-beyond 3-being 3-tallNCA
(Intended: “My mother is taller than my younger sister).”
c. Comment: “You don’t want to use nineez. This implies that they’re tall
people. You’d have to use ’áníɬnééz.”
By contrast, in a context like (28) where both individuals under comparison meet
the standard for tall, both nca- and ca-marked adjectives were felicitous:
(28) a. Context: You are describing your friend’s family. Her mother and younger
sister are both tall women. Her mother is 6’4” and her sister is 6’2”.
bimá bidéézhí yilááh ’át’éego nineez
her.mother her.younger.sister 3-beyond 3-being 3-tallNCA
b. Comment: “This sounds good. You can say ’áníɬnééz, too.”
In both English and Navajo, adjectives that occur in the bare form are oblig-
atorily norm-related. In both languages, norm-relatedness is sometimes retained
when adjectives appear in comparative constructions. Kennedy and McNally
(2005) cite the presence of minimum endpoints on adjectival scales as a source of
norm-related comparative constructions. In Navajo, competition between nca-
and ca-marked predicates is another source of norm-relatedness in comparative
constructions. If a ca-marked form of a particular predicate exists, then a com-
parative construction containing the nca-marked form of that predicate is oblig-
atorily norm-related.
4.3 Constructing the contexts
We saw above that the distribution of ca and nca morphology is linked to Bier-
wisch ’s (1989) distinction between dimensional and evaluative properties. Dimen-
sional properties are expressed by stems with both ca- and nca-marked forms.
Evaluative properties are expressed by stems for which only nca morphology is
generally available. In the study described above, the second author probed for
norm-relatedness in comparative constructions with dimensional and evaluative
predicates. In this section, we discuss the elicitation materials used by the second
author to obtain the reported judgments. Visual and tactile materials were found
to be best-suited for dimensional properties while we suggest that verbal contexts
can be valuable for studying evaluative properties.
In initial elicitation sessions examining adjectival predicates with dimen-
sional meanings (e.g., -neez “tall”), verbal contexts like (29) were used on their
own. Before a judgment about a Navajo target sentence was given, the fieldworker
and consultant agreed that the two individuals were short. This was established by
giving each individual’s height and referring to them as “short women.”
(29) Context: You are describing your family to me. Your mother is 5’2” tall and
your younger sister is 4’11”. Your mother and your younger sister are both
short women but there is a difference in their heights.
As discussed for Washo in section 3, adjectives like tall are compatible with
verbal contexts because the contact language (English) permits us to precisely talk
about heights using measure phrases. However, targeted visual materials were used
in later elicitation sessions. Although felicity judgments were consistent regardless
of whether visual materials were used or not, visual materials helped consultants
D
C
B
A
Figure 4.2 “Height chart” visual stimulus for norm-relatedness judgments
to keep track of individuals under comparison. After many contexts like (29) have
been discussed in succession, it is easy to lose track of which individual is meant
to be the taller one and which the shorter one. Figure 4.2 was presented to consul-
tants in conjunction with contexts like (29). Names and identifying information
were added below the relevant individuals in accordance with the context. Each
individual was indicated by the fieldworker when mentioned in the context.
To this point, discussion of elicitation materials for both Washo and Navajo
has focused on dimensional properties. A different set of challenges is presented
by evaluative properties like pretty. First, pretty and its Navajo translation nizhóní
lack a single dimension of measurement: in contrast with tall, individuals’ levels
of beauty cannot be described precisely with measure phrases. One potential solu-
tion to this problem would be to use visual and tactile materials like those used to
study the predicate translated as big in Washo. Recall that big not only lacks a pre-
cise measurement system but is also potentially associated with many dimensions.
A similar challenge is presented by pretty: while pretty is not clearly associated
with multiple dimensions, prettiness is a subjective property (Lasersohn 2005,
Stephenson 2007). Different speakers may take different factors into account when
determining whether an object counts as pretty or nizhóní. As a result, speakers
may disagree about whether the same object is “pretty” or not.
Second, because of the subjectivity of pretty and nizhóní, it proved difficult to
locate ideal objects for use in visual or tactile contexts. While pine cones that dif-
fered minimally in size could be found, fieldworkers may find it difficult to locate
objects that both the consultant and the fieldworker agree to be “pretty.” It is even
more difficult to locate pairs of objects that only differ slightly in their beauty.
Given these considerations, we argue that verbal contexts are particularly

suitable for the study of evaluative properties. The following verbal contexts were
used to elicit the judgments reported in (24) and (25), respectively.19
(30) Context: We’re looking at rugs at a trading post. There is a wide range of
rugs, from very attractive to very unattractive. You hold up two rugs that
are pretty. Their wool is attractively dyed and the weaving was done very
straight. One of the rugs is even prettier than the other. I ask you what you
think of them.
(31) Context: We’re looking at rugs at a trading post. There is a wide range of
rugs, from very attractive to very unattractive. You hold up two rugs that
are not good-looking: their wool is dyed in strange colors, the weaving was
not done in straight lines, and the design is not complex. One is of slightly
better quality than the other, however. I ask what you think of the two rugs.
By presenting the context verbally, the fieldworker is able to invoke a degree
of subtlety that might not be possible if real-world objects were used. Further-
more, the verbal context in (31) explicitly establishes what qualities the rugs have
(crooked weaving, strange coloration, etc.). These qualities were independently
confirmed by consultants to be attributes of unattractive rugs. By making back-
ground assumptions explicit in this way, the consultant and fieldworker can come
to a clear agreement on whether the objects under comparison meet the standard
of comparison for the property in question.
A final note can be made about the study of bare adjectives. In order to elicit
judgments about sentences like (32), contexts were presented verbally rather
than visually. This was a deliberate decision, even though the predicate in (32) is
dimensional.
(32) a. Context: You are describing your family to me. Your mother is tall: she is
5’11”. You are telling me about her.
b. shimá nineez
my.mother 3-tallNCA
Comment: “You can say this if your mother is tall.”
Visual materials depicting more than one object—the target (‘my mother’)
and a standard of comparison (e.g., a tree or building)—were initially constructed
to test sentences like (32). However, consultants found it most natural to describe
*“the” these pictures with comparative constructions rather than unmodified ad-
jectives. If the standard of comparison were eliminated from the visual materials,
In context (30), the English adjective pretty was named in the context. In other contexts, the
19
English adjective does not appear. The presence or absence of the English adjective in the context did
not seem to affect consultants’ judgments.
it was still necessary for the context to be clarified verbally. Verbal presentation
allows the fieldworker and consultant to come to agreement about the context
without the additional complication introduced by visual materials.
The distinction between dimensional and evaluative properties is relevant to
the distribution of adjectival morphology in Navajo. While dimensional adjecti-
val stems can bear either ca or nca morphology, evaluative properties are only
expressed by nca-marked adjectival predicates. The distribution of ca and nca
morphology was in turn shown to be relevant to determining whether a com-
parative construction is obligatorily norm-related or not. In many—perhaps even
most—languages, the distinction between evaluative and dimensional properties
may not affect the distribution of adjectival morphology. However, it is still impor-
tant for the fieldworker to be aware of the distinction between dimensional and
evaluative meanings. Different types of elicitation materials (verbal, visual, and
tactile) are best suited to the two classes of adjectives.
5 Conclusion
The conclusions we have drawn can be categorized into the theoretical and the
methodological, although there is, of course, interaction between the two cat-
egories. Considering first the theoretical conclusions, the fieldworker should be
aware of predictions made by theoretical accounts of scale structure and predi-
cate meaning. The application of Kennedy and McNally ’s (2005) theory permitted
both authors to make initial predictions about the types of scale structures that
might appear in their languages of study. Two of Kennedy and McNally’s scale
structural diagnostics—the distribution of modifiers and norm-relatedness in
comparatives—were applied to gradable predicates in Washo and Navajo. Both the
modifier and norm-relatedness findings suggest interesting differences between
Washo, Navajo, and English. For instance, both Washo and Navajo were shown
to have modifiers translated as English very but with different distributions from
very. We argued that the modifiers in both languages were sensitive to aspects
of adjectival meaning that were not relevant to the distribution of English very.
Further cross-linguistic studies of the semantics of gradable predicate and degree
constructions will likely find constructions sensitive to other aspects of adjectival
meaning, and may also uncover additional complexities and patterns that should
be incorporated into the typological picture of adjectives.
On the methodological side, we have drawn two conclusions. First, fieldwork
on a relatively restricted subject—such as the semantics of gradable predicates—
still benefits if the fieldworker has access to a broader base of knowledge about the
language of study. In both Navajo and Washo, norm-relatedness of adjectives arose
in comparative constructions containing adjectives for which n orm-relatedness
is not predicted by Kennedy and McNally (2005). For both case studies, these
departures suggested a reanalysis of gradable predicates. Their reanalysis in each
language was supported by additional knowledge about the language. In Washo,

the findings for norm-relatedness in adjectives were considered alongside the un-
availability of crisp judgments in comparative constructions. Both sets of facts
follow if gradable predicates are analyzed as vague predicates (Bochnak 2013a,b).
In Navajo, the patterns of norm-relatedness became clear once the distribution of
adjectival morphology (ca vs. nca) was taken into account.
Second, even if a given way of classifying adjectives does not seem relevant
to the language of study, different classes of adjectives may still be best suited to
different types of elicitation materials. We found that for dimensional predicates,
both visual and tactile contexts were effective. For evaluative adjectives, however,
verbal contexts permitted more careful definition of the context and more explicit
agreement between fieldworker and consultant about the context’s crucial aspects.
While we have not exhausted all the possible avenues of research on gradable
predicates and degree constructions that a fieldworker may undertake, we hope to
have provided a basic overview of some of the issues that may arise and the meth-
odologies we have found useful in investigating this area of meaning in the field.
Though we have already mentioned it, it bears repeating that this is an area of se-
mantics that is ripe for making new discoveries based on cross-linguistic research,
which can then inform future typological and formal research.
References
AnderBois, Scott, and Robert Henderson. This volume. Linguistically establishing dis-
course context: Two case studies from Mayan languages.
Baker, Mark C. 2003. Lexical Categories: Verbs, Nouns and Adjectives. New York: Cambridge
University Press.
Beck, Sigrid, Sveta Krasikova, Daniel Fleischer, Remus Gergel, Stefan Hofstetter, Christiane
Savelsberg, John Vanderelst, and Elisabeth Villalta. 2009. Crosslinguistic variation in
comparative constructions. In Linguistic Variation Yearbook, vol. 9, Jeroen van Crae-
nenbroeck and Johan Rooryck (eds.), 1–66. Philadelphia: John Benjamins.
Bierwisch, Manfred. 1989. The semantics of gradation. In Dimensional adjectives: Gram-
matical Structure and Conceptual Interpretation, Manfred Bierwisch and Ewald Lang
(eds.), 71–261 Berlin: Springer-Verlag.
Bobaljik, Jonathan David. 2012. Universals in Comparative Morphology. Cambridge, MA:
MIT Press.
Bochnak, M. Ryan. 2012. Managing vagueness, imprecision and loose talk in Washo: The
case of šemu. In Proceedings of Semantics of Under-represented Languages of the Ameri-
cas 6, Elizabeth Bogal-Allbritten (ed.), 1–14. Amherst, MA: GLSA Publications.
Bochnak, M. Ryan. 2013a. Cross-linguistic variation in the semantics of comparatives. PhD
diss., University of Chicago.
Bochnak, M. Ryan. 2013b. The non-universal status of degrees: Evidence from Washo. In
Proceedings of North-East Linguistics Society 42, Stefan Keine and Shayne Sloggett
(eds.), 79–92. Amherst, MA: GLSA.
Bogal-Allbritten, Elizabeth. 2008. Gradability and degree constructions in Navajo. Bach-

elors thesis, Swarthmore College.
Bogal-Allbritten, Elizabeth. 2010. Splitting pos: Evidence from Navajo for two pos mor-
phemes. Paper presented at the Workshop on Comparatives, MIT.
Bogal-Allbritten, Elizabeth. 2013. Decomposing notions of adjectival transitivity in Navajo.
Natural Language Semantics 21: 277–314.
Bogal-Allbritten, Elizabeth. To appear. Slightly coerced: Processing evidence for adjectival
coercion by minimizers. In Proceedings of the 48th Meeting of the Chicago Linguistics
Society: Parasession on Meaning and Cognition. Chicago: CLS.
Cresswell Max, J. 1976. The semantics of degree. In Montague Grammar, Barbara Partee
(ed.), 261–292. New York: Academic Press.
Dixon, R. M. W. 1982. Where have all the adjectives gone? In Where Have All the Adjectives
Gone? And Other Essays in Semantics and Syntax 1–62, New York: Mouton.
Eckardt, Regine. 2009. Nothing compares to you. In Proceedings of Semantics and Linguistic
Theory (SALT) XIX, Ed Cormany, Satoshi Ito, and David Lutz (eds.), 142–164. Ithaca,
NY: CLC Publications.
Fara, Delia Graff. 2000. Shifting sands: An interest-relative theory of vagueness. Philosophi-
cal Topics 28: 45–81 Originally published under the name “Delia Graff ”.
Francez, Itamar, and Andrew Koontz-Garboden. 2013. Semantic variation and the grammar
of property concepts. Unpublished Manuscript, University of Chicago and University
of Manchester.
Galant, Michael. 1998. Comparative constructions in Spanish and San Lucas Quiaviní Za-
potec. PhD diss. University of California, Los Angeles.
Heim, Irene. 1985. Notes on comparatives and related matters. Unpublished Manuscript,
University of Texas at Austin.
Hohaus, Vera. 2012. Directed motion as comparison: Evidence from Samoan. In Proceed-
ings of Semantics of Under-represented Languages of the Americas 6 and SULA-bar,
Elizabeth Bogal-Allbritten (ed.), 335–348. Amherst, MA: GLSA.
Jacobsen, William H. 1964. A grammar of the Washo language. PhD diss., University of
California, Berkeley.
Kamp, Hans. 1975. Two theories of adjectives. In Formal Semantics of Natural Language
Edward Keenan (ed.), 123–155. Cambridge: Cambridge University Press.
Kennedy, Christopher. 1999. Projecting the Adjective: The Syntax and Semantics of Grada-
bility and Comparison. New York: Garland. 1997 PhD diss., University of California,
Santa Cruz.
Kennedy, Christopher. 2007a. Modes of comparison. In Proceedings of CLS 43, Malcolm
Elliott, James Kirby, Osamu Sawada, Eleni Staraki, and Suwon Yoon (eds.), Chicago:
Chicago Linguistic Society.
Kennedy, Christopher. 2007b. Vagueness and grammar: The semantics of relative and abso-
lute gradable adjectives. Linguistics and Philosophy 30(1): 1–45.
Kennedy, Christopher, and Louise McNally. 2005. Scale structure, degree modification and
the semantics of gradable predicates. Language 81(2): 345–381.
Klein, Ewan. 1980. A semantics for positive and comparative adjectives. Linguistics and Phi-
losophy 4(1): 1–46.
Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal
taste. Linguistics and Philosophy 28: 643–686.

Mithun, Marianne. 1999. The Languages of Native North America. New York: Cambridge
University Press.
Morzycki, Marcin. 2012. Adjectival extremeness: Degree modification and contextually re-
stricted scales. Natural Language and Linguistic Theory 30(2): 567–609.
Neeleman, Ad, Hans van de Koot, and Jenny Doetjes. 2004. Degree expressions. The Lin-
guistic Review 21: 1–66.
Paradis, Carita. 2001. Adjectives and boundedness. Cognitive Linguistics 12: 47–65.
Pearson, Hazel. 2010. How to do comparison in a language without degrees: A semantics for
the comparative in Fijian. In Proceedings of Sinn und Bedeutung 14, Martin Prinzhorn,
Viola Schmitt, and Sarah Zobel (eds.), 356–372. Vienna, http://www.univie.ac.at/sub14/
proc/pearson.pdf.
Rett, Jessica. 2007. Antonymy and evaluativity. In Proceedings of Semantics and Linguistic
Theory (SALT) XVII, Tova Friedman and Masayuki Gibson (eds.), 210–227. Ithaca, NY:
CLC Publications.
Sapir, Edward. 1944. Grading: A study in semantics. Philosophy of Science 11: 93–116.
Sawada, Osamu. 2009. Pragmatic aspects of implicit comparison: An economy-based ap-
proach. Journal of Pragmatics 41: 1079–1103.
Seuren, Pieter. 1978. The comparative revisited. Journal of Semantics 3: 109–141.
Stassen, Leon. 1985. Comparison and Universal Grammar. Oxford: Basil Blackwell.
von Stechow, Arnim. 1984. Comparing semantic theories of comparison. Journal of Seman-
tics 3: 1–77.
Stephenson, Tamina. 2007. Towards a theory of subjective meaning. PhD diss., MIT.
Ultan, Russell. 1972. Some features of basic comparative constructions. Working Papers on
Language Universals 9: 117–162.
Young, Robert W. and William Morgan. 1987. The Navajo Language: A Grammar and Col-
loquial Dictionary. Albuquerque, NM: University of New Mexico Press.
5
Targeted Construction Storyboards

in Semantic Fieldwork
Strang Burton and Lisa Matthewson
1 Introduction
The goal of this paper is to describe, illustrate, and advocate for the targeted con-
struction storyboard methodology for semantic fieldwork.* Storyboards are pic-
torial representations of stories, which consultants are asked to tell in their own
words. The criterial property of targeted construction storyboards is that the story
is designed to include at least one targeted context that can be used to test hypoth-
eses about the relation between linguistic forms and that context. The storyboards
thus combine the advantages of spontaneous speech with the benefit of being able
to test hypotheses about particular linguistic elements or constructions. The find-
ings reported here draw on an ongoing storyboard project by researchers at the
University of British Columbia; sample storyboards created for that project are
available for download at http://www.totemfieldstoryboards.org/.
In section 2, we outline the motivation for using targeted construction story-
boards, illustrating our claims with a case study where the storyboard technique
has proved useful. Section 3 gives a detailed description of the storyboard method-
ology, including how to create the storyboards and how to use them in a fieldwork
setting. In section 4 we illustrate further results of our methodology, as it has been
used in testing hypotheses about modals in three unrelated languages: Gitksan
*
Many thanks to our language consultants for the languages reported on in this paper: Carl Alex-
ander, Beatrice Bullshields, Vincent Gogag, the late Gertrude Ned, and Barbara Sennott. Thanks also
to the UBC Totem Field Storyboards Team (Sihwei Chen, Heidi Johnson, Tracy Lam, Patrick Littell,
Valerie Marshall, Andy Matheson, Stacey Menzies, Katie Sardinha, and Daniella Zandbergen), to UBC
Gitksan researchers Clarissa Forbes, Michael Schwan, Alyssa Satterwhite, and Savanna van der Zwan,
to Henry Davis and Martina Wiltschko, and to the reviewers for this paper, Ryan Bochnak and Amy
Rose Deal. Funding for the Totem Field Storyboards Project came from SSHRC grants #410-2010-2627
and #410-2011-0431, a UBC AURA grant, a UBC Hampton Fund Grant, a UBC HSS Grant, and a Jacobs
Research Fund Grant. 135

(Tsimshianic), St’át’imcets (Salish), and Blackfoot (Algonquian). Section 5 reports

on a pilot experiment designed to test the naturalness of the speech produced
using storyboards (as perceived by native speakers); the result is that storyboard
speech is judged to be either as natural, or only marginally less natural, than spon-
taneous speech. Section 6 concludes.
2 Why Storyboard?
In this section we present the motivations for the targeted construction storyboard
methodology, illustrating the discussion with one particular storyboard, “Feeding
Fluffy.”
We take it as given that useful fieldwork on semantics must be hypothesis-
driven—that is, it must seek to answer some research question. In some cases, the
fieldworker begins with an initial hypothesis (based, for example, on prior results on
other languages) and attempts to falsify that hypothesis via empirical testing. In other
cases, the fieldworker begins with a “fishing” expedition by gathering some data in
the language under investigation, and then forms an initial hypothesis. In either case,
however, hypothesis-formation is a critical step that drives further testing, because it
makes concrete empirical predictions (see also Murray this volume for this point).
Suppose, for example, that the research question we are interested in is whether
epistemic modals can have a past temporal perspective. The relevant readings for
this question are those where a situation was possible or necessary given the agent’s
epistemic state at some past time (but potentially no longer at the utterance time).
For example, can (1) be uttered if the speaker’s knowledge at some past time was
consistent with them having won the game, but the speaker knows at the utterance
time that they lost?
(1) They might have won the game. (Condoravdi 2002)
These past epistemic readings are usually claimed to be absent for English (Groe-
nendijk and Stokhof 1975, Cinque 1999, Drubig 2001, Condoravdi 2002, Stow-
ell 2004, Hacquard 2006, 2011, Borgonovo and Cummins 2007, Demirdache and
Uribe-Etxebarria 2008, Laca 2008, among others), although a small minority of
authors argue that they are possible (von Fintel and Gillies 2008, Matthewson and
Rullmann 2012, Rullmann and Matthewson 2012 (see also Iatridou 1990 on related
constructions)). Beyond English, epistemic modals with a past temporal perspec-
tive have been argued to exist in languages such as French, Norwegian, Dutch, and
German (Fagan 2001, Boogaart 2007, Eide 2003, Homer 2010, Martin 2011, among
others; see also Portner 2009:222–236 for discussion). There is very limited dis-
cussion of the issue beyond Indo-European, and the disagreement about even the
English facts shows the difficulty and delicacy of the judgments.
If we are interested in whether epistemic modals in a fieldwork language can
have past temporal perspectives, a useful initial hypothesis is that they do have
Targeted Construction Storyboards 137
such readings. We will then attempt to elicit such readings, and if we find that the
relevant sentences are rejected in the past epistemic contexts, we will have to revise
(or completely throw out) our initial hypothesis.
This leads to a second important desideratum for fieldwork methodology: it
must enable the generation of negative data. That is, the methodology must enable
the researcher to discover what sentences cannot mean, as well as what they can,
and the situations in which utterances are infelicitous, as well as felicitous. For
arguments in support of both these desiderata, see Davis et al. (in press), and the
introduction to this volume, among many others.
Given these desiderata for data-collection methodology, targeted fieldwork is
not only a legitimate, but also an indispensible, part of the semanticist’s method-
ological toolbox. If we are following a scientific approach that consists of testing
hypotheses about semantics, we cannot proceed by gathering only spontaneously
created material.1 The researcher must be able to target particular forms and obtain
information about their (in)felicity in a specified range of discourse contexts. Our
methodology satisfies both these desiderata because it crucially involves targeted
storyboards; it differs from, for example, Chafe’s (1980) pear stories or Berman
and Slobin’s (1994) frog stories, which elicit narratives without targeting particular
linguistic structures. And our methodology allows the collection of negative evi-
dence, through follow-up elicitation.
However, there are non-trivial challenges and drawbacks to traditional direct
elicitation methodologies, especially when one is researching discourse-sensitive
semantic phenomena like the possibility for past temporal perspectives for modals.
Standard semantic elicitation techniques include three main types of tasks (setting
aside here, for the reasons given above, the collection of “spontaneous” untargeted
narratives). First, we have translation tasks: the speaker is asked to translate a sen-
tence from the contact language into their native language, or vice versa. Second,
we have judgment tasks, where the speaker is asked to judge whether a sentence
sounds acceptable and/or true in a specific discourse context. Third, there are elic-
ited production tasks, in which the speaker is asked to produce an utterance in
their language as a response to some stimulus.
All three of these tasks require the researcher to convey to the consultant a
discourse context or a stimulus of some sort. Note that this is true even of trans-
lation tasks in most cases, because a semantic fieldworker almost never asks for
a translation of a sentence given in isolation. While describing a discourse con-
text verbally works fairly well with many consultants, the descriptions can become
very complex, particularly when discourse-sensitive phenomena are being investi-
gated, as the past-epistemic case noted above illustrates. Lengthy descriptions can
tax the consultants’ memory, and/or be difficult to understand. It can be difficult
even to explain some necessary aspects of the context, as these may have to do
1
This includes both truly spontaneous material, as well as “spontaneous” narratives elicited during
fieldwork. See Bohnemeyer (this volume) and Cover (this volume) for discussion.
with what knowledge all the interlocutors already have, or what has been under
discussion so far.
With verbal context descriptions, the issue also arises of which language to
present the context in. Some researchers have argued that a discourse context
or stimulus presented in the contact language does not tarnish the results (Mat-
thewson 2004, Krifka 2011). AnderBois and Henderson (this volume) argue that
this claim is partially correct. They provide evidence that in some cases, elicited
production data can be influenced by a contact-language context description.
The alternative—presenting contextual discussion in the language targeted for
investigation—carries its own risks, including the difficulty of avoiding priming
of lexical items or structures.
A typical verbal context description is illustrated in (2) for our case study. (2)
presents data from Gitksan, a highly endangered Interior Tsimshianic language
spoken in Northern British Columbia. The elicitor here chose to present the dis-
course context in English.2
(2) Context: The Canucks were playing last night. You weren’t watching the game
but you heard your son sounding excited and happy from the living room where
he was watching the game, so you thought they were winning. You found out
after the game that the Canucks lost 20–0, and your son was happy about
something else that his friend had told him on his cellphone.
yugw=imaa=hl xsdaa-diit
ipfv=epis=cn win-3pl.ii
“They might have been winning.” [according to my evidence last night]
The context for (2) makes clear that it is no longer compatible with the speaker’s
knowledge at the utterance time that the Canucks were winning last night. The
acceptance of the sentence in this context therefore suggests that the epistemic
modal imaa does allow a past temporal perspective.
What are the drawbacks to this verbal elicitation method? One issue is that
the reading we are trying to get may not exist for the English sentence (given that
it is controversial whether English allows past-temporal perspectives for epistemic
modals like might). This means that the elicitor is hard-pressed to provide an Eng-
lish version of the sentence they are testing, and there is the potential for con-
fusion for the consultant. Other problems relate to the required complexity and
length of the context description. For these reasons, we have found that fieldwork-
ers often feel un-confident about the results obtained by the methodology in (2).
As pointed out also by Louie (this volume), consultants may show signs of having
2
Abbreviations not covered by the Leipzig glossing rules: i: series i pronoun, ii: series ii pronoun,
circ: circumstantial, cn: common noun connective, conj: conjunction, contr: contrastive, coun-
ter: counter to expectation, deon: deontic modal, epis: epistemic modal, exis: assertion of existence,
necess: necessity modal, pn: proper name, possib: possibility modal, prosp: prospective aspect, prt:
particle, red: reduplication, t: argument-structure adjuster.
misunderstood the intended discourse context. Consultants may accept the sen-
tences in the context given, but then offer comments suggesting they are thinking
of a slightly different context. Or they may offer English translations that do not
match the intended meaning.
There is one final issue that arises particularly with production data collected
via direct elicitation. While such data are in a sense spontaneously produced and
while they are likely to be fully grammatical and appropriate, they are nonetheless
open to the charge of non-naturalness.3 Among other issues, the task requires the
consultant to imagine themselves in a context they are not in, to imagine what they
would say if they were in that different context. This introduces a certain amount
of doubt about whether the response given is what the speaker would actually say,
were they to find themselves in that context.
The solution, we believe, to all these challenges is to supplement standard elic-
itation techniques with targeted construction storyboard elicitations. As we show
in this paper, storyboards enable the gathering of spontaneous, natural utterances,
with minimal or no contact-language influence or translation interference. Tar-
geted construction storyboards preserve the advantages of enabling the researcher
to target particular forms, and, when combined with follow-up elicitation, they
allow the collection of negative data. Storyboard elicitations therefore allow
hypothesis-testing, while minimizing contact-language influence and maximizing
naturalness of the utterances produced.
The targeted construction storyboard methodology also goes at least some
way toward solving the problems of verbal context descriptions. As the storyboard
method conveys the discourse context via pictures rather than verbally, it can be
more efficient at characterizing a context precisely. This means that we can mini-
mize the extent to which the context can be misunderstood, as well as the extent
to which consultants silently extrapolate from the context description. The issue of
which language the context is presented in is minimized (although not completely
eliminated). As will be seen in section 3, the storyboard methodology does con-
tain some discussion of the story in the contact language, to ensure understanding.
However, when consultants actually tell the story, they rely only on the visual pres-
entation of the context (or at least, they are not prompted or otherwise spoken to
during the storytelling by the researcher).
Let’s now see how the targeted construction storyboard technique helps with
our epistemic modals case study. The storyboard “Feeding Fluffy” was designed
to create a context in which it is absolutely clear that the speaker believed at some
earlier time that something was possible, but no longer believes it at the speech
time. In “Feeding Fluffy,” Pat asks his friend Stacey to pet-sit for him while he is
away. Stacey visits the store and realizes she doesn’t know what kind of pet Pat
3
For example, Mosel (2012) writes that “there is always the danger that created examples sound
unnatural,” and “by definition, elicitation only provides examples of decontextualized isolated sen-
tences, whereas natural speech is always embedded in the context of a particular speech situation.”
has. She thinks it could be a cat, a dog, or a rabbit, so she buys food for all those
animals. She later finds out that Fluffy is actually a snake. When Pat questions her
about why she bought cat-, dog- and rabbit-food, she replies that Fluffy might have
been a cat, a dog, or a rabbit.
Three pictures from “Feeding Fluffy” are illustrated in Figure 5.1 (in reduced size).
(For the full storyboard, see http://www.totemfieldstoryboards.org/stories/feeding_
fluffy/.) These panels show Stacey in the pet store thinking that Fluffy might be a dog,
deciding to buy a bone, and telling Pat afterward that Fluffy might have been a dog.
Figure 5.1 Targeted construction sequence from Feeding Fluffy

Some Gitksan results for “Feeding Fluffy” are given in (3–4). In this language,
epistemic modals are easily obtained at the relevant spots. They are sometimes
spontaneously offered, but if not, are accepted on follow-up elicitation. Although
there is no overt past-tense morphology in the sentences (because there is no overt
past-tense morphology in the language), the storyboard context ensures that the
modal is interpreted at a past time.4
(3) yugw=imaa=hl duus-t diya-t Stacey

ipfv=epis=cn cat-3sg.ii say-3sg.ii Stacey
“‘He might have been a cat,’ said Stacey.” (Feeding Fluffy)
(4) yugw=imaa=hl gax

ipfv=epis=cn rabbit-3sg.ii
“He might have been a rabbit.” (Feeding Fluffy)
Based on “Feeding Fluffy,” we are confident that the epistemic modal imaa in Gitk-
san allows a past temporal perspective.
Blackfoot (Algonquian) is another language in which “Feeding Fluffy” has
been tested. The consultant does not spontaneously offer plain epistemic modals
when telling the “Feeding Fluffy” story. Instead, she offers the modal embedded
under a higher attitude verb. However, on follow-up elicitation the consultant ac-
cepts an unembedded epistemic modal, as shown in (5) (thanks to Meagan Louie
for the data).
(5) matonni nimaatssksini’pa otaanista’pssi piiksiksinaa

matonni ni-maat-ssksini-’p-wa ot-aanist-a’pssi-wa piiksiksinaa-wa
yesterday 1-neg-know.vti-loc:0-nonaff 3-manner-be.vai-3 snake-3
“Yesterday, I didn’t know it was a snake,”
aahkamomitaa
aahkam-omitaa-wa
epis-dog-3
“I thought it was a dog.”
Blackfoot researchers consistently report that pictures are important for obtaining
these past epistemic readings with confidence.
The comparison between Gitksan and Blackfoot highlights another advantage
of the storyboard methodology, namely that it allows entirely consistent contexts
4
English translations of data throughout the paper were either offered or agreed to by the consul-
tant in a subsequent elicitation session.
Table 5.1
Why Storyboard?
Challenge Standard elicitations Targeted construction storyboard elicitations
Easy to convey discourse Can be difficult to explain Context given primarily by the story
contexts? verbally
Contact-language interference? Potentially/sometimes Very minimal to none
Natural speech? Potentially non-natural More natural
Easily replicable by other In theory, yes. In practice, Yes (sometimes with changes for
researchers? only if the researcher local cultural context)
provides a full description of
techniques and all contexts
Easily transferable to No Yes
curriculum materials?
to be presented to different speakers, even in different languages and by different

researchers. The methodology therefore meets the important scientific criterion of
the potential for replicability.5
Storyboard elicitations have two further advantages that do not relate directly
to the scientific endeavor but which are welcome side-effects. The first is that they
enable fast transfer to teaching materials. Language learners benefit from and
enjoy reading and hearing connected text, and storyboards provide pre-illustrated
stories that can quickly be integrated into language-teaching curricula. Finally,
storyboards are fun, not only for language learners, but for consultants during
the elicitation itself. Most of our stories contain an amusing twist, and consultants
often smile or laugh while telling the stories.
Table 5.1 summarizes the claims of this section. The “Standard elicitations”
column refers to research conducted only via standard techniques (excluding
asking for “spontaneous” narratives). The storyboard methodology includes a
standard elicitation sub-part; we are therefore not advocating abandoning standard
elicitation techniques. Recall also that both standard elicitations and storyboard
elicitations share the core advantages of being able to support hypothesis-driven
testing and to provide negative data.
3 Targeted Construction Storyboard Methodology
As noted above, the crucial feature of a targeted construction storyboard is that

it presents a story in pictures that leads up to and establishes a desired narrative
5
There is a proviso here: it is difficult to draw pictures that are fully culturally neutral. Researchers
may need to redraw storyboards for use in certain cultural contexts. In some cases, the story itself may
need to be altered to suit local traditions or technologies. However, the core plot of the story, and the
critical aspects of the context, can usually be preserved.
context; having built up to that context, the linguist can then elicit positive and
negative data from the speaker about permissible forms in that context. This sec-
tion explains from a practical perspective how to construct and use storyboards of
this type. We outline first how the storyboards are created and organized, and then
the methodology for using them.
3.1 ORGANIZING STORYBOARDS INTO NARRATIVES
Although it is often possible to present a context in targeted construction story-

boards with just one or two pictures, we have found that speakers generally re-
spond better to stories, i.e., presentations with a simple narrative arc. Although
there are certainly many effective narrative structures, our targeted storyboards
generally follow this narrative pattern, which seems to work reasonably well for a
variety of speakers and cultures:
¤ An introduction to the characters, ideally in a relationship the speaker

can relate to, such as a family or friend relationship; for example, Figure
5.2 shows (in reduced size) the introductory frames for “Woodchopper”
(http://www.totemfieldstoryboards.org/stories/woodchopper/), in which
we introduce the two characters and explain that their house is beside a
river.
Figure 5.2 Introductory panels for Woodchopper
¤ A logical series of events that leads up to the targeted structure. For

example, Figure 5.3 shows the lead-in from “Woodchopper”: John says
he’s going to chop some wood to make fire; Mary says don’t do it, as he’ll
drop some wood and she’ll fall in the river; John ignores her, and chops
the wood anyway; he does drop a piece on the path, and when she goes
out at night she trips over it and falls in the river.
Figure 5.3 Lead up to the targeted structure

Figure 5.4 The targeted context, tested twice
¤ One or more iterations of the targeted narrative context (often we find

three repetitions works well, in a reasonably short story). For example,
Figure 5.4 shows two panels that target past-counterfactual forms like
If you hadn’t chopped the wood, I wouldn’t have fallen in the river, and I
wouldn’t have gone over the waterfall.
As a reviewer points out, repetitions may not strictly speaking lead to independent
data points, due to potential priming effects. For example, a speaker may choose
one modal over another in the first iteration, and then continue to use the same
one in subsequent repetitions, even though an alternative modal would also have
been felicitous. However, repetitions do provide examples of the construction with
more than one set of lexical items; this is important in case particular lexical items
have idiosyncrasies. Priming effects can be mitigated during follow-up elicitation,
by asking the consultant to substitute the alternative form.
¤ An engaging conclusion such as a simple plot twist on the third repetition of
the context. For example, Figure 5.5 shows the conclusion to “Woodchopper,”
where John says something like “Well, if I hadn’t chopped the wood, you
wouldn’t have (had) this nice fire to get warm with!” (in this case again
potentially eliciting another past counterfactual in the appropriate context).
Figure 5.5 The conclusion with a “twist”

Generally we find that a simple story of this pattern can be told in a series of
10–20 images, which leads to roughly 1–2 minutes of narrative. Although the
establishment of a full story takes slightly longer than just presentation of a con-
text in one or two pictures, we believe the added effort is worthwhile: by engag-
ing the speaker, we ensure that they become involved in telling the story, rather
than focusing on the linguistic issues, and the storytelling comes more naturally.
3.2 CREATING STORYBOARDS
The following steps are an effective way to develop new targeted construction
storyboards.
1. Decide on the targeted structure, and plan out a narrative (in notes, without
drawings) that you think will work well to elicit that structure.
2. Do quick and rough drawings, such as stick-figure drawings, and try them
with a speaker.
3. Adjust the rough drawings, adding and removing frames where anything
did not work well with the narrative, and test again with a speaker to
confirm they are working well.
4. If you find you have a highly effective storyboard, consider arranging
higher-quality drawings, and sharing them with other linguists for cross-
linguistic comparisons.
The initial rough drawings can be done simply on paper, and no real artistic talent
is required at that stage: speakers can generally follow stories even when they are
drawn with stick figures.
3.3 PRESENTING STORYBOARDS TO SPEAKERS
In the methodology we have used with targeted construction storyboards, we have

found the following series of steps to be generally effective.
1. First, the linguist goes through the pictures while telling the story to the
speaker in the contact language. This helps the speaker to see what is
intended in each picture.
2. Optional: The speaker practices telling their own version in the target
language. At this stage the speaker may need to reflect for a short time on
vocabulary items and phrasing.
3. Once the speaker is comfortable telling the story, the linguist records the
story with the speaker.
4. The linguist will then usually have follow-up questions, going back
through the story eliciting further positive and negative data at various
points; these follow-up questions may be done immediately, or at the next
elicitation session, or a combination.
Follow-up elicitation usually starts with reminding the consultant about the
story, and reviewing the sentence they originally gave. Speakers almost always
have good recall of the plotline of the stories, as they are much more memora-
ble than unconnected standard elicitations; thus it becomes relatively very easy
to reproduce the same context, and ask follow-up questions where the speaker
judges whether alternative versions of the sentence are also acceptable in the
context.
4 Further Examples of Results
The Totem Fields Storyboards project can be found at http://www.totemfieldsto

ryboards.org/. A sub-set of the Totem Fields storyboards (called the Hampton Sto-
ryboard Collection) is a set of storyboards designed to test questions about mo-
dality. “Feeding Fluffy,” discussed in section 2, belongs to this collection. In this
section we summarize some empirical and theoretical results obtained via two
more Hampton storyboards: “Chore Girl” (http://www.totemfieldstoryboards.
org/stories/chore_girl/) and “Sick Girl” (http://www.totemfieldstoryboards.org/
stories/sick_girl/).
4.1 “CHORE GIRL” AND “SICK GIRL”
“Chore Girl” and “Sick Girl” are designed to investigate whether and how lan-
guages distinguish between different types of modality and different strengths of
modality. Thus, these storyboards elicit the contrast (or lack of one) between de-
ontic possibility (permission, as in (6a)), non-deontic circumstantial possibility
(ability, as in (6b)), and deontic necessity (obligation, as in (6c)). Both storyboards
also elicit interactions of modals with negation.
(6) a. I can come out to play (I am allowed to).

b. I can come out to play (I’m not sick or injured).
c. I have to do my chores.
The basic plot line of both “Chore Girl” and “Sick Girl” involves a girl whose
friends repeatedly ask her to come out and play. In “Chore Girl,” she is not
allowed by her mother to go out and play until after she has done her three
chores. After she finally completes her chores, she falls down and breaks her
leg. At this point, she is permitted to play, but is physically unable to play. The
three final pictures from this storyboard are given in Figure 5.6 (in reduced
size) for illustration. When asked by her friends to come out and play, the girl
reports that her mother has allowed her to go, but she is unable to, due to her
broken leg.
4 p.m.
Figure 5.6 The conclusion of Chore Girl
In “Sick Girl,” a girl repeatedly injures herself or gets sick, so is unable to go out
to play. When she finally recovers from all her ailments, she is physically able to
play, but is now not permitted to go out and play, because her mother says she has
to do her homework. Three sample pictures from this storyboard are illustrated in
Figure 5.7 in reduced size.
Figure 5.7 Sample pictures from Sick Girl
“Chore Girl” and “Sick Girl” have been elicited by the Totem Field Storyboards
team in American Sign Language, Atayal (Formosan), Blackfoot (Algonquian),
Coeur D’Alene (Salish), English, German, Gitksan (Tsimshianic), Halkomelem
(Salish), Kwak’wala (Wakashan), Mandarin, Okanagan (Salish), Russian, and
St’át’imcets (Salish). Results from “Chore Girl” and “Sick Girl” are also being used
by a number of other researchers; see for example Vander Klok (2012) on Javanese.
Through using “Chore Girl” and “Sick Girl,” our team has found that many of
the languages we investigated possess a general circumstantial possibility modal,
which can cover both ability and permission readings. Such languages may also
possess a separate dedicated permission modal (cf. English can vs. be able to). We
have also established that in almost all the languages looked at, negation scopes
over a possibility modal. That is, negation plus a possibility modal has the inter-
pretation “not possible” (not permitted / not able), rather than the interpretation
“possibly not” (permitted not to / able not to). In our sample, the one exception to
this generalization was ASL, in which there is a separate lexical item for negated
circumstantial possibility.
If the storyboards do not initially elicit any distinction between ability and
permission, follow-up elicitation is used to elicit the dedicated permission modal.
If distinct modals are obtained within the stories, follow-up elicitation will try to
switch the two forms in the relevant sentences (to obtain negative evidence about
what each modal cannot do).
In the following sections we briefly report on specific results from two lan-
guages: Gitksan and St’át’imcets. For results from other languages, including ASL,
Atayal, Kwak’wala, Mandarin, and Russian, see Burton and Matthewson (2011).
4.1.1 Gitksan
“Chore Girl” and “Sick Girl” reveal that Gitksan has a lexical distinction between
permission (marked by the verb anook) and ability (marked by the verb da’akxw),
and that the plain prospective aspect marker dim is often used for deontic neces-
sity (obligation). Permission is shown in (7), and ability in (8).6
(7) a. ii-t anook-s nox-t

and-3sg.ii deon.possib-pn mother-3sg.ii
“And her mother let her.” (Sick Girl)
b. ii-t anook-s nox-t dim ma’us-t galk

and=3sg.ii deon.possib-pn mother-3sg.ii prosp play-3sg.ii outside
“Her mother let her play outside.” (Sick Girl)
(8) a. ii nee=dii da’akxw dim ma’us-t

and neg=contr circ.possib prosp play-3sg.ii
“And she was not able to play.” (Sick Girl)
b. ii nee-dii-n da’akxw dim xsaw-’y

and neg-contr-1sg.i circ.possib prosp go.out-1sg.ii
“And I am not able to go out.” (Chore Girl)
(9) shows that the modal da’akxw can be used for permission as well as ability. We
analyze this element as a general circumstantial possibility modal (Matthewson 2013).
6
The same consultant offered (7a, 7b and 8a) in a single telling of the “Sick Girl” story, and (8b, 9,
10a and 10b) in a single telling of the “Chore Girl” story.
(9) ii he-s Mary: “mahl-d-i-s no’o-’y dim da’akxw dim

and say-pn Mary tell-t-tr-pn mum-1sg.ii prosp circ.possib prosp
ma’us-’y.”
play-1sg.ii
“And Mary said: ‘My mother told me I could play.’” (Chore Girl)
(10) shows that the ordinary prospective aspect marker dim can be used in con-
texts of deontic necessity (obligation). Whether dim actually expresses deontic
necessity in its lexical semantics is a matter of ongoing investigation; see Peterson
(2011), Matthewson (2013). In follow-up elicitation, the sentence in (11) containing
the dedicated circumstantial necessity modal, sgi, was offered to the consultant in
the context of the story and accepted.
(10) a. dim saks-in-n=hl no’ohl

prosp clean-caus-sg.ii=cn dishes
“You will wash the dishes.” [Target: “You have to . . . .”] (Chore Girl)
b. ii he-s Mary: “yukw dim yats=hl lakw.”

and say-pn Mary ipfv prosp chop=cn fuel
“And Mary said: ‘I’m going to chop wood.’”
[Target: I have to . . . .] (Chore Girl)
(11) sgi mi=dim saks-in=hl no’ohl

circ.necess 2sg.i=prosp clean-caus=cn dishes
“You have to do the dishes.”
4.1.2 St’át’imcets
St’át’imcets (a.k.a. Lillooet) is a highly endangered Northern Interior Salish lan-
guage spoken in the southern interior of British Columbia. “Chore Girl” and “Sick
Girl” show that in St’át’imcets, the circumfix ka- . . . -a is used in both permission
and ability contexts. This is illustrated in (12–14).
(12) tsut ta=skícza7-s=a cw7áoz=ka séna7 kw=á=su

say det=mother-3sg.poss=exis neg=irr counter det=ipfv=2sg.poss
ka-sáy’sez’-a t’u tsekw-tsukw i=kalhás=a
circ-play-circ until red-finish det.pl=three=exis
s-7álkt-su
nmlz-work-2sg.poss
“Her mother said, ‘You can’t play until your three chores are finished.’”
(Chore Girl)
(13) tsut s=Mary, wa7 tsut ta=n-skícez7=a

say nmlz=Mary ipfv say det=1sg.poss-mother=exis
kw=en=s=wá7 ka-nás-a sáy’sez’
det=1sg.poss=nmlz=ipfv circ-go-circ play
“Mary said, ‘My mother said I could go play.’” (Chore Girl)
(14) t’u7 cw7aoz kw=en=wá ka-sáy’sez’-a, xán’=lhkan nilh

but neg det=1sg.poss=ipfv circ-play-circ hurt=1sg.sbj foc
s=qácwecw n-sq’wáxt=a
nmlz=break 1sg.poss-foot=exis
“But I can’t play, I hurt myself and broke my foot.” (Chore Girl)
Two different speakers consistently expressed obligation in these storyboards by

using the prospective aspect marker huy’, rather than the deontic modal ka. Al-
though ka is independently known to allow obligation readings,7 findings such as
these can reveal information about alternative ways to express obligation.
(15) tsut kw=s=Mary, “cw7aoz, húy’=lhkan n-ts’áw’-cal

say det=nmlz=Mary neg prosp=1sg.sbj loc-wash-act
“Mary said, ‘No, I’m going to wash dishes.’” [Target: “I have to . . . .”]
(Chore Girl)
(16) tsut kw=s=Mary cw7aoz káti7 húy’=lhkan=t’u7 t’cw-áy’lup

say det=nmlz=Mary neg deic prosp=1sg.sbj=prt sweep-floor
múta7 ts’áw’-lap
and wash-floor
“Mary said, ‘No way, I’m going to sweep and wash the floor.’”
4.1.3 Summary of “Chore Girl” and “Sick Girl”

We have found the “Chore Girl” and “Sick Girl” storyboards to be very useful in
determining how a language can mark the notions of deontic possibility, deontic
necessity, and ability, as well as negated ability and permission. The results include
negative data, not given here for reasons of space; the fieldworker can easily, for
example, establish that a dedicated obligation modal is inappropriate in a permis-
sion context.
7
As in, for example, (i).
(i) qwatsáts=ka ta=sísqa7-sw=a
leave=deon.necess det=uncle-2sg.poss=exis
“Your uncle should leave.” (Davis 2012:ch. 24)
5 Pilot Study on Naturalness
Because storyboard elicitations involve a distinct form of interaction with the

speaker, the question arises of how natural storyboarded data samples are in com-
parison with other forms of elicitation. Specifically, we may wonder how the re-
sults of the storyboard methodology compare with other narratives with respect
to factors including intonation patterns, vocabulary choice, narrative connectives,
and discourse particles. Because our methodology involves a presentation of the
story in the communication language, prior to the elicitation, the question also
arises of whether translation interference may arise.
As a preliminary exploration of the comparative naturalness of storyboards,
we have therefore conducted a small pilot study, comparing a narrative elicited
with storyboards with a narrative elicited without storyboard prompts. For this
pilot, we used the following methodology. Using a native-speaking Japanese con-
sultant, with English as the communication language between the recorder and
the speaker, we recorded two short narratives, of similar length. One story was
recorded using the “Woodchopper” storyboard, following the methodology de-
scribed above in this paper. The other story was a more “spontaneous” narrative,
recorded by asking the speaker to recollect a series of events from recent memory.
The second story was recorded with no prompts (though, as with storyboards, the
speaker was allowed to practice until she was comfortable prior to recording). We
then asked five other native Japanese speakers to listen to and compare the audio
recordings for the two narratives, and to evaluate the naturalness of the record-
ings, defining “natural” to mean that the narrative contained forms that they felt
a Japanese speaker would normally use in that context. We specifically asked the
evaluators to evaluate each of the two narratives for naturalness (as defined here)
with respect to four factors: (i) vocabulary choice, (ii) intonation patterns, (iii)
narrative transitions, (iv) use of discourse sensitive markers such as discourse par-
ticles. Because our storyboard methodology involves some discussion in English
prior to recording, we also asked the reviewers to evaluate whether they felt they
could detect translation interference in either narrative. On each item, for both
forms of elicitations, reviewers were asked to rate the narrative on a scale of a–d,
where “a” is completely natural, and “d” is clearly unnatural and distorted speech.
(The reviewers were also asked to identify and comment on specific issues identi-
fied in the narratives, though those comments are not reproduced here.) Because
we referred in our study to linguistic terms, the reviewers for this pilot study were
all linguists.
The results of the pilot study are presented in Table 5.2. In the table, FM =
From Memory, FS = From Storyboard. The response “a” marks most natural, while
“d” marks very unnatural usage.
The results suggest that the storyboarded narrative, though somewhat less
“natural” overall than narratives elicited from memory, is quite close, and com-
parable. In 16 out of 24 comparisons, the storyboarded narrative was rated as the
Table 5.2
Results of Pilot Study on Naturalness of Storyboarded Elicitations Versus Story Told from
Memory with No Aids
Reviewer 1 Reviewer 2 Reviewer 3 Reviewer 4 Reviewer 5
FM FS FM FS FM FS FM FS FM FS
Vocabulary choices b b b a b b n/a a a b

Intonation patterns a a a b a a a b a a
Narrative transitions b b a a b b a b a b
Discourse sensitive items b b a b b b b b a b
Freedom from apparent b b b a c a b b a c
translation interference
Table 5.3
Summary of Pilot Study Results (average scores)
vocabulary intonation narrative discourse freedom from
choices patterns transitions sensitive items translation interference
From Memory 1.75 1.4 1.6 1.6 2

From Storyboard 1.6 1.8 2 2 1.8
same or higher in naturalness than the completely spontaneous narratives; in all

but one case it was rated as within one point; and in just one case it was rated two
points lower than the story from memory. The storyboard story was rated higher
in naturalness than the story from memory in 3 out of 24 cases.
Table 5.3 summarizes the pilot study results numerically. We assigned all “a”
responses a value of 1, “b” responses a value of 2, and so on, and then averaged the
responses over all five participants. Cases where the storyboard narrative were on
average rated as more natural than the original narrative are highlighted.
Unfortunately, for this pilot study we were not able compare the two narratives
with narratives obtained via other elicitation methods, such as phrase-by-phrase
translations from English, but it seems plausible that the storyboard technique will
give better results than this method. If correct, this suggests that storyboarded nar-
ratives are at a high intermediate level of naturalness, not quite as natural as speech
with no prompt, but also not highly degraded from it, and plausibly much more
natural than speech elicited phrase by phrase.
6 Conclusion
Good linguistic fieldwork will always involve a variety of elicitation methods, each
with advantages and limitations; in this paper, we have outlined how targeted con-
struction storyboards can add to that variety. The advantages of this methodology
include the fact that it allows us to collect relatively spontaneous speech while
targeting specific linguistic constructions, and in contexts that other researchers

can easily reproduce in any language. Using follow-up elicitation, the same sto-
ryboards can also be used to collect negative evidence. And finally, perhaps most
important of all, many consultants find that storyboards are an easy, even fun, way
for them to work with linguistic researchers.
References
Berman, R., and D. Slobin (eds.). 1994. Relating Events in Narrative: A Crosslinguistic Devel-
opmental Study. Hillsdale, NJ/Hove UK: Lawrence Erlbaum.
Boogart, R. 2007. The past and perfect of epistemic modals. In Recent Advances in the
Syntax and Semantics of Tense, Aspect and Modality, L. de Saussure, J. Moeschler, and
G. Puskas (eds.), 47–70. Berlin: Mouton de Gruyter.
Borgonovo, C., and S. Cummins. 2007. Tensed modals. In Coreference, Modality, and Focus,
J. Eguren and O. Fernández Soriano, (eds.), 1–18. Amsterdam: John Benjamins.
Burton, S., and L. Matthewson. 2011. www.TotemFieldStoryboards.org. Paper presented at
the 46th International Conference on Salish and Neighbouring Languages, University
of British Columbia.
Chafe, W. (ed.). 1980. The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of Narra-
tive Production. Norwood, NJ: Ablex Publishing.
Cinque, G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford:
Condoravdi, C. 2002. Temporal interpretation of modals. In The Construction of Meaning,
D. Beaver, S. Kaufmann, B. Clark, and L. Casillas (eds.), 59–87. Stanford, CA: CSLI.
Demirdache, H., and M. Uribe-Etxebarria 2008. On the temporal syntax of non-root
modals. In Time and Modality, J. Guéron and J. Lecarme (eds.), 79–113. Dordrecht:
Springer.
Davis, H. 2012. Teaching grammar of Upper St’át’imcets. Unpublished manuscript, Univer-
sity of British Columbia.
Davis, H., C. Gillon, and L. Matthewson. In press. Towards a scientific approach to linguis-
tic typology. Language (Perspectives section).
Drubig, H. 2001. On the syntactic form of epistemic modality. Unpublished manuscript,
Universität Tübingen.
Eide, K. 2003. Modals and tense. In Proceedings of Sinn und Bedeutung 7, M. Weisgerber
(ed.), 120–135. Konstanz: Fachbereich Sprachwissenschaft der Universität Konstanz.
Fagan, S. 2001. Epistemic modality and tense in German. Journal of Germanic Linguistics
13: 197–230.
von Fintel, K. and A. Gillies 2008. CIA leaks. Philosophical Review 117: 77–98.
Groenendijk, J., and M. Stokhof. 1975. Modality and conversational information. Theoretical
Linguistics 2: 61–112.
Hacquard, V. 2006. Aspects of modality. PhD diss., MIT.
Hacquard, V. 2011. Modality. In Semantics: An International Handbook of Natural Language
Meaning, K. von Heusinger, C. Maienborn, and P. Portner (eds.), 1484–1515. Berlin: de
Gruyter.
Homer, V. 2010. Epistemic Modals: High ma non troppo. Proceedings of NELS 40. Amherst,
MA: GLSA.
Iatridou, S. 1990. The past, the possible, and the evident. Linguistic Inquiry 21: 123–129.
Krifka, M. 2011. Varieties of semantic evidence. In Semantics: An International Handbook
of Natural Language Meaning, K. von Heusinger, C. Maienborn, and P. Portner (eds.),
242–267. Berlin: de Gruyter.
Laca, B. 2008. On Modal Tenses and Tensed Modals. Unpublished manuscript, Université
Paris 8 / CNRS.
Martin, F. 2011. Epistemic modals in the past. In Romance Languages and Linguistic Theory,
Berns, J., H. Jacobs, and T. Scheer (eds.), 185–202. Oxford: Benjamins.
Matthewson, L. 2004. On the methodology of semantic fieldwork. International Journal of
American Linguistics 70: 369–415.
Matthewson, L. 2013. Gitksan modals. International Journal of American Linguistics 79:
349–394.
Matthewson, L., and H. Rullmann. 2012. Epistemic modality with a past temporal perspec-
tive. Paper presented at the Modality @ Ottawa U. Workshop.
Mosel, U. 2012. Morphosyntactic analysis in the field: A guide to the guides. In The Oxford
Handbook of Linguistic Fieldwork, N. Thieberger (ed.), 72–89. Oxford: Oxford Univer-
sity Press.
Peterson, T. 2011. The role of the ordering source in Gitksan modals. Paper presented at Se-
mantics of Under-Represented Languages in the Americas 6, University of Manchester.
Portner, P. 2009. Modality. Oxford: Oxford University Press.
Rullmann, H., and L. Matthewson. 2012. Epistemic modals can scope under past tense.
Paper presented at the Texas Linguistic Society.
Stowell, T. 2004. Tense and modals. In The Syntax of Time, J. Guéron and J. Lecarme (eds.),
621–636. Cambridge, MA: MIT Press.
Vander Klok, Jozina 2012. Tense, aspect, and modal markers in Paciran Javanese. PhD diss.,
McGill University.
6
Reasoning About Equivalence in Semantic

Fieldwork
Amy Rose Deal
Semantic fieldwork is about figuring out the truth and felicity conditions of
expressions of a natural language on the basis of the behavior of other human
beings.* You can be a semantic fieldworker far from home, or in your own com-
munity. You can concentrate on your native language, or on a language very differ-
ent from what you know. When you are doing fieldwork, though, your project is
always centered on other people. It’s about figuring out what they know about how
meaning is conveyed in a language that belongs to them.
There’s a challenge in that. We all know that speakers aren’t endowed with any
direct access to the generalizations they employ in the creative use of language. Nor
do they have direct access to the content of the lexical items that tend to be most in-
teresting to semanticists—tenses, modals, articles, discourse particles, and the like.
This means that particular methodologies for semantic fieldwork have to be built up
around indirect approaches to speaker knowledge. We go about it in various ways.
We ask for translations by asking speakers questions like “How would you say that
in your language?” or (frequently) “What does that mean?” We ask for judgments
by producing a sentence for speakers in the context of some linguistic or nonlin-
guistic information, and asking, in view of that information, “Could you say that?”
or “Would that be an okay thing to say?” Then we reason from the answers speakers
give to a theory of the ways in which meanings are conveyed in their language.
The subject of this paper is that process of reasoning. How do we get from
some data points about speaker behavior to a view of linguistic meaning? It’s a
question we must take seriously in thinking about methodological matters. After
*
I have the happy responsibility to thank my friends and colleagues in three groups: (1) Ryan
Bochnak and Lisa Matthewson for organizing the session at the 2012 LSA meeting from which this
paper grew; (2) Florene Davis and Bessie Scott, who helped me appreciate the Nez Perce facts that are
reported here and in my other work on Nez Perce modals (Deal, 2011); (3) Sandy Chung, Ryan Boch-
nak, Lisa Matthewson, and an anonymous reader, for helpful comments on the manuscript. 157

all, we spend the time to go out and collect particular data points in view of a
theory of how one might reason from such data points to wider conclusions. The
theory usually remains implicit. What I want to do here is try to make explicit
some of the hypotheses that are at stake, in particular as concerns translation and
judgment tasks. These hypotheses matter because they are useful and natural but
also imperfect. Their imperfections must be weighed against their usefulness in
the practice of research.
Let me get started by walking through two particular hypotheses a field-
worker might use in reasoning about translation and about judgment. That
should illustrate the point about usefulness and naturalness. It takes us to the
issue of imperfection, however, which is the topic of sections 2 (on translation)
and 3 (on judgment). In these sections I will emphasize how the reasoning we
apply to basic fieldwork data can influence the types of semantic analyses at
which we arrive. We stand the best chance of navigating toward the types of
theories we want to produce when we keep this multiplicity of possible interpre-
tations in mind.
1 Two Hypotheses
We’ve just met, you and I. After chatting a little bit and getting to know one an-
other, we’ve agreed that you will help me learn about your community’s language.
So I ask you, “How do you say ‘cat’ in your language?” and you respond, “Picpic.”
What can I conclude from that response?
Nothing, without a theory of translation. Fortunately, I have such a theory,
and it’s pretty easy to state it explicitly. It’s something like this.
(1) Equivalent translations hypothesis (ETH)
The input to translation and the output of translation are equivalent in
meaning.
Without the ETH or some alternative hypothesis playing a parallel role, all I
have gathered is a data point about how you responded to a question. That’s not
in itself something that’s very useful for linguistics research. But I hypothesize
that we both understand that I have asked you for a translation, that you’ve re-
sponded in a cooperative way, and that we are in agreement that the task works
as the ETH proposes. Now I can conclude something useful, namely that cat in
our shared language (English) and picpic in your language (Nez Perce) mean the
same thing.
I don’t want to stop there, of course. Once we’ve been chatting a little bit
longer, I might ask you a more complex question, like, “How could you say, ‘There’s
a cat outside’?” Here’s your response:
(2) Picpic hiiwes ’eemtii.
Reasoning About Equivalence 159
This response, in view of the ETH, is very useful indeed. I’ve asked you for a trans-
lation of an English sentence that quantifies existentially over cats, and if I’ve rea-
soned correctly, you’ve given me back a Nez Perce sentence that does that, too.
Now I can start to think about all the sorts of things we think about in doing se-
mantics. I can investigate the formal properties of that quantification and how they
are related to the sentence’s words and structure. I can investigate how that quan-
tification behaves in discourse and in reasoning. I am started on a research proj-
ect. Your answers to my questions didn’t get me to that point on their own. They
got me there together with a theory of how those answers can be interpreted—a
theory in which the ETH plays a central role.
Let’s fast forward a bit to a different sort of scenario. Now that we’ve been
meeting with each other for a little while, I can start to construct and produce sen-
tences in your language that you understand and you judge to be grammatically
well-formed. This means we can start to talk about semantic judgments. We are
sitting in the town of Lapwai, and I tell you a little story about our friend Harold.
Harold, I say, is down the road in the town of Clarkston, and he is wondering
where I am. “If that’s so,” I say, “could I say
(3) Pay’s Harold hinekise wees Clarkstonpa.
Would that make sense?” Yes, you say. What can I conclude from that?
I need a theory again. You haven’t translated anything from one language to
another, so the reasoning doesn’t turn on the ETH. A major hypothesis I could use
is rather something like this.
(4) Equivalent judgments hypothesis (EJH)
In a particular context, speakers accept/reject sentences expressing the same
range of propositions regardless of what language they are speaking.
I hypothesize that we both understand that I have asked you, in view of a context,
to accept or reject a sentence that expresses some particular proposition (or family
of propositions, should it be an ambiguous sentence), and that you’ve responded
cooperatively. Your acceptance lets me conclude, by the EJH, something about
what the proposition expressed could possibly be. It is not the proposition ex-
pressed by these English sentences:
(5) Maybe Harold thinks I am here in Clarkston.
(6) Harold is not in Clarkston.
I know this because English speakers reject these sentences in the context I’ve
described. (I can ask you about the English sentences to confirm that.) Accord-
ing to the EJH, there will be no context where someone would reject a sentence
expressing proposition ϕ if speaking English, but accept a sentence expressing ϕ
if speaking Nez Perce. To figure out what proposition it is that (3) expresses, then,
I should consider what is expressed by English sentences that are accepted in this
context, like
(7) Maybe Harold thinks I am in Clarkston.

(8) Harold is in Clarkston.
(9) Harold is wondering whether I am in Clarkston.
Since I have some idea of which propositions are expressed by which English sen-
tences, I can start to narrow down the range of possible propositions that could be
expressed by Nez Perce sentence (3).
Now I want to ask you about a sentence that is like (3), but contains an addi-
tional word. “What about this,” I say. “In that same situation where Harold is in
Clarkston wondering where I am, could I say this sentence?
(10) Pay’s Harold hinekise wees kine Clarkstonpa.”
You shake your head no. “You could only say this if you were in Clarkston,” you
say. I can now entertain the possibility that (10) expresses the same proposition as
one of the sentences in (5)–(6), which are likewise rejected in this context. And
what you’ve helpfully told me about the sentence lets me go even further. Sen-
tence (5) is a sentence that is rejected in this context, but which would have been
accepted if I were saying it while in Clarkston. So sentence (5) is an especially
plausible candidate for a sentence that expresses the same proposition as does this
Nez Perce sentence (10).
2 What Can Go Wrong: Translation
Those who are keeping track will have noticed that I’ve now assumed quite a
number of important things. I’ve assumed the ETH and the EJH as hypotheses
about how translation and judgment work. I’ve also made a number of assump-
tions about you as a speaker and about our conversation. It’s no surprise that
those latter assumptions could be incorrect, and that there would be potentially
disastrous results in that scenario for my research project. If you are joking
with me in a way I haven’t figured out, or if you have a different idea than I do
about what my English questions mean, I may be totally wrong about how your
responses relate to the questions I am trying to answer about meaning in your
language. But for right now I want to grant that we are communicating and co-
operating in pretty much the way I thought we were. What needs to be shown
is that even in this scenario, I might end up with faulty conclusions about your
language simply by virtue of assuming that we will always stick to the ETH
and EJH.
Consider what could go wrong in my reasoning about the situation of transla-
tion. If we are communicating and cooperating normally, what reason could you
have to avoid giving me a translation that expresses the same meaning as the origi-
nal material I asked you to translate? Both semantic and pragmatic factors can get
in the way. In semantic terms, your language might simply not make it possible to
express the meaning I am asking you to express. (More on this just below.) In prag-
matic terms, it might be possible to express a certain meaning in your language,
but there might be unwelcome consequences of that expression in situations of
actual conversation. If you’re not a semanticist (and even potentially if you are), I
can’t count on you to say, “The task you’ve asked me to perform is not possible, or
not practical, in view of such-and-such facts about my language.” Most likely you
will simply give me a translation (and if I’m lucky, say something insightful about
it), and I will have to figure out how that data point fares in view of the space of
possible hypotheses about what translation data means.
Both semantic and pragmatic factors that can interfere with reasoning via the
ETH come up in dealing with fieldwork data from Nez Perce. The following three
sections feature examples from real life.
2.1 When equivalence of meaning is not possible
Before we come to the first example, a few words are in order concerning this sec-
tion’s title. How could I claim that equivalence of meaning between two languages
might not be possible? This claim of mine seems to run contrary to the received
wisdom we so often encounter in our introductory training, where we are faced
with the view from such luminaries as Jakobson:
All cognitive experience and its classification is conveyable in any existing
language. Whenever there is deficiency, terminology may be qualified and
amplified by loanwords or loan-translations, neologisms or semantic shifts,
and finally, by circumlocutions. . . . No lack of grammatical device in the lan-
guage translated into makes impossible a literal translation of the entire con-
ceptual information contained in the original. (Jakobson 1959, 234–235)
Jakobson’s remark makes for a useful clarification of what I am concerned with
here. It could well be that every language has the grammatical means to describe
the same range of experiences and their classification. It is certainly not the case
that every language has the lexical means to do this, though, at least not at any
given time. Fortunately, it is easy for communities to gradually modify the lexicon,
to “qualify” and “amplify” in the ways that Jakobson suggests, when it becomes
useful for speakers to talk about some aspect of experience they have not had oc-
casion to talk about before. When Jakobson speaks of language here, it’s language
in this temporally extended sense he seems to have in mind.
It’s language in a different sense that’s encountered in a conversation at a par-
ticular time with a particular consultant. If, for instance, Nez Perce speakers found
it useful to talk about particle physics in their language, over a period of time one
might expect the lexicon of Nez Perce to change in ways that make it possible to
accurately convey notions of mass, charge, chirality, and all the rest in fluent Nez
Perce. But someone who masters the grammar and lexicon of Nez Perce as it exists
right now cannot express the precise set of propositions that are needed for that task.
What an unreasonable request it would be for me to ask you to qualify and amplify
on the spot, inventing new words as needed in order to translate! I can only ask you
to translate into your language using the grammar and lexicon that you and your
community have antecedently agreed upon. And that means that I should not be
surprised if you cannot give a semantically equivalent translation of every possible
type of sentence I might present you with in an elicitation session. Your language as
it exists right now might not give you the ingredients that would be required.
I think that is compatible with Jakobson’s view. In the long run, speakers of
any language will be able to find a way to use their language to talk about any
kind of thing. In the short run, there may be ‘deficiencies,’ or lexical gaps, and
this can affect the sorts of messages that speakers are able to straightforwardly
convey. English has gaps of various sorts (some of which we will see a bit later
on), and other languages do too. The potential for gappiness on both sides of
the process of translation has to be taken seriously as a cause for deviation from
the ETH.
The relevance of lexical limitations comes up in a serious way in studying Nez
Perce modals. In Deal 2011, I argued that one fact about the lexicon of Nez Perce
as it exists right now is that it doesn’t have any simple non-epistemic necessity
modals. The only simple non-epistemic modal is a suffix, o ’qa, which conveys pos-
sibility, not necessity. That means that translators can stick to the ETH when they
are translating possibility modals from English into Nez Perce, but not when they
are translating necessity modals. For a sentence like
(11) Employees must wash their hands before returning to work.
there is no semantically equivalent Nez Perce translation.

Let’s see how this plays out in reasoning about translations of modal sen-
tences. Here are two translation tasks I presented to consultants, and the responses
they gave (with my glosses).
(12) ARD: How could a mother say to her kid, “You can eat candy after the
meal”?
C(onsultant): Tepelwéeku’s-ne ’aa-p-ó’qa hip-naaq’í-t-pa.
candy-obj 3obj-eat-modal eat-finish-part1-loc
(13) C: [having just commented in English on the size of hamburgers at Burger

King] I can finish a small one.
ARD: How would you say that in Nez Perce?
C: Hiinaq’i-yó’qa kúckuc.
finish-modal small
If I reason via the ETH, I conclude that the Nez Perce sentences I’ve just been
given express possibility, like the English originals do. The one morpheme in
common is the suffix (y)o’qa (the glide being phonologically predictable), so it

seems reasonable to conclude that that morpheme is a possibility modal.
What about necessity sentences? Here are two more translation prompts and
the responses from Nez Perce speakers.
(14) ARD: We have to get home before it gets dark.

C: Kíye pe-ckilíi-toq-o’qa kulaawit-’ásx.
we S.pl-return-back-modal dark-before
(15) ARD: According to the rules, I should leave.

C: Tamáalwit-ki ’aat-ó’qa.
rule-inst go.out-modal
(Comment:) That’s not really saying I should go out. It’s just saying I
could go out.
If I reason via the ETH again, I take the translation of (14) to express necessity.
Curiously, o’qa shows up here as well. Something funny is going on in (15). The
speaker has translated the English necessity sentence with o’qa, as in (14), but she
felt it necessary to follow that up with a serious hedge. Her hedge casts doubt on
whether she really thinks that (15) and its Nez Perce translation express the same
proposition. It sounds like she takes her Nez Perce sentence to express not neces-
sity, but possibility.
How should we proceed in the face of this seeming inconsistency? Perhaps we
could just temporarily discount this funny fact about sentence (15). A reasonable
thing to hypothesize would be that o’qa is a modal, but one that’s lexically ambigu-
ous. It can express either possibility or necessity. There are more funny facts about
the hypothesized necessity modal o’qa, though—in fact, so many that it becomes
very difficult to discount them. One major curiosity is that when speakers translate
o’qa sentences from Nez Perce to English, they always translate with a possibility
sentence if o’qa is in a non-upward-entailing environment—in the scope of ne-
gation, in a conditional antecedent, or in the restrictor of a universal quantifier.
These are environments that share the logical property of not supporting infer-
ences from necessity to possibility.
A few words are in order on this logical behavior. Suppose I tell you that
you must enter this pie-eating contest. That’s a necessity claim, and it says
something about the set of possible worlds that are compatible with the rules:
you enter the contest in all of the best of those worlds. It follows from that, of
course, that you also enter the contest in some of the best of the rule-following
worlds. From the fact that your entry is necessary, it follows that it is possible.
The key thing to observe is that that inference disappears when the necessity
sentence is embedded in certain types of grammatical environments. Sentence
(16) does not entail sentence (17), for instance. Plausibly, you are permitted but
not required to compete.
(16) It’s false that you must enter this pie-eating contest.
(17) It’s false that you may enter this pie-eating contest.
That shows us that negation removes the inference from necessity to possibility.
The same holds for restrictors of universal quantifiers, and for conditional ante-
cedents, which we see in (18) and (19). The first of these sentences certainly does
not entail the second.
(18) If you are required to enter, you will win.
(19) If you are permitted to enter, you will win.
Now, all this is relevant for the analysis of Nez Perce modal sentences in
view of a correspondence between these inference patterns and the translations
of sentences with o’qa. The environments where necessity sentences aren’t logi-
cally stronger than possibility sentences are also the environments where o’qa isn’t
translated with English necessity modals. When o’qa is embedded within a con-
ditional antecedent like in (20), for instance, the result is always translated with a
possibility modal, not a necessity modal—“if I can enter” or “if I could enter,” not
“if I should enter” or “if I have to enter.”
(20) c’alawí ’ac-ó’qa, kaa ’ac-ó’

if enter-modal then enter-prosp
If I can/could enter, I will enter.
The fact that this translation pattern is quite systematic makes it doubtful that o’qa
sentences are lexically ambiguous in the simplest sense. Their two possible transla-
tion types are not distributed randomly, but rather correlated with a logical prop-
erty. If I insisted on sticking strictly to the ETH in the face of these data, I might
be able to propose, at the limit, that o’qa is ambiguous between a possibility modal
that imposes no special requirement and a necessity modal that is a positive polar-
ity item. That might account for the missing translations of sentences like (20). But
I am still left with no good analysis of simple sentences like (15), where speakers for
some reason find it necessary to severely hedge.
The alternative would be to fail to apply the ETH to sentences (14) and (15).
This is essentially what the speaker’s hedge on (15) seems to be telling us to do. That
comment suggests that there’s something difficult about the task of translation for
this example. On the view I initially outlined, the difficulty is that there is in fact
no possibility of a semantically equivalent translation of According to the rules, I
should leave into Nez Perce. That sentence expresses a flavor of weak necessity,
and in translating (15), the speaker has done her best to approximate that meaning
given the lexical resources her language provides. What her comment suggests to
us is that the o’qa sentence she has given in fact expresses possibility. It might be
a good enough translation for the necessity sentence—it will be true whenever the
necessity sentence is true—but the match is not a perfect one.
This account explains not only the hedge on (15) (and parallel hedges that
show up in other discussions) but also what we see in (20) in translation from
Nez Perce into English. When we translate into English, we have both possibility
and necessity modals at our disposal. Suffix o’qa is strictly a possibility modal, so
an English possibility modal can be used to translate it. There is no lexical gap to
cause us to deviate from the ETH.
At this point, this case study points to two things that can be said about the
ETH. On one hand, it’s not the case that speakers always keep to this hypothesis
in performing the task of translation. Granting that leads to a simpler and more
explanatory view of the meaning of o’qa. On the other hand, in a situation where
linguists and speakers have come to a pretty good understanding of what the task
of translation ought to be, speakers’ comments can give a sense for when the ETH
should be taken off the table. Lisa Matthewson has very aptly emphasized the
status of speaker comments as data points for semantic fieldworkers (Matthewson
2004). This case study shows how such data points may inform our reasoning
about the rest of what speakers are doing.
2.2 When equivalence of meaning is not practical, part I
Now I want to illustrate how pragmatic factors can interfere with equivalence in
translation. I’ll first give an example that continues the discussion of Nez Perce
modals. Then, in the following section, I’ll give an example from the domain of
time and tenses before returning to the modal system once again. The plan here is
to get some further ideas about why the ETH might not hold in particular trans-
lation situations, and also to understand how one might reason instead in those
situations.
What I’ve said so far about the suffix o’qa leads us to expect that o’qa will never
be translated with English necessity modals. But in point of fact it is sometimes
translated that way—only when it occurs in upward-entailing contexts, i.e., con-
texts where necessity entails possibility. Here is a sentence that was translated in
a conversation about how the speaker finds it difficult sometimes to understand
younger people. She first uttered the sentence in Nez Perce, and then translated it
into English.
(21) ’I’yéwki hi-pa-c’íix̂-no’qa.

slowly 3subj-S.pl-speak-modal
They should speak slowly.
This looks like a gratuitous violation of the ETH. If o’qa (here in morphologically
conditioned form no’qa) expresses possibility, why isn’t it translated with a possi-
bility modal? English provides the speaker with plenty of options to choose from.
Why has she chosen a translation with should instead of with could or can?
We have to think carefully about what the Nez Perce sentence expresses, and
how that compares with what would have been expressed by various English trans-
lations. The Nez Perce sentence makes a possibility claim about slower speaking
by the relevant group. Since there’s no necessity modal that could have been used
instead of o’qa, there’s no scalar implicature that comes from using a possibility
modal. So the Nez Perce sentence simply expresses a possibility claim of the type
we record in propositional logic as ◊ϕ (proposition ϕ is possible). The sentence is
true and appropriate if there is at least one accessible world in which the proposi-
tion ϕ (that they speak slowly) is true, which would still be the case even if all ac-
cessible worlds are that way. This is not quite the situation in English. If the speaker
had translated (21) with a possibility sentence like “They could speak slowly,” she
would have chosen a sentence that comes with a scalar implicature. It has an en-
riched meaning which we record in propositional logic as ◊ϕ ∧ ¬ ¨ϕ: it’s possible
for them to speak slowly, but it isn’t necessary. So the Nez Perce sentence can be
translated with either of two choices in English, considering enriched meanings:
◊ϕ ∧ ¬ ¨ϕ (ϕ is possible but not necessary, in this case expressed by “They could
speak slowly”) or ¨ϕ (ϕ is necessary, in this case expressed by “They should speak
slowly”). Neither one of those perfectly matches the meaning of the Nez Perce
sentence. So the ETH can’t be fully followed.
We must go one step further concerning sentences like (21), where o’qa is in an
upward-entailing environment. If the speaker is giving a translation that doesn’t keep
to the ETH, what exactly is she doing? She’s not pairing Nez Perce and English sen-
tences at random, to be sure. She makes her choice among English translations using
a principle that is weaker than the ETH. If the speaker believes that ¨ϕ is true, only
one of the English translations expresses what she takes to be a true proposition. In
choosing among potential translations, she seems to be following reasoning like this:
(22) Lower Bound on Good Enough Translation

Within a particular context, if the original sentence is true and felicitous,
its translation is true and felicitous.
That’s how she’s getting to an English necessity sentence as the translation of a Nez
Perce possibility sentence. If ¨ϕ is true in this context, a Nez Perce speaker may
use an o’qa sentence to truthfully and felicitously express ◊ϕ, which of course fol-
lows from ¨ϕ. An English speaker uttering a simple sentence does not have a clear
route to ◊ϕ. Following this “lower bounding” rule, her fallback plan should be to
express some proposition in the neighborhood of the original that is both true and
felicitous. A sentence which expresses ¨ϕ is the most natural choice.
But wait. The scalar implicature borne by the English possibility sentence is,
after all, famously cancelable. Couldn’t the speaker give one of these translations
for (21)?
(23) They could talk slowly. In fact they should.

(24) It is at least possible that they talk slowly.
The answer should be clear—one would be quite surprised to find a consultant who
gives this type of translation. These are complex and somewhat technical expres-
sions in English. It is possible though not practical to translate this way. It’s really
the impracticality of this type of translation that pushes the speaker back into the
choice between ◊ϕ ∧ ¬ ¨ϕ (could-translation) or ¨ϕ (should-translation).
Now we have a clear explanation for one final detail that has been left myste-
rious. Why exactly should the logical property of upward-entailingness correlate
with necessity translations for o’qa? Why don’t necessity translations show up for
sentences like (25), where o’qa is in a non-upward-entailing environment?
(25) c’alawí ’ac-ó’qa, kaa ’ac-ó’

if enter-modal then enter-prosp
If I can/could enter, I will enter.
The solution lies in the fact that possibility modals bring scalar implicatures only in con-
texts where they are logically weaker than necessity modals. In a non-upward-entailing
environment like the antecedent of a conditional, an English possibility modal triggers
no scalar implicature. That means that just in these cases, the English possibility modal
can provide an equivalent translation for the Nez Perce modal o’qa. In this example,
Nez Perce sentence (25) expresses a conditional of the type [◊ϕ]→ ϕ (if ϕ is possible,
then ϕ holds true; in this case ϕ is the proposition that I will enter). When the speaker
translates with the English sentence “If I can enter, I will,” she chooses a sentence that
carries this very meaning. There is no unwanted enriched meaning found in the Eng-
lish but not in the Nez Perce. So, there is no cause for deviation from the ETH.
We should dwell a bit on this issue of enriched meanings, since it bears on
the question of when we expect speakers to deviate from the ETH. Sentence (24)
plausibly expresses the very same proposition as (21), so it is not a lack of lexical
means that leads the speaker not to choose a fully semantically equivalent transla-
tion for that example. There seems to be an important pragmatic principle at stake.
Some additional examples might make it clearer how this worry about practicality
relates to familiar types of pragmatic concerns.
2.3 When equivalence of meaning is not practical, part II
The starting point for these next examples is Jakobsonian: languages are clearly
different in the types of information they require their speakers to express. An
English speaker, for instance, is required to pick a tense in order to utter a finite
sentence, whether or not particulars of time are important to what that speaker
wants to say. A speaker of Mandarin or St’át’imcets is not always required to make
a parallel choice (Lin 2006, Matthewson 2006). Speakers of those languages can
grammatically utter finite sentences that leave unspecified whether the events they
describe are occurring in the present or have already occurred in the past. English
makes it possible, but less practical, to do this.
We see a version of this issue in comparing the expression of time in Eng-

lish to the corresponding system in Nez Perce. Both languages distinguish present
from past tenses, but Nez Perce has two past tenses, the distinction between which
has something to do with remoteness. Events in the distant past are described
using verbs with the tense suffix ne/na. Events in the more recent past are de-
scribed using verbs with the tense suffix qa.
(26) hani-sa-qa
make-imperf-rec.past
I was making something (recent)
(27) hani-sa-na
make-imperf-rem.past
I was making something (remote)
This means that an English speaker can easily utter a temporally vague past sen-
tence like “I was making something,” whereas Nez Perce makes it less practical to
do so. The simplest Nez Perce sentences along this line are more temporally precise
than are their English counterparts.
How, then, should I expect you to respond if I ask you to translate a tempo-
rally vague sentence like ‘I was making something’ into Nez Perce? Certainly you
could in principle say something like this:
(28) hani-sa-qa ’iitq’o hani-sa-na
make-imperf-rec.past or make-imperf-rem.past
That would plausibly conform to the ETH. But of course you are exceedingly un-
likely to respond in this way. That’s probably because the pragmatics of the original
English sentence and the pragmatics of this potential disjunctive translation are
markedly different. The disjunctive translation calls attention to the time in a way
that the original sentence does not—it brings into play the maxim of manner. Is
the speaker unsure about the time at which she made something? Does she not
want to share that information? There’s no way to leave the past time entirely vague
and not draw the issue into the spotlight. The situation arises because Nez Perce
requires its speakers to provide more information about location in the past than
does English.
There is an aspect of the grammar of modals where English requires less
precision than Nez Perce does, and the same situation arises here, too. English
modal auxiliaries are famously flexible in what type or “flavor” of modality they
express. The same sentence with modal may, for instance, expresses epistemic
possibility in the discourse in (29) but non-epistemic possibility in the discourse
in (30).
(29) John eats a lot of meat, but I’ve spotted his car outside this vegan
restaurant. John may eat here.
(30) John is coming to town, and he’s a strict vegan. Fortunately, this is a vegan
restaurant. John may eat here.
Nez Perce, on the other hand, uses different modal morphemes for epistemic and
non-epistemic modality. Words like pay’s are used for epistemic possibility; o’qa is
reserved for non-epistemic possibility. What kind of Nez Perce translation should
I expect, then, for a sentence like “John may eat here,” which leaves open which
sort of possibility is at stake? Certainly not like this:
(31) pay’s Caan hi-ip-teetu kine ’iitq’o hi-ip-o’qa kine

maybe John 3subj-eat-habitual here or 3subj-eat-modal here
John mayepistemic eat here or he maynon.epistemic eat here.
That kind of translation would conform to the ETH, of course. Like the English
original, it’s non-committal about whether the modality in question is epistemic
or non-epistemic. Once again, this absence of precision can be produced in Nez
Perce only by a disjunctive translation that calls attention to the very matter that
the English sentence conveniently sidestepped.
If we take this type of foregrounding of unwelcome issues to constitute infe-
licity in a particular context, then examples (28) and (31) will be ruled out by the
Lower Bound on Good Enough Translation. Alternatively, we might conclude that
speakers are using some additional principle to guide their translations away from
such choices, perhaps something like this.
(32) Equivalent implicatures hypothesis (EIH)
The input to translation and the output of translation are equivalent in
what they implicate.
The important bit about the EIH is that it seems to outrank the ETH in a number
of situations. Where speakers can’t give translations that are equivalent in both
content and implicature, they sacrifice equivalence of content to make sure that
certain types of implicatures are avoided. That’s what stops speakers from translat-
ing simple, everyday sentences into complex technical explications that preserve
the content but not the implication.
3 What Can Go Wrong: Judgment
The interpretation of translation is complex. It’s more complex than we counte-

nanced in section 1, where all we needed was the ETH. Let’s now consider how
things look for the other hypothesis that came up in that section, namely the hy-
pothesis EJH about judgment.
Certainly, judgment poses its own challenges of the type we’ve mostly set
aside thus far. If I ask you to judge an utterance and you reject it, I have to wonder
whether you’ve rejected it because of what it means, or because it’s simply not well
formed on grounds that are syntactic, morphological, phonological, or prosodic.
(This is another point that’s nicely made by Matthewson 2004.) But even leaving
such cases, I want to show that we must be careful about the EJH itself. The reasons
have to do with the coming together of semantic and pragmatic information.
Why might you reject a sentence expressing ϕ in English but accept a sentence
expressing ϕ in some other language? Perhaps because the former sentence carries
an enriched meaning that makes it false in the particular situation, even though ϕ
itself is true. To exemplify this I propose we turn to a data set featuring modal sen-
tences once again. To change things up a little bit, the particulars for this example
come from the work on Gitksan, a Tsimshianic language, by Tyler Peterson (2010).
The situation with modals in that language is very much similar to Nez Perce,
though the analysis Peterson arrives at is rather different from the one I’ve outlined
above and in Deal 2011. What I want to do is to show how Peterson’s analysis is influ-
enced by the EJH, and how a potentially simpler analysis comes into view once our
reasoning about judgments is qualified to take pragmatic information into account.
Let’s focus on Gitksan sentences involving the clitic ima, which relates to
epistemic modality and inference. Peterson presents speakers’ judgments on such
sentences in two types of contexts. In one context, there is strong evidence for a
particular inference. Here is one of his examples of an ima sentence that is ac-
cepted in such a context.
(33) Context: You hear that Alvin’s truck broke down on the way up to the
Suskwa. It’s a very reliable truck, but someone suggests that the problems
he’s having starting it indicate a problem with the fuel pump.
n̍idimahl g̱an wilt

n̍it=ima-hl ḵan wil-t
3sg=mod=cnd coord do.something-3sg
“That must’ve been why it happened.” ≻“That might’ve been why it
happened.”
Along with the example itself, Peterson presents two alternative English transla-
tions. The sentence to the left of ≻ is judged by his consultant to be more felicitous
than the sentence to the right, given the context. Reasoning by the EJH, Peterson
concludes that the Gitksan sentence could not express the same proposition as the
less felicitous translation. It could, however, express the same proposition as the
more felicitous translation. All this points to the conclusion that the ima sentence
expresses some version of epistemic necessity.
In other contexts, this line of reasoning leads to a different result. Here is a
context where there is not particularly strong evidence for a particular inference.
An ima sentence is felicitous again, but the range of felicitous English translations
is different.
(34) Context: You see your uncle stopped at the intersection talking to some
people through the window of his pickup. You and your friends don’t
recognize the people.
wilaayimas nibib-y̍ (n̍idiit)

wilaa-i-(t)=ima=s nipip-y̍ (n̍itiit)
know-tr-3=mod-pnd mother’s.brother-1sg 3pl
“My uncle might know them.” ≻ “My uncle must know them.”
Following the EJH, we would have to conclude that this ima sentence cannot ex-
press the version of epistemic necessity expressed by the less felicitous English
translation. It could, however, express the same proposition as the more felicitous
translation. This points to the conclusion that an ima sentence expresses some
version of epistemic necessity in certain contexts (e.g., (33)) but some version of
epistemic possibility in other contexts (e.g., (34)). Thus Peterson concludes that
this curious clitic “has variable modal force. . . [it] can be interpreted as might or
must” (p. 166).
There is another way to think about what is happening here, and something
pragmatic—scalar implicature—plays an important role. Let’s suppose that ima
only expresses epistemic possibility. Gitksan doesn’t have an epistemic necessity
modal to serve as its dual, however, and so ima sentences don’t carry scalar impli-
catures. An ima sentence is therefore possible in a context supporting epistemic
mere possibility, like (34), but also in a context supporting epistemic necessity,
like (33). The English translations, of course, don’t work the same way. The must
translation of (34) is disfavored in its context because it expresses a necessity claim
which is false. The might translation of (33) is disfavored in its context even though
its narrow content is clearly true. It’s the implicature the sentence carries that is
false.
The key test of this alternative proposal would have to come from non-
upward-entailing environments, where possibility modals fail to trigger scalar
implicatures. Matthewson (2013) observes that there are difficulties producing
the relevant data, owing to the syntax of Gitksan epistemic particles. (Rullmann,
Matthewson, and Davis 2008 make a similar observation concerning the syntax
of modal elements in St’át’imcets, which likewise are acceptable in both possibil-
ity and necessity contexts.) These difficulties, of course, should not obscure the
methodological point to be made on the interpretation of judgments. If we think
that scalar implicatures are not part of propositional sentence content, then this is
indeed a situation where we would do well not to totally trust in the validity of the
EJH. In (33), we plausibly have a context where someone would reject a sentence
expressing ◊ϕ if speaking English (given the scalar implicature associated with ◊ϕ
in that language) but accept a sentence expressing ◊ϕ if speaking Gitksan.
This kind of worry motivates at least a partial retreat to a somewhat more aus-
tere theory of judgment. Unlike translation, judgment does not inherently trade
on a notion of equivalence between expressions of two languages. Judgments can
be understood simply as revealing some aspect of the pairing between acceptable
utterances and the way the world is.
(35) Austere Theory of Judgment

In a particular context, speakers accept sentences that are both true and
felicitous and reject sentences that are false and/or infelicitous.
Let’s take this revised theory to our initial example of a judgment scenario, where I
sat with you in Lapwai and we talked about our friend Harold down the road. I’ve
explained the backstory and asked, “could I say
(36) Pay’s Harold hinekise wees Clarkstonpa.

Would that make sense?” You’ve said yes. If I don’t reason by the EJH, it’s no use to
sit and think about what sorts of English sentences might be uttered appropriately
in this scenario. All I conclude is that sentence (36), whatever it might mean, is
true and felicitous in this little context.
4 Conclusions
The bread and butter of empirical semantics is the pairing of such-and-such sen-
tence with such-and-such meaning. We take these pairings as our basic data when
we set out to build and evaluate semantic theories. What the fieldworker must
confront is that this data can ultimately hardly be taken as basic at all. The basic
data are speaker behavior, and to get from this behavior to something that is useful
for semantic theory, we have recourse to typically implicit theories of translation
and of judgment. It is these auxiliary theories that make it possible to reason from
the particulars of observation in the field to the abstractions in which our theories
deal. This makes it possible to conduct empirical investigations in semantics.
This situation on an abstract level is by no means unique to one particular
corner of scientific investigation. It is much the same in syntax, for instance, where
the fieldworker again must bridge a chasm between the type of basic data the
theory demands—in this case, distinctions between grammatical and ungram-
matical sentences—and the ultimate type of facts that can be confronted in the
field. The challenge is fittingly described by Judith Aissen:
While both linguistic theory and fieldwork deal in data, the sort of data they
deal in is absolutely and fundamentally different in kind. Linguistic theories
or analyses make predictions about sentences as abstract objects—not utter-
ances used in particular situations or by particular people. On the other hand,
what gets elicited in informant sessions is very much a concrete object: it is a
judgment of grammaticality offered by one specific person to another specific
person in a particular setting on a particular day and in a particular place.
From this point of view, it seems quite clear that data of one type cannot ar-
ticulate directly with data of the other type. Elicited data has got to be cooked
to be brought into meaningful relation with theory. (Aissen 1988)
It is a “myth of fieldwork,” Aissen concludes, that speaker responses can be applied

directly to analyses in syntactic theory. Between the speaker response and the
analysis stands the fieldworker and his or her auxiliary theories. The particulars
of those theories can very much influence the ultimate type of syntactic analysis at
which the fieldworker arrives.
The essential outlines of this situation stretch well beyond linguistics. Ais-
sen’s description brings us awfully close, for instance, to the description of rea-
soning in physics by the physicist and philosopher of science Pierre Duhem
(Duhem 1914 [1982], 166). Duhem points out that a physicist who wants to apply
Boyle’s law (which relates the temperature, volume, and pressure of constant
masses of gas) is not faced with temperature, volume, and pressure as basic data.
His basic data, like that of the linguist, are remarkably less abstract than is the
currency of his theory. If he is faced with a particular gas, he is faced with a
concrete substance that is more or less warm and more or less voluminous and
more or less pressurized. To get from the particulars of these things to useful
information, the physicist must have not only instruments (like a thermom-
eter) but also theories of how those instruments work. The level of mercury in
a thermometer is not a particularly useful point of data considered all by itself.
It’s only in view of a hypothesis about how such information corresponds to the
abstract notion of temperature that the mercury level becomes worth seeking
out and recording.
Our situation is of course somewhat messier than that of Duhem’s physi-
cist. Our basic data come from interactions with people, not thermometers.
There is no theory of the actions of a human being that has yet been stated
in fully predictive form, and so we must consider different possible hypoth-
eses that could explain and render useful the basic data we’re provided. That
is why there is no one single theory of judgment or translation being used by
semantic fieldworkers. For translation, it is also why Matthewson(2004) has
so thoroughly emphasized that a translation is “a clue, not a result.” There are
many ways to interpret clues. Those ways are alternative hypotheses about the
workings of translation.
What should not get lost in all this is that simple theories of transla-
tion like the ETH and judgment like the EJH are incredibly useful when they
work. Certainly, they don’t take us to the best conclusions all of the time. But
we want our theories to be simple and natural enough to be used on the fly,
in conversing with our speakers and modifying (if necessary) the questions
we’ve planned to ask them. If they’re useful enough, even theories that are
sometimes flawed can play an important role. What we have to do is to routinely
check the way we have been reasoning about our collected basic data. What
other reasonable hypotheses could be applied to the data at hand? How
does that change the ultimate result we take away to our project of theory
construction?
References
Aissen, Judith. 1988. Myths of fieldwork. Paper presented at the Center for the Study of the
Native Languages of the Plains and Southwest, University of Colorado, April 1988.
Deal, Amy Rose. 2011. Modals without scales. Language 87: 559–585.
Duhem, Pierre. 1914 [1982]. The Aim and Structure of Physical Theory. Princeton, NJ: Princ-
eton University Press.
Jakobson, Roman. 1959. On linguistic aspects of translation. In On translation, Reuben
A. Brower (ed.), 232–239. Cambridge, MA: Harvard University Press.
Lin, Jo-Wang. 2006. Time in a language without tense: The case of Chinese. Journal of Se-
mantics 23: 1–53.
Matthewson, Lisa. 2006. Temporal semantics in a superficially tenseless language. Linguis-
tics and Philosophy 29: 673–713.
Matthewson, Lisa. 2013. Gitksan modals. International Journal of American Linguistics 79:
349–394.
Peterson, Tyler. 2010. Epistemic modality and evidentiality in Gitksan at the semantics-
pragmatics interface. PhD diss., University of British Columbia.
Rullmann, Hotze, Lisa Matthewson, and Henry Davis. 2008. Modals as distributive indefi-
nites. Natural Language Semantics 16: 317–357.
7
Investigating D in Languages
With and Without Articles
Carrie Gillon
1 Introduction
The semantics of articles is an important locus of variation cross-linguistically, and

yet it is difficult to uncover the meaning of articles.* It is often assumed that articles
are (in)definite, but the situation is more complicated: some languages have (in)
definite articles, while some have articles that are not (in)definite (Matthewson
1998, Gillon 2013). Further, there are many articleless languages, where bare nouns
often share the same meanings as nominals with articles in articleful languages. In
this chapter, I first provide an overview of elements that look like articles (as op-
posed to, say, demonstratives), describe their function and position(s), and briefly
show that their semantics can vary wildly. I then provide an overview of the empir-
ical questions that should be addressed when examining the semantics of articles
as well as the semantics of bare nouns in articleless languages, and describe some
tests that were designed to answer those questions. I show how these tests can be
applied to languages with overt articles, as well as to articleless languages.
In this chapter, I use five different languages to show how to test for the se-
mantics of articles as well as the semantics of bare nouns: English, Skwxwú7mesh
*
I would like to thank all of my consultants. Skwxwú7mesh: the late Lawrence Baker, the late
Tina Cole, the late Lena Jacobs, the late Yvonne Joseph, the late Eva Lewis, Margaret Locke, and the
late Frank Miranda. Innu-aimun: Marilyn Martin, Basile Penashue, Kanani Penashue and Anne Rich.
Inuktitut: Holda Zarpa and Solomon Semigak. Lithuanian: Solveiga Armoskaite (who is also a collabo-
rator on work on Lithuanian). The research on Innu-aimun and Inuttut was supported by funding from
the Department of Linguistics (Social Sciences and Humanities Research Council of Canada grant
#410-2008-0378, awarded to Julie Brittain and Social Sciences and Humanities Research Council of
Canada grant #833-2004-1033, awarded to Marguerite MacKenzie) and the Faculty of Arts at Memorial
University of Newfoundland and by an ISER research grant. I would also like to thank Lisa Matthew-
son, Phil Branigan, Julie Brittain, Marguerite MacKenzie, Hotze Rullmann, Martina Wiltschko, Ryan
Bochnak, an anonymous reviewer, and the audiences at many conferences, but especially the audience
at the LSA 2012. 175

(Salish), Lithuanian (Baltic), Innu-aimun (Algonquian), and Inuttut (Labrador

Inuktitut; Eskimo-Aleut).
This chapter has the following structure. In section 2, I discuss articles and
their syntactic and semantic behavior(s). In section 3, I briefly describe the behav-
ior of bare nouns in articleless languages. In section 4, I discuss the empirical ques-
tions that need to be addressed when looking for articles and their semantics as
well as the semantics of bare nouns in articleless languages. In section 5, I provide
the tests that I have developed to examine those empirical questions. In section
6, I provide the results for languages with articles (section 6.1) and those without
(section 6.2). Section 7 concludes.
2 What Are Articles?
When investigating the semantics of articles, it is first important to judge which

elements could be articles. What makes an element an article? Does an element
have to be associated with a particular meaning, or have a particular feature (e.g.,
[±def]) in order for it to be an article? I provide an overview of my assumptions
about what counts as an article in terms of its function and syntax. I also provide a
caution about the semantics of articles: they can vary.
2.1 THE FUNCTION OF ARTICLES
Articles minimally create an argument out of a predicate (Higginbotham 1985; Sz-

abolsci 1987, 1994; Stowell 1989; Longobardi 1994). This is what allows nominals to
occupy argument position (in the normal case). Singular N(oun) P(hrase)s cannot
occupy argument position, but an NP + an article can.
(1) a. I saw the/a cat.

b. *I saw cat.
Articles are different from other elements that allow nominals to occupy ar-
gument position. For example, demonstratives appear to have a similar function
(albeit with an extra deictic function).1
(2) I saw that cat.
However, demonstratives behave differently from articles in that they can occur on
their own, without a following noun.2
1
Deixis does not by itself prevent an element from functioning as an article, since many Salish
articles are deictic (Matthewson 1998; Gillon 2013, 2009a; see also section 2.3).
2
Further, in some languages demonstratives and articles can co-occur: Michif (Rosen 2003) and
St’át’imcets (Matthewson 1998), for example. That is something else that should be tested in a particular
language.
Investigating D in Languages With and Without Articles 177
(3) a. I saw that.

b. *I saw the/a.
Thus, when looking for articles in a language, we look for elements that create
arguments out of predicates that also require a following noun.
2.2 THE POSITION(S) OF ARTICLES
Articles are part of the functional superstructure of a nominal. In much of the

traditional syntactic and semantic literature on English, articles are considered to
occupy the same position as other functional elements that can precede a noun.3
(4) a. I saw the/a/one/each/every/that cat.

b. I saw the/two/those cats.
Abney (1987) argues that all of these functional elements occupy the same posi-
tion: D.
(5) Determiners as head of D
DP
D/Q/Num/Dem NP
In previous work, however, I have argued that the and a occupy different po-
sitions (Gillon 2013, following Epstein 1999, Lyons 1999, Borer 2005): a occupies
a lower position than the. The definite article the occupies D and the indefinite
article a occupies Num.
(6) a. b.
DP NumP
D NumP Num NP
the ... a ...
Thus, when investigating article semantics, we might be comparing elements

from different parts of the tree. This has implications for the kinds of semantics
associated with each article. For example, we expect to only find definiteness asso-
ciated with an element that occupies D, rather than something that occupies Num.
Minimally, articles can occupy different syntactic positions, and therefore might
display different semantics as a result. However, even elements within the same
head (D) can display different semantics, as I describe below.
3
Note that these functional elements do not include adjectives, as they occur between the func-
tional elements like the and the noun.
2.3 The Semantics of Articles
As is already known, articles vary in their semantics (Matthewson 1998). We

cannot assume that articles will be either definite or indefinite (see section 6.1 for
more discussion). For example, in most Salish languages, articles encode deictic
information (Matthewson 1998; Gillon 2013, 2009a). In Skwxwú7mesh,4 the article
kwa can only be used for referents who are not near the speaker and who are also
invisible. The article ti can only be used for referents that are close to the speaker
(within arm’s reach).
(7) Skwxwú7mesh5
a. Kw’áy’ kwa Bill.
get.hungry det Bill6
“Bill is hungry.” (Bill not in room and not visible)
b. Na mi púm ti-n s7átsus.
realis come swell det-1sg.poss face
“My face is puffy/swollen.”
These are truly articles and not demonstratives, since they cannot occur on their own.
(8) Skwxwú7mesh
a. *Kw’áy’ kwa.
get.hungry det
(Intended: “He is hungry.”)
b. *Na mi púm ti-n.
realis come swell det-1sg.poss
(Intended: “It is swollen.”)
True demonstratives, like kwiya “that,” can occur on their own.
(9) Skwxwú7mesh
Chen tkwáya7n kwíya.
1sg.s hear dem
“I heard it.”
4
Skwxwú7mesh (a.k.a. Squamish) is a Central (or Coast) Salish language spoken in southwestern
British Columbia. There are fewer than 10 native speakers remaining. Fieldwork on Skwxwú7mesh
began in 1997.
5
All data are from my own fieldwork, except where noted.
6
I use the following abbreviations: 1 = first person, 2 = second person, 3 = third person, 3’ = third
person obviative, x>y = x is the subject, y is the object, abs = absolutive, acc = accusative, art = article,
det = determiner, dem = demonstrative, erg = ergative, gen = genitive, intns = intensifier, neg = nega-
tive, obv = obviative, pl = plural, pref = prefix, pres = present, q = question marker, refl = reflexive, s =
subject, and sg = singular. Because I discuss five different languages, I have simplified many of the glosses.
There are other possibilities for the semantics of articles. For example, articles in
Samoan encode specificity, rather than definiteness (Mosel and Hovdhaugen 1992).
The article le is used to refer to specific individuals within the discourse (10)a. The ar-
ticle se, on the other hand, is used in cases where the individual might not exist (10)b.
(10) Samoan
a. ‘O le povi. b. ‘O se povi lale?
pres art cow pres art cow dem
“It’s a cow.” (specific) “Is that a cow?” (non-specific)
(Mosel and Hovdhaugen 1992)
I do not address the question of how to test for specificity here,7 but instead focus
on definiteness (or lack thereof).
While it appears that articles can vary in whatever ways they like ((in)defi-
niteness, (non-)specificity, deixis), I have argued in other work that most articles
(those that occupy D) do share a common core: they all express domain restric-
tion (Gillon 2013, 2009a).8 Other features of D-articles can vary from language to
language.
(11) All Ds: domain restriction [± feature(s)]
Domain restriction is essentially the context that a nominal must be inter-

preted within. More formally, domain restriction is a variable that restricts the set
of individuals that match the NP description to those within the context of the
discourse (Westerståhl 1984, von Fintel 1994, 1998, 1999, Martí 2003).
As I show in section 6.1, English and Skwxwú7mesh (Salish) articles behave
very differently from one another. Skwxwú7mesh articles do not display (in)defi-
niteness. Thus, when looking for articles in any language, we cannot assume that
they will behave like English the or a. Instead, we expect that articles (at least those
in a particular location within the nominal) will be context-sensitive. I describe how
to test for domain restriction (context sensitivity) in section 6.1.4 and section 6.2.4.
3 Bare Nouns in Articleless Languages
The varying semantics of articles (and their accompanying NPs) and the semantics
of bare nouns intersect in interesting ways. In this section, I describe the problem
of bare nouns in languages without articles.
See Matthewson (1998) and Ionin (2013) for possible tests for specificity.
7
Cappelen and Lepore (2005) argue that there is no such thing as domain restriction. It is beyond
8
the scope of the chapter to defend the existence of domain restriction. See also Giannakidou (2004),
Etxeberria (2005), and Giannakidou and Etxeberria (2010) for a slightly different view of domain re-
striction and D.
As is well known, bare nouns in articleless languages can be found in novel or

familiar contexts. For example, in Lithuanian,9 the same bare noun peruką “wig”
may be used in a novel context (12)a, and in a familiar context (12)b.
(12) Lithuanian
a. Tad nusipirko tikrų žmogaus plaukų
hence pref.refl.buy.past.3sg real.gen.pl man.gen.sg hair.gen.pl
peruką.
wig.acc.sg
“Hence he bought himself a wig made of real human hair.” (novel)
b. Kas čia juokinga, kad pametei peruką?

what here funny that lose.past.3sg wig.acc.sg
“What’s so funny about losing the wig?” (familiar)
(Gillon and Armoskaite 2013)
The major question about languages without articles is whether they have
covert articles or not. There are three logical possibilities for the structure of bare
nouns in articleless languages: (i) they are always NPs (Bošković 2007, 2008a,b,
2009, 2010; Chierchia 1998; Bošković and Gajewski 2011), (ii) they are always
D(eterminer) P(hrase)s (Longobardi 1994, Progovac 1998), or (iii) they vacillate
between NP and DP (Franks and Pereltsvaig 2004, Ajíbóyè 2006). In the next sec-
tion, I describe the questions that must be addressed in order to decide between
these three possibilities.
4 Empirical Questions
We can now turn to the empirical questions that need to be tested when we inves-
tigate the semantics of articles and bare nouns.
Clearly, definiteness is a feature that an article might display, since many
languages do display a definite/indefinite split. However, there is a question as to
what definiteness actually is. While the semantic contribution of definiteness is
not agreed upon, most authors agree that nominals are either definite or indefinite.
The debate is mainly divided into two camps. Many researchers argue that some
form of uniqueness drives the definiteness effects we see (Frege 1997[1892]; Russell
1998 [1905]; Hawkins 1978, 1991; Kadmon 1992; Abbott 1999; among many others).
Others argue that familiarity is encoded by definite nominals (Christophersen
9
Lithuanian is a Baltic language spoken mainly in Lithuania, with approximately 3.4 million
speakers worldwide. Work with Solveiga Armoskaite began in 2011.
1939, Prince 1981, Prince 1992, Heim 1988, among others).10 One question that
fieldworkers will ask themselves when encountering a new article will be “is this
article definite?”
Another factor that fieldworkers should take into account is scope. Different
kinds of nominals take different scopes (Carlson 1980, among many others). Defi-
nite nominals escape scope, but other nominals vary as to whether they only take
wide scope (e.g., some St’át’imcets nominals, Matthewson 1999), only take narrow
scope (e.g., some Māori nominals, Chung and Ladusaw 2004), or take either (e.g.,
a nominals in English, Carlson 1980). Thus, the scope-taking ability of nominals
can vary cross-linguistically, and can give us clues as to the semantics associated
with the article(s) under investigation.
Given that articles are associated with creating arguments out of predicates,
we expect that any nominal introduced by an article would be of type e (i.e., they
refer to individuals). However, this is not always the case. Definite nominals are
of type e, but other nominals can vary with respect to their type. For example, he
nominals in Maori have been argued to be of type et (Chung and Ladusaw 2004).
Thus, uncovering the type of a nominal can tell us something about the meaning
of the article that introduces it.
Finally, must nominals sharing the same NP description refer to the same
referent within the same context? If they must, that also tells us something about
the semantics of the article that introduces the NP.
Here are some questions that need to be answered in order to determine the
meaning of a new article:
¤ Is this article definite? What is the nature of definiteness? Is it familiarity

or uniqueness?
¤ What scope does a nominal introduced by this article take?
¤ What type does a nominal introduced by this article take?
¤ Must any nominal introduced by this article which shares the same noun
refer to the same referent within a particular context?
Another set of related questions arise when we investigate bare nouns in articleless
languages.
¤ Do articleless languages have covert articles?

¤ Is the presence of a (c)overt article universal?
¤ If we find evidence of a covert article in articleless languages, does this
covert article have the same properties as languages with overt articles?
10
However, there are some who argue that more features are required to describe English (e.g., de
Jong 1987). De Jong (1987) in particular claims that there are three categories of DPs in English: definite,
indefinite, and something in between. Definiteness for her must be decomposed into two features:
uniqueness and presupposition of existence.
¤ What semantic features can covert articles have (if any)?

¤ If two articleless languages have covert articles, do those articles share the
same semantics?
In the next section I provide tests that are designed to answer each of those questions.
5 The Tests
Here I provide the tests I have used to answer the empirical questions raised in
section 4. Each of these tests can be applied to potential articles or to bare nouns
in languages that lack articles.
5.1 Definiteness
If familiarity is encoded by definite articles, it should be a possible way to test for

definiteness. The easiest test for familiarity is to look in texts to see how referents
get introduced versus how they are referred to later on in the text. In English, a
is used for novel referents, whereas the is used for familiar referents (Heim 1988).
For example, in The Frog King, or Iron Henry, the king is introduced by a king; only
later on in the text is that king referred to by the king.
(13) In olden times when wishing still helped one, there lived a king whose
daughters were all beautiful . . . (novel)
The next day when she had seated herself at table with the king and all the
courtiers . . . (familiar)
(The Frog King, or Iron Henry; Grimm’s Tales, translated by D.L. Ashliman)
In any language, we can use texts to uncover potential patterns: articles that are
only used for first mentions are good candidates for indefinites, and articles that
are only used for first mentions are good candidates for definites.
However, to show that this is a real pattern, negative data will still need to
be collected. For example, a can never be used in a familiar context; the is dispre-
ferred in novel contexts (unless there is strong contextual support). In (14)b, the
hearer is likely to ask “which book?” if the speaker utters it out of the blue.
(14) a. I saw a dog and a cat. #A cat was meowing.

b. #I read the book last night. (novel)
Similar tests will need to be applied to the articles in the language of study. If an
article can only be used in familiar contexts, then it is likely a definite article.
A word of caution: this test can only be applied to overt articles, since bare
nouns in articleless languages can be used in both novel and familiar contexts, as
described in section 3.
If uniqueness (otherwise referred to as “maximality” for plurals, follow-

ing Kadmon 1992) is encoded by definite nominals, it too should be a possible
way to test for definiteness. Crucially, uniqueness/maximality must be tested
in a carefully controlled context. In (15), for example, a cat and a dog are both
introduced into the context. The dog must refer to that same dog introduced by
a dog.11
(15) A dogi and a cat fought. The dogi won.
If there are multiple dogs in the context, the dog is infelicitous.
(16) Two dogs and a cat fought. #The dog won.
For plural definites, the nominal must refer to the maximal individual in the
context (the entire set matching the NP description). For example, in (17), six
bears are introduced into the context. In the subsequent sentence, the bears must
refer to the maximal individual in the context (all six bears). It cannot refer to a
subset of the bears in the context.
(17) I saw 5 caribou and 6 bears. I killed the bears, #but one escaped.
Definite nominals in English must therefore always refer to the maximal individ-
ual in the context matching the NP description.
We can therefore use this as a test for definiteness: if a plural nominal must
refer to the entire set of entities that match the NP description introduced into
the context, then the nominal must be definite. This test can be applied to overt
articles and to bare nouns in articleless languages.
5.2 SCOPE
Definite nominals always escape scope. For example, in (18), there must be a
unique ghost in the context; therefore, the ghost cannot take scope under negation.
(18) I didn’t see the ghost.
= there is a unique ghost such that I didn’t see it
≠ I didn’t see any ghosts
Since there is a ghost in the context, that ghost can be referred to using it in a fol-
lowing sentence (19).
(19) I didn’t see the ghosti. Iti must have been hiding.
Other nominals, however, can take scope with respect to negation. For ex-
ample, a-indefinites can take either wide or narrow scope.
11
The indices indicate (co-)reference; if two nominals share the same index, they refer to the same
referent. If they have different indices, they refer to different referents.
(20) I didn’t see a ghost.

= There is a ghost that I didn’t see (wide)
= I didn’t see any ghosts (narrow)
When an a-indefinite takes wide scope, it introduces a referent into the discourse.
This referent can then be referred to (Karttunen 1976).
(21) I didn’t see a ghosti. Iti must have decided to stay hidden.
On the other hand, if an a-introduced nominal takes narrow scope, no referent
is introduced into the discourse. In this case, the speaker can explicitly state that
there are no individuals in the universe that match the NP description.
(22) I didn’t see a ghost, because ghosts don’t exist.
We can use scope tests to help us determine the semantics of an article. If
the nominal + article must escape/take wide scope, the article might be definite.
If, however, the nominal + article can take both wide and narrow scope, or only
narrow scope, then it cannot be definite.
We can also use scope to help us decide whether a bare noun has a covert
article. Some nominals can only take narrow scope. In English, bare plurals only
take narrow scope (Carlson 1980).
(23) I didn’t see ghosts.
The nominal ghosts does not introduce a referent into the discourse. (23) cannot
be followed up by another sentence where the discourse referent is referred back
to with they.
(24) I didn’t see ghosts. They were hiding.
It is not only nominals that lack articles that are forced to take narrow scope.
For example, he-introduced nominals in Māori also must take narrow scope.
(25) Māori
a. Kāore he tangata i waiata mai.
t.not art person t sing to.here
“No one at all sang.” (narrow; Chung and Ladusaw 2004:41)
b. Kāore he take kotahi.

t.not art reason one
“There’s no reason at all.”
(lit. there is not one reason) (narrow; Chung and Ladusaw 2004:43)
Thus, narrow scope does not tell us that there is no article. However, I have argued
in other work that the ability to take wide scope is only available for nominals
that are introduced by articles (Gillon 2013). Further, cross-linguistically, the
more functional superstructure associated with the nominal, the more likely the
nominal will take wide scope (Borthen 2003). Therefore, if a bare noun can take
wide scope, it is likely associated with functional superstructure.
5.3 LAW OF CONTRADICTION
Since articles create arguments out of predicates, we expect at least a subset of

nominals to be of type e (i.e., to refer to individuals). Since elements of type e obey
the law of contradiction (Russell and Whitehead 1910–1913, Barnes 1969, Heim and
Kratzer 1998, Löbner 2002), we can use it as a test for the semantics of an article.
In English, the nominals pass the law of contradiction.
(26) #The cat meowed and the cat didn’t meow.
In (26), the cat in each instance must refer to the same individual. Therefore, the cat
must be of type e. Conversely, a-indefinites do not pass the law of contradiction.
(27) A cati meowed and a catj didn’t meow.
(There’s a cat that meowed and there’s a cat that didn’t meow)
Thus, a-indefinites must not be of type e (and must be of some other type instead).
This test can help us decide what kind of article we are investigating. It can
also help us decide if a bare noun is associated with a covert article or not.
5.4 DOMAIN RESTRICTION
Finally, I have argued elsewhere that domain restriction is the defining feature of D
(Gillon 2013, 2009b). If this is correct, articles that obligatorily invoke domain re-
striction must occupy D. We can also use domain restriction (within a nominal)12
to test for the presence or absence of D for bare nouns.
Domain restriction must be tested in a context (by definition). In Eng-
lish, we can see the effects in examples like (28). DPs introduced by the must
refer to same individual as a previous mention with the same NP description.
A dog introduces an individual dog into the context; the dog must refer to that
same dog.
(28) a. A dogi and a catj were fighting. (Introduces {dogi, catj} into the context)
b. The dog won. Cthe dog = {dogi}
c. [[ the dog ]] = dogi
In (29), four dogs introduces four dogs into the context; the dogs must refer to all
four dogs.
12
Domain restriction is also introduced by other, non-nominal elements; I am focusing on the
presence of domain restriction within a nominal only.
(29) a. Four dogs and a cat were fighting.

(Introduces {dogi, dogj, dogk, dogl, catm} into the context)
b. The dogs won. Cthe dogs = {dogi, dogj, dogk, dogl}
c. [[ the dogs ]] = dogi+dogj+dogk+dogl
If nominals must refer to individuals within the context (and not introduce
new individuals), then domain restriction is present, and, I argue, the article must
occupy D. In articleless languages, if the bare nouns must refer to individuals
within the context (and not introduce new individuals), then the bare nouns must
have a covert D.
6 Results
In this section, I provide the results from five different languages: English, Skwx
wú7mesh, Lithuanian, Innu-aimun, and Inuttut.
6.1 RESULTS IN LANGUAGES WITH OVERT ARTICLES
English and Skwxwú7mesh have very different results when it comes to definite-
ness and scope tests, but are remarkably similar when it comes to the law of con-
tradiction and domain restriction tests.
6.1.1 Definiteness
As already discussed in section 5 (and as is well known), English the is definite. It
is strongly preferred in familiar contexts, and must only be used when the nominal
refers to a unique/maximal entity in the context.
Salish languages in general (and Skwxwú7mesh in particular) do not have
definite articles (Matthewson 1998, Gillon 2013, 2009b). In Skwxwú7mesh, the
same articles can be used for novel and familiar referents. For example, the article
ta can be used to introduce nominals in both novel and familiar cases. In (30)b,
ta k’ek’i7as si7ich’ ta mlashis “a barrel of molasses” is used to introduce a novel refer-
ent; in (30)b, ta sitn “the basket” is used to refer to a familiar referent.
(30) Skwxwú7mesh
a. Uyulhshitemwit ta k’ek’i7as si7ich’ ta mlashis.
put.aboard.for.3pl det barrel full det molasses
“A barrel of molasses was put aboard for them.” (novel; Kuipers 1967: 238)
b. . . . ses men tsexwstas ta staw’xwelh txwta7 t-ta sitn.

3 just throw.3>3 det child into obl-det basket
“. . . and she threw the children in the basket . . .”
(familiar; Kuipers 1967: 219–220)
(The same effects can be seen in elicitation contexts.) Articles in Skwxwú7mesh

therefore do not have the same familiarity restrictions that the has in English.
Skwxwú7mesh articles do not behave like the definite article the (or like the in-
definite article a).
Skwxwú7mesh articles also do not force nominals to refer to a unique entity
in the context. In example (31) below, ta lapat “a/the cup” can refer to either cup
in the context.
(31) Skwxwú7mesh
Mí7shits chexw ta lapát.
bring.me 2sg.s det cup
“Bring me one of the cups.”
(translated as “bring me the cup”)
Context: 2 identical cups, side by side
Consultant’s comment: “You’re not asking for a specific one.” (Gillon 2009b)
In plural cases, a similar phenomenon occurs. Plural nominals can refer to the
maximal individual in the context, but crucially they do not have to. For example,
in (32)b, ta mexmixalh “the bears” refers to all four bears introduced by tsi xa7utsn
mixalh “four bears.” However, this implicature of maximality can be canceled, as
in (32)c.
(32) Skwxwú7mesh
a. Chen nam ch’áatl’am kwi chel’áklh.
1sg.s go hunt/track det yesterday
Chen kw’áchnexw tsi xa7útsn míxalh.
1sg.s look det four bear
“I went hunting yesterday. I saw four bears.”
b. Sen men kwélasht ta mexmíxalh.
1sg.s just shoot det bears
“I shot (and killed) the/*some of the bears.” (= all 4 bears)
c. . . . welh na tl’íw’numut ta nch’u7 míxalh.

but realis escape det one bear
“. . . but one of them escaped.” (3/4 bears were killed)
Skwxwú7mesh articles do not care about familiarity, and do not obligatorily

refer to a unique/maximal referent in the context. They therefore lack definiteness.
6.1.2 Scope
As described in section 5, English the forces the nominal to escape scope. English
a-nominals can take either wide or narrow scope.
There are two types of Skwxwú7mesh articles: deictic articles, which can take
either wide or narrow scope, and non-deictic articles, which only take narrow scope
(Gillon 2013). In (33)a, the deictic nominal, ta sts’úkwi7 “a/the fish,” can take either
wide or narrow scope. When the nominal takes narrow scope, the hearer can re-
spond with (33)b, and when it takes wide scope, the hearer can respond with (33)c.
(33) Skwxwú7mesh
a. A: Nú chexw silh7án ta sts’úkwi7?
realis.q 2sg.s buy det.deictic fish
“Did you buy a/the fish?” (wide or narrow scope)
b. B: Háw, háwk sts’úkwi7.
neg be.not fish
“No, there weren’t any fish.”
c. B: Háw, an tl’í7.
neg very dear
“No, it was too expensive.”
In (34)a, the non-deictic nominal, kwi sts’úkwi7 “a fish,” can only take narrow
scope. The hearer can only respond with (34)b, not (34)c.
(34) Skwxwú7mesh
a. A: Nú chexw silh7án kwi sts’úkwi7?
realis.q 2sg.s buy det.non-deictic fish
“Did you buy a fish?” (narrow scope)
b. B: Háw, háwk sts’úkwi7.

neg be.not fish
“No, there weren’t any fish.”
c. B: #Háw, an tl’í7.
neg very dear
“No, it was too expensive.” (Gillon 2013)
There are two kinds of articles in Skwxwú7mesh: those that allow the nominal
to take wide scope, and those that do not. Thus, narrow scope cannot be taken as
evidence of a lack of structure. However, the ability to take wide scope is likely to
involve functional superstructure.
6.1.3 Law of Contradiction

As described in section 5, English the-nominals obey the law of contradiction; a-
nominals do not. Skwxwú7mesh deictic nominals also obey the law of contradic-
tion. In (35), ta swi7ka “the man” cannot both leave and not leave.
(35) Skwxwú7mesh
#Na huyá7 ta swí7ka i háw k’as i huyá7 ta swí7ka.
realis leave det man and neg 3 leave det man
#“The man left and the man didn’t leave.” (Gillon 2013)
Consultant’s comment: “It’s a contradiction.”
The article ta thus creates an element of type e when it combines with a noun.
6.1.4 Domain Restriction
As described in section 5, English the crucially introduces domain restriction to
the nominal. In Skwxwú7mesh, the same effect can be seen. While nominals need
not refer to the unique/maximal individual in the context (section 6.1), they still
do refer to an individual in the context (if there is one). For example, nominals
introduced by ta must refer to same individual as a previous nominal with the
same NP description has introduced earlier in the discourse. In (36), ta mixalh
introduces a bear into the context; the next use of ta mixalh must refer to that
same bear.
(36) Skwxwú7mesh
a. Chen kw’áchnexw ta míxalh. (Introduces {beari} into the context)
1sg.s saw det bear
“I saw a bear.”
b. Sen men kw’élasht ta míxalh Cta mixalh = {beari}
1sg.s just shoot det bear
“Then I shot the bear.”
≠ “Then I shot another bear.”
c. [[ ta mixalh ]] = beari
Crucially, the second use of ta mixalh cannot refer to a new, different bear.
This effect can also be seen with plural nominals. For example, in (37)a, ta
xa7útsn mixalh “four bears” introduces four bears into the context. In (37)b, ta
mexmixalh “the bears” refers to all four bears that were previously introduced.
(37) Skwxwú7mesh
a. Chen kw’áchnexw ta xa7útsn míxalh.
1sg.s saw det four bear
“I saw four bears.” (Introduces {beari, bearj, beark, bearl} into the context)
b. Sen men kw’úynexw ta mexmíxalh.
1sg.s just die det bears
“I saw a beari, and then I shot the beari.” Cta mexmixalh = {beari, bearj, beark,
bearl}
c. [[ ta mexmixalh ]] = beari+bearj+beark+bearl
With the non-deictic article kwi, it also must introduce domain restriction ef-
fects, despite behaving differently from the deictic articles in terms of scope.
(38) Xa7útsn slhánay’ na mi úys.

four woman realis come inside
Chen kwíkwi-s kwi slhánay’.
1sg.s talk-caus det.non-deictic woman
“Four women came in. I talked to one of the women.”
≠ “Four women came in. I talked to a (different) woman.”
The article kwi, when used in a context, cannot be used to introduce a new refer-
ent. Thus, all D-articles must encode domain restriction. All nominals with DP
structure must refer to an individual within the discourse if there is an individual
that can be referred to.
Articles need not be definite, but if they occupy D, we expect them to display
domain restriction effects. That is, nominals that have DP structure will normally
refer to the maximal individual in the context even in a language that does not
have definiteness. Only when it is clear that the speaker is not referring to the
maximal individual can this effect be canceled. When investigating a new article,
domain restriction should be a feature that is tested for.
6.1.5 Summary
Articles can be definite or not, can force an element to take narrow scope or allow
it to take wide scope, can change the type of a nominal to e (from et), but crucially,
if an article occupies D, it must involve domain restriction. We can now turn to
languages without overt articles and use these tests to probe for the presence or
absence of D.
6.2 RESULTS IN LANGUAGES WITHOUT OVERT ARTICLES
In this section, I show how the tests can help us decide if bare nouns have covert
structure or not. Lithuanian looks a lot like English: Lithuanian bare nouns behave
the same as English the-nominals with respect to all of the tests. Lithuanian ap-
pears to have a covert article that is very much like the. Inuttut bare nouns look
more like Skwxwú7mesh nominals when it comes to all of the tests, except for
scope. In terms of scope, Inuttut bare nouns behave more like St’át’imcets nomi-
nals (cf. Matthewson 1999). Inuttut appears to have a covert article that is very
much like a St’át’imcets article. Innu-aimun is the only language where the evi-
dence for a covert article is more equivocal.
6.2.1 Definiteness
Before we can apply the tests for definiteness to bare nouns, we must remember
that there are two features commonly associated with definiteness: familiarity and
uniqueness. Bare nouns in languages without articles can be used for both novel
and familiar referents, as I discussed for Lithuanian in section 3. This is also true in
Inuttut13 and Innu-aimun14. In (39)a, tuttuk introduces a caribou into the context.
In (39)b, tuttuk refers back to that same caribou.
(39) Inuttut
a. Tuttuk takujaga.
caribou see.1>3
“I saw a caribou.” (novel)
b. Tuttuk Kukijaga
caribou shoot.1>3
“I shot the caribou.” (familiar)
In (40)a, mashkua introduces a bear into the context. In (40)b, mashku refers to
that same bear.
(40) Innu-aimun
a. Mashkua ka-utinikushâpanua, nitishinuâu.
bear.obv preverb-taken.3’>3 perceive.1>3past
“He was taken by a beari, I dreamed.” (novel)
b. Uîpat mâ kâtshî panâkuneuâkanit anite itâkanû

soon intns after remove.snow.from.den dem say.3
mashku, ekue unuîpaniut . . .
bear and.then make.go.outside.3
“Soon after the snow was removed, it is said, the beari ran out . . .”
(familiar)
(Gillon 2011)
I therefore set aside familiarity as a test for potential covert articles. Instead, I focus
on uniqueness.
The relevant question is: When bare nouns are used anaphorically, must they
refer to the maximal individual in the context? If they must, this means that the
bare nouns are definite in anaphoric contexts. If not, they are not.
Only Lithuanian bare nouns can be called definite. When anaphoric, bare
nouns must refer to the maximal individual. In (41)b, meškas “bears” must refer to
all five previously introduced bears.
13
Inuttut (Labrador Inuktitut) is an Eskimo-Aleut language spoken in Labrador. There are ap-
proximately 500 speakers. Fieldwork on Inuttut began in 2007.
14
Innu-aimun is an Algonquian language spoken in Labrador and Quebec. There are approxi-
mately 12,000 speakers. Fieldwork on Innu-aimun began in 2007.
(41) Lithuanian
a. Pamačiau penkias meškas ir septyni vilkus . . .
saw five bears and seven wolves
“I saw five bears and seven wolves . . .”
b. . . .Užmušiau meškas, #bet viena (meška) paspruko.

killed bears but one (bear) escaped
“. . . I killed the bears, #but one escaped.”
Unlike Lithuanian bare nouns, Inuttut bare nouns can refer to the maximal
individual, but they do not need to. In (42)b, tuttuit “caribou” refers to all six cari-
bou in the context. However, the addition of (42)c cancels this reference and in-
stead tuttuit refers to a subset of the original six.
(42) Inuttut
a. Tallimat adlait amma sâksit tuttuit napâttulinii.
five bears and six caribou.pl forest.loc
“There were 5 bears and 6 caribou in the forest.”
b. Tuttuit Kukijaka
caribou.pl shoot.1>3
“I shot (and killed) the caribou.” (all 6, not 5/6)
c. Illangit Kimâjut
some flee.3
“Some escaped.” (therefore less than 6)
This is reminiscent of the behavior of Skwxwú7mesh nominals. Inuttut bare nouns

refer to the maximal individual until the context says otherwise.
Similarly, Innu-aimun bare nouns need not refer to the maximal individual
in the context. In (43)b, mashkuat “bears” refers to all five bears introduced by pa-
tetât mashkuat. However, the addition of (43)c cancels this reference, and instead
mashkuat refers to four out of the original five.
(43) Innu-aimun
a. Patetât tâuat mashkuat mâk kutuâsht atîkuat uâpamakâu . . .
five exist.pl bears and six caribou.pl see.1>3pl
“There were 5 bears and 6 caribou that I saw . . .”
b. Nipâssueuat mashkuat.
shoot.1>3.pl bear.pl
“I shot (and killed) the bears.” (all 5)
c. . . . Peiku na mashku tshîtshipâtâu

one dem bear leave.by.running.3
“. . . One of them escaped/ran away.” (only 4 bears were killed)
(Gillon 2011)
This test shows that languages without articles can vary in their nominal se-
mantics in the same way that languages with articles do. The putative covert article
in Lithuanian must be definite, while the putative covert articles in Inuttut and
Innu-aimun cannot be.15 The rest of the tests must be run to see what other prop-
erties the putative covert articles in each language have.
6.2.2 Scope
Recall that nominals with articles can escape scope, only take wide scope
(St’át’imcets), take wide or narrow scope (Skwxwú7mesh), or only take narrow
scope (Māori). The ability to take wide scope (or to escape it) is suggestive of struc-
ture, since in general, DPs take wider scope than NPs do (Borthen 2003). Fur-
ther, in many languages, bare nouns obligatorily take narrow scope (e.g., Brazilian
Portuguese: Müller 2005; Blackfoot: Glougie 2000). Scope can therefore allow us
to test for potential structure. Note that by itself, this diagnostic is not enough
(narrow scope does not automatically mean that there is no article), but it is sug-
gestive of a default structure.
In Lithuanian, bare nouns strongly prefer to take wide scope, even in out-of-
the-blue contexts.
(44) Lithuanian
Kiekviena moteris pabučiavo vaiką.
every woman pref.kiss.past.3sg child
“Every woman kissed a child.”
(i) = every woman kissed a particular, same child (wide)
(ii) =?? every woman kissed a different child (?? narrow)
This is suggestive of some kind of functional superstructure. It also fits with the
analysis of the covert article as definite.
In Inuttut, bare nouns always take wide scope (see also Wharram 2003;
Compton 2004), even in out-of-the-blue cases.
15
Cyr (1993) argues that Innu-aimun has a definite article ne. I analyze ne as a demonstrative,
rather than as a definite D (see Gillon 2011 for arguments).
(45) Inuttut
Atunit adlaup nigijagit iKaluk
each bear.erg eat.3>3 fish.abs
“Each bear ate a fish.”
(i) = there’s a fish that each bear ate (wide)
(ii) ≠ each bear ate their own fish (*narrow)
Bare nouns in Inuttut are therefore likely always associated with functional
superstructure.
In Innu-aimun, on the other hand, bare nouns prefer to take narrow scope. In
out-of-the-blue contexts, bare nouns like auâssa must take narrow scope.
(46) Innu-aimun
Kassinû ishkueu shuenimepan auâssa.
every woman kiss.3>3’past child.obv
“Every woman kissed a child.”
(i) = every woman kissed a different child (narrow)
(ii) ≠ there’s a child that every woman kissed (*wide) (Gillon 2011)
However, there is a way to force wide scope. If the bare noun is anaphoric, then
wide scope becomes possible.
(47) Innu-aimun
. . . apû tût pâssuk mashku.
neg past shoot.1 bear
“. . . I didn’t shoot the bear.” (wide)
Context: a particular bear has already been introduced into the discourse
(Gillon 2011)
In previous work, I have argued that anaphoric bare nouns are associated with
more structure than non-anaphoric bare nouns in Innu-aimun (Gillon 2011).
Semantic fieldwork reveals a three-way split: Lithuanian bare nouns prefer
to take wide scope, Inuttut bare nouns always take wide scope, and Innu-aimun
bare nouns take narrow scope in out-of-the-blue cases, but wide in familiar cases.
Inuttut and Lithuanian likely have more structure than Innu-aimun in out-of-the-
blue cases.
6.2.3 Law of Contradiction

Recall that elements of type e obey the law of contradiction. If bare nouns obey the
law of contradiction, they could come out of the lexicon as type e (Chierchia 1998),
or they could have a D that creates something of type e from the NP predicate. In
only one of the languages do the bare nouns act like they are of type e directly from
the lexicon.
In Lithuanian, bare nouns obey the law of contradiction, but only in certain
positions. In preverbal position, bare nouns obey the law; in postverbal position,
they do not.16
(48) Lithuanian
a. #Katė buvo didelė ir katė buvo maža.
cat be.past big and cat be.past small
#“The cat was big and the cat was small.”
b. Buvo didelė katė ir buvo maža katė.

be.past big cat and be.past small cat
“There was a big cat and there was a small cat.”
Because bare nouns only obey the law of contradiction in some positions, it cannot
be the case that they come out of the lexicon as type e. Instead, they must be NPs
in postverbal position, and DPs in preverbal position.
Inuttut is the only language where the bare nouns could conceivably come out
of the lexicon as type e (see Compton 2004 for arguments for this). Inuttut bare
nouns always obey the law of contradiction.
(49) Inuttut
a. #Angutik takijuk amma angutik takijulungituk.
man.abs tall.3 and man not.tall.3
#“The man is tall and the man is not tall.”
b. #Takijuk angutik amma takijulungituk angutik.

tall.3 man and not.tall.3 man
#“ The man is tall and the man is not tall.”
This behavior is also consistent with the bare nouns always being associated with a
covert article. In the interests of space, I do not explore why a DP analysis is prefer-
able to a type e analysis.17
In Innu-aimun, bare nouns do not obey the law of contradiction.
(50) Innu-aimun
a. Tshinuashkushiu nâpeu mâk apû tshinuashkushit nâpeu.
tall.3 man and neg tall.3 man
“There’s a man who’s tall and a man who isn’t.”
16
I am simplifying here. This is true for copulas. It is also correlated with the preferred interpreta-
tion of the bare nouns in intransitives: preverbal bare nouns usually receive a definite interpretation
and postverbal bare nouns usually receive an indefinite interpretation (Gillon and Armoskaite 2012).
Other verbs are more complicated and beyond the scope of this chapter.
17
See Gillon and Wharram (2008) for arguments against Compton (2004).
b. Nâpeu tshinuashkushiu mâk apû tshinuashkushit nâpeu.

man tall.3 and neg tall.3 man
“There’s a man who’s tall and a man who isn’t.” (Gillon 2011)
Therefore, bare nouns in Innu-aimun cannot come out of the lexicon as type e. It
looks as though Innu-aimun bare nouns are always NPs. However, in anaphoric
cases (where nâpeu “the man” is familiar), (43) would be a contradiction. Innu-
aimun once again lacks covert articles in novel cases, but projects an article when
the referent is familiar. Thus, not only do we need to use the law of contradiction
test for nominals in different positions (post- and pre-verbal), but also in novel
and familiar cases.
6.2.4 Domain Restriction
If my hypothesis is correct (that D introduces domain restriction), then we should
see the effect of domain restriction in anaphoric uses of bare nouns.
In Lithuanian, bare nouns nearly always18 refer to a previously introduced
referent, where there is one. In (51)a, peruką introduces a wig into the context. In
(51)b, peruką refers back to that same wig.
(51) Lithuanian
a. . . . nusipirko tikrų žmogaus plaukų peruką.
refl.buy.past.3sg real.gen man.gen hair.gen wig.acc
“. . . he bought himself a wigi made of real human hair.”
(Introduces {wigi} into the context)
b. Kas čia juokinga, kad pametei peruką, . . .
what here funny that lose.past.3sg wig.acc.sg
“What’s so funny about losing the wigi . . .” Cperuką = {wigi}
c. [[ ØD peruką ]] = wigi
This is also true of plural bare nouns. In (52)a, penkias meškas “five bears” intro-
duces five bears into the context. In (52)b, meškas “bears” refers back to those same
five bears.
(52) Lithuanian
a. Pamačiau penkias meškas ir septynis vilkus.
saw five bears and seven wolves
“I saw five bears and seven wolves.” (Introduces {beari, bearj, beark, bearl,
bearm, wolfn, wolfo, wolfp, wolfq, wolfr , wolfs, wolft} into the context)
18
It is possible to use the same bare noun to refer to two different referents, but only when the bare
nouns are sufficiently far apart in the text.
b. Užmušiau meškas.
killed bears
“I killed the bears.” Cmeškas = {beari, bearj, beark, bearl, bearm}
c. [[ ØD meškas ]] = beari+bearj+beark+bearl+bearm
In Inuttut, just as in Lithuanian, bare nouns will usually refer to a previously

introduced referent, if there is one. In (53)a, tuttumik “caribou” introduces a cari-
bou into the discourse. In (53)b, tuttuk “caribou” refers back to that same caribou.
(53) Inuttut
a. tuttumik takuKaujunga . . .
caribou.acc see.recent.past.1
“I saw a caribou . . .” (Introduces {cariboui} to the context)
b. . . . tuttuk Kukijaga
caribou.abs shoot.1sg>3sg
“. . . I shot the caribou.” Ctuttuk = {cariboui}
c. [[ ØD tuttuk ]] = cariboui
This is also true of plural bare nouns. In (54)a, sâksit tuttuit “six caribou” intro-
duces six caribou into the context. In (54)b, tuttuit “caribou” refers back to those
same six caribou.
(54) Inuttut
a. Tallimat adlait amma sâksit tuttuit napâttulinii.
five bears and six caribou forest.loc
“There were 5 bears and 6 caribou in the forest.”
(Introduces {beari, bearj, beark, bearl, bearm, cariboun, caribouo, cariboup,
caribouq, caribour, caribous} into the context)
b. Tuttuit Kukijaka
caribou shoot.1sg>3pl
“I shot (and killed) the caribou.”
Ctuttuit = {cariboun, caribouo, cariboup, caribouq, caribour, caribous}
c. [[ ØD tuttuit ]] = cariboun+caribouo+cariboup+caribouq+caribour+caribous
In Innu-aimun, just like in Lithuanian and Inuttut, bare nouns will usually refer
to a previously introduced referent, if there is one. In (55)a, mashkua “bear” intro-
duces a bear into the context. In (55)b, mashku “bear” refers back to that same bear.
(55) Innu-aimun
a. Mashkua ka-utinikushâpanua . . .
bear.obv preverb-taken.3’>3
“He was taken by a beari . . .” (Introduces {beari} into the context)
b. Uîpat mâ kâtshî panâkuneuâkanit anite itâkan-û mashku,

soon intns after remove.snow.from.den dem say-3 bear
ekue unuîpaniut itâkan-û . . .
and.then make.go.outside.3>3 say.3
“Soon after the snow was removed, it is said, the beari ran out it is told. . .”
Cmashku = {beari}
(Gillon 2011)
c. [[ ØD mashk ]] = beari
u
This is also true of plural bare nouns. In (56)a, patetât . . . mashkuat “five bears”
introduces five bears into the context.19 In (56)b, mashkuat “bears” refers back to
those same five bears.
(56) Innu-aimun
a. Patetât tâuat mashkuat mâk kutuâsht atîkuat uâpamakâu.
five exist.pl bear.pl and six caribou.pl see.1>3pl
“There were five bears and six caribou that I saw.”
(Introduces {beari, bearj, beark, bearl, bearm, cariboun, caribouo, cariboup,
caribouq, caribour, caribous} into the context)
b. Ni-pâssueuat mashkuat.
I-shoot.1>3pl bear.pl
“I shot (and killed) the bears.” Cmeškas = {beari, bearj, beark, bearl, bearm}
(Gillon 2011)
c. [[ ØD mashkuat ]] = beari+bearj+beark+bearl+bearm
However, unlike Lithuanian and Inuttut, Innu-aimun bare nouns are more flex-
ible: they can introduce new referents, even where there is a previously introduced
referent with a matching NP description. In (57) below, the second atîkua “caribou”
must refer to a new caribou, unless the first caribou somehow came back to life.
(57) Innu-aimun
Tshân pâssuepan atîkua utâkushît mâk pâssuepan anûtshîsh
John shoot.3.past caribou.obv yesterday and shoot.3.past today
atîkua
caribou.obv
“John shot a cariboui yesterday and he shot a caribou*i/j today.”
This ability for bare nouns to introduce new referents so easily is unique to Innu-
aimun. This could be due to a lack of a covert article at all, or it could be that the
covert article is less likely to be inserted in Innu-aimun than in Lithuanian or Inuttut.
19
Discontinuous nominals are common in Innu-aimun and Algonquian in general (Russell and
Reinholtz 1995, Kathol and Rhodes 1999, Gillon 2011, among others).
Despite some drastic differences in the meaning of bare nouns (definite vs.
not definite), in all three languages, bare nouns can be used anaphorically. Where
they are used anaphorically, bare nouns (preferentially/must) refer to the maxi-
mal individual in the context. This behavior is unexplained without reference to
domain restriction.
6.2.5 Summary
These tests can provide us with evidence of hidden structure. In both Inuttut and
Lithuanian, it appears that there is a covert article. In Lithuanian, the covert article
looks like the; whereas in Inuttut, the covert article looks more like a St’át’imcets
article. The tests, when applied to Innu-aimun, fail to conclusively show us that
there is covert article. New tests need to be developed to show if either it com-
pletely lacks a covert article or the covert article is just less likely to be inserted
than it is in Lithuanian or Inuttut. These tests are meant as a way to begin to ex-
plore the semantics of bare nouns and their putative articles.
7 Conclusions
In this chapter, I have shown that semantic tests can be useful for showing how
much languages vary. I used these tests to address two main issues: (i) what ar-
ticles can (and do) mean in different languages (and how they can vary) and (ii)
whether articleless languages in fact have covert articles (or not). I showed that the
same tests can be applied to languages with articles and to those without. There is
evidence that all three articleless languages investigated here have covert articles
(and therefore covert D), but that the semantics of these covert articles varies from
language to language.
These tests highlight the fact that meanings are shared between languages
with and without articles: Lithuanian (anaphoric) bare nouns and English definite
nominals receive the same interpretation. They are both definite. Similarly, Innu-
aimun and Inuttut bare nouns and Skwxwú7mesh deictic nominals receive many
of the same interpretations: they can be used in novel and familiar contexts, and,
in familiar contexts, they can refer to the maximal individual in the context, but
they are not obliged to (and are therefore not definite).
Thus, there are two types of covert articles, with the same denotations as overt
articles: the-type articles (Lithuanian), and Salish-type articles (Inuttut and po-
tentially Innu-aimun). Definiteness is not found in all languages, and therefore
cannot be used as a diagnostic for (c)overt articles. However, it should still be
tested for, in order to determine the meaning of the articles (overt or covert) in
each language.
I argue that testing for domain restriction (the only obligatory meaning of D)
is crucial. All three articleless languages demonstrate sensitivity to the context in
ways that domain restriction predicts. The other tests are useful for determining
how often a covert article must be inserted: scope and the law of contradiction
show that Lithuanian and Inuttut bare nouns have articles (some of the time in
Lithuanian, and all the time in Inuttut), but that Innu-aimun bare nouns prefer to
remain articleless whenever possible.
The diagnostics that I have developed are not the end of the story, however.
The three articleless languages vary in some way that is not explored here. More
syntactic tests (probably language-internal and not applicable across these three)
are required. However, semantic tests can be useful for showing how much lan-
guages vary, even though their nominal syntax is superficially the same.
References
Abbott, Barbara. 1999. Support for a unique theory of definiteness. Proceedings from Seman-
tics and Linguistic Theory 9: 1–15.
Ajíbóyè, Oládiípò. 2006. Topics on Yorùbá nominal expressions. PhD diss., University of
British Columbia.
Ashliman, D. L. 1999. Translation of Grimm, Jacob and Wilhelm. Der Froschkönig oder der
eiserne Heinrich, Kinder- und Hausmärchen, no. 1.
Barnes, Jonathan. 1969. The law of contradiction, Philosophical Quarterly 19: 302–309.
Borer, Hagit. 2005. In Name Only. Structuring Sense. Vol. I. Oxford: Oxford University Press.
Borthen, Kaja. 2003. Norwegian Bare Singulars. PhD diss., Norwegian University of Science
and Technology.
Bošković, Željko. 2007. On the clausal and NP structure of Serbo-Croatian. In Formal
Approaches to Slavic Linguistics: The Toronto Meeting, 2006, R. Compton, M. Goled
zinowska, and U. Savchenko (eds.), 42–75. Ann Arbor: Michigan Slavic Publications.
Bošković, Željko. 2008a. The NP/DP analysis and Slovenian. In Proceedings of the Novi Sad
Generative Syntax Workshop 1, 53–73. Novi Sad: Filozofski fakultet u Novom Sadu.
Bošković, Željko. 2008b. What will you have, DP or NP? In Proceedings of NELS 37.
Bošković, Željko. 2009. More on the no-DP analysis of article-less languages, Studia Lin-
guistica 63, 187–203.
Bošković, Željko. 2010. On NPs and clauses. Unpublished manuscript, University of
Connecticut.
Bošković, Željko and Jon Gajewski. 2011. Semantic correlates of the NP/DP parameter. In
Proceedings of NELS 39.
Cappelen, Herman, and Ernie Lepore. 2005. Insensitive Semantics. A Defense of Semantic
Minimalism and Speech Act Pluralism. Malden, MA: Blackwell Publishers.
Carlson, Greg N. 1980. Reference to Kinds in English. New York: Garland Publishing.
Chierchia, Gennaro. 1998. Reference to kinds across languages. Natural Language Semantics
6: 339–405.
Christophersen, Paul. 1939. The Articles. A Study of Their Theory and Use in English. Copen-
hagen: Munksgaard.
Chung, Sandra, and William A. Ladusaw. 2004. Restriction and Saturation. Cambridge: MIT.
Cyr, Danielle. 1993. Cross-linguistic quantification: Definite articles vs. demonstratives.
Language Sciences 15(3): 195–229.
Epstein, Melissa A. 1999. On the singular indefinite article in English. In Syntax at Sunset 2,
Gianluca Storto (ed.), 14–58. Los Angeles: UCLA Working Papers in Syntax.
Etxeberria, Urtzi. 2005. Quantification and domain restriction in Basque. PhD diss., Uni-
versity of the Basque Country.
von Fintel, Kai. 1994. Restrictions on quantifier domains. PhD diss., University of Mas-
sachusetts, Amherst.
von Fintel, Kai. 1998. The semantics and pragmatics of quantifier domains. Unpublished
manuscript, MIT. Handout of a talk given at the Vilem Mathesius Center in Prague,
March.
von Fintel, Kai. 1999. Quantifier domains and pseudo-scope. Unpublished manuscript,
MIT. Handout of a talk given at Cornell Context-Dependence Conference, March 28.
Franks, Steven, and Asya Pereltsvaig. 2004. Functional Categories in the Nominal Domain.
In Proceedings of FASL.
Frege, Gottlob. 1997 [1892]. On sense and reference. In Readings in the Philosophy of Lan-
guage, Peter Ludlow (ed.), 563–583. Cambridge, MA: MIT Press.
Giannakidou, Anastasia. 2004. Domain Restriction and the Arguments of Quantificational
Determiners. In Proceedings of SALT XIV, Robert B. Young (ed.). CLC Publications.
Giannakidou, Anastasia and Uribe Etxeberria. 2010. Definiteness, contextual domain re-
striction, and quantifier structure: a crosslinguistic perspective. Unpublished manu-
script, University of Chicago and IKER/CNRS.
Gillon, Carrie. 2013. The Semantics of Determiners: Domain Restriction in Skwxwú7mesh.
Newcastle Upon Tyne: Cambridge Scholars Publishing.
Gillon, Carrie. 2009a. Deixis in Skwxwú7mesh. International Journal of American Linguis-
tics 75, 1–27.
Gillon, Carrie. 2009b. The semantic core of determiners: Evidence from Skwxwú7mesh.
In Determiners: Variation and Universals, Jila Ghomeshi, Ileana Paul, and Martina
Wiltschko (eds.), Linguistik Aktuell. Amsterdam: John Benjamins.
Gillon, Carrie. 2011. Bare nouns in Innu-aimun: What can semantics tell us about syntax?
Proceedings for the Workshop on the Structure and Constituency of the Languages of the
Americas 16, Alexis Black and Meagan Louie (eds.), 29–56. Vancouver: University of
British Columbia Working Papers in Linguistics (UBCWPL). Available at http://www.
linguistics.ubc.ca/wscla/16.
Gillon, Carrie, and Solveiga Armoskaite. 2012. The semantic import of (c)overt D. In
Proceedings of the 29th West Coast Conference on Formal Linguistics, Jaehoon Choi,
E. Alan Hogue, Jeffrey Punske, Deniz Tat, Jessamyn Schertz, and Alex Trueman (eds.),
337–345. Available at http://www.lingref.com/cpp/wccfl/29/index.html.
Gillon, Carrie, and Solveiga Armoskaite. 2013. Diagnosis: D. On getting a second opinion
for Lithuanian. In Proceedings for NELS 42, S. Keine and S. Sloggett (eds.), 169–182.
Amherst, MA: GLSA Publications.
Gillon, Carrie, and Douglas Wharram. 2008. Bare nouns in Inuktitut: To D or not to
D. Paper presented at Canadian Linguistics Association, University of British Colum-
bia, May 31–June 2.
Glougie, Jennifer. 2000. Blackfoot “Indefinites”: Bare nouns and non-assertion of existence.
WCCFL 19: 125–138.
Hawkins, John A. 1978. Definiteness and Indefiniteness: A Study in Reference and Grammati-
cality Prediction. London: Croom Helm.
Hawkins, John A. 1991. On (in)definite articles: Implicatures and (un)grammaticality pre-

diction. Journal of Linguistics 27: 405–442.
Heim, Irene. 1988. The Semantics of Definite and Indefinite Noun Phrases. New York: Garland.
Heim, Irene, and Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford:
Blackwell.
Higginbotham, James. 1985. On Semantics. Linguistic Inquiry 16: 547–593.
Ionin, Tania. 2013. Pragmatic variation among specificity markers. In Different Kinds of
Specificity Across Languages. Studies in Linguistics and Philosophy, vol. 92, Cornelia
Ebert and Stefan Hinterwimmer (eds.), 75–103. Dordrecht: Springer.
de Jong, Franciska. 1987. The compositional nature of (in)definiteness. In The Representa-
tion of (In)definiteness, Eric Reuland, and Alice ter Meulen (eds.), 286–317. Cambridge,
MA: MIT Press.
Kadmon, Nirit. 1992. On Unique and Non-unique Reference and Asymmetric Quantification.
New York: Garland.
Kathol, Andreas, and Rhodes, Richard Alan. 1999. Constituency and Linearization of
Ojibwe Nominals. In Proceedings of Workshop on Structure & Constituency in Native
American Languages, Marion Caldecott, Suzanne Gessner, and Eun-Sook Kim (eds.),
75–91. Vancouver: Dept. of Linguistics, University of British Columbia.
Karttunen, Lauri. 1976. Discourse referents. In Syntax and Semantics 7: Notes from the Lin-
guistic Underground, J. McCawley (ed.), 363–385. New York: Academic Press.
Kuipers, Aert H. 1967. The Squamish Language: Grammar, Texts, Dictionary. The Hague:
Mouton.
Löbner, Sebastian. 2002. Understanding Semantics. London: Arnold Publishers.
Longobardi, Giuseppe. 1994. Reference and proper names: A theory of N-movement in
syntax and logical form. Linguistic Inquiry 25: 609–665.
Lyons, Christopher. 1999. Definiteness. Cambridge: Cambridge University Press.
Martí, Luisa. 2003. Contextual variables, PhD diss., University of Connecticut.
Matthewson, Lisa. 1998. Determiner Systems and Quantificational Strategies: Evidence from
Salish. The Hague: Holland Academic Graphics.
Matthewson, Lisa. 1999. On the interpretation of wide-scope indefinites. Natural Language
Semantics 7: 79–134.
Mosel, Ulrike and Even Hovdhaugen. 1992. Samoan Reference Grammar. Oslo: Scandina-
vian University Press.
Müller, Ana. 2005. Bare Nominals, Number and the Count-Mass Distinction in Brazilian
Portuguese. Talk given at the University of British Columbia, October 14.
Prince, Ellen F. 1981. On the Inferencing of Indefinite-this NPs. In Elements of Discourse
Understanding, Aravind K. Joshi, Bonnie L. Webber, and Ivan A. Sag (eds.), 231–250.
Cambridge: Cambridge University Press.
Prince, Ellen F. 1992. The ZPG letter: Subjects, definiteness, and information-status. In
Discourse Description: Diverse Linguistic Analyses of a Fund-Raising Text, William
C. Mann and Sandra A. Thompson (eds.), 295–325. Amsterdam: Benjamins.
Progovac, Ljiliana. 1998. Determiner phrase in a language without determiners, in Journal
of Linguistics 34: 165–179.
Rosen, Nicole. 2003. Demonstrative position in Michif. The Canadian Journal of Linguistics/
La Revue Canadienne de Linguistique 48: 39–69.
Russell, Bertrand. 1998 [1905]. On Denoting. In Definite Descriptions: A Reader, Gary Os-
tertag, (ed.), 35–49. Cambridge, MA: MIT Press.
Russell, Bertrand, and Alfred North Whitehead (1910–1913), Principia Mathematica, 3 vols.,
Cambridge: Cambridge University Press.
Russell, Kevin, and Charlotte Reinholtz. 1995. Hierarchical structure in a non-configurational
language: Asymmetries in swampy Cree. In Proceedings of the Fourteenth West Coast
Conference on Formal Linguistics, 431–445.
Stowell, Tim. 1989. Subjects, specifiers and X-bar theory. In Alternative Conceptions of
Phrase Structure, Mark Baltin and Anthony Kroch (eds.), 232–262. Chicago: The Uni-
versity of Chicago Press.
Szabolcsi, Anna. 1987. Functional categories in the noun phrase. In Approaches to Hungar-
ian, vol. 2, Istvàn Kenesei (ed.), 167–189. Szeged: JATE.
Szabolcsi, Anna. 1994. The noun phrase. In The Syntactic Structure of Hungarian, Syntax
and Semantics, vol. 27, Stephen R. Anderson and Katalin F. Kiss (eds.), 179–275. New
York: Academic Press.
Westerståhl, Dag. 1984. Determiners and Context Sets. Generalized Quantifiers in Natural
Languages, Johan van Benthem and Alice ter Meulen (eds.), 45–71. Dordrecht: Foris.
Wharram, Douglas. 2003. On the Interpretation of (Un)certain Indefinites in Inuktitut and
Related Languages. PhD diss., University of Connecticut.
8
Linguistically Establishing Discourse Context:

Two Case Studies from Mayan Languages
Scott AnderBois and Robert Henderson
1 Introduction
In her recent article, “On the methodology of semantic fieldwork,” Matthewson (2004)
describes in detail the primary1 method of direct elicitation to the researcher’s dis-
posal: truth-value and felicity judgment tasks (henceforth, simply “judgment tasks”).*
In these tasks, the researcher presents the native speaker consultant with two things: a
sentence that the researcher already knows to be grammatical in the language under
investigation, and a context against which the truth/felicity of the sentence is to be
judged. By “context” here, we mean simply the presumed facts of the world and the
conversation within which the sentence is uttered. We see this illustrated in (1–2) for
the two languages we will work with below: Yucatec Maya and Kaqchikel.2
*
First of all, thanks to the native speaker consultants for the two case studies reported on here:
Ricardo Caba nas Haas, Samuel Canul Yah, Rosi Couoh Pool, Nercy Chan Moo, Alberto Poot Cocom,
Aarón Puc Chi, and Jael Vazquez Tuun for Yucatec Maya and Flora Simon, Ana Lopez de Mateo, Juan
Ajsivinac Sian, and (especially) Magda Sotz Mux for Kaqchikel. Thanks also to Ryan Bochnak, Seth
Cable, Lisa Matthewson, Wilson Silva, Stavros Skopeteas, Judith Tonhauser, one anonymous reviewer,
and participants at the 2012 LSA Special Session on Semantic Fieldwork Methodology for comments
on aspects of the present chapter. For the work described in the two case studies themselves, we are
also grateful to a long list of other people and audiences; see AnderBois 2012 and Henderson 2011 for
the full list.
1
Matthewson, in fact, makes the stronger claim that “these are the only legitimate types of seman-
tic judgment.” It should be noted as well that Matthewson also accepts the validity of grammaticality
judgments themselves, but classifies this as a syntactic judgment, rather than a semantic one.
2
Abbreviations used for Yucatec Maya glosses: Imp: imperfective aspect, Pfv: perfective aspect,
Stat: “status” suffixes, Top: topic marker. For agreement morphology, we follow the terminological
tradition among Mayanists, referring to Set A (≈ Ergative) and Set B (≈ Absolutive) markers, e.g.,
A3 = 3rd person Ergative. Abbreviations used for Kaqchikel glosses: A: Set A (≈ Ergative), Ag: agentive
nominalization, B: Set B (≈ Absolutive), Com: completive aspect, Dat: dative, Dem: demonstrative, Pl:
plural derivation, Plrc: pluractional derivation. Note that for both Yucatec Maya and Kaqchikel, we
follow the respective standard orthographies. We refer the reader to the primary works cited in each
case study for more details. 207

(1) a. Context: Another speaker has asked you whether it is going to rain.
b. Sentence whose felicity is to be judged:
K-in tukl-ik-e’ yan u k’áax-al ja’.
Imp-A1 think-Stat-Top will A3 fall-Stat water
≈“I think it’s going to rain.”
(2) a. Context: We have to examine some weavings very closely. One worker
looks through half of them one by one. The other looks through the
other half one by one.
b. Sentence whose truth is to be judged:
Ka’i’ samaj-el-a’ x-Ø-ki-nik’o-la’ ri kem.
two work-Ag-Pl Com-B3s-A3p-look.through-Plrc Dem weaving
“Two workers looked through the weavings one by one.”
In this paper, we examine in detail one particular aspect of these tasks: the
choice of what language to use for presenting the context to the consultant. Note
that for the sake of illustration, our example contexts in (1–2) are written here in
English, which is neither the language under investigation nor the prevalent lan-
guage of wider communication in our two case studies. While we recognize that
practical and/or sociolinguistic factors (discussed in section 2) may prove deter-
minate in any given case, our primary focus is on the more purely linguistic factors
which influence this decision. To this end, we present two in-depth case studies
from Mayan languages—attitude reports and parentheticality in Yucatec Maya
(section 3) and distributive pluractionality in Kaqchikel (section 4)—illustrating
linguistic factors that may favor one language or the other. Both case studies come
from the authors’ own fieldwork experiences. In particular, the (alphabetically)
first author has worked regularly with speakers of Yucatec Maya in Mexico and
the United States since 2007. The (alphabetically) second author has been doing
linguistic fieldwork in Kaqchikel-speaking communities in Guatemala since 2005.
(Hereafter, first author and second author will be used to single out the authors in
alphabetical order.)
What we hope to show with these case studies is that there is a complex array
of linguistic factors one must consider in choosing the language in which to pres-
ent the discourse scenario. That is to say, there is no hard and fast rule or algorithm
to determine what language ought to be used to establish the discourse context for
judgment tasks. Therefore, our central conclusion is a methodological one:
(3) Best practices for linguistically establishing discourse contexts in
judgment tasks:
1. Researchers should disclose what language was used to establish the
discourse context.
2. Researchers should disclose the reasons why a given language was
chosen, especially when these reasons are purely linguistic in nature.
Linguistically Establishing Discourse Context 209
Before proceeding, a brief piece of terminology is needed. There are two

possible choices of which language3 to use to establish the discourse context:
the language that is itself under investigation or some other language spoken by
both the research and native speaker consultant. Following Matthewson’s lead
we will refer to this first language as the object language—OL—as it is the object
of study. We will refer to the second language as the “language of wider commu-
nication”—LWC—following Grenoble and Whaley (2006)’s work on language
revitalization.
There are several reasons we prefer this term to other available options. First,
as Matthewson notes, her term of choice—“meta-language”—is already a tech-
nical term in formal semantics and logic and therefore may lend itself to confu-
sion. Second, while our term of choice does convey something about the status of
the two languages in most fieldwork situations, it avoids the assumptions implicit
in terms like “contact language,” which seems inappropriate to describe native
speaker linguists, since no situation of “contact” is present. We believe that even
for native speaker linguists, there may well be semantic/pragmatic research ques-
tions where the use of LWC is nonetheless preferable, especially given the wide-
spread bilingualism in many such scenarios.
2 Practical and Sociolinguistic Factors
As discussed in depth by Matthewson (2004), researchers dating at least to

Harris and Voegelin (1953) have suggested that elicitation must always be con-
ducted in the object language in order to avoid influencing the results. Mat-
thewson argues convincingly that this concern is overstated, claiming that the
use of LWC to establish the discourse context “has only a negligible influence
on the consultant.” The central point she makes is that there does not seem to
be any evidence that speakers in translation tasks provide sentences that are
ungrammatical in OL, rather than ones that may be less frequent or may have
particular pragmatic functions (whether or not these are shared with the orig-
inal sentence in LWC). This latter possibility does mean that the results of a
translation task likely do not provide reliable evidence regarding the frequency
of particular structures or word orders. It also is one of the reasons why Mat-
thewson cautions repeatedly that “translations should always be treated as a
clue rather than a result.” However, neither of these caveats gives us reason to
doubt that translation tasks from LWC into OL are reliable ways to produce
grammatical sentences of OL.
Moreover, even if there were evidence that speakers do in fact produce un-
grammatical sentences in OL in translation tasks, this does not necessarily mean
3
The possibility of establishing the discourse context by non-linguistic means is discussed in section 2.
that OL must be used to present the discourse context in a truth-value/felicity

judgment task. Unlike a translation task, judgment tasks do not particularly ask
speakers to attend to the details of the language used to construct the context,
but rather to the information it conveys. As we will see in the two case studies we
present below, the ability to control this information in a fine-grained way is one
of the main factors dictating the choice of which language to use for presenting
the context.
Having established that there is no inherent reason why either OL or LWC
must be used across-the-board, we turn now to a more detailed examination of
the factors that may be relevant for this decision in a given case. Before introduc-
ing our two case studies, we first briefly discuss practical and/or sociolinguistic
factors that may supersede the linguistic ones in certain circumstances. By prac-
tical/sociolinguistic factors, we simply mean factors relating to the circumstances
of the particular investigation and investigators in question, that is, ones that may
not be intrinsic to the topic being examined and that may or may not be relevant
to potential replications. Given this, we will briefly discuss some of these factors
here, though we refer the reader to more general introductions to linguistic field-
work (e.g., Bowern 2008; Newman and Ratliff 2001; Sakel and Everett 2012), as
many of these issues are not particular to fieldwork in formal semantics/pragmat-
ics. Before proceeding, we will also briefly discuss the potential for non-linguistic
presentation of discourse contexts, which we argue is limited primarily by practi-
cal considerations.
2.1 Non-linguistic presentations of discourse context
While we are focused primarily on the choice of which language to use in pre-
senting discourse contexts, we would be remiss not to address a third possibil-
ity for presenting discourse contexts: non-linguistic methods such as pictures,
videos, figurines, and so forth. For certain areas of semantic/pragmatic re-
search, this method may well be ideal. However, there are several limitations
to non-linguistic presentation of discourse contexts that make it impractical,
and maybe even impossible for certain kinds of semantic/pragmatic fieldwork.
First, creating videos or animations for establishing contexts non-linguistically
can be quite cumbersome logistically compared to creating a brief dialogue of
paragraph of text. Obviously, the degree to which this is the case will vary sig-
nificantly depending on the phenomenon under investigation. For instance,
the contributions by Bar-el (2014) and Burton and Matthewson (2014) in this
volume show that presenting contexts visually can be quite fruitful for certain
investigations.
While the logistics of showing a movie or animation are far less challeng-
ing than in the days before laptops and other modern technologies, there may
be unforeseen cultural challenges to this methodology. For example, DuBois
(1980) describes the many challenges faced in showing the famous “Pear Film”
to speakers of Sakapultek (a K’iche’an branch Mayan language of Guatemala).

The pear story is a six-minute film about a group of children who steal some
pears, with no dialogue. The researcher asks the participant to narrate the story
found in the film in his or her language. Among the challenges faced was the
fact that many of the prospective consultants in the community were members
of evangelical Protestant sects that had prohibitions on watching movies (along
with dancing and certain other activities), thus straining relations with com-
munity members.
Beyond this extreme example, consultants may have trouble understanding
the concept of videos, in particular that the events portrayed in them are supposed
to be treated as real. For example, in working to elicit evidentials using videotaped
scenarios, Wilson Silva (personal communication) reports that a native speaker
consultant of Desano (a Tukanoan language spoken in the Vaupés River basin)
would consistently use the hearsay evidential, rather than considering the source
of the evidence within the video itself. That is, the degree to which speakers men-
tally situate themselves within the context of the film may vary across cultures,
speakers, and so forth.
Second, careful thought must be given to the question of what aspects of the
scenario consultants are paying attention to. Presenting the context nonverbally
forces the researcher to include certain information that could otherwise be omit-
ted and makes it more challenging4 to ensure that consultants are paying atten-
tion to the aspects of the context that are important to the researcher (see Müller
et al. 2011 for an example of this sort from the acquisition literature). The variation
across consultants, languages, and cultures in tasks such as the “pear stories” (e.g.,
Chafe 1980) also highlights the impracticality of this task for the investigation of
particular linguistic forms. Indeed, it is for precisely this reason that the pear sto-
ries are useful tools for investigating other higher-level topics such as linguistic
and cultural variation in narrative structure.
The above two concerns regarding non-visual presentation of the discourse
context do not necessarily argue against the use of non-linguistic discourse con-
texts altogether. However, they do suggest that they will often be impractical for
many researchers and that their use in semantic/pragmatic fieldwork may be
somewhat limited. A third concern, however, is far more significant in our view.
As researchers in semantics/pragmatics, we are often interested in expressions that
are intrinsically related to discourse itself (e.g., the shared or private knowledge of
speakers, what has been discussed in prior discourse, etc.) That is to say that in
cases where the relevant contextual information is crucially linguistic in nature,
non-linguistic presentation of the discourse context may be quite difficult or even
impossible.
4
There are of course ways of meeting these challenges in some cases; see Bar-el (this volume) and
Burton and Matthewson (this volume) for discussion of some such methods.
Even for contextual information that is not intrinsically linguistic, we often

wish to investigate the sensitivity of linguistic expressions to things like mental
states, which are often not readily represented nonverbally, at least not without de-
vices like “thought bubbles,” which (i) might be culturally dependent and (ii) could
themselves be more complicated than meets the eye (cf. Abusch 2012’s work on
the intricacies of individual and temporal reference in even simple comics without
thought bubbles or the like).
In many cases, the necessary context that we really wish to provide are mental
representations of previous discourse and therefore subject to both worries.
For example, in Case Study 1, the crucial context is a Question Under Dis-
cussion, or QUD, which is a shared mental representation of the issues being
considered by a group of speakers at a given moment in conversation (i.e., the
goals of the conversation) as determined in part by prior conversation. In this
case, then, the use of non-linguistic means to establish the discourse is likely
to be quite fraught (we return to this particular case briefly in section 3.2). To
sum up, although non-linguistic presentation of discourse contexts may well
be a possible and even preferred method in certain cases, there are both prac-
tical and substantive reasons why it may be infeasible or impossible in many
cases too.
2.2 Practical and sociolinguistic factors favoring the use of OL
The first and most obvious practical reason to use OL is if the native speaker con-
sultant is not fluent or is otherwise uncomfortable using a LWC. For instance, the
second author has done elicitation with a woman from Santiago Sacatepéquez (Pa
K’im) who speaks Spanish, but came to the language late in life and just does not
like it much. She is proud to be Kaqchikel and explicitly dislikes that conversa-
tions often “default” to Spanish. To illustrate her feelings, the second author once
introduced her in Kaqchikel to a fellow linguist who only spoke Spanish. When
the linguist replied to her “please to meet you” in Spanish, the woman turned her
shoulder to him and continued the entire conversation in Kaqchikel. It would be
impossible to work with this speaker in LWC, not because she would not neces-
sarily understand, but because she would not want to do the work through the
medium of Spanish. Regardless of any purely linguistic reasons to present contexts
in LWC, this would be a case where OL is clearly to be preferred for sociolinguistic
reasons.
Beyond this, the use of the OL for talking about the linguistic data at hand
(i.e., for metalinguistic uses) frequently yields more speaker commentary in OL,
which is often illuminating and can reveal new phenomena to investigate. In
principle, it would be possible to present discourse contexts in LWC, but use OL
for metalinguistic discussion. However, such a split does not seem to be a likely
state of affairs. In our experience, it generally is the case that whatever language if
used for the discourse context (and other sorts of metalinguistic discourse such as
describing the task itself) is the language that the consultant will most naturally
use for metalinguistic discussion (though of course individual consultants will
vary in their preference).
As in the case above, the use of the LWC may at times be met with hostility or
otherwise make consultants less comfortable than would the use of OL. Depend-
ing on the researcher’s fluency in OL, the use of LWC as a basis for translation of
the context may be inescapable. However, once a given context has been translated
into OL, it may again be possible to simply present the context in OL, avoiding the
need for LWC. This may make the use of OL somewhat more challenging logisti-
cally, but does not in principle force the use of LWC.
One final potential practical reason for favoring the use of OL is if language
documentation is a concurrent goal of the research. In addition to producing more
metalinguistic commentary in the OL as discussed above, this method may aid
documentation in two further ways. First, if the researcher is not fluent in OL,
the translation of the contexts themselves provides additional material in the lan-
guage. Second, to the extent that it is true that the choice of OL for the discourse
context does produce more natural (though equally grammatical) target sentences,
as discussed in the beginning of this section, the resultant sentences may provide
a more natural picture of the language as it is used. That said, translations are less
than ideal forms of language documentation, so this benefit may not be worth the
time and other resources it takes.
2.3 Practical and sociolinguistic factors favoring the use of LWC
Just as we saw for OL, perhaps the most compelling practical factor favoring
the LWC is the level of fluency of the researcher in OL. If the researcher is
less than fluent in OL, this may result in unnatural or unintelligible discourse
contexts in OL. While this problem is surmountable in principle by predeter-
mining the felicity of the discourse contexts in OL, this may prove a significant
burden in practice. Relating to this is the fact that even if the researcher is
fully bilingual, it is generally (though not always) the case that crucial lin-
guistic constructions used in the discourse context have been better studied
in LWC. In such a case, the use of OL may result in a discourse context that
contains unintended ambiguities or otherwise fails to produce exactly the in-
tended meaning.
Consider Kaqchikel, for instance. Though the second author is a good
second-language speaker, building a context that required carefully manipulating
the discourse properties of definite DPs would be too difficult for him to do in OL.
The difficulty is that the definite article in Kaqchikel has a wide distribution. It can
co-occur with names, the indefinite article, and in some first-mention NPs. While
English and Spanish allow some of these co-occurrences, it is not hard to find
examples that just do not sound right when they are directly translated into either
of these languages. The fact is that the second author just does not understand the
meanings of definite DPs well enough, from either a formal or informal perspec-
tive, to be sure that all of the relevant factors could be controlled for. Here it would
be safer to use LWC. By taking the other route, we would risk creating an infelic-
itous discourse context or one whose properties are not completely understood.
2.4 Indeterminacy of practical and sociolinguistic factors
In some cases, the aforementioned practical and/or sociolinguistic factors may

well force the researcher’s hand as to which language is to be used for the discourse
context. However, since many fieldwork situations are characterized by heavy bi-
lingualism, these factors will frequently not provide a definitive answer. In cases
where these factors are not determinate, then, purely linguistic factors have an
important, yet complex, role to play.
As we will see in the two case studies, there is no single procedure for de-
termining whether to use OL or LWC. Rather, the researcher must give careful
thought to the inventory of linguistic features in the two languages that is rele-
vant for the particular question at hand. We believe that the decision of which
language to use for linguistically establishing the discourse context is rooted in
Jakobson (1959)’s oft-quoted observation that “languages differ essentially in what
they must convey and not in what they may convey.” That is, the language chosen
will oblige the researcher to include certain information in the discourse context.
If the researcher wishes to omit this information from the context (i.e., wishes
to underspecify the context), then this will force the use of one language or the
other. Conversely, if the language chosen lacks a grammaticized way of encoding a
particular kind of information, it may be cumbersome (though possible in princi-
ple) to include this information in the discourse context. In the remainder of this
paper, we present two case studies illustrating these two pressures.
3 Case Study 1: Attitude Reports in Yucatec Maya
Thus far, we have discussed a variety of practical and sociolinguistic factors that
can influence the choice of how to present discourse contexts to speakers. While
researchers certainly must keep these factors in mind, we turn now to our prin-
ciple focus: the purely linguistic factors that influence this decision. As noted in
the introduction, these factors are often intertwined with the particular semantic/
pragmatic research question being addressed. For this reason, we find that detailed
case studies will be the best way to demonstrate the key issues involved. Section
5 will make some attempt to generalize across various cases, but as stated at the
outset, our principal claim is simply that the choice of which language is to be
used for establishing discourse context is a vital part of the research methodology
of any given study and therefore ought to be disclosed and discussed in semantic/
pragmatic fieldwork.
3.1 The Phenomenon
The first case study reports on ongoing research by the first author on the
syntax, semantics, and pragmatics of attitude reports in Yucatec Maya (YM).5
See AnderBois 2011, 2012 for further detail on these constructions and their
analysis.
In addition to several other constructions that we do not consider here, YM
has the two kinds of attitude reports seen in (4). Descriptively, we will refer to
examples like (4a), where the topic marker -e’ does not appear, as Bare Clause
reports and refer to examples like (4b), where it does appear, as Topic + Clause
reports.
(4) a. K-in tukl-ik yan u k’áax-al ja’.

Imp-A1 think-Stat will A3 fall-Stat water
≈ “I think it’s going to rain.” Bare Clause
b. K-in tukl-ik-e’ yan u k’áax-al ja’.
≈ “I think it’s going to rain.” Topic + Clause
In terms of the surface string, then, the two sentences differ only the absence
or presence of the clause-final clitic -e’. Outside of attitude reports, -e’ occurs in a
wide variety of topic constructions in the language, including individual topics,
(5a); temporal topics, (5b); and conditional antecedents, (5c).
(5) a. A kiik-e’ t-in wil-aj jo’oljeak.

A2 older.sister-Top Pfv-A1 see-Stat yesterday
“As for your big sister, I saw her yesterday.” Brody (2004)
b. Jo’oljeak-e’ t-in wil-aj a kiik.
Yesterday-Top Pfv-A1 see-Stat A2 older.sister
“Yesterday, I saw your big sister.” Brody (2004)
c. Wáa k-u jant-ik ba’al Juan-e’ k-u weenel.
if Imp-A3 eat-Stat thing Juan-Top Imp-A3 sleep
“If Juan eats something, he falls asleep.”
In terms of their semantics, the two types of attitude reports in (4) appear
to have approximately the same truth-conditions. This rough equivalence is
supported by the results of a translation task from Spanish. When speakers
were presented with a Spanish sentence Pienso/Creo que va a llover “I think/
believe that it will rain,” they produced either (4a) or (4b) in similar frequency
to one another. Based on such observations, previous literature (Hanks 1990;
5
See Section 3.2 for demographic and sociolinguistic information about the language.
Bohnemeyer 2002; Verhoeven 2007; Gutiérrez-Bravo 2010) has taken Topic +

Clause and Bare Clause attitude reports to be equivalent in their semantics
and has further assumed that they are equivalent in their syntax as well. That is
to say, that the presence or absence of -e’ has been treated as a case of more or
less free variation.
While these assumptions are, of course, consistent with the data thus far, there
is reason to expect that any semantic differences between the two forms would be
non-truth-conditional in nature. Indeed, this seems to be quite clearly the case for
individual topics, as the comparison between (6)—where no topic is present—and
(6) T-in wil-aj a kiik jo’oljeak

Pfv-A1 see-Stat A2 older.sister yesterday
“I saw your older sister yesterday.” Brody (2004)
the topicalized sentences in (5a) and (5b) makes clear. The propositional content
of all three sentences is identical.
As Matthewson (2004) argues in detail, however, non-truth-conditional
meaning is often ignored by speakers in translation tasks. Therefore, the ap-
parent truth-conditional equivalence of the Topic + Clause and Bare Clause
constructions revealed by the translation task is also consistent with there
being a systematic non-truth-conditional difference in the semantics of pairs
like (4).
Outside of attitude reports, topics both within YM and across languages are
often described as being “backgrounded.” There is a variety of different ways of
understanding what is meant by this term and there is of course cross-linguistic
variation in how topics behave. Here, we pursue the specific hypothesis that the
topic marker -e’ in YM marks information that is “backgrounded” in the sense of
being semantically not-at-issue. We adopt the now commonplace assumption that
discourse is structured around a hierarchically organized set of Questions Under
Discussion or QUDs (e.g., Roberts 1996). The basic idea is that whether or not a
given utterance responds to an overt question, speakers rely on the assumption
that there is nonetheless some implicit question whose resolution is their imme-
diate goal. This allows us in turn to characterize at-issueness in terms of the QUD
as follows (see Simons et al. 2011; AnderBois et al. 2011, and others for related
discussion):
(7) Only semantically at-issue content can felicitously respond to the
(immediate) QUD.
To take a concrete example, AnderBois et al. (2011) claim that appositive rel-
ative clauses are semantically (i.e., conventionally) marked as not-at-issue, with
main clause material, such as that underlined in (8), being at-issue. The generaliza-
tion in (7), then, holds that (8) can be felicitously used to respond to a QUD about
who is being treated at the hospital, but not respond to a QUD about what disease
Tammy’s husband has. This is not to say that all of the at-issue content necessarily
resolves a QUD, simply that it has this potential.
(8) Tammy’s husband, who had prostate cancer, was being treated at the
Dominican hospital.
The hypothesis we wish to test, then, is that Topic + Clause and Bare Clause
reports differ in the QUDs to which they (most) felicitously respond. In particu-
lar, our proposal that -e’ marks not-at-issue content leads us to expect that the two
forms in question will differ in what part of the sentence can naturally be used to
respond to the QUD, as seen in (9) (material that is hypothesized to be semanti-
cally at-issue is underlined).
(9) a. K-in tukl-ik yan u k’áax-al ja’.

≈ “I think it’s going to rain.” Bare Clause
b. K-in tukl-ik-e’ yan u k’áax-al ja’.

≈ “It’s going to rain, I think.” Topic + Clause
To test this hypothesis, consultants were asked to judge the felicity of attitude
reports of the two sorts in various discourse contexts that differ in their QUDs.
Returning to our central concern of methodology, the question is how the QUD
ought to be presented to consultants: in the LWC, Spanish, or in the OL, Yucatec
Maya.6 Before examining the semantic/pragmatic factors influencing this decision
in section 3.3, we first present a brief discussion of practical and sociolinguistic
factors at play.
3.2 Sociolinguistic and practical factors in Yucatec Maya
According to the 2005 census report on indigenous languages (INEGI 2009),

Yucatec Maya (known to speakers as maaya t’aan or more commonly, maaya)
is spoken by an estimated 759,000 speakers throughout the Yucatán peninsula.
While there are sizable populations in all three states of the peninsula (Campeche,
Quintana Roo, and Yucatán), the census reports that only 5.3% are monolin-
guals. One notable point of variation in rates of monolingualism is gender: 6.6%
of female speakers are monolinguals, whereas only 4.0% of male speakers are
monolinguals. The most significant factor determining rates of monolingualism
(and to a lesser extent the total number of speakers) is economic region. Pfeiler
and Zámišová (2006) report (based on slightly older data) that the percentage
6
The relative impracticality of non-linguistically establishing the QUD in this case is discussed
in section 3.3.
of monolingual speakers is highest in the corn- and citrus-growing areas, some-

what lower in the livestock- and henequen (agave fiber)–producing regions, and
essentially zero in coastal regions where fishing and, increasingly, tourism are the
primary economic activities.
Across all regions, though, intergenerational transmission of the language is
on the decline. Children are still learning the language (alongside Spanish), but
the rate at which it is being learned is declining (see Pfeiler and Zámišová 2006
for details and discussion of the various reasons for this decline). In sum, while
the number of speakers and intergenerational transmission of the language is
large enough that the language is considered “safe” according to most criteria for
language endangerment, its long-term vitality will depend on achieving a stable
bilingualism.
It is certainly possible, therefore, for researchers to find monolingual consul-
tants if this is desired. However, it is also the case that YM-speaking regions of
the Yucatán are typified by high rates of bilingualism. Therefore, most potential
consultants are accustomed to regularly speaking both languages and more or less
equally comfortable in both languages.
The consultants for the case study described here are seven undergraduate
students in the LICENCIATURA EN LINGÜÍSTICA Y CULTURA MAYA (“Pro-
gram in Maya Language and Culture”) at the Universidad de Oriente in Valladolid,
Yucatán. All seven consultants were native speakers of YM and are also natively
fluent in Spanish. In informal conversations (e.g., between classes, at lunchtime),
consultants consistently use both languages, with the exact mixture varying sig-
nificantly by individual and the surrounding social setting. While all of the stu-
dents in the program are native or heritage speakers of YM, the coursework is
conducted primarily in Spanish. The upshot of this is that the consultants for
this study are not only fluent in LWC, they are comfortable using LWC to talk
about OL.
Given the sociolinguistic situation of YM and the linguistic background of the
consultants for this particular study, the use of either Spanish or YM is both practi-
cal and culturally appropriate. Therefore, it seems safe to say that practical and so-
ciolinguistic factors are indeterminate in this case. As we noted in the introduction
to this chapter, this sort of widespread bilingualism is far from rare when dealing
with endangered and understudied languages. While particular circumstances
may lead to one language or the other a strong preference for, the indeterminacy
we see in the case of Yucatec Maya is not at all uncommon.
3.3 The test
Thus far, we have seen that YM has two types of attitude reports that differ super-
ficially in the presence or absence of the topic marker, -e’. We seek to test the hy-
pothesis that despite their apparent truth-conditional equivalence, the two forms
differ not just in this superficial way, but also in what sorts of QUDs they most
readily respond to in discourse (again, see AnderBois 2012 for further details and
analysis). To test this hypothesis, then, we need to present consultants with dis-
course contexts that establish particular QUDs and test which forms are (most)
felicitous in these contexts.
Since QUDs need not be implicit, the clearest way to linguistically establish a
QUD is to construct a dialogue where Speaker A asks Speaker B an explicit ques-
tion with the native speaker consultant acting as Speaker B. Before proceeding to
the question of whether to use the OL or the LWC (Yucatec Maya or Spanish), we
will first briefly discuss the prospects of non-linguistically establishing this con-
text. Since the QUD by its nature is concerned with discourse itself, it cannot be
directly established by non-linguistic means. However, it is possible to present
non-linguistic scenarios such that particular issues are more salient and therefore
presumably ought to be more likely QUDs in discourses about the scene.
Lewis et al. (2012)’s work on the interpretation of attitude reports by
English-speaking children uses just such a technique. They perform two experi-
ments where children are asked to perform a truth-value judgment task, differ-
ing in the non-linguistic context. The non-linguistic context for both experiments
involves an animation of a hide-and-seek game where a character on the screen,
Swiper, has hidden somewhere. The target sentence would be something like
“Dora thinks that Swiper is behind the toy box,” and the child is asked whether the
sentence is true or false in the animation. In order to establish different QUDs in
the two experiments, then, Lewis et al. (2012) manipulate whether there is a single
seeker, Dora, or multiple seekers, Dora and some other character. The plausible
assumption being made is that when there is a single seeker, children take the most
relevant aspect of the animation to be whether or not the seeker is right, while in
the case with multiple seekers, the differences between the beliefs of the seekers
become more salient.
While the study does suggest that it is possible to produce different QUDs
non-linguistically, applying this method in a fieldwork setting is subject to the
potential drawbacks mentioned in section 2.1, and in particular the concern that
the QUD is an intrinsically discourse-related notion. While clever scenarios like
this hide-and-seek game (or more adult-appropriate variations) may make a given
QUD more likely, they potentially introduce uncertainty about what the partici-
pant takes the QUD to be. While such complications are likely necessary for work-
ing with children, this uncertainty can be avoided with adult speakers by using an
overt question to produce the QUD, since the relationship between these is quite
direct.
Returning to linguistic means of establishing the QUD, the central question
is which language we should use to do it. Here, there is a clear linguistic reason to
favor the use of Spanish, the LWC: it is difficult or impossible to avoid the use of
the target constructions in the description of the discourse context itself. To see
this, let’s first look in depth at the task and results obtained using LWC to establish
the discourse context.
Consultants were presented with an overt question in Spanish as the QUD

having to do either with the attitudinal object7 (in (11–12), the rain itself) or the
mental state of some attitude-holder. In addition to this background, consultants
were presented with a set of four possible responses. First, two additional types of
attitude reports, (10), were used as filler items. While we leave the analysis of these
forms to future work, they appear to differ truth-conditionally from the two test
items as indicated in the glosses.
(10) Filler attitude reports

a. ??K-in tuklik u k’áax-al ja’.
Imp-A1 think A3 fall-Stat water
“??I plan for it to rain.”8 Dependent Clause
b. K-in tuklik káa k’áax-ak ja’.

Imp-A1 think for rain-Subj water
“I think/fear it could rain” Irrealis Clause
In addition to these fillers, consultants were presented with the two test items,
the Topic + Clause and Bare Clause attitude reports. Speakers were then asked9
to judge two things: (i) the naturalness/appropriateness of each response given
a particular QUD and (ii) which response is the best one, again given the QUD.
With respect to task (i), speakers varied a good deal with some speakers gener-
ally accepting both forms regardless of QUD and others indicating the pattern of
judgments indicated below. With respect to task (ii), however, speakers were more
or less unanimous that there was an asymmetry between the two reports and that
this asymmetry matches the judgment offered by some speakers in task (i). These
judgments were as follows:
(11) Question: ¿Va a llover? (“Is it going to rain?”)

a. K-in tukl-ik-e’ yan u k’áax-al ja’.
“It’s going to rain, I think.” Topic + Clause
b. # K-in tukl-ik yan u k’áax-al ja’.

“I think that it’s going to rain.” Bare Clause
7
We use this term rather than “complement clause” to remain neutral here as to its syntax. Indeed,
AnderBois (2012) argues that the Bare Clause does involve syntactic complementation, while the
Topic + Clause does not.
8
This example happens to be pragmatically odd because it sounds like the speaker must be god-
like in order to control the rain.
9
These instructions themselves were generally given primarily in Spanish, though for reasons of
expedience (i.e., the researcher is more fluent in Spanish than YM and participants were equally fluent
in both) rather than anything substantive.
(12) Question: ¿Piensas (tú) que va a llover? (“Do you think it’s going to rain?”)
a. #?K-in tukl-ik-e’ yan u k’áax-al ja’.
“It’s going to rain, I think.” Topic + Clause
b. K-in tukl-ik yan u k’áax-al ja’.
“I think that it’s going to rain.” Bare Clause
If the question and therefore the QUD has to do with the rain itself—the at-
titudinal object—the Bare Clause is dispreferred by all and marked as entirely
infelicitous by some.10 If the question/QUD has to do with the mental state of the
speaker herself, the Topic + Clause is dispreferred, whereas the Bare Clause is
judged optimal.
Taken together, these judgments provide clear support for the hypothesis that
Topic + Clause and Bare Clause reports differ in their at-issue content, as de-
fined by the QUDs to which they (most) felicitously respond. Had the presence or
absence of -e’ simply been a matter of truly free variation (as other authors seem to
suggest), there is no reason to expect that the choice of form would vary with the
QUD in the way we have found.
Given the subtlety of these judgments, there are two methodological aspects
worth highlighting that are unrelated to the choice of language for the QUD. First,
the study used judgments from seven native-speaker consultants, a relatively large
number for traditional fieldwork methods. Second, in addition to judging the fe-
licity of the sentence, speakers were asked to judge the relative felicity of the test
items in the context. The results from this additional task strongly suggest that the
grammar of all participants differentiates the forms in question, despite the varia-
tion in response to task (i).
Having seen the basic empirical results obtained using the LWC, Spanish, to
present the discourse context, we can ask ourselves: what would have happened if
we had instead used the OL, Yucatec Maya, for this purpose? To do this, we would,
of course, have to construct the relevant QUDs in YM. The problem we face,
though, is that questions about attitudes necessarily involve the use of attitude
reports. And attitude reports in YM necessarily involve one of the constructions
under investigation. The result is that questions about attitude reports also make
use of either the Topic + Clause or Bare Clause forms, as seen in (13). As noted
above, consultants do not necessarily attend to the difference between these forms
in translation tasks, and therefore may provide either of the two forms or both.
10
The fact that the Bare Clause in (11b) is actually rejected does not follow directly from the
division of at-issue content. For example, the English gloss we have given to (12b) is felicitous in re-
sponse to both sorts of QUDs (Simons 2007 and references therein). In AnderBois 2012, this infelicity
is attributed to pragmatic competition between Topic + Clause and Bare Clause forms, rather than
semantic at-issueness.
(13) a. K-a tukult-ik-wáa yan u k’áax-al ja’?

Imp-A2 think-Stat-Q will A3 fall-Stat water
≈ “Do you think it will rain?” Bare Clause
b. K-a tukult-ik-e’ yan-wáa u k’áax-al ja’?

Imp-A2 think-Stat-Top will-Q A3 fall-Stat water
≈ “Do you think it will rain?” Topic + Clause
There are two problems, then, with using these forms to provide the question/
QUD for a felicity judgment task. First, while the translation task confirms that
both these forms are grammatical, we still do not know precisely what they mean.
Given this, regardless of which form we choose, the discourse context will contain
whatever information it is that -e’ (or its absence) conveys. Even if we do the test
twice, using both forms as the QUD, the context will still convey certain things
and we as the researchers will not know a priori what these will be since this is
exactly the research question we are investigating.
Second, if we did find that speakers show preferences for including or not
including -e’ in the target items, we can’t really know that this result is due
to the semantics/pragmatics of the QUD as we have hypothesized, or is due
more directly to its form. It is worth recalling here that the position of the
previous literature was that the presence/absence of -e’ was not indicative of a
syntactic/semantic/pragmatic difference, but was free variation of some sort. If
the presence of -e’ were simply free variation, we might then expect that ques-
tions with -e’ would still receive responses with -e’ and that questions without
-e’ would receive responses without -e’. While this result would be consistent
with our hypothesis, it could also be explained by appeal to a low-level prim-
ing or entrainment effect or to a syntactic requirement of the question-answer
relationship.
3.4 Lessons from Case Study 1
In this case study, we have seen that there is a compelling linguistic reason to
prefer the use of the LWC, Spanish, to establish the discourse context, rather than
the OL, Yucatec Maya. As stated earlier, this preference stems from Jakobson’s ob-
servation that “languages differ essentially in what they must convey and not in
what they may convey.” In YM, attitude reports in both assertions and questions
must indicate whether the attitude is at-issue or not through the use (or non-use)
of the topic marker, -e’. Since the relevant context for the felicity judgment task in-
volves a QUD about an attitude, using OL for this obliges the researcher to express
this difference in the context. In Spanish (like English), attitude reports with em-
bedding (e.g., Creo que va a llover. ‘I think it will rain.’) are regularly used for both
purposes. That said, both languages do have parenthetical attitude constructions,
such as Slifting—as in (14) from Haverkate 2002—and therefore can semantically
encode this distinction (or at least a similar one). However, unlike in YM, the use
of an embedding attitude report does not itself convey whether the attitude is par-
enthetical or not.
(14) Esta se nora está vinculada, creo, a la mejor burgues’ia local
‘This lady is linked, I believe, to the upper local middle class.’
Note that this is one case where translation tasks are, in fact, revealing or
at least suggestive. The fact that consultants frequently provide Topic + Clause
reports as translations of ordinary embedding attitude reports in Spanish reflects
the salience of the construction in the language. In contrast, it is hard to imagine
an English or Spanish speaker providing a sentence with Slifting in such a task.
In AnderBois (2012), the first author argues that this salience plays a crucial role
in creating Gricean competition between the two forms in YM. In contrast, upon
hearing an embedding attitude report in English or Spanish, the corresponding
Slift is not a pragmatic competitor.
One final point to stress is that the fact that OL obliges the use of one of the
target forms is not a peculiarity of this example, but is something one encounters
quite frequently. For example, Matthewson (2004) discusses a quite similar situ-
ation arising in a study of clefts in St’át’imcets. Another similar example of this
dynamic in a quite different empirical domain can be found in Cable (2013)’s work
on “graded tense” in Gĩkũyũ, a Bantu language spoken in Kenya. The language dis-
tinguishes three different morphemes that are traditionally described as Current
Past (event occurred earlier today), Near Past (event occurred recently, but not
today), and Remote Past (event did not occur recently).
One of the key empirical questions Cable addresses is which of these past
forms is appropriate in case the speaker is uncertain when in the past the event
took place. To do this, consultants are provided with a context in English (the
LWC), which describes visiting a friend’s house and seeing a new TV and indicat-
ing “You have no idea when he bought the TV.” The target utterances for the felicity
judgment task, then, are three questions in Gĩkũyũ, each translated as “When did
you buy that TV?”, but differing in which of the three past tense forms is used. Had
Gĩkũyũ, the OL, been used to establish the discourse context, the Gĩkũyũ trans-
lation of bought would itself have to be marked with one of the three past tense
morphemes since this distinction is obligatory in the language. Just as in Case
Study 1, then, the use of the LWC allows the researcher to avoid including certain
information in the context and avoid using the target forms in a natural way.
4 Case Study 2: Distributive Pluractionality in Kaqchikel
The previous section presented a clear case where LWC is the right choice for
setting up a judgment. The problem with OL was that it made distinctions
that LWC did not, meaning the former could not be used without using the
target forms in the context itself. In this section, we consider examples where
the relationship between LWC and OL is similar, but making the fine distinc-
tions available in OL is necessary for setting up the judgment. In particular, we
will see that OL makes distinctions that are hard to encode in LWC (either in
principal or for practical reasons), but controlling these distinctions is critical
for determining the source of a negative judgment, thus necessitating the use
of OL.
4.1 The phenomenon
Our second case study grows out of ongoing work on pluractionality in Kaqchikel
conducted by the second author, which has been reported on in Henderson 2011
and Henderson 2012. These works, especially the latter, present the data and their
analysis in detail.
A strict definition takes pluractionality to be verbal derivational morphology
that generates predicates that cannot be satisfied in single event scenarios (Cusic
1981; Wood 2007). Kaqchikel has a variety of pluractional morphemes, but one
in particular, shown in bold below, forces distributive readings of plural internal
arguments. One way to think of -la’ is that it ensures a plurality of events by requir-
ing the internal argument to be interpreted distributively.
(15) a. X-e’-in-q’ete-j ri ak’wal-a’.

Com-B3p-A1s-hug-Tv Dem child-Pl
“I hugged the children.” Dist/Coll
b. X-e’-in-q’ete-la’ ri ak’wal-a’.
Com-B3p-A1s-hug-PLRC Dem child-Pl
≈ “I hugged the children one by one.”
≈ “I hugged each child”
≈ “I hugged the children individually” Dist
(i.e., False if I give the children a group hug.)
Faced with a morpheme like -la’, we wish to determine the similarities and
differences between pluractional distributivity and other kinds of expressions that
force distributive readings of nominal arguments. For example, one dimension
along which distributive expressions can vary is whether they also allow for cu-
mulative readings (Schein 1993, among others). To take an example from English,
although every and each are arguably both universal quantifiers, the former but
not the latter can have the scopeless cumulative reading in object position. That is,
both (16a) and (16b) have the wide scope universal readings, namely, for each mis-
take there are three (possible different) copy editors who caught it (individually /
collectively). They both also have the narrow scope universal reading, namely,
there are three copy editors who (individually / collectively) caught every mistake
in the manuscript. But only the sentence with every can be true in the cumulative
context in (16).
(16) Suppose that, between them, three copy-editors caught all the mistakes in a
manuscript.
a. Three copy editors caught every mistake in the manuscript.
b. #Three copy editors caught each mistake in the manuscript.
We now want to know whether the distributive dependencies established by -la’

are more like those accompanying each or every, which would provide impor-
tant evidence for its analysis. We can test for cumulative readings of distributively
interpreted internal arguments of pluractional predicates using a simple test for
entailment. As in case study 1, however, the question arises: should the context be
presented in LWC or OL, namely, Spanish or Kaqchikel? Before presenting our
answer, first let’s consider some of the sociolinguistic and practical factors as they
are in some ways different than in Yucatec Maya.
4.2 Sociolinguistic and Practical Factors for Kaqchikel
Kaqchikel is a K’ichean branch Mayan language spoken in the western high-

lands of Guatemala to the east of lake Atitlán. Kaqchikel has well over 500,000
speakers, with most being bilingual in Spanish (Richards 2003). While Kaq-
chikel is threatened, it is not endangered. Despite systematic, state-sponsored
violence against Mayas in recent decades, including Kaqchikel speakers, there
is a high rate of intergenerational transmission, especially in rural communities
(Richards 2003). In these same communities it is still quite easy to find mon-
olingual Kaqchikel speakers and speakers who are otherwise uncomfortable
speaking Spanish. Moreover, even in towns where Kaqchikel had been on the
decline there is a resurgence in interest in the language, especially among herit-
age speakers (Maddox 2011). There is also now a constitutional right to access to
education in Mayan languages, as well as an independent directorate for Mayan
language education, DIGEBI (Maxwell 2011), and an active umbrella organiza-
tion for community- and parent-directed private Kaqchikel languages schools
(Greebon 2011).
Thus, while either LWC or OL could be used most of the time in principle,
there are strong forces at work in Guatemala to make conducting fieldwork in
Kaqchikel the preferred choice. In rural areas, speakers are often significantly
more dominant in OL, but in urban areas, where LWC is more widespread, speak-
ers express the conscious preference to live and work in Kaqchikel as a statement
of being Maya. These preferences carry over to the mechanics of doing fieldwork,
namely building discourse contexts.
The consultants for the study we report on here are four native speakers of
Kaqchikel, all 30+ years old, and all of whom are proficient in Spanish. That being
said, none of the participants particularly like to speak in Spanish. For instance,
the second author has witnessed three of the speakers continue to use Kaqchikel
in conversations with monolingual Spanish participants present. From a practical
standpoint, therefore, either language could be used. From a sociolinguistic stand-
point, Kaqchikel is preferred.
4.3 The test
Recall that Kaqchikel has a pluractionality marker -la’, which generates distributive
entailments about an argument. We want to know whether this argument can be
interpreted cumulatively with respect to a higher-scoping quantifier. The relevant
test that must be run is a truth-value judgment task relative to a context, where the
context describes the cumulative scenario. The following context–sentence pair
exemplifies the kind of test that was run. The context was presented in Kaqchikel,
followed by a Kaqchikel test sentence. We then asked whether a speaker would
have been speaking truthfully had she said the test sentence in the context de-
scribed. It is particularly important to note that not only do we use OL to present
the context in (17), but that the context contains the construction -la’ itself. Thus,
everything is reversed from the first case study, where LWC was avoided precisely
to avoid using the OL construction under investigation. Both of these choices will
be defended in what follows.
(17) K’o chi niqanik’oj jujun täq kem ütz ütz. Jun samajel xunik’ola’ nik’aj. Jun
chïk samajel xunik’ola’ nik’aj chïk. We have to examine some weavings very
closely. One worker looks through half of them one by one. The other looks
through the other half one by one.
a. Ka’i’ samaj-el-a’ x-Ø-ki-nik’o-la’ ri kem.

two work-Ag-Pl Com-B3s-A1p-look.through-Plrc Dem weaving
“Two workers looked through the weavings one by one.”
The test sentence (17a) is judged true in the context in (17), showing that
distributively interpreted objects of pluractional sentences can have cumulative
interactions with subject quantifiers. The reason is that the distributive surface
scope reading of (17a) is false in the context in (17) because each worker did not
examine all of the weavings individually.11 Similarly, the collective surface scope
reading of (17a) is also false since the two workers did not, as a group, examine
each individual weaving. The inverse scope readings are similarly false because
the there are at most two workers in the given context. The only available read-
ing for (17a) in this context is the cumulative one, which all four speakers agree
it has.
11
The demonstrative ri ensures that the object denotes the maximal set of weavings salient in the
context. Without the demonstrative, that is, with a bare plural object, this reading of (17a) would be
true in the relevant context.
Having seen the basic set-up for the truth value judgment task, we can ask
what would have happened had the context been presented in LWC, Spanish?
Note that in setting up the context, we would have to paraphrase the pluractional
predicate. We have a few choices: uno por uno, individualmente, separadamente,
etc. While all of these options are close—with “uno por uno” one by one proba-
bly being the closest—none of these distributors completely overlap with -la’. For
instance, -la’ can distribute over parts of an atomic individual, while the proposed
Spanish translations cannot.
(18) X-Ø-u-k’utu-la’ mostró el tejido uno por uno.

Dat.1s show det weaving one by one
“He showed me the weaving one by one.”
(19) *Me mostró el tejido individualmente.

Dat.1s show det weaving individually
“He showed me the weaving individually.”
(20)
X-Ø-u-k’utu-la’ ch-w-e’ ri jun kem.
Com-B3s-A3p-show-Plrc P-A1s-Dat Dem one weaving
“He showed me (the various parts of) the weaving one by one.”
The general problem is that Spanish and Kaqchikel divide up the space of distri-
butive meaning in different ways, making simple direct paraphrases difficult. We
cannot select any of the candidate translations without failing to capture part of
the meaning of -la’. Suppose we ignored this problem and used LWC to set up the
context anyway. The problem is that if the entailment did not go through, it would
not be possible to tell if it were due to a lack of cumulative readings or an inade-
quate paraphrase. By using OL, we can be sure that the entailments follow from the
semantic contribution of the relevant construction.12
While we cannot safely use LWC to present the context in this situation, could
we have established the discourse context visually? While it is possible to do this,
it is also important to note that we actually run into a related problem that signif-
icantly complicates the process of doing so. To represent the context in (17), for
instance, we would need a video or a series of pictures of two people inspecting
weavings. How should we depict those people inspecting weavings? This is actu-
ally a serious problem because pluractionals like -la’ have non-trivial cardinality
and manner components. For instance, -la’ usually requires, not just distributivity,
but a large cardinality of events of the appropriate type. If our video showed people
inspecting weavings, but they did not happen to inspect enough, the test sentence
12
Another option, of course, is to just enumerate inspecting events in Spanish, e.g., John inspected
weaving 1, John inspected weaving 2, John inspected weaving 3, Mary inspected weaving 4, John inspected
weaving 5, etc. The primary problem with this tack is that it gets cumbersome fast, both for the inves-
tigator and the speaker.
would be rejected, regardless of the facts concerning cumulativity. Thus, in order

to use nonverbal methods to investigate the interaction of -la’ and higher-scoping
quantifiers, we first need an adequate account of the semantic interaction between
-la’ and various verbs. There are ways to get around this, of course. We could give
speakers the context as a script, and have them act it out for later presentation to
subjects. At that point, though, we might as well give the script to the subjects as
the context, which is exactly what we do in (17).
4.4 Lessons from Case Study 2
Our second case study presents strong linguistic arguments, going far beyond
the sociolinguistic factors, for presenting the context in OL. The issue is that OL,
Kaqchikel, makes distinctions that are hard to make via simple direct paraphrases
into LWC. If setting up the context requires making such a paraphrase, we cannot
safely use LWC. The problem cuts to the heart of truth-value judgments in a con-
text. If a speaker accepts a sentence in a context, we know that it is both true and
felicitous. If a speaker rejects a sentence, it could be false or infelicitous for a vari-
ety of reasons, some of which the investigator might not be expecting. By using OL
in examples like those above, we can more carefully control the factors that might
lead to the rejection of a sentence.
Although we have only presented one example, the problem of adequate para-
phrases is widespread. It arises particularly often when we are interested in the
entailments of particular open-class lexical items and their derivations. Consider,
for instance, a second type of pluractional contributed by the morpheme löj in (21)
and discussed in detail in Henderson (2012).
(21) X-e-b’oj-löj ri aj.

COM-B3p-show-PLRC2 Dem fireworks
“The fireworks kept going off.”
While this pluractional derives a variety of roots, it most readily targets po-
sitionals, which are a root class unique to Mayan languages, and which describe
physical properties and configurations (Haviland 1994; Tummons 2010). One of
the striking features of positionals is how highly specific their descriptive content
is. The descriptions in (22) present some representative examples of positionals in
their stative predicate form.
(22) a. jewël ‘seated uncomfortably (like with one leg to the side)’
b. tziyïl ‘heaped long and fine things (like pine needles)’
c. lab’äl ‘hanging but also thin and smooth (like nylon)’
d. nak’äl ‘stuck, but without any sort of glue’
It is difficult to set up contexts in LWC to uncover facts about the pluractional

forms of expressions like these precisely because they have such fine-grained
descriptive content. Consider, for instance, setting up a context to test whether

(22a) can be satisfied in a distributive scenario after it has been derived by the
pluractional in (21). To do so, we might say that John is seated uncomfortably with
his leg to the side, Mary is seated uncomfortably with her leg to the side, and Sue is
seated uncomfortably with her leg to the side. The sentence presented for a truth-
value judgment in this context would have a plural subject with John, Mary, and
Sue predicated of the pluractional form of (22a). The worry is that if this sentence
were judged false, it might be because seated uncomfortably with one leg to the side
just does not adequately capture the meaning of jewël, not that the pluractional
cannot describe this kind of scenario. The safer option, then, would be to set up the
context with jewël, which avoids the problem of understanding exactly what the
root means. This is an especially important consideration when working expres-
sions like positionals, which are morphologically simple expressions that have no
morphologically simple paraphrases in the LWC.
While it is always possible to do additional elicitation to discover the reason
why sentences are rejected, it is better to reduce complications if possible. Doing
semantic fieldwork is hard enough without creating new obstacles that are unre-
lated to the research question. The Kaqchikel case study has shown one example
where presenting the context in LWC is more likely to create obstacles than using
OL. By using OL, we can set up a context for a truth-value judgment, while ab-
stracting away from those aspects of the semantic content of expressions that are
not relevant for the study at hand.
5 Conclusion
This paper has presented two case studies that examine in detail how one might
choose whether to present a discourse context in OL or LWC. Beyond the issues
the two case studies raise, the central conclusion of this work is that the decision
to choose OL or LWC is potentially a crucial component of the methodology and
one that often goes undiscussed.
(23) Best practices for linguistically establishing discourse contexts in

judgment tasks:
1. Researchers should disclose what language was used to establish the
discourse context.
2. Researchers should disclose the reasons why a given language was
chosen, especially when these reasons are purely linguistic in nature.
It is worth pointing out that this proposal presupposes that researchers are
indeed providing the discourse context used for truth-value/felicity judgment in
the first place. As Matthewson (2004) has argued, the inclusion of the context
for these tasks is a crucial component of the raw linguistic data. Our proposal
builds on this, arguing that the choice of which language to present the discourse
context in is a core component of the methodology and therefore also ought to be
disclosed. Presenting the logic behind the choice to select LWC over OL, or vice
versa, is equivalent to properly detailing the methods in an experimental study.
Not only does it clarify the logic behind the study itself, it aids in replicability.
We want future researchers to be able to test and retest empirical claims and to
readily understand how to conduct parallel inquiries in other languages. Knowing
whether or not to present the context in LWC or OL could be crucial for setting up
a successful replication.
Beyond these broader methodological points, what our case studies show is
that the key factor in determining which language to use is the ability to control
the information in context. It has long been noted that languages differ not in what
they can convey, but in what they must. For establishing discourse contexts, then,
the relevance is that elements that are obligatory (or whose absence triggers prag-
matic inferences on the part of the hearer) cannot be avoided. Both case studies,
which are summarized below, play off of this point.
¤ Case 1: (Yucatec Maya attitude reports) LWC is used to avoid making
distinctions in the discourse contexts that OL requires.
¤ Case 2: (Kaqchikel pluractionals) OL is used to make distinctions in the
discourse context that are hard to make in LWC.
In each case, making the wrong choice could potentially cloud the outcome of
the judgment. In Case Study 1, using Yucatec Maya means disambiguating the
ambiguity the test sentence relies upon. In Case Study 2, using Spanish means
potentially missing the source of a negative judgment. In both cases, then, the OL
makes distinctions that the LWC does not, and the difference lies in whether these
distinctions are desirable for the research question at hand. Of course, it should
also be said that the reverse situation is equally possible—the LWC may make
distinctions that the OL is lacking. In fact, the contribution by Deal in this volume
provides an illustrative example of exactly this kind of situation. The LWC, English
in this case, obliges speakers to choose the quantificational force of modals (e.g.,
might vs. must) whereas the OL, Nez Perce, does not. In conclusion, while the
complex range of linguistic considerations precludes a more determinate proposal
for choosing which language to use in establishing discourse contexts, we hope
that this chapter has shown that this choice can be quite complex and as such a key
element of the methodology of semantic/pragmatic fieldwork.
References
Abusch, Dorit. 2012. Applying discourse semantics and pragmatics to co-reference in pic-
ture sequences. In Proceedings of Sinn und Bedeutung 17: 19–25.
AnderBois, Scott. 2011. Las atribuciones actitudinales en maya yucateco: Sintaxis y semán-
tica, presentation at V Congreso de Idiomas Indígenas de Latinoamérica.
AnderBois, Scott. 2012. At-issueness and embedding in Yucatec Maya attitude reports. Un-
published manuscript, University of Rochester.
AnderBois, Scott, Adrian Brasoveanu, and Robert Henderson. 2011. Crossing the appos-
itive/at-issue meaning boundary. In Proceedings of Semantics and Linguistic Theory
(SALT) 20, eLanguage, 328–346.
Bar-el, Leora. Documenting and classifying aspectual classes across language. In this
volume.
Bohnemeyer, Jurgen. 2002. The Grammar of Time Reference in Yukatek Maya. Munich:
LINCOM Europa.
Bowern, C. 2008. Linguistic Fieldwork: A Practical Guide. New York: Palgrave Macmillan.
Brody, Michal. 2004. The fixed word, the moving tongue: Variation in written Yucatec Maya
and the meandering evolution toward unified norms. PhD diss., University of Texas
at Austin.
Burton, Strang, and Lisa Matthewson. Targeted construction storyboards in semantic field-
work. In this volume.
Cable, Seth. 2013. Beyond the past, present and future: Towards the semantics of graded
tense in Gĩkũyũ. Natural Language Semantics. 21(3): 219–276.
Chafe, Wallace (ed.). 1980. The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of
Narrative Production. Norwood, NJ: Ablex Publishing.
Cusic, David. 1981. Verbal plurality and aspect. PhD diss., Stanford University.
Deal, Amy Rose. Reasoning about equivalence in semantic fieldwork. In this volume.
DuBois, John. 1980. Introduction—the search for a cultural niche: Showing the pear film in
a Mayan community. In The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of
Narrative Production, Wallace Chafe (ed.), 1–7. Norwood, NJ: Ablex Publishing.
Greebon, David. 2011. Educación primaria bilingüe desde el aula. In Más que Desarrollo:
Memorias de la Primera Conferencia Bienal sobre Desarrollo y Acción Comunitaria,
Peter Rohloff, Anne Kraemer Díaz, & Juan Ajsivinac Sian (eds.), 125–135. Bethel, VT:
Wuqu’ Kawoq.
Grenoble, Lenore, and Lindsay Whaley. 2006. Saving Languages: An Introduction to Lan-
guage Revitalization. New York: Cambridge University Press.
Gutiérrez-Bravo, Rodrigo. 2010. Los complementos oracionales en maya yucateco.
Lingüística Mexicana 5: 5–31.
Hanks, William. 1990. Referential Practice: Language and Lived Space Among the Maya. Chi-
cago: University of Chicago Press.
Harris, Zelig, and C. F. Voegelin. 1953. Eliciting in linguistics. Southwestern Journal of An-
thropology 9: 59–75.
Haverkate, Henk. 2002. The Syntax, Semantics and Pragmatics of Spanish Mood. Philadel-
phia: John Benjamins.
Haviland, John Beard. 1994. Te xa setel xulem [The buzzards were circling]: Categories of
verbal roots in (Zinacantec) Tzotzil. Linguistics 32(4–5): 691–742.
Henderson, Robert. 2011. Pluractional distributivity and dependence. In Proceedings of
SALT 21, Neil Ashton, Anca Chereches, and David Lutz, (eds.), 218–235. Ithaca, NY:
CLC Publications.
Henderson, Robert. 2012. Ways of pluralizing events. PhD thesis, University of California,
Santa Cruz.
INEGI, Instituto Nacional de Estadística y Geografía. 2009. Perfil sociodemográfico de la
población que habla lengua indígena. http://www.inegi.org.mx.
Jakobson, Roman. 1959. On linguistic aspects of translation. In On Translation. Reuben

A. Brower (ed.), 232–239. Cambridge, MA: Harvard University Press.
Lewis, Shevaun, Valentine Hacquard, and Jeff Lidz. (2012). The semantics and pragmatics
of belief reports in preschoolers. In Proceedings of SALT 22, A. Chereches (ed.). Ithaca,
NY: CLC Publications.
Maddox, Marc C. 2011. Chwa’q chik iwonojel: Language affect, ideology, and intergenera-
tional language use patterns in the Quinizilapa Valley of highland Guatemala. PhD
diss., Tulane University.
Maxwell, Judith. 2011. Bilingual bicultural education: Best intentions across a cultural
divide. In Mayas in Postwar Guatemala: Harvest of Violence Revisited, T. J. Smith and
W. E. Little (eds.), 84–95. Tuscaloosa: University of Alabama Press.
Müller, Anja, Petra Schulz, and Barbara Höhle. 2011. Pragmatic children: How German
children interpret sentences with and without the focus particle ‘only.’ In Experimen-
tal Pragmatics/Semantics, J. Meibauer and M. Steinbach (eds.), 79–100. Philadelphia:
John Benjamins.
Newman, P. and M. S. Ratliff. 2001. Linguistic Fieldwork. New York: Cambridge University
Press.
Pfeiler, Barbara, and Lenka Zámišová. 2006. Bilingual education: Strategy for language
maintenance or shift of Yucatec Maya. In Mexican Indigenous Languages at the
Dawn of the Twenty-First Century, M. Hidalgo (ed.), 281–300. Berlin: Walter de
Gruyter.
Richards, Michael. 2003. Atlas Lingüístico de Guatemala. Instituto de Lingüísticoy Edu-
cación de la Universidad Rafael Landívar.
Roberts, Craige. 1996. Information structure in discourse. In OSU Working Papers in Lin-
guistics, revised 1998 version, retrieved 8/20/09. http://www.ling.ohio-state.edu//
~croberts/infostr.pdf.
Sakel, J. and D. L. Everett. 2012. Linguistic Fieldwork: A Student Guide. New York: Cam-
Schein, Barry. 1993. Plurals and Events. Cambridge, MA: MIT Press.
Simons, Mandy. 2007. Observations on embedding verbs, evidentiality, and presupposition.
Lingua 117: 1034–1056.
Simons, Mandy, Judith Tonhauser, David Beaver, and Craige Roberts. 2011. What projects
and why. In Proceedings of Semantics And Linguistic Theory (SALT) 20, eLanguage,
309–327.
Tummons, Emily. 2010. Positional Roots in Kaqchikel. Master’s thesis, University of Kansas.
Verhoeven, Elisabeth. 2007. Experiential Constructions in Yucatec Maya. Philadelphia: John
Benjamins Publishing Company.
Wood, Esther Jane. 2007. The semantic typology of pluractionality. PhD diss., University of
California, Berkeley.
9
Semantic Fieldwork on TAM

Rebecca T. Cover
1 About This Chapter
Many linguists find daunting the prospect of investigating the semantics of tense,
aspect, and modality (TAM) in languages they do not speak, for several reasons.
First, with respect to these categories, judgments of grammaticality, truth, and fe-
licity are notoriously subtle, hard to access, and variable across speakers. Second,
the three categories are often semantically interdependent, such that one may not
be able to capture the semantics of imperfective aspect, for instance, without also
understanding how modality and/or tense work in the language of interest. Third,
because languages divide up the TAM space so differently, translation—always
a limited tool in the semantic fieldworker’s canon, as argued by Matthewson
(2004)—is particularly insufficient when it comes to TAM semantics, requiring
the linguist to develop supplementary approaches.
In this chapter, I hope to convince the reader that high-quality research into
TAM semantics is entirely feasible, by laying out a strategy for semantic fieldwork
on temporal and aspectual/modal reference. Specifically, I argue that a thorough
understanding of the semantics of a given TAM category is best obtained by ap-
plying three research methods, each of which builds upon the others: semantic
elicitation, text collection and analysis, and participant observation.
I illustrate these methods with case studies from my own fieldwork on Badi-
aranke, an under-described Atlantic (Niger-Congo) language. While this paper
focuses on TAM, the methods I have advocated are broadly applicable to many
semantic domains.*
*
I gratefully acknowledge the assistance of my Badiaranke consultants, Bouly Niabaly, Boubacar
(“Babacar”) Niabaly, Kady Niabaly, and Sountouré Niabaly. I also thank Ryan Bochnak, Lisa Matthew-
son, Christy Melick, and an anonymous reviewer for their helpful suggestions on an earlier draft of
the chapter. 233

For purposes of discussion, I assume the following informal working defini-

tions of tense, aspect, and modality. Tense is a grammaticalized linguistic form
that serves to restrict the temporal location of the time being talked about (e.g.,
English -ed), and aspect is a grammaticalized linguistic form that positions the
time spanned by an eventuality with respect to some other contextually salient
time (e.g., English be + -ing). (See Cover and Tonhauser, this volume, for more
formal definitions.) Modality, meanwhile, is a way of talking about possibility, ne-
cessity, probability, and so on; languages often have both grammatical and lexical
mechanisms for expressing modality.
2 About the data

2.1 The Language
Badiaranke (a.k.a. Badyaranke or Badyara) is an under-documented, under-

described language in the North Atlantic branch of Niger-Congo. It is spoken
in the border area between Senegal, Guinea, and Guinea-Bissau; the number of
speakers is uncertain but probably falls somewhere between 12,700 (Lewis 2009)
and 15,000 (based on a 2008 Badiaranke-internal census of speakers in Senegal).
In southern Senegal, the primary location of my fieldwork, Badiaranke has bor-
rowed extensively from Mandinka and Malinke, regionally dominant Mande lan-
guages, and to a lesser degree from Fula (Fulacounda and Pular Fuuta dialects),
the dominant Atlantic language in that zone; most Badiaranke speakers in Senegal
are fluent in both Fula and Mandinka.
Badiaranke is a head-marking language with basic Subject-Verb-Object (SVO)
word order. Overt tense marking is often absent, in part because aspect—which
is obligatorily marked—has default mappings to temporal reference, depending
on the sentence’s Aktionsart (inherent aspectual properties of the lexical content;
see Cover and Tonhauser, this volume). A perfective-marked clause, for instance,
generally receives a relative past interpretation when the eventuality description is
eventive (1), while a perfective-marked stative clause most often receives a relative
present interpretation (2).
(1) wub- iː de.
cough- 2sg.perf aff.decl
“You coughed.” (8.4, #1, elicitation)1
1
In examples, numbers in parentheses indicate the source of the example; e.g., (5.33, #1, elicitation)
means the example is on page 33 of notebook 5, spoken by Consultant #1 during elicitation. Numbers
are used instead of initials to adhere to human subjects requirements. Bolding is used to highlight
the predicate of interest and underlining to emphasize the relevant morpheme. I use the following
abbreviations in glosses: AFF.DECL=affirmative declarative; ben=benefactive; cond=conditional;
cop=copula; detrans=detransitivizer; dem=demonstrative; det=determiner; habit=habitual;
imper=imperative; impf=imperfective; indep=independent pronoun; inf=infinitive; interj=
interjection; irr=irrealis; loc=locative; narr=narrative morphology (in certain sequences of
Semantic Fieldwork on TAM 235
(2) Amerik ŋam- ə̃ de.

America be.far- 3sg.perf aff.decl
“America is far away.” (4.11, #2, elicitation)
Syntactic embedding, temporal adverbials, past tense suffixes, and principles

of discourse coherence can all (within limits) override these default temporal
readings. Thus in (3), paːki “yesterday” forces a past interpretation of the stative
perfective-marked clause, while in (4), the perfective-marked antecedent of a con-
ditional refers to a future time.
(3) paːki Faatu mitʃ- ĩ de Maimuna mpə- nij kaːs
yesterday Faatu think- 3sg.perf aff.decl Maimuna 3sg.impf- break cup
sẽ de bari nij- r- a.
det aff.decl but break- neg.perf- 3sg.perf.neg
“Yesterday Faatu thought that Maimuna would break a cup, but she didn’t
break [it].” (10.21, #2, elicitation)
(4) fajaːr fã nĩ rẽ mã- tʃe- woː de.

mouse det if/when come.3sg.perf 1sg.impf- fear- pass aff.decl
“If/when the mouse comes, I’ll get scared!” (8.27, #1, elicitation)
In (5), meanwhile, a non-tense-marked, perfective-marked stative clause, part of

the speaker’s account of the previous day’s activities, describes a state that obtained
at the past time already under discussion, not at utterance time.
(5) tẽbe niːna waina be- fali sẽ rəsə- bə̃ de mi-
time that at.that.time pl- donkey det be.tired- 3pl.perf aff.decl narr-
jeːr- ũ.
go.home- vent.1sg.perf
“Then, since the donkeys were tired, I came home.”
(2.58, #1, “what I did yesterday” text)
Despite the frequent absence of tense, Badiaranke is not a “tenseless language”

in the sense of Tonhauser (2010). In fact, it has two past tense markers, -ako- and
-akəd-, illustrated in (6)–(7).
clauses); neg=negative; nmlz=nominalizer; nperf=non-perfective (occurring in those clauses that

are not perfective); nsbj=non-subject (object or possessor); p=preposition; pass=passive; past=past;
perf=perfective; pl=plural; pluract=pluractional; prog=progressive; quot=quotative;
redup=reduplication; rel.habit=habitual marker in subject relative clauses, focus constructions,
and wh- questions; rel.impf=corresponding imperfective marker; rel.perf=corresponding per-
fective marker; vent=ventive (expressing motion toward the speaker)
(6) par- akõ ka- kam- e de bari kam- ə̃ de.

refuse- past.3sg. inf- dance- inf aff.decl but dance- 3sg.perf aff.decl
“She had refused to dance but she danced.” (8.37, #2, elicitation)
(7) puannə pã naŋa. piːsidoː patʃae maːe daːtuaː, siːmã patʃae maːe.
bride det self every.day chicken two lunch dinner chicken two
haː pidʒaːda kobəda-ŋka-maːe. wũ i kəde- ri- akəd-
until day seven dem cop rel.habit- do- past.irr-
aː- o.
detrans- pass
“The bride herself. Every day two chickens for lunch; dinner, two chickens.
Until the seventh day. That’s what used to be done.”
(3.26, #1, marriage customs text)
Both -ako- and -akəd- are discussed in more detail below.

Aspectual reference, meanwhile, is rarely isolated in a single morpheme;
rather, each aspect is characterized by several morphosyntactic properties, includ-
ing those listed in (8).
(8) Morphosyntactic factors distinguishing Badiaranke aspects in general:
1. Which series of subject markers is used in the clause;
2. What the subject marker attaches to (main verb, auxiliary verb, or
other);
3. The presence or absence of other morphemes (e.g., habitual kəd-,
affirmative declarative de, auxiliary verbs); and
4. The overall syntax, including relative positions of the auxiliary, main verb,
and subject markers.
Detailed discussions of the morphology and semantics of Badiaranke aspect can
be found in Cover (2010, 2011).
2.2 The fieldwork
I collected the data discussed here during four periods of fieldwork between 2004
and 2012. My primary field site was Paroumba, a predominantly Badiaranke-
speaking village in southern Senegal (Department of Velingara, Kolda Region).
In 2006 and 2008, I also worked in other nearby villages, including Patin Couta,
Pakour, and Koufambora in Senegal, as well as Sounkoutou in Guinea. For most
elicitation, I used Pulaar (a dialect of Fula) as a metalanguage, although in 2008
and 2012 I did a fair amount of monolingual elicitation. In collecting data from
texts and naturally occurring discourse, I used only Badiaranke.
2.3 The phenomena of interest
In illustrating the use of semantic elicitation, text collection, and participant ob-
servation, I will focus on two Badiaranke TAM categories. These are the imper-
fective aspect—whose semantics is heavily modal—and the discontinuous past
tense markers -ako- and -akəd-. (The term “discontinuous past” is borrowed from
Plungian and Auwera 2006.)
2.3.1 The Imperfective

Affirmative declarative clauses in the morphosyntactic category I call “imper-
fective” follow the schematic template in (9): an imperfective subject agreement
marker from the paradigm in (10) is prefixed to the verb, which is sometimes fol-
lowed by the affirmative declarative clitic de. (This is only the most basic version of
the imperfective template. Overt subjects precede the verb; affixal object markers
follow the verb; and overt non-subject participants follow the verb, but can pre-
cede or follow de.)
(9) SUBJECT-V (de)
(10) Subject marker series in the imperfective
1sg: mã-
2sg: k-
3sg: mp-
1pl: bõ-
2pl: nũ-
3pl: bẽ-
In imperfective clauses, the presence or absence of de depends on both syntax
and semantics. Syntactically, de can only occur in matrix and complement clauses;
semantically, futurate imperfective clauses and conditional consequents usually
contain de, whereas progressive-type imperfectives do not.
Let us now look at the semantic range of Badiaranke’s imperfective (Cover
2010, Cover 2011). Like imperfectives in many languages, it can convey both ha-
bitual (11) and progressive (12) meaning.
(11) birĩ duniaː sẽ fẽt- ə̃ piːsidoː pidʒaːda pẽ mpə- bədd-

since world det begin- 3sg.perf every.day sun det 3sg.impf- go.out-
uː de.
vent aff.decl
“Since the world began, every day the sun comes out.”
(7.89, #2, elicitation)
(12) dattaː- bə̃ de, Kumba mp- doːkũ.

asleep- 3sg.perf aff.decl Kumba 3sg.impf- work
“They’re asleep, [while] Kumba is working.” (10.15, #1, elicitation)
Unlike typical imperfectives, however, the Badiaranke imperfective can also convey
futurate meaning (13), as well as strong epistemic probability (14).
(13) mã- tʃifəd- ã de

1sg.impf- marry- 3sg.nsbj aff.decl
“I’ll marry her.” (4.12, #2, elicitation)2
(14) katʃud- eː toː sẽ mpə- duːdə fe Amerik.

morning- of today det 3sg.impf- enter p America
“This morning she’ll likely have entered America.” (8.22, #2, elicitation)
As argued in Cover (2011), this range of readings can be captured with a single
semantics that is partially modal, partially aspectual. Specifically, the Badiaranke
imperfective is used to talk about eventualities whose realization is likely at some
contextually salient time—with the precise kind of likelihood depending on con-
text. If the context suggests that an eventuality is ongoing and likely to continue,
the imperfective receives a progressive or habitual reading. If the eventuality has
not started but, given the way the world typically works, is likely to occur, then the
futurate reading arises instead. The imperfective can also be used if event realiza-
tion (in the past, present, or future) is judged likely based on what the speaker
knows—in which case the type of likelihood is epistemic. (For formal details of the
proposal, see Cover (2011:1329). For a cross-linguistic analysis in a similar vein, see
Arregui et al. (2014).)
2.3.2 The Discontinuous Past

As already mentioned, Badiaranke has two past-tense suffixes, -ako- and -akəd-.
Both are examples of what Plungian and van der Auwera (2006) call “discontinu-
ous pasts,” that is, tense markers that express both past time reference, and the
eventuality’s lack of “relevance” at the present time (e.g., the result state of a telic
event no longer holds). The -ako- suffix is used in clauses that are perfective (15a)
or progressive (15b), while -akəd- is used in the scope of deontic modality (16a), in
counterfactual consequents (16b) and some counterfactual antecedents (16c), and
with the periphrastic habitual aspect (16d).
(15) a. walaː, be- dʒalã sẽ, jak- ako- bə̃ de, bari silã lislam
voila pl- fetish det be- past- 3sg.perf aff.decl but now Islam
sẽ ka- ŋabət- e k- ə̃ be- dʒalã.
det inf- reduce- inf be- 3sg.prog pl- fetish
“Voila, the fetishes, they used to exist, but now Islam is reducing fetishes.”
(9.2, #7, magic and the supernatural narrative)
2
While subject and object agreement are indifferent to gender, with the verb tʃifəd- “marry”,
active voice can only be used with a masculine subject (and feminine object); for a female subject, the
passive suffix -o- must be added.
b. ka- sadʒ- e jak- akõ bẽ- pə- kamə.

inf- clap- inf be- past.1sg.perf 3pl.impf- impf- dance
“I was clapping [while] they were dancing.” (10.14, #1, elicitation)
(16) a. nteː dãt- i- akəd- ã.
neg.imper say- ben- past.irr- 3sg.nsbj
“You shouldn’t have told him.” (8.29, #1, elicitation)
b. paːki nĩ kab- ə̃ to kama dʒitt- ẽ fe
yesterday if know- 1sg.perf cond neg.impf.1sg. get- 2sg.nsbj p
telefon, mã- watt- akəde de kũpia.
telephone 1sg.impf- wait- past.irr.nperf aff.decl tomorrow
“Yesterday if I’d known I wouldn’t reach you on the phone, I would
have waited till tomorrow.”
c. paːki nĩ kab- akədõ kama dʒitt- ẽ,
yesterday if know- past.irr.perf.1sg. neg.impf.1sg. get- 2sg.nsbj
kama lẽb- ẽ.
neg.impf.1sg. call- 2sg.nsbj
“Yesterday if I had known I wouldn’t reach you, I wouldn’t have called you.”
d. weː bẽ- tʃimə wẽ maːm- e- mma a-
3sg.indep 3pl.impf- sing det grandparent- of- 1sg.indep 3sg.habit-
pə- tʃim- akəd- ã de.
impf- sing- past.irr- 3sg.nsbj aff.decl
“What they are singing, my grandparent used to sing it.”
Cover (2010:156–203) argues that both -ako- and -akəd- presuppose pastness and
implicate that the eventuality is no longer relevant (an “anti-perfect” implicature).
Reality status governs their distribution: -ako- occurs in realis environments and
-akəd- in irrealis ones.3
3 Methodology: Overview
3.1 Why worry about methodology?
While the semantics of TAM categories can certainly be investigated, linguists

conducting such investigations must control for a number of methodological
3
As Lisa Matthewson (personal communication) points out, it is not obvious why the progressive
should count as realis with respect to -ako- while the habitual is treated as irrealis. Possible reasons for
this discrepancy are explored in Cover (2010:199–200).
challenges. Foremost among these is the fact that translation is both inadequate
and unreliable for working out TAM semantics. It is inadequate because TAM is
highly language-specific: there is rarely a one-to-one correspondence between
TAM categories in two languages, and even when there is, the particular semantics
of a given category is likely not identical in both languages. For instance, a field-
worker using French as a metalanguage while researching Badiaranke would make
a fatal error if she assumed, just because some imperfective Badiaranke sentences
receive a French translation using the imparfait (imperfective), that the Badi-
aranke “imperfective aspect” has the same semantics as the French “imperfective
aspect.” The French imperfective is associated with past time reference (Hacquard
2006:29), while Badiaranke’s is not; meanwhile, the Badiaranke imperfective has
futurate and epistemic readings unmatched by its French counterpart. Even worse,
this imaginary fieldworker might take an imperfective-marked Badiaranke sen-
tence with a futurate reading—particularly common in elicited data—and, when
the French translation employed a future tense, assume that the Badiaranke form
is also a future tense, a conclusion contradicted by plentiful data (Cover 2010,
2011). Translation is also unreliable, since the fieldworker and/or consultant may
have less-than-full fluency in the metalanguage, or they might speak different dia-
lects. Even if neither of these conditions obtains, the linguist and consultant might
differ as to their interpretation of a given sentence in the metalanguage—and this
difference in interpretation will not emerge through translation alone. For all these
reasons, over-reliance on translation runs the risk of overly simple, and probably
wrong, claims about the meanings of TAM categories.
None of these complications mean that fieldworkers should ban translation
from their methodological toolkits. To the contrary, translation is still invaluable.
It allows collaborators to step outside the bounds of the target language in discuss-
ing meaning, in a language that both participants understand; and while it cannot
pinpoint the semantics of a TAM category—or most semantic categories—it cer-
tainly helps narrow down the possibilities. (For instance, if a sentence in the target
language receives a past-tense translation, it is safe to assume that the correspond-
ing category in the original sentence is not a future tense!)
How, then, can we take advantages of the strengths of translation, but avoid
over-reliance on it? Both fieldworker and consultant need to learn to talk about
the meaning of sentences—paraphrasing the meaning in the target language and/
or metalanguage, often in ways that “ordinary” speakers would find highly un-
natural.4 This does not mean that consultants must be versed in technical linguis-
tic terminology; rather, fieldworkers and consultants develop a particular way of
4
I recently got into a debate about how to express, in a descriptively grammatical English sen-
tence, that the speaker is currently unable to remember something: I forget or I forgot? As I defended
my position, I found myself saying things like, “When you say I forget, you’re saying that you are cur-
rently in the state of forgetting.” I quickly realized this was not the way non-linguists talk in non-
elicitation settings.
talking about acceptability and unacceptability that works for them. (Bohnemeyer,
this volume, also addresses this point.) Because this skill takes time and practice to
develop, fieldworkers may find that data on TAM semantics obtained early in their
collaborations with consultants, and/or early in their fieldwork experience, need
to be reinvestigated or verified at a later stage.
Two examples, one pertaining directly to TAM and one to elicitation more
generally, serve to illustrate the need to develop mutual understanding between
consultant and fieldworker. An extreme example comes from my first meeting
with one of my consultants. When I asked him (without providing any context)
how to say “I danced”, he replied with (17), which I dutifully transcribed in my
notebook.
(17) kam-iː de.
Immediately afterward, however, I asked the same consultant how to say “you (sg.)
danced”—and received the same response. I soon figured out that the form in (17)
actually had a 2sg. subject, and was not, in fact, the translation for “I danced”. Evi-
dently, in the first case, the consultant imagined a context in which I had previously
danced and he was telling me what had happened—whereas in the second case, he
chose to simply translate the sentence. Once I understood that I was not getting
the data I thought I was, and identified the reason, I developed various strategies
to avoid this confusion, including adding independent pronouns for clarity and,
of course, constructing discourse contexts to clarify the intended meaning. Even a
“simple” translation task, it turns out, is not necessarily so simple.
Also during my early fieldwork, I elicited what I intended to be paradigms of
sentences with habitual semantics, using a translation task. I tried, for instance,
to get a Badiaranke equivalent for “Every day I go to the field”—but using Pulaar,
not English, as a metalanguage. To enforce the habitual meaning, I used a habitual
adverb, as in (18).
(18) piːsidoː mã- raː de fe pədao.

every.day 1sg.impf- go aff.decl p field
Intended: “Every day I go to the field.”
Unfortunately, the Pulaar sentence that I used to prompt (18) (ñande fof mi yahat
gese) can be used to talk about an ongoing habit or about a future recurring
event. I was therefore unable to draw reliable conclusions about whether this
imperfective-marked sentence can actually get a present habitual reading. Over
subsequent months of fieldwork, I developed more surefire strategies. One of
the simplest involved eliciting similar sentences, but choosing adverbial expres-
sions that rule out future time reference (e.g., “every day last year”). Another was
to present a discourse context (as discussed below) and ask whether a certain
sentence could be used in that context. However, it turned out that I needed
to phrase that question in different ways for different consultants: one of them
would sometimes reject a sentence if it failed to capture every detail of the sce-
nario I had described. I learned to let that consultant offer a thorough transla-
tion of the context I had offered, then ask whether the scenario that he had now
detailed in Badiaranke could be truthfully described by the target sentence. This
method was more roundabout, but more reliable—and had the bonus of provid-
ing supplemental Badiaranke data.
Indeed, an important implication of the limitations of translation is that
discourse context becomes absolutely crucial. That is, once the grammaticality
of a sentence has been established, the linguist must investigate its felicity and
truth in various scenarios (see section 4.1.1 below), and what it would mean
in each.
But the construction of contexts is itself complex. First, the slightest acciden-
tal shift in discourse context can lead to a total reversal of speaker judgments, or
to completely opposite judgments by two different speakers. In such cases, it may
be indecipherable to the linguist whether she or he described the context in differ-
ent ways, or whether speakers interpreted the same context differently, or whether
they really have conflicting grammars. One of my consultants, for instance, offered
as a felicitous context for (19) a scenario in which a tree had been transplanted
to higher ground, effectively making the tree appear taller. Yet when I described
(what I intended to be) the same scenario to another consultant, the second con-
sultant rejected utterance of (19) in that context.
(19) matə mã sar- raːn- ə̃ de.

tree det tall- pluract- 3sg.perf aff.decl
“The tree became tall.” (11.67, #15, elicitation; 12.33, #2, elicitation)
Conversely, only the second consultant approved (20) as a description of an ex-
tremely tall man.
(20) saraːsar- ə̃ de.

tall.redup- 3sg.perf aff.decl
“He’s super-tall.” (10.100, #2, elicitation; 11.67, #15, elicitation)
A second complication, which may help explain such discrepancies in judgments,

is that context can never be described in every detail; proffered contexts are neces-
sarily sketchy and underspecified. As a result, consultants implicitly fill in addi-
tional details, especially any they construe as relevant to the sentence in question;
and the linguist, who rarely doubles as a mind-reader, can never be sure of how the
consultant is filling in the blanks. And third, as already mentioned, it is not enough
to come up with one context per sentence: to probe the semantics of the TAM
category in question, one needs to make adjustments to the initial context and test
how those changes affect a sentence’s acceptability. To do this well in the course of
an elicitation session, the fieldworker needs to be able to absorb information on
the fly and, equally quickly, make up new variations on the context to see if and
how judgments change.5 For an in-depth discussion of other considerations in
presenting discourse contexts, see AnderBois and Henderson (this volume).
In choosing methodology for data collection, fieldworkers researching TAM
also need to consider factors beyond the semantics of the TAM categories them-
selves. The semantic effect of a particular TAM marker or construction is often
intertwined with syntax, the semantics of other elements in the sentence, and
pragmatic (world-knowledge) concerns. Syntactic structure—and information
structure—frequently affect TAM grammaticality. As a result, if a speaker rejects a
sentence when it contains a particular TAM construction or marker, that reaction
alone does not warrant semantic conclusions: the problem may be syntactic or
pragmatic rather than semantic. Indeed, the distribution and semantics of TAM
markers may depend in part on the clause type in which they occur. In (21), for
instance, the consultant rejected as ungrammatical the progressive use of the im-
perfective in a matrix clause in isolation (21a), where he preferred the periphrastic
progressive construction (21b). He accepted the clause in (21a), however, when
embedded under jetʃ-ĩ de “I hear/heard” (21c)—in which environment he dispre-
ferred the periphrastic progressive (21d).
(21) a. *nəse nə̃ mpə- tʃimə.

child det 3sg.impf- sing
Intended: “The child is singing.” (8.1, #1, elicitation)
b. nəse nə̃ ka- tʃim- e k- ə̃.
child det inf- sing- inf be- 3sg.prog
“The child is singing.” (8.1, #1, elicitation)
c. jetʃ- ĩ de nəse nə̃ mpə- tʃimə.

hear- 1sg.perf aff.decl child det 3sg.impf- sing
“I heard the child singing.” (8.1, #1, elicitation)
d. ?jetʃ- ĩ de nəse nə̃ ka- tʃim- e k- ə̃.

hear- 1sg.perf aff.decl child det inf- sing- inf be- 3sg.prog
Intended: “I heard the child singing.” (8.1, #1, elicitation)6
As for semantic interference, lexical semantics, including Aktionsart, can
affect TAM grammaticality and/or meaning. This point was already illustrated
with the effect of stativity and eventivity on the semantics of Badiaranke’s perfec-
tive ((1)–(2) above); it is highlighted by the incompatibility of the achievement in
(22) with the progressive imperfective.
5
Mous (2007:3) makes a similar point.
6
I use an asterisk to mark ungrammatical sentences, while a question mark indicates that the
sentence was judged less than perfectly grammatical (but not ungrammatical). The use of two question
marks, as in (22), indicates even lower acceptability, verging on ungrammaticality.
(22) ??dʒeːnə- mãnã de mpə- nij kaːs.

see- 1sg.on.3sg. aff.decl 3sg.impf- break cup
Intended: “I saw him breaking a cup.” (10.13, #1, elicitation)
Beyond purely linguistic factors, world knowledge can also get in the way of
a “pure” grammaticality judgment (indeed, the knowledge that breaking a cup is
usually instantaneous causes problems in (22) as well). For instance, the fact that
with the possible exception of cats, living things usually die only once means that
while (23) can get a multiple-event reading, (24a) was simply rejected. (The plural
counterpart of (24a) is perfectly grammatical, as shown in (24b).)
(23) dajaːdaj- õ de.

fall.redup- pass aff.decl
“He fell over and over.”
(Appropriate context: A person with epilepsy has had seizures and fallen
several times that day.) (13.46, #15, elicitation)
(24) a. *kuɲiɲi kũ sadaːsad- ə̃ de.

stinging.ant det die.redup- 3sg.perf aff.decl
Intended: “The ant died and died.” (Not grammatical in English
either.) (11.83, #15, elicitation)
b. be- kuɲiɲi kũ sadaːsadə- bə̃ de.

pl- stinging.ant det die.redup- 3pl.perf aff.decl
“All the ants died.”
(Appropriate contexts: an ant disease or a strong pesticide has
decimated the ant population, one at a time or all at once.)
These methodological complexities mean that high-quality fieldwork on
TAM requires a great deal of patience and persistence from both fieldworker
and consultant. Good results also demand a certain level of experience from
consultants: while initial investigation into TAM can begin at any stage, a deep
investigation of TAM semantics can only succeed when the consultant is used
to the type of questions the linguist is asking, understands the significance
of small semantic or pragmatic variations and has developed the vocabulary
to talk about them, and, above all, has gained practice in accessing his or her
judgments.
3.2 Methodological strategy
Despite the alarming tone of the previous section, the purpose of this paper is
actually not to discourage semanticists and fieldworkers by convincing them of
the difficulty of TAM fieldwork, but rather to encourage them about the feasibility
of such fieldwork given sound and careful methodology. Specifically, I argue that
for best results, linguists investigating the semantics of TAM in any given language
should interweave three methods:
1. One-on-one elicitation;
2. Text collection and analysis; and
3. Participant observation of natural discourse.
Crucially, these three techniques should be interdependent. On the one hand,
elicitation can draw the linguist’s attention to TAM constructions to look for in
recorded texts and naturally occurring discourse; on the other, the linguist can use
elicitation to probe naturally produced data. (Both Rice 2006 and Chelliah 2001
contribute insightful discussions of the relationship between text collection and
elicitation.)
4 Methodology: Application
In this section, I discuss the three strategies—semantic elicitation, text collection,

and participant observation—in more detail. For each method, I lay out its defin-
ing properties, argue for its importance, and then present two case studies illus-
trating its value in illuminating each of the Badiaranke phenomena introduced
above (the imperfective aspect and the discontinuous past suffixes).
4.1 Semantic elicitation
First let us turn to semantic elicitation. I focus here on the strengths and weak-
nesses of this research method; for more detailed guidance on elicitation tech-
niques, the reader is referred to Matthewson (2004) and the papers in this volume
by AnderBois and Henderson; Bohnemeyer; Cover and Tonhauser; and Louie.
4.1.1 What is “Semantic Elicitation”?

I use the term semantic elicitation to designate a type of elicitation geared
toward teasing out the semantics of a particular semantic category or categories.
Like other kinds of elicitation, semantic elicitation involves asking targeted ques-
tions of, and engaging in dialogue with, a native speaker consultant, while tran-
scribing and/or audio-recording their responses. (See Matthewson 2004:393 on
the importance of noting all consultant comments.) This is a more complex and
broader view of elicitation than that of authors like Mosel (2012), who assumes
that “elicitation” involves only the gathering of expressions in the target language:
“Elicitation means collecting linguistic data by asking native speakers to produce
words, phrases, or sentences that can serve as data for the analysis of a particu-
lar linguistic phenomenon” (Mosel 2012:79). Instead, my definition of elicita-
tion, which treats dialogue with the consultant about these sentences to be part
of elicitation proper, is more in keeping with the view of Mous (2007:2) that “Elici-
tation is guided conversation about language data. It is not a questionnaire to be
filled out.”7 This back-and-forth between fieldworker and consultant is particularly
critical for semantic elicitation, which also requires some distinctive strategies and
skills.
Strategically, semantic elicitation is a more complex, more intellectual ver-
sion of the game Twenty Questions (where one player has to guess the entity the
other player has in mind by asking no more than twenty yes/no questions). In
both the game and in semantic elicitation, after the first question, the guesser has
only eliminated a fraction of the possible answers, leaving thousands or millions
of options. But with each subsequent yes-no question, those initial possibilities
are pared down more and more. With skillful participants, the questions are not
random, but targeted in such a way as to narrow the possibilities down to one.
In semantic elicitation, as in the game, the person answering questions is some-
times unable to provide a clear or definitive answer, but cannot (or will not, in the
game scenario) explain why. Naturally, there are key differences between semantic
elicitation and Twenty Questions. In elicitation, neither participant is restricted to
simple polar questions or answers, and the consultant is always free to elaborate.
Furthermore, in Twenty Questions, the questions are, in late stages of the game,
very direct: Is the answer X? Is it Y? In elicitation, in contrast, it is at best futile and
at worse misleading to ask directly: Is this suffix a realis past tense marker? Does
this construction mark imperfective aspect?
4.1.2 Why Do Semantic Elicitation?

Several linguists have questioned the value of using elicitation as a tool for lan-
guage documentation and description, and many of their critiques apply equally
to semantically oriented elicitation. Their concerns—e.g., Mithun’s (2001:48) ad-
monition that “the introduction of invented data into the literature can distort the
record of the language,” especially if that language is endangered—are legitimate
and worthy of caution. Nonetheless, the benefits of semantic elicitation outweigh
the risks. These benefits include the following:
1. Only elicitation guarantees paradigmatic data (see Louie, this volume,
on elicitation and paradigmaticity). Reliance on texts and naturally
occurring discourse, while having the advantage of being more
naturalistic, also has the weakness of leaving data to chance. The
fieldworker has little to no control over the syntactic structure, lexical
content, or TAM of the sentences that arise in these contexts. In order
to draw reliable semantic conclusions from data, it is best to make sure
7
I do not mean to say that questionnaires are never a legitimate component to linguistic elicita-
tion, merely that elicitation should not be limited to such tools. Certainly questionnaires can be useful,
especially early in the data-gathering process. I thank Ryan Bochnak and an anonymous reviewer for
helping me to clarify this point.
that data is as comprehensive and paradigmatic as possible, just as in

phonological or morphological analysis. In order to accurately describe
a category’s semantic breadth, the linguist needs to know not only where
it happens to be usable, but also the limits of its usability—where it butts
up against other categories. On the flip side, to know the full range of
readings available to a particular TAM construction, the semanticist
needs to explore how it interacts with different adverbials, different lexical
items, and so on—a diversity that can be ensured only through elicitation.
To take a particularly simple example, if one happened to come across
only perfectives of eventive predicates in Badiaranke texts, one might
assume that all perfectives, including with statives, have past time
reference—and therefore, one might be tempted to analyze the perfective,
wrongly, as a past tense.
2. Only elicitation allows control over syntax and lexical semantics, both
of which interact semantically with TAM. For instance, as discussed
in section 4.1.5 below, tenses/aspects/modal elements in superordinate
clauses can affect the TAM interpretation of subordinate clauses, and the
meaning of a particular T/A(/M) marker, especially aspectual ones, may
vary with Aktionsart. Moreover, TAM markers in subordinate or non-
declarative clauses may be more—or less—restricted, or have different
(default) readings from their counterparts in matrix clauses; see (21)
above.
3. Only elicitation allows direct investigation of semantic anomaly
and infelicity. That is, it is safe to assume that with the exception of
occasional speech errors or disfluencies, texts and naturally occurring
discourse contain only grammatical sentences, felicitous in the context
in which they are uttered. These grammatical, felicitous sentences
provide important positive evidence as to the semantics of the target
categories. However, the only way to know (a) what it is about those
contexts that allows the TAM-marked sentences to be true and felicitous
there, and (b) what the semantic limits of these categories are, is to test
grammatical and lexical variations of the sentences, as well as slightly
different contexts. Only through elicitation can we systematically obtain
minimal pairs of sentences that differ in semanticality (to borrow a
term from Pustejovsky 1995), and/or minimally differing contexts that
affect an utterance’s felicity, allowing us to pinpoint key ingredients in the
entailments, presuppositions, and implicatures introduced by the TAM
category of interest.
4. A dialogue between fieldworker and consultant about sentences and the
speaker’s reactions to them can take place only in elicitation, not in the
middle of a text or a naturally occurring conversation. The persuasiveness
of this argument depends, of course, on whether one accepts speaker
judgments at all as a legitimate source of data. To some extent, the
acceptance, or lack thereof, of judgment tasks is simply a philosophical

dividing line between schools of fieldworkers, with, for example, Mithun
(2001) rejecting them and Rice (2006) defending their importance.
As Rice (2006:238) points out, the more fundamental question is what
one considers to be the aim of linguistic description: Should we try to
describe the language as it is naturally produced (i.e., performance), or,
in addition, what speakers of the language know subconsciously about
its grammatical structure (i.e., competence)? (See Chelliah and de Reuse
2010, especially chapters 2, 3, and 12, for thoughtful discussion on this
point.) For those who do consider this subconscious, often abstract,
system of rules to be a key target of language description, it is essential
to gather native speaker judgments of grammaticality, semanticality,
truth, and felicity—and elicitation is the only way to directly probe these
judgments. Furthermore, while acknowledging Mithun ’s (2001:48)
concerns about the fieldworker making up sentences, I have found that
consultants are quite willing to laugh or tell me outright when a sentence
is bad—and that, as discussed above, seeing whether they will repeat it
themselves verbatim is a good check on whether it would actually be
uttered by a native speaker.
5. When identifying the semantics of some category, the distinction
between implicature and entailment is clearly an important one. Texts,
however, rarely allow the linguist to distinguish between the two, unless
an implicature is explicitly canceled. Instead, to determine whether A
entails or implicates B, the semanticist must ask questions designed
to tell whether or not B is cancelable—and not only in the context in
which A was spontaneously uttered. Such central questions constitute
an indispensable kind of semantic elicitation. (Tonhauser et al. 2013
provide valuable insight on how to probe “projective content,” including
implicature and presupposition, through elicitation.)
6. A fringe benefit of elicitation is that for expositional purposes, elicited
data tends to be cleaner, and therefore clearer, than textual data. The
latter, besides including more disfluencies, tends to be more complex
with respect to both propositional content and grammar. This complexity
requires an author to devote space to the cultural context, or to argue
why, despite initial appearances, the sentence is not actually an exception
to the argument s/he has been making.
4.1.3 Strategies for Semantic Elicitation

In the pursuit of semantic answers, most clues are indirect (cf. Matthewson 2004):
the truth conditions of sentences containing the categories of interest—and there-
fore the semantics of the categories themselves—emerge through the discourse
contexts in which they can (and cannot) be truthfully and felicitously used. Indeed,
this is the defining property of semantic elicitation: discussing various conceivable

contexts with consultants, in order to determine (a) whether a given sentence can
be used in each of those contexts and (b) if not, why not.8
In many cases, the fieldworker will want to carefully construct the contexts
him- or herself, in such a way that the sentence’s hypothesized truth conditions are
either present or absent. At other times, as in (25), it may be the consultant who
suggests a context. (In (25) and other examples, the dialogue about the Badiaranke
sentences took place in Pulaar and Fulacounda.)
(25) Me: Could you say:
dʒaːnaː- raːn- ə̃ de.
eat- pluract- 1sg.perf aff.decl
Consultant: Yes, you can say that.
Me: What does that mean?
Consultant: dʒaːnaː-raːn-ə̃ de. What that means. It’s like this—you didn’t
want to eat but then you went and ate.
Me: You didn’t want to?
Consultant: Yes, you didn’t want to. Now, you ate. You didn’t foresee it.
That’s dʒaːnaː-raːn-ə̃ de.
Me: Like you had refused to eat, or you hadn’t planned to?
Consultant: Yes, you hadn’t planned to. Voila.
Me: Like if you were fasting?
Consultant: Yes, while you were fasting, very good! It’s like this: You’re
there fasting. You didn’t want to break the fast. But then you eat, you forget
and eat. Then you can say—?i! dʒaːnaː-raːn-ə̃ de.
Two points about (25) deserve highlighting here. First, in this dialogue, the
consultant conveniently repeated the sentence of interest after I had introduced it.
I have found that when the consultant does not repeat the sentence spontaneously,
it is a good practice to ask the consultant to do so before delivering judgments.
This is because some consultants have a tendency to inadvertently “improve” the
sentence in their minds to make it more grammatical, true, and/or felicitous in the
given context. If the consultant’s “repetition” differs from the intended sentence,
it is necessary to repeat the latter and try again, until one can establish for certain
whether or not it can actually be used in the context provided.
A second point about (25), pointed out by an anonymous reviewer, is that
the consultant and I discussed the discourse context only after the sentence was
8
Generally, this “why” must also be investigated through indirect means; but consultants some-
times volunteer explanations which, in my experience, can reveal key semantic insights. Thus I would
not state the case so strongly as Nida (1947:139), quoted by Matthewson (2004, 386), who wrote: “The in-
vestigator has no justification for asking the why of any such linguistic phenomenon.” I would soften this
statement to read: “The investigator has no justification for expecting consultants to explain the why. . . .”
elicited; this method contrasts with, for example, that of Matthewson (2004),
where context is consistently provided before elicitation of the sentence. In my
own work, I use both strategies, depending on the aim of the particular line of
questioning. If I am trying to probe the semantics of a particular morphologi-
cal marker or construction, then I will sometimes elicit a specific sentence that
includes that morphology before discussing possible discourse contexts. When
these particular data were gathered, the aim of my elicitation was to explore the
grammaticality and the semantics of two kinds of morphological marking—the
suffix -raːn- and verb stem reduplication—when used with a range of predicates.
Consequently, I asked first whether the form was grammatical before investigating
contexts in which it was felicitous and true. An advantage of this method was that
only by asking consultants in what contexts a given sentence could be uttered was
I able to discover the vast range of these contexts, and thus the range of meanings
that these two pluractionality markers convey.
Once I have established that a sentence is usable in a particular context, I
attempt to pin down the “active ingredients” in that context by altering some rel-
evant contextual factor. For instance, (26), was an effort to refine the key elements
of the context the consultant provided in (25).
(26) Me: Once I was fasting, but then I got dizzy and I had to eat. Could I say
about that:
dʒaːnaː- raːn- ə̃ de.
eat- pluract- 1sg.perf aff.decl
Consultant: No. In that case you could only say:
dʒaːna- ã de.
eat- 1sg.perf aff.decl
“I ate.” (13.46, #15, elicitation)
A consultant’s rejection of a sentence-context pair can be followed up with a dis-

cussion of what sentence the consultant would prefer in the context. In a variation
on this procedure, the consultant may reject the first context offered for a sentence,
then suggest or approve alternative contexts in which it would be acceptable.
As (25) above illustrates, the contexts that make up the clues to the puzzle can
be provided by consultants themselves (a point also made by Bohnemeyer, this
volume). This option is valuable not only because it more actively involves the
consultant, but also because the fieldworker’s imagination is necessarily limited.
When the consultant provides a context, unexpected but significant components
of the semantics may emerge; additionally, the context that comes to the consul-
tant’s mind reveals what semantic components are most salient for the consultant
(and therefore most important to include in an illustrative context).
Another fruitful strategy for semantic elicitation involves establishing read-
ings for a particular sentence, then substituting a different lexical item and testing
whether or not the same readings are available for the grammatical categories of
interest. Before discussing (27), for instance, I had already established that several
verbs of motion, including jaːs- ‘walk/travel’ and tʃatt- “speed-walk,” take on a
semantic component of spatial distributivity (moving from place to place to place)
when the pluractional suffix -raːn- is added, including with perfective aspect. I
therefore asked:
(27) Me: Can you say:
toː raŋə- raːn- ə̃ de.
today go- pluract 3sg.perf aff.decl
Intended: ‘Today s/he went+pluract.’
Consultant: Uh-uh (i.e., ‘no’).
Me: Like if the person went one place, then left there and went somewhere
else, then left there and went somewhere else, like that?
Consultant: No.
Me: What about, you know there are some kids, if you try to send them
somewhere (on an errand) they refuse to go. But then you say, “If you go,
I’ll give you a cookie.” Or, “If you go, I’ll give you money.” So he gets up
and goes. Then can you say:
raŋə- raːn- ə̃ de.
go- pluract 3sg.perf aff.decl
Consultant: Yes. raŋə-raːn-ə̃ de. [Then he offers:]
birĩ dʒeːn- ə̃ kodi kõ raŋə- raːn- ə̃ de.
since see- 3sg.perf money det go- pluract 3sg.perf aff.decl
“As soon as s/he saw the money, s/he went+pluract.”
Here, it was the fieldworker who came up with a felicitous context, by modifying a
scenario (performing an action after having refused to do so) that the consultant
had come up with earlier for other predicates.
Matthewson (2004:380) observes that some authors consider the technique il-
lustrated in (27)—providing a sentence and asking for judgments—an illegitimate
elicitation technique (though Matthewson argues against that stance). Admittedly,
when the consultant rejects a novel sentence as produced by the fieldworker, it is
not always obvious to the fieldworker why it is rejected—the consultant may have
dismissed it for reasons orthogonal to what the fieldworker is trying to investigate.
This imperfection should not, however, lead us to totally reject elicitation. Instead,
it should make us more cautious and thorough in our application of the methodol-
ogy, and in particular in the follow-up questions that we ask.
In addition to investigation of the contexts in which a given sentence can
be uttered, another key component of semantic elicitation is the comparison
of minimally differing sentences. Minimal pairs of lexical items, of course,
reveal what phonetic properties the language’s phonological system treats as
contrastive, i.e., meaningful; in the same vein, pairs of sentences whose lexical
content is similar or identical, but whose TAM markings differ in just one way
can provide clues to the meanings of the categories in question. The contrast
between (28a) and (28b), for instance, reflects the fact that -akəd- not only shifts
perspective time into the past (Aysatu’s arrival was a live possibility with respect
to some past time, but no longer), but also forces an irrealis reading of the clause
(Aysatu did not, in fact, arrive in America).
(28) a. kãtʃĩ kãkã, Aysatu mpə- səmə de fe Amerik.

time dem Aysatu 3sg.impf- arrive aff.decl p America
“By this time, Aysatu has most likely arrived in America.”
b. kãtʃĩ kãkã, Aysatu mpə- səm- akəde de fe

time dem Aysatu 3sg.impf- arrive- past.irr.nperf aff.decl p
Amerik.
America
“By this time, Aysatu would have arrived in America.”
(Felicitous if and only if Aysatu has not arrived in America).
As the examples above illustrate, semantic elicitation differs from other kinds
with respect to the type of data collected. When the goal of elicitation is semantic,
generally the data elicited will be entire sentences—rarely will individual words
or phrases be sufficiently informative (see Matthewson 2004:383 on this point).
Moreover, for the reasons discussed in section 3.1, the fieldworker must take care
to collect a sufficiently wide range of sentence types, with respect to both syntax
and semantics, to thoroughly answer the research question. This diversity can be
obtained in two ways. First, the fieldworker should prepare for each elicitation
session by coming up with a number of sentences, or types of sentences, to ask
about. To know which sentences to include, he or she must review previously col-
lected data and figure out what gaps need to be filled, and what broad hypotheses
need to be refined. The second way to ensure sufficient variety is by encouraging
consultants (implicitly as well as explicitly) to share sentences that they themselves
think of during elicitation, inspired by the sentences already under discussion.
This technique has the same advantages as encouraging consultants to provide
contexts for felicitous and truthful utterance.
Note, finally, that sentences constructed by the fieldworker or consultant—in
whatever language—are not the only source of data for semantic elicitation. The
linguist may also wish to ask questions about constructions observed in naturally
occurring speech and in texts. Using the actual sentence uttered as a jumping-
off point, the fieldworker can then modify the original sentence with respect to
aspect, tense, or whatever other semantic category she or he is researching, check
the new version’s grammaticality, and investigate contexts in which it is and is not
true and felicitous. This option is discussed further in section 4.2.2 below.
4.1.4 Elicitation Case Study #1: The Badiaranke Imperfective

As noted in section 2.3.1, the Badiaranke imperfective has numerous possible
readings–readings that are disambiguated by syntactic and pragmatic context. In my
fieldwork, elicitation played a critical role in establishing which pragmatic and syn-
tactic ingredients play a role in this disambiguation. The dialogue in (29) exemplifies
my strategy for teasing out possible versus impossible readings of the imperfective.
(29) I ask what this Badiaranke sentence means:9
mã- səmə de utʃaːfe wẽ mpə- rəs.
1sg.impf- encounter aff.decl woman det 3sg.impf- be.tired
Consultant: “I’ll encounter her [the woman] getting tired.”
Me: When could you say that?
Consultant: When I get to my maid, I’ll find her tired from all the work
she has.
Me: Could you say that if when you get to her, she won’t have started
getting tired yet?10
Consultant: No. (8.4, #1, elicitation)
Furthermore, the Badiaranke imperfective construction is more reliably ob-
tained in elicitation than in texts (see point 6 above in section 4.1.2). This is be-
cause many sequences of clauses require a different construction with somewhat
different semantics (so-called narrative morphology; see Cover (2010:106–
118)). The sentence pair in (30) illustrates this contrast between elicited and textual
data. Both (30a) and (30b) receive a habitual reading, but (30a), produced in an
elicited, isolated matrix clause (translated from Pulaar), uses the “ordinary” (non-
narrative) imperfective, while (30b), which occurs in a text about what women
typically do at a performance of the traditional mambasaːmbas dance, uses the
narrative imperfective.
(30) a. Elicited data:

pədaːtuaːna woː bẽ- kamə de.
lunchtime each 3pl.impf- dance aff.decl
“Whenever people are eating lunch, they dance.” (10.14, #1, elicitation)
9
As with the data in (25), the target of the elicitation in this case was a particular morphosyntactic
construction—the use of imperfective morphology in subordinate clauses—leading me to construct a
sentence first and discuss contexts of use afterward. Note that consultant’s initial approval and transla-
tion of the sentence in the absence of any context tells us very little about its semantics, only about its
grammaticality. Only after the discussion of contexts in which the statement would be felicitous and
true can we begin to draw semantic conclusions.
10
Although I did not do so in this case, the best practice is to repeat the sentence each time a new
context is discussed.
b. Text data:
be- ri- a be- kam- a be- ni:n- a ha:
3pl.narr- come- irr 3pl.narr- dance- irr 3pl.narr- play- irr until
ma- pi:s- a.
3sg.narr- dawn- irr
“They come, they dance, they play until dawn.”
(7.5, #1, mambasaːmbas text)
Since texts are full of such sequences, it is hard to find enough textual examples of
the “ordinary” imperfective with the precise contexts and lexical semantics needed
to piece together an analysis.
Lest the reader fear that the “ordinary” imperfective is somehow less descrip-
tively valid, I note here that the ordinary imperfective is also attested in texts, as
discussed in section 4.2.3 below. Nonetheless, because such instances are relatively
few and far between, it is important to use elicitation to work through their seman-
tics more systematically.
4.1.5 Elicitation Case Study #2: The Discontinuous Past Suffixes

Here I discuss two ways in which elicitation proved crucial for my investigation
into the semantics of Badiaranke’s two discontinuous past markers, -ako- and
-akəd-.
First, in order to distinguish between competing analyses of tense seman-
tics in a given language, one needs data on sequence of tense effects or lack
thereof (see, e.g., Enç 1987). That is, one needs to compare the semantics of tense
in different kinds of embedded clauses (complement clauses vs. relative clauses),
marked with different tense-aspect combinations, embedded under matrix clauses
of different tenses. In my own fieldwork, for instance, I made sure to check all
the possible matrix clause-subordinate clause combinations in Table 9.1 for both
complement and relative clauses, using a range of predicates selected for their as-
pectual properties.
Table 9.1
Matrix/subordinate tense-aspect-Aktionsart combinations to test
Subordinate clause
Grammatical aspect Aktionsart Tense
perf IMPF Stative Eventive None Past
perf
Aspect
impf
Matrix clause
Stative
Aktionsart
Eventive
None
Tense
Past
As discussed in section 4.1.2 above, such systematic, paradigmatic data can

only be reliably obtained through elicitation. The minimal pair in (31), for exam-
ple, illustrates the effect of embedding an -ako-marked, perfective-marked, stative
complement clause under an imperfective-marked, eventive matrix clause.
(31) a. Faatu mpi- jimə de Maimuna kẽdan- akõ

Faatu 3sg.impf- say aff.decl Maimuna healthy- past.3sg.perf
de.
aff.decl
“Faatu will say that Maimuna was healthy.” (10.7, #1, elicitation)
b. Faatu mpi- jimə de Maimuna kẽdan- ə̃ de.

Faatu 3sg.impf- say aff.decl Maimuna healthy- 3sg.perf aff.decl
“Faatu will say that Maimuna is healthy.” (10.7, #1, elicitation)
In (31a), subordinate tense can receive only a past-shifted reading in the sense
of Stowell (1995): that is, the past time picked out by -ako- must precede the time
of the matrix event—that is, Faatu’s utterance—but need not precede the time at
which (31a) itself is uttered. In (31b), which lacks past tense marking, Faatu’s claim
is predicted to be about Maimuna being healthy at the time of the matrix speech
event—but again, no claim is made about Maimuna’s state of health at the time
(31b) is being spoken.
The second way in which elicitation proved important for the discontinuous
past suffixes is that it enabled me to investigate its anti-perfect implicature
(Cover 2010)—and, crucially, to determine that this component of meaning is
indeed an implicature and not an entailment.11 I call the implicature “anti-perfect”
because while some perfects, for example, in English, can indicate that the perfect-
marked eventuality is still relevant at reference time (the so-called current rele-
vance meaning; see, e.g., Portner 2003), Badiaranke’s discontinuous past suggests
that the eventuality is no longer relevant (32): the eventuality is over, and/or the
result state no longer holds.
(32) a. toina kab- ə̃, mma naŋa kab- ako- re-

that.day know- 1sg.perf 1sg.indep self know- past- neg.perf-
mã wũ. . . toina kab- ə̃ waː i kujoːmu.
1sg.neg.perf dem that.day know- 1sg.perf what cop kujoːmu
“It’s that day that I learned about that. I myself didn’t know that. . . .
It’s that day that I came to know what a kujoːmu [funerary ritual for a
respected leader] is.” (8.16, #1, codification festival text)
11
Unlike the discontinuous pasts in other languages investigated by Plungian and van der Auwera
(2006).
b. waːti sãp- ə̃ fe jaːr- eː- bõ, pə- dʒitt- i

time pass- 3sg.perf p town- of- 1pl.nsbj loc- get- 2sg.perf
woː kəd- i mən- akəde ka- rak-
each habit- 2sg.habit be.able- past.irr.nperf inf- smoke-
e de (bar taːme kaː siɲi woː te- rak-
inf aff.decl but now neg.impf.3sg. place each rel.perf- smoke-
oː).
pass
“There was a time in our town when wherever you got a chance,
you could smoke (but now it’s not just anywhere that smoking is
allowed).” (8.54, #1, elicitation)
While the anti-perfect meaning emerges from examples in both texts (32a)
and elicitation (32b), elicitation has two advantages. Only in elicitation could I
directly probe the anti-perfect meaning and whether it arises; and only through
elicitation could I determine that the anti-perfect meaning that shows up so often
is only an implicature, because it was cancelable—as in (33).
(33) a. paːki paːdə pã dobb- akõ de bari haːtoː

yesterday room det dirty- past.3sg.perf aff.decl but still
Mariaama re:- r- a peːs- a jã paːdijã
Mariaama come- neg.perf- 3sg.neg.perf sweep- irr here her.room
pã.
det
“Yesterday the room was dirty, but Mariaama still hasn’t swept here in
her room.” (8.67, #2, elicitation)
b. peːs Mamadu rã- kõ de Itali. jeːr- u-

last.year Mamadu go- past.3sg.perf aff.decl Italy go.home- vent-
ãkən- t- a.
still- neg.perf- 3sg.neg.perf
“Last year Mamadu went to Italy. He hasn’t come home yet.”
In (33a), since the dirty room has not yet been cleaned, it is still dirty; in (33b),
the traveler has not yet returned, so the result state of going to Italy still ob-
tains. That is, despite the presence of -ako-, any anti-perfect implicature has
been ruled out.
4.2 Text collection
4.2.1 What is text collection?

By text collection, I mean the recording of any sequence of sentences in the
language of interest.12 For the purposes of semantic description and analysis, it
does not much matter what kind of texts are collected, as long as they provide
a sufficient quantity of data bearing on the phenomena of interest. Many se-
mantic fieldworkers, however, are working with languages that lack adequate
documentation and description, and may even be endangered. In such cases,
it is imperative—and no more difficult—to record texts representing as many
genres as possible, in order to foster the most thorough documentation and
most accurate description. Such genres include stories, songs, riddles and
jokes, oral histories, monologues about future plans and past activities, ex-
planations of cultural and religious practices, conversations between native
speakers, and so on. These recorded texts should then be transcribed, glossed,
and translated.
Before arguing for the importance of using text sources, I should note that
the discussion in this section reflects my experience with Badiaranke, a very
poorly documented language with few or no written texts available for study. For
a language with a longer history of written records and therefore a larger body of
written texts, the risk that an artificial picture of the language will emerge from
elicitation is not as great, reducing the need for the fieldworker to spend time
collecting, transcribing, and translating texts. Nonetheless, as sources of data to
inform accurate description, texts in well-studied and/or written languages are no
less valuable than those in lesser-studied and unwritten ones. Regardless of who
has collected the text and whether it was originally written or oral, it should always
be noted which kind of text the examples come from, due to questions of register,
style, dialect, authorship, and so on.
4.2.2 Why Use Texts?

In many ways, it is more difficult to extract valuable data from texts than from
elicitation notes. In collecting texts, the fieldworker has virtually no control
over what the speaker says (though the use of storyboards, as discussed by
Burton and Matthewson in this volume, can alleviate this disadvantage). Text
data contain more disfluencies, speech errors, and self-interruptions than do
individual elicited sentences, and the linguist—unless she or he is a native
speaker—is less likely to understand everything that is said in the text. The
12
These texts may be either offered spontaneously by the speaker, or produced at the fieldworker’s
request. See Bohnemeyer (this volume) for a discussion of the differences between “staged” and com-
pletely spontaneous texts.
fieldworker often needs speakers to repeat unclear passages verbatim—a skill

that does not come naturally to all consultants. Additionally, before extract-
ing any relevant data from a text, the linguist must get through the work of
transcribing and translating it, an extremely time-consuming project. So why
bother with texts at all?
First, as argued by Mithun (2001) and Dixon (2007), among others, text
data, being more spontaneous and created by the speaker, constitutes a more
natural, and therefore arguably more accurate, representation of how the
language is actually spoken—a fuller and more accurate picture of speaker
performance. (Elicitation, in turn, presents a fuller picture of speaker com-
petence; see, e.g., Rice 2006:248.) In addition, texts can reveal more complex,
fuller data than elicitation alone—in part because sequences of sentences in-
teract in complex ways (e.g., reference tracking, anaphora, advancement of
reference time, and modal subordination). Perhaps most importantly, kinds of
data tend to emerge in texts that are simply unlikely to come up in elicitation,
as argued by Chelliah (2001). Badiaranke phenomena of this sort include (but
are not limited to) the narrative morphology discussed in section 4.1.4, inter-
jections and exclamatives, and the shifting of temporal reference by discourse
context.
Related to this last point, over-reliance on elicitation runs the risk of lim-
iting the data collected to topics and phenomena that occur to the linguist,
who is inevitably influenced by result of his or her theoretical training and
knowledge about other languages. Treating speakers as collaborators and en-
couraging them to volunteer additional information and suggest directions for
sessions to go in helps alleviate this problem, but only to an extent. Texts, in
contrast, can introduce data that the linguist never thought to ask about and
that the speaker never thought of as worthy of investigation, or as a phenom-
enon to be investigated. Morphologically encoded pluractionality, a signifi-
cant aspectual phenomenon in Badiaranke, met this description in my own
fieldwork.
When the fieldworker finds an interesting sentence in a text, that sentence
and the TAM categories it involves can be investigated further through elicitation.
The linguist might, for instance, change the tense or aspect marking, or change the
context slightly, then ask if the sentence, or some version of it, could still be ut-
tered in the new context. The data in (34)–(35) illustrate how I used this procedure
in working out the semantic effects of reduplication—one type of pluractionality
marking in Badiaranke—and its interaction with various grammatical aspects. I
encountered the example in (34), where pluractionality is marked by reduplica-
tion, in a story several years ago. While working on pluractionality during my
most recent field trip, I used elicitation to experiment with modifying the subject
(“woman” vs. “tree”) and aspect (habitual vs. perfective, for example); some of the
permutations are shown in (35).
(34) awa, fa͂kama dĩtaː- j- a͂, wusia i wu͂, kaː pə-

interj ruler peer- of- 3sg.nsbj man cop dem neg.impf.3sg. impf-
rəm. . . tʃaːfe wẽ kaː pə- rəmə tʃaːfe, wusia tũ
give.birth woman det neg.impf.3sg. impf- give.birth female male only
kəd- ə̃ rəmaːrəm.
habit- 3sg.habit give.birth.redup
“Well, the ruler, his friend, that was a man, didn’t give birth. . . (his) wife
never gave birth to girls; she kept bearing boys.”
(6.35, #5, ruler with two “wives” story)
(35) a. wusia tũ kəd- ə̃ rəmaːrəm.

male only habit- 3sg.habit give.birth.redup
“It’s only boys that she gave birth to over and over.”
Consultant’s context: The woman gives birth every couple of years,
always to boys.
b. darĩtʃa sẽ a- pə- rəmaːrəmə de.

orange.tree det 3sg.habit- impf- give.birth.redup aff.decl
“The orange tree constantly bears fruit.” (12.24, #1, elicitation)13
Consultant’s explanation: This sentence could be truthfully spoken
about a certain tree in the village that always has fruit on it, in every
season—not about a tree that bears many fruits once a year.
c. rəmaːrəm- ə̃ de.
give.birth.redup- 3sg.perf aff.decl
“She (has) had a lot of kids at short intervals.” (12.24, #1, elicitation)
Consultant’s judgments: Acceptable for a woman who has given birth to
four or more children every couple of years—not for a woman who has
had only two children, no matter how close their births were in time,
and not for one who has had many children separated by several years.
4.2.3 Texts Case Study #1: The Imperfective
Textual data proved particularly important when dealing with the Badiaranke
imperfective. In elicitation, consultants rarely use the general (underspecified)
imperfective construction discussed in section 2.3.1 to express habitual or pro-
gressive meaning. It is true that, as noted in section 4.1.4, I found the general
imperfective, in its ordinary (non-narrative) form, to be easier to find through
elicitation than through texts. Nonetheless, for reasons discussed below, it can
13
The final schwa on the verb is epenthetic, due to the following /d/.
still be difficult to elicit this aspect in its non-futurate senses—or to be sure

which reading the consultant has in mind. Accordingly, textual examples are
critical for confirmation that the less common uses are indeed attested in natu-
ral speech.
The issue is that the general, underspecified imperfective competes in Badi-
aranke with two other, semantically narrower aspects: the dedicated habitual and
progressive aspects. Like the imperfective, these pop up in both elicited and textual
data. However, to translate Pulaar sentences—at least matrix declarative clauses—
with unambiguously habitual or progressive semantics, consultants tend to use the
equally unambiguous aspectual constructions (see also Cover 2010:135–136). Simi-
larly, when asked about imperfective-marked Badiaranke sentences, consultants
often initially get a futurate reading.
These preferences are illustrated in (36)–(37). My consultant accepted (36)
with a habitual reading, namely in a context where the children have been, for the
last two years, repeatedly and regularly getting dirty.
(36) birĩ paːrə peːs bepoːse pãpã piːsidoː bẽ- dobbə de!
since two.years.ago children dem every.day 3pl.impf- dirty aff.decl
“Since two years ago, those children have gotten dirty every day.”
When I suggested the same sentence without the adverbials “since last year” and
“every day,” however, the consultant reported only a futurate reading.14 To express
the habitual reading of (36), he said he preferred (37), which uses the dedicated
habitual aspect:
(37) bepoːse pãpã kədə- bẽ dobbə de!

children dem habit- 3pl.habit dirty aff.decl
“These/those children get dirty!” (8.24, #2, elicitation)
Even in the presence of habitual adverbials, this preference for the less ambiguous
habitual construction remained (though the general imperfective version was ac-
ceptable as well).
Similar facts obtain for the imperfective’s progressive use in matrix clauses:
although consultants will sometimes approve it in elicitation, they rarely
produce it spontaneously as a translation. Without supplemental text data,
14
Since the two versions of (36) were elicited one after another, the consultant may have been
primed to find a contrast between them (Ryan Bochnak, personal communication). As pointed out by
(Matthewson 2004:405), there may be a tendency for consultants to latch onto a single reading once
they have identified it; thus a consultant’s judgment that a sentence is unambiguous cannot be taken
as definitive. Indeed, the emergence of the habitual reading in the presence of certain adverbials shows
that it is indeed available for this construction, as do examples from texts like (38b) below.
therefore, one would have to regard the elicitation data with skepticism: Do
consultants’ approval of progressive and habitual readings reflect actual usage,
or merely kindness toward the fieldworker? On the other hand, their prefer-
ence for producing the dedicated habitual and progressive constructions in
elicitation may or may not reflect a real preference in natural speech; it could
stem from the translation task, or even from the elicitation setting itself, since
consultants might be attuned to my interest in meaning differences between
different TAM categories.
Texts eliminate this translation confound. In texts, speakers are (ideally)
producing language the way they normally would when they produce that
genre of speech. Additionally, for any sentence of interest in a text, the context
tends to be rich enough that it is clear which imperfective reading is intended.
The linguist thus has clearer, more naturalistic evidence that the imperfective
can receive a certain reading in a certain type of context. (As discussed above,
elicitation can then help flesh out what about the context licenses this use of the
imperfective.)
The text-based examples in (38) provide reassuring confirmation that the
Badiaranke imperfective really does have all the functions that emerge through
elicitation. Examples (38a), (38b), and (38c) illustrate the habitual, progressive, and
futurate uses, respectively.
(38) a. mpə- ŋaːrəŋaːrən- ẽ jã de. . .

3sg.impf- knock.down.redup- 2sg.nsbj there aff.decl
“It [the jinn] knocks you down over and over until. . .”
Context: The speaker is describing various kinds of supernatural
entities. In this section, he has been describing one kind of spirit
(jinn) that has lived in his village for ages; here, he explains what the
troublemaking spirit does when it catches a human.
(9.1, #7, magic and the supernatural narrative)
b. maː tʃaːfe u- wak- a jak- ako- se jã. taːme

quot woman nmlz- run.away- irr be- past- rel.perf there now
mpə- raː ka- wak- e.
3sg.impf- go inf- run.away- inf
“It’s said there was a runaway woman. Now she was going to run away.”
Context: These were the first two sentences of a story. The first sentence
merely asserts the existence of the woman; the second states that at
the time the story begins, the woman is in the process of running
away. That she is in the process and not planning to run away becomes
unambiguously clear with the sentences that follow, which describe her
actions as she is running away (encountering a river, climbing on the
back of a crocodile to cross. . .). (6.45, #3, runaway woman story)
c. bõ- raː jã de ka- raː tʃad- i- e sadaː . . .

3pl.impf- go there aff.decl inf- go sacrifice- ben- inf sacrifice
“We’ll go there to go participate in the sacrifice.”
While the text primarily describes habitual behaviors of the Badiaranke
people, in this section the consultant is laying out his immediate plans
to visit the home of someone who has died in another village.
(2.60, #1, “what happens when someone dies” narrative)
Such examples show that the progressive, habitual, and futurate uses are all at-
tested in natural speech.
4.2.4 Texts Case Study #2: The Discontinuous Past Suffixes

As with the imperfective, texts helped me to establish the semantics and pragmat-
ics of Badiaranke’s two past-tense suffixes.
First, texts clarified the semantics of -ako- and -akəd- by revealing where and
why they appear in various kinds of discourse. (This advantage pertains to textual
data more generally—one would not obtain reliable results simply by asking a con-
sultant, “At what point in a conversation would you say a sentence with -ako- in
it?”) For -ako-, one such environment is in opening sentences of stories. A very
common opening sentence, the equivalent of English Once upon a time . . . , is
shown in (39):
(39) maː nda jak- ako- se jã.

quot thing be- past- rel.perf there.
“It’s said that once something happened.”
(6.1, #3, Daamaseree story) (and many other stories)
This usage shows that although -ako- encodes precedence with respect to a con-
textually salient time, that time need not be overtly expressed. In these story-
initial sentences of stories, the contextually salient time is simply the time of
utterance.
Equally importantly, texts revealed an environment where past-tense mark-
ing is not used in Badiaranke. While these introductory, presentational sentences
always contain past-tense marking (see also (38a) above), the subsequent sen-
tences do not—even when they describe states that obtained at the past time when
the story (ostensibly) took place, as in (40), which shows the first two sentences of
a story about a boy called Balã Tamba.15
15
In this example the second sentence happens to be stative, but the same generalization applies
to eventive sentences.
(40) maː nəse ujaːra jak- ako- se jã. nəse nə̃, nəse nə̃ mə-
quot child other be- past- rel.perf there child det child det narr-
niŋ- a- õ pəroke.
pleasant- detrans- pass.3sg.perf fishing
“It’s said there was a child there. The child, the child liked fishing.”
(6.59, #3, Balã Tamba story)
The past suffixes occur text-medially only when the speaker begins talking about a
time in the even more distant past, preceding the time she or he has been discuss-
ing at that point in the narrative. In (41), for instance, the crocodile character is
reprimanding his emissary, a bird, for having stuffed himself with food the woman
provided him on the way to the river, causing him to arrive late.
(41) maː dʒaːdʒa dʒõkə- mã wẽ wũ iː jak- akõ ka-

quot food give- 3sg.on.3sg. det dem cop be- past.3sg.perf inf-
dʒaː- e.
eat- inf
“He said the food she had given him, that’s what he had been eating.”
(6.57, #3, runaway woman story)
Such data provided evidence that -ako- and -akəd- express temporal precedence
with respect to the local perspective time (Cover 2010:180–187) rather than with
respect to the actual time of utterance.
4.3 Participant observation
The (dis)advantages of elicitation versus texts have been discussed relatively often
in the literature (e.g., Chelliah 2001; Mithun 2001; Matthewson 2004; Rice 2006).
A method that has received much less attention in the linguistic fieldwork litera-
ture is the third one I advocate here: participant observation.
4.3.1 What Is Participant Observation?

By participant observation, I mean observing and documenting constructions
of interest that are produced by native speakers during naturally occurring speech.
The linguist does not ask for or prompt these utterances, and might not have re-
cording equipment (or even a notebook) on hand. However, the linguist may well
be a part of the conversation she or he is observing—hence the label participant
observation. With this method, the traditional barrier between linguist-as-ob-
server and speaker-as-observee is less transparent (though the linguist is still the
observer, and the native speaker still the observee).
Participant observation may not be feasible or practical for all fieldworkers
or all fieldwork situations. To make it work, the linguist must learn the language,
ideally through a combination of elicitation and immersion (trying to speak the

language him- or herself as much as possible). Once the linguist has learned the
language well enough to pick up on what is being said around (and not just directly
to) him or her, the goal is to observe naturally occurring sentences illustrating the
categories of interest—as well as the contexts in which they were uttered—and
then write them down as soon as possible.
Beyond the difficulty of gaining adequate language proficiency, participant
observation raises several other potential obstacles. Naturally, it requires enough
speakers for conversation to occur. For an extremely moribund language, this
setup may be impractical. Another challenge is that often, by the nature of par-
ticipant observation, the linguist is not audio-recording the conversation she
or he is observing; instead, she or he must be able to remember and accurately
write down the sentence and context as soon as possible. Finally, ethical concerns
should always be kept in mind. Writing down anonymous linguistic data obtained
through observation of public behavior is generally unproblematic ethically.16
Nonetheless, use of such data should be avoided if there is any reason to believe
that the speaker(s) in question object to the work being done, or to documentation
of their utterances for posterity.
4.3.2 Why Do Participant Observation?

Like text collection, participant observation creates the possibility of serendipi-
tous discoveries, whose semantics and pragmatics can be probed in more depth
through elicitation. Much of the time, participant observation is simply a way to
extend and branch out from the kind of naturalistic data recorded in texts. None-
theless, it provides a valuable supplement to text collection, both because it brings
in a different genre of speech, and because new (or clearer) meanings of the cat-
egories might emerge. Moreover, text and elicitation data are both necessarily lim-
ited; but if the language is spoken on a daily basis in the fieldworker’s environment,
then naturally occurring discourse constitutes a rich, virtually endless source of
data that would be unfortunate to ignore. By using this third method, the linguist
can catch and write down instances of the phenomena in question that can either
confirm or challenge the working hypothesis.
4.3.3 Observation Case Study #1: The Imperfective

While it may be buried in an as-yet-untranscribed text, I have not noticed the
epistemic use of the Badiaranke imperfective in those texts I have transcribed.
Instead, I discovered this usage only by listening to the interlocutors in Badiaranke
conversations going on around (and with) me. One such epistemic imperfective in
a spontaneous utterance is shown in (42).
16
See, for instance, Wilson (2012) and UC Berkeley’s guidelines on exempt research at http://cphs.
berkeley.edu/exempt.pdf.
(42) kab- iː naddə- se jã te- riːt- ã.

know- 2sg.perf lay.down- rel.perf there rel.impf- remove- 3sg.nsbj
“You know whoever put it down there must have taken it.”
(10.32, spoken by my interlocutor, in response to my comment that I
didn’t know where a certain comb had come from.)
When I first observed such epistemic imperfectives in conversation, I was not

sure whether I had heard or interpreted them correctly, nor was I certain that the
imperfective had been used only for epistemic reasons. I therefore turned to elic-
itation to establish whether consultants would approve the imperfective for epis-
temic use. While ultimately successful, this confirmation took a certain amount
of experimentation. Epistemic imperfectives produced the same challenge as the
dedicated progressive and habitual aspects: consultants never spontaneously pro-
duced epistemic imperfectives as translations of Pulaar sentences, regardless of
how much context was provided. A different strategy turned out to work better:
carefully constructing contexts in which such a reading might be favored (includ-
ing specifying the epistemic state of the speaker regarding the proposition), then
constructing a sentence with imperfective marking but no other epistemic mark-
ing on it and asking about (a) whether the sentence was “healthy” (grammatical),
and (b) whether it could be said in that context (i.e., was felicitous and true).
To be sure that the imperfective could be used even in the absence of futurate,
progressive, and habitual semantics, I needed to ensure that these readings were
ruled out by the context. Examples of sentences that helped confirm the truly epis-
temic use of the imperfective are given in (43)–(44).
(43) A: e faː k- ə̃ Waalibo? ka- r- a fe paːdi-

p where be- 3sg.perf Waalibo be- neg.perf- 3sg.neg p room-
jã.
3sg.nsbj
“Where’s Waalibo? He’s not in his room.”
B: (mitʃ- ĩ de) fe bitiki mpi- jak.

think- 1sg.perf aff.decl p store 3sg.impf- be
“(I think) he’s probably in the store [now].” (8.9, #2, elicitation)
(44) katʃud- eː toː sẽ mpə- duːdə fe Amerik.

morning- of today det 3sg.impf- enter p America
“This morning she’ll likely have entered America.” (8.22, #2, elicitation)
In (43), B uses the imperfective in response to A’s question about where Waalibo
is at the moment—not in the future and not habitually. Similarly, (44), the ad-
verbial “this morning” delimits the time of the conjectured event; the consultant
specifically confirmed that the sentence could be spoken at noon, such that the
time being talked about is in the past. Thus in both cases, the imperfective is used
solely for its epistemic function.
4.3.4 Observation Case Study #2: The Discontinuous Past Suffixes

While not as striking as the discovery of an entire new function of the imper-
fective, the semantics of the past-tense suffixes also became clearer through par-
ticipant observation. I knew from elicitation that one common environment for
-akəd- is in counterfactuals—where the -akəd-marked eventuality is never actually
realized. Armed with this general knowledge, I came across, through participant
observation, a counterfactual class I had not thought to investigate—those with
verbless antecedents, roughly translatable as “If not for X, Y”. The sentence in ques-
tion, shown in (45), was produced as part of the speaker’s description of a time
when she was almost stung by a scorpion, but (fortunately) was not.
(45) nĩ kaː torsə sẽ, mpə- saf- akəd- ã

if neg.impf.3sg. flashlight det 3sg.impf- sting- past.irr.nperf- 1sg.nsbj
de.
aff.decl
“If it weren’t (for) the flashlight, it [the scorpion] would have stung me.”
(5.8, heard “in the wild”)
This type of counterfactual—later probed through elicitation, as well as texts—

provided additional evidence about the time with respect to which -akəd- ex-
presses precedence: this time need not be overtly stated, nor need it be marked
as the event time of another eventuality described in the sentence. It need only be
in the past of utterance time, a time at which the eventuality marked with mp- . . .
-akəd- was still a live possibility.
5 The Upshot
A mixture of data collection methods needs to be used to maximize the thorough-

ness, reliability, and accurate interpretation of data. At a minimum, both texts and
elicitation should be used extensively; to the extent possible, naturally occurring
data observed in discourse should be used to enrich these data sources and pro-
vide evidence in support of or against the linguist’s working hypotheses. Crucially,
none of these methods is independent from the others: to be maximally useful,
data collected via texts or through participant observation should be fleshed out
in elicitation, while elicited data may receive serendipitous confirmation—or un-
expected complications may be introduced—through the more natural, connected
speech in texts or discourse.
While this paper has focused on TAM, the methods I have advocated are
broadly applicable to many semantic domains (indeed, to non-semantic domains
as well). Both quality and quantity of data are critical for any fieldwork aimed
at obtaining a linguistic analysis; this need for high-volume, high-quality data is
especially evident when the category whose semantics is being investigated is as
subtle and complex as TAM, but no less pressing for other semantic topics.
With respect to elicitation in particular, I have argued that as long as it is
done well, elicitation can and should be used to work through the semantics of
TAM categories (in concert with the other methods, of course). The linguist must
take care not to rely too heavily on translation, and must carefully construct the
questions in such a way that the speaker reveals the answers to the often very
subtle questions being investigated. These questions can rarely or ever be asked
directly (one cannot ask, “What is this mp- morpheme and why can it be used for
statements about future as well as habitually recurring or in-progress events?”);
as a result, elicitation requires extra care in preparing questions that will help the
linguist piece together the answers through indirect clues. At the same time, rigid
adherence to pre-selected questions can be detrimental to the investigation. The
linguist must learn to be flexible and think of follow-up questions on the fly—a
skill that develops over time, with experience. Consequently, the early results of
TAM elicitation, while they should never be thrown out, may be less valuable,
detailed, and/or reliable than those collected after both fieldworker and consultant
have gained experience with this type of work.
References
AnderBois, Scott, and Robert Henderson. This volume. Linguistically establishing dis-
course context: Two case studies from Mayan languages.
Arregui, Ana, María Luisa Rivero, and Andrés Pablo Salanova. 2014. Cross-linguistic varia-
tion in imperfectivity. To appear in Natural Language and Linguistic Theory. 32(2):
307–362.
Bohnemeyer, Jürgen. This volume. A practical epistemology for semantic elicitation in the
field and elsewhere.
Burton, Strang, and Lisa Matthewson. This volume. Targeted construction storyboards in
semantic fieldwork.
Chelliah, Shobhana L. 2001. The role of text collection and elicitation in linguistic fieldwork.
In Linguistic Fieldwork, Paul Newman and Martha Ratliff (eds.), 153–165. Cambridge:
Cambridge University.
Cover, Rebecca T. 2010. Aspect, Modality, and Tense in Badiaranke. PhD diss., UC Berkeley.
Cover, Rebecca T. 2011. Modal aspects of Badiaranke aspect. Lingua 121: 1315–1343.
Cover, Rebecca, and Judith Tonhauser. This volume. Theories of meaning in the field: Tem-
poral and aspectual reference.
Dixon, R.M.W. 2007. Field linguistics: A minor manual. Sprachtypologie und Universalien-
forschung 60: 12–31.
Enç, Mürvet. 1987. Anchoring conditions for tense. Linguistic Inquiry 18: 633–657.
Hacquard, Valentine. 2006. Aspects of Modality. PhD diss., MIT.
Lewis, M. Paul. 2009. Ethnologue: Languages of the World. 16th ed. Dallas, TX: SIL
International.
Louie, Meagan. This volume. The problem with no-nonsense elicitation plans (for semantic
fieldwork).
Mithun, Marianne. 2001. Who shapes the record: The speaker and the linguist. In Linguis-
tic Fieldwork, Paul Newman and Martha Ratliff (eds.), 34–54. Cambridge: Cambridge
University Press.
Mosel, Ulrike. 2012. Morphosyntactic analysis in the field: A guide to the guides. In The
Oxford Handbook of Linguistic Fieldwork, Nicholas Thieberger (ed.), 72–89. New York:
Mous, Maarten, 2007. Language documentation as a challenge to description. Paper pre-
sented at the 38th Annual Conference on African Linguistics, Gainesville, FL, March
25, 2007. http://home.planet.nl/~gongg010/veldwerk/Language%20documentation%20
as%20a%20challenge_article.pdf.
Plungian, Vladimir A., and Johan van der Auwera. 2006. Towards a typology of discontinu-
ous past marking. Sprachtypologie und Universalienforschung 59: 317–349.
Portner, Paul. 2003. The (temporal) semantics and (modal) pragmatics of the perfect. Lin-
guistics and Philosophy 26: 459–510.
Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, MA: The MIT Press.
Rice, Keren. 2006. Let the language tell its story? The role of linguistic theory in writing
grammars. In Catching Language: The Standing Challenge of Grammar Writing, Felix
Ameka, Alan Dench, and Nicholas Evans (eds.), 235–268. Berlin: Mouton de Gruyter.
Stowell, Tim. 1995. What do the present and past tenses mean? In Temporal Reference,
Aspect, and Actionality, Vol. 1: Semantic and Syntactic Perspectives, Pier Marco Berti-
netto, Valentina Bianchi, James Higginbotham, and Mario Squartini (eds.), 381–396.
Torino: Rosenberg and Sellier.
Tonhauser, Judith. 2010. Is Paraguayan Guaraní a tenseless language? In Proceedings of Se-
mantics of Under-Represented Languages of the Americas (SULA) 5, Suzi Lima (ed.),
227–242. Amherst, MA: GLSA.
onomy of projective content. Language 89: 66–109.
Wilson, Aidan, June 12, 2012. Participant observation: A LIP discussion. http://www.
paradisec.org.au/blog/2012/06/participant-observation-a-lip-discussion/. Accessed
September 4, 2012.
10
Deriving Topic Effects in Kiowa With Semantics

and Pragmatics
Andrew McKenzie
1 Introduction
This chapter employs semantic fieldwork to offer a new perspective on dislocated

topical D(eterminer) P(hrase)s in the Kiowa language.* These topics are often dis-
located to the left periphery of the clause. For instance, in (1), the object báò, “(the)
cat,” is placed to the left of the subject chégùn, “(the) dog,” even though overt ob-
jects typically follow subjects.
(1) Báò, chégùn ā́lḗ.1
báo tségũn Ø–áːléː
cat dog 3sA:3sO–chase.pf
“The dog chased the cat.” (lit. “The cat, the dog chased (it)”)
Previous studies of Kiowa have been unable to discern a motivation for this
dislocation, since it lacks the clear signs of syntactic movement found in more com-
monly studied languages. These studies have also been unable to use the meaning
of topichood to motivate the dislocation in Kiowa, because it is very difficult to
*
The work conducted here was funded by an NSF Doctoral Dissertation Improvement Grant
(#BCS-0843901), while at the University of Massachusetts Amherst, and later during a post-doctoral
fellowship at the University of Texas at Arlington. Many thanks go to the Kiowa-speaker consultants,
especially Christina Simmons, George and Marjorie Tahbone, and all others whose help has been
crucial.
1
Examples are glossed as follows: The Kiowa examples are written in the Kiowa orthography
developed by Parker McKenzie (Meadows and McKenzie 2001, Watkins and Harbour 2010), and in
IPA. On IPA glosses, low tone is left unmarked. The following abbreviations will be used: A: agent,
d: dual, D: dative, DS: different subject/situation marking, evid: indirect evidential, hab: habitual, i:
inverse number, impf: imperfective aspect, inv: inverse number, mir: mirative, neg: negation, O: direct
object, pa: plural animate, pf: perfective aspect, pn: plural inanimate, px: plural exclusive, r: reflexive,
s: singular, S: intransitive subject.
All Kiowa examples are from my field notes, unless noted. 269

elicit for topics. This chapter will propose that we instead derive topichood from
dislocation, rather than the other way around. The hypothesis offered here is that
the dislocation occurs to remove a certain possible ambiguity arising from a DP
in a discourse-neutral position. The DP does not have any particular role as topic,
but a sort of discourse prominence emerges from the pragmatics— the DP did not
have to be dislocated, so a listener can interpret the change in position as a sign
of the DP’s importance, which appears to us as a kind of topichood. The hypoth-
esis has two advantages: first, it explains the data and makes fruitful predictions.
Second, it allows us to test topic effects using well-understood methods of seman-
tic and pragmatic fieldwork.
The rest of this chapter will lay out this hypothesis. The next section will dis-
cuss the relevant facts concerning dislocation, showing that a standard syntactic
account is untenable. Section 3 will discuss the way that DPs are interpreted rel-
ative to various intensional operators and discuss the hypothesis semi-formally.
Section 4 will offer a situation semantics account that formalizes the hypothesis,
and Section 5 rounds up the discussion with an outlook on cross-linguistic testing.
2 A Purely Syntactic Account Does Not Suffice

2.1 Kiowa word order
The language focus of this chapter is the Kiowa language, an endangered member
of the Kiowa-Tanoan family. It is currently spoken by a few dozen elder members
of the Kiowa tribe of Oklahoma. A few hundred others speak it as a heritage lan-
guage, out of a tribal population of over 12,000. Nascent efforts are underway to
revitalize the language among children, and it has long been taught as a second
language at a number of schools and colleges in the state, but its future as a living
language is still in doubt.
Kiowa has a neutral word order of S-IO-DO-V (2a, Watkins 1984:206). How-
ever, DPs can be dislocated (2b,2c) or dropped (2d) based on the discourse (also
see Watkins 1990).
(2) a. Bill chê hā́ugà. b. Chê Bill hā́ugà.
̂
bi l tse͂ː Ø–hɔ́ːɡjæ c. Bill hā́ugà chê.
d
B. horse 3sA:3sO–buy.pf d. Hā́ugà. “He bought it.”

“Bill bought the horse.”
The apparently free word order is not truly free. In neutral word order, DP place-
ment reflects argument structure, but otherwise, placement reflects the roles DPs
play in the discourse. The most common roles are topic and focus. Kiowa has
no morphemes dedicated to expressing these discourse roles, so dislocation (and
perhaps prosody) is often the only overt signal that these DPs bear these roles. For
instance, topical objects can be found to the left of the subject, like báò “cat” in
Deriving Topic Effects in Kiowa 271
(1) or chê “horse” in (2b). Also, backgrounded DPs can be placed in a post-verbal
position (2c).
2.2 The limits of a syntactic approach
In a book (Adger et al. 2009) and a paper (Harbour et al. 2012), the trio of David
Adger, Daniel Harbour, and Laurel Watkins (henceforth AHW) reveals many new
facts about word order in Kiowa. These facts will form the starting-off point for
this investigation. One fact (from AHW 2012) is that the dislocation does not seem
to be linked to specific syntactic positions. Rizzi (1997) discovered that dislocation
in Italian involved movement from an argument position to a specific functional
projection in the left periphery (TopP for topics, FocP for focus). His evidence was
the relative ordering of topical and focused DPs. The fixed projections provide a
location for dislocated DPs, and the Top° and Foc° heads bear syntactic features
that motivate the DPs’ movement. Rizzi’s methodology has been successfully ap-
plied to many different languages, but AHW find no reliable evidence of ordering
of the same sort in Kiowa.
The second fact comes from AHW (2009): Certain adverbials occupy fixed
positions in the clause. These “selective particles” are associated with functional
projections in the inflectional layer, and they select for particular values of verbal
inflection on the head of the respective projection. For instance, the negation par-
ticle hàun, “not,” is at NegP, and it must co-occur with verbal negation marking at
Neg° (3).2
(3) a. Hàun dé ā́umâu. ́ mḗ.
b. *Hàun dé āu
hɔ͂n dé–ɔ͂́m–ɔ̂ː ́ –éː
hɔ͂n dé–ɔ͂m
not 1sA:3iO–do–neg not 1sA:3iO–do–pf
“I didn’t do it.”
The generic/habitual particle àn is at AspP and selects for imperfective aspect

with eventive verbs (4).
(4) a. Àn dé ā́umā́u. b. *Àn dé ā́umḗ.
ãn dé–ɔ͂́m–ɔ͂́ː ãn dé–ɔ͂́m–e͂́ː
hab 1sA:3iO–do–impf hab 1sA:3iO–do–pf
“I do it/I often do it/I usually do it.”
2
The selection requirement is one-way. For instance, the imperfective occurs regularly without
àn. Negation has been observed without the negation particle, although so rarely that occurrences
might be speech errors. Also, the particle can appear without the selected morpheme if the paradigm
does not have a specific form. The habitual particle àn does not trigger imperfective marking on stative
verbs or negated verbs, both of which lack aspect marking (6a). These observations may suggest that
the selection has more to do with compatibility of truth-conditions than a simple syntactic relationship,
but pursuing this suggestion is beyond the scope of this chapter.
The somewhat mirative particle béthàu ([béthɔ]) is at EvidP and selects

for indirect evidential marking. It’s used to describe events that the speaker
had been unaware of. AHW gloss it as “unbeknown” in 2009 and as mir
in 2012. It’s usually translated as “I didn’t know (that P)” or “I didn’t realize
(that P).”
(5) a. Béthàu bé ā́umhêl. b. *Béthàu bé ā́umḗ.
béthɔ̀ bé–ɔ͂ḿ –hêl béthɔ̀ bé–ɔ͂ḿ –e͂́:
mir 2sA:3iO–do.pf–evid mir 2sA:3iO–do–pf
“I didn’t realize you did it.”
AHW (2009) demonstrate the location of the selective particles, observing
that they cannot be postverbal, and that when there is more than one, they appear
in a fixed order.
́ mâuhèl.
(6) a. Béthàu hàun àn bé āu
béthɔ̀ hɔ͂n ãn bé–ɔ͂́m–ɔ͂̂:–hêl
mir not hab 2sA:3iO–do–neg–evid
“I didn’t know you didn’t do it. / I didn’t know you never did it.”
b. *Béthàu àn háun bé ā́umâuhèl e. *Àn béthàu hàun bé ā́umâuhèl
c. *Hàun béthàu àn bé ā́umâuhèl f. *Àn hàun béthàu bé ā́umâuhèl
d. *Hàun àn béthàu bé ā́umâuhèl
AHW derive the particles’ fixed order by placing their projections in a fixed
order in the inflectional layer : Evid – Neg – Asp. This order is further demon-
strated by the fact that the verbal inflection in this head-final language mirrors the
order of the selective particles.3
(7) EvidP
béthàu
MIR
NegP Evid°
–hèl
hàun
not
AspP Neg°
–âu
àn
HAB vP Asp°
-
àum– —
¯ do
3
Aspectual marking is neutralized under negation, so it is missing from the tree structure shown.
Also, this tree differs a bit from AHW’s model, though not in any way significant to this argument.
Notably, they use left-headed trees, while (7) uses a right-headed one for clarity.
Table 10.1
Apparent clause domains in Kiowa
Preparticular Domain Particles Postparticular Domain Verb Postverbal Domain
The third fact that AHW discover is that the selective particles seem to delimit
domains in the structure that correspond to discourse roles. Adger et al. (2009:134)
describe three domains, listed in Table 10.1.
The key section of interest in this chapter is the preparticular domain. The
postverbal domain is used for backgrounding; topicalized, focused, and wh-DPs
are barred from this domain. The postparticular domain between the selective
particles and the verb is neutral with respect to discourse.
AHW (2012) point out that it is unlikely that the domains actually exist. In-
stead, the particles’ fixed position allows them to serve as a milestone for dislo-
cation, especially when there is only one overt DP. Furthermore, they find that
DPs in the preparticular domain can be interpreted differently than those in the
postparticular domain. In (8), the subject DP fā́, “some people” is in the scope of
the habitual particle àn. It is not interpreted as a topic.
(8) bôt àn [châu fā́ ét gìsā̀ujejàu.]

bôt ãn tsɔ̂ː páː ét–gı̃ː+sɔte+tɔ
because hab thus some 3iA:3pnO–night+work+act
“as some people are wont to work at night.” (Harbour et al. (2012):113)
The same goes for the embedded subject êlchòhyòp “old women” in (9); the
elderly women are brought up as part of the generic event, in a parenthetical con-
text. They are not mentioned again, so they are not topical.
(9) Yá háigádā́u châu àn [êlch0̀hyòp gà àunfṑgà ]

jæ̃́ –háigà+dɔː tsɔ̂ː ãn êltsõ-hyop gja–ɔ͂n+põː-gja
1sD:3pnO–know+be thus hab old woman-inv 3paD:3pnO–foot+sound-impf
tóòba gà dā́uḕ.
thóoba ɡja–dɔ̀ː=e͂ː
quiet 3p–be=when.DS
“I know how elderly women sound walking in moccasins when it is silent.”
(Adger et al. 2009:138)
In contrast, the subject in (10), chát “door”, is in the preparticular domain and
has topicality.
(10) . . . chát àn hábé pā́qùh� dáu dāu ́

h ́
tsát ãn hábé p ãː+k u=he͂ː
ʔ
dɔ́–dɔ́ː
. . . door hab sometimes lock+lie=without 1pxD:3sS–be
“Our doors4 were sometimes unlocked.” (Adger et al. 2009:138)
The fourth fact (AHW 2009/2012) is that dislocation of topics to the prepar-
ticular domain is preferred but not obligatory. Three pieces of evidence support
this fact: prominent DPs can be situated in the postparticular domain; a DP can
land between selective particles; and speakers prefer preparticular dislocation.
Apart from wh-words, no DP is obliged to be in the preparticular domain.
Thus, some DPs that are clearly the topic of discussion are located in the postpar-
ticular domain. These DPs cannot be postverbal.
(11) Qāh́ î gàu mā̀yí è hḗbà, . . .

kˀā́ːhıː ɡɔ maːyĩ́ e–héːba
man and woman 3d–enter.pf
“A man and a woman came in, . . .”
a. nàu héjáu háun mā̀yí kī́gâu.

nɔ͂ hétɔ́ hɔ͂́n maːyĩ́ Ø–khíːɡɔ̂ː
and:DS still not woman 3s–exit.neg
“. . . and the woman still hasn’t left.”
b. nàu mā̀yí héjáu háun kī ́gâu.

c. *nàu héjáu háun kī ́gâu mā̀yí. (Harbour et al. 2012:106)
Even some focused DPs are in the postparticular domain. For instance, an-
swers to wh-questions can be postparticular, even though the wh-word itself must
be dislocated. Focused DPs cannot be postverbal.
(12) a. Hâundé àn à fáutjàu? b. *Àn hâundé à fáutjàu?

hɔ̂ndé an a–pɔ́t–tɔ
what hab 2sA:3sO–eat–impf
“What do you usually eat?”
(13) a. Cī́ àn gàt fáutjàu. b. Àn cī́ gàt fáutjàu.

kíː an gjat–pɔ́t–tɔ c. *Àn gàt fáutjàu cī́.
meat hab 1sA:3pnO–eat–impf (Harbour et al. 2012:105)
“I usually eat meat.”
4
I cannot explain why the translation of “door” is plural, while the Kiowa form is singular.
Second, some DPs dislocate to a location between selective particles. For in-
stance, in (14), the DP câugàuàl qṓbáu, “other old people,” is placed between the
mirative béthàu at EvidP and the habitual àn at AspP.5 It is interpreted as a topic.
(14) Béthàu câugàuàl qṓbáu châu àn á hotgûyī ̀

béthɔ kɔ̂–ɡɔ=al kʔṍːbɔ̀ tsɔ̂ː an á–hotɡũ̂–yi
mir other-inv=too elder.inv thus hab 3pa–run around–impf.evid
“We didn’t realize that other old people ran around like that.”
(Harbour et al. 2012:67)
Third, although discourse-prominent DPs need not be dislocated to the pre-

particular domain, AHW report a clear speaker preference for doing so. This pref-
erence suggests that the dislocation is subject to a pragmatic effect.
AHW have greatly enhanced our understanding of Kiowa phrase structure.
This subsection has presented four of the facts they uncovered. The first is that
there is no clear evidence for a syntactic movement account following Rizzi (1997).
The second is the existence of fixed adverbials in the inflectional layer. The third is
AHW’s use of these adverbials as guideposts to delimit apparent domains, which
are broadly linked to discourse configuration. Finally, the dislocation of topics is
preferred but not obligatory. These facts demonstrate that there is not a one-to-one
link between domains and discourse roles, so the explanation for the dislocation
and domain effects must lie elsewhere.
2.3 A new direction of inquiry
While AHW (2012) rule out a purely syntactic motivation for these facts, they
leave the question open without rashly speculating a more definitive proposal.
They also do not attempt to base their inquiry on any particular notion of topic-
hood. Throughout, they rely on an intuitive sense of topic, noting the extreme
difficulty of attempting to elicit judgments based on accounts of topicality found
in the literature (Reinhart 1981; Gundel 1988; Frascarelli and Hinterhölzl 2007).
Given the numerous different types of topics, we might wonder if there is even
such a thing as a “topic” in a formal sense.
This chapter suggests taking a different tack on the matter. The use of direct
elicitation of topichood to explain dislocation makes a crucial assumption: that
the DP’s meaning is derived from its status as topic. But what if it’s the other way
around—what if the DP’s topic status is derived from its ordinary meaning? Taking
this approach would allow us to use well-tested techniques of semantic fieldwork
to determine the meaning of the DPs. Based on those meanings, we can then try
to derive the source of a sense of topicality.
5
The adverbial châu “thus” is not selective and its location is not fixed, as (8–9) demonstrate.
The hypothesis of this chapter is that Kiowa speakers dislocate these DPs
solely to disambiguate them.6
Specifically, the dislocation rules out one possible type of meaning, called
“opaque” (see the next section). Crucially, the dislocation is not itself tied to any
particular discourse role like topic. Any sense of topichood is pragmatic and
emerges from the prominence associated with the dislocation. Essentially, disloca-
tion signals to the listener that the DP is referential and prominent enough to be
targeted for disambiguation. The listener can conclude that if it’s prominent, it may
well be the topic.
The next section will lay out this hypothesis in a semi-formal fashion. The sec-
tion after that will work it out more thoroughly using situation semantics.
3 Transparency and Emergent Topicality

3.1 Transparent and Opaque Readings
The hypothesis relies on the distinction between transparent and opaque read-
ings of DPs. These readings are types of meaning that reflect the way a DP is inter-
preted, and often make sentences ambiguous. For instance, the proposition Alissa
is looking for a book is ambiguous. On the opaque reading, Alissa would be happy
to find any book. In fact, there might turn out not to be a book at all. On the trans-
parent reading, Alissa is looking for a particular book presumed to exist.7
One way to distinguish these readings is to complete them with different fol-
low-up sentences. Example (15a) is acceptable under the opaque reading, but not
the transparent one. Conversely, (15b) is acceptable under the transparent reading,
but not the opaque one, because it refers to a specific book. Also, (15b) is acceptable
if Alissa is looking through a full bookshelf, but (15a) is not.
(15) a. Alissa was looking for a book, but couldn’t find one.
b. Alissa was looking for a book, but couldn’t find it.
The opaque reading gets that name because the DP a book is interpreted relative
to the possible worlds or situations where the search is successful described by the
verb look for is successful. It is cut off from anaphoric reference, as if it were hidden
behind a screen. In the transparent reading, the DP is not interpreted relative to
look for, so it is visible to an antecedent and can refer to a particular book.
6
This account assumes, along with AHW, that dislocation of Kiowa DPs actually involves move-
ment from an A-position, based on the typical diagnostics. That said, in the semantics, moving the
DP would have the same meaning as base-generating it high and having it bind a silent pronoun in an
A-position. Thus, the actual mechanism of dislocation is not a concern in this chapter, so this chapter
will neutrally describe the change in placement as “dislocation.”
7
In formal terms, opaque DPs are intensional, while transparent ones are extensional. In the
philosophical and semantic literature, the readings are often called de dicto and de re, respectively.
The transparent/opaque distinction is most obvious with indefinite DPs, but it

also applies to definite ones as well. With definites, transparency distinguishes be-
tween anaphoric readings and intensional or bound variable readings. For instance,
the phrase Alissa thinks Bill is handsome is ambiguous. On the transparent reading,
the definite DP Bill denotes the Bill in the real world. On the opaque reading, Bill
denotes the guy who Alissa thinks is Bill, but who might be someone else. That is,
Bill here is interpreted with respect to Alissa’s beliefs, introduced by the verb think.
The transparency distinction applies to all intensional operators, including
those associated with aspect. And that’s where it can apply to this study. For in-
stance, in (8), the DP fā́ ([páː]), “some people,” is in the scope of the habitual op-
erator àn (marked in brackets). In this case it is interpreted opaquely (i.e., relative
to that operator). The speaker is describing a scenario where he is at the office late,
noting that there were usually people there after dark.
(8) bôt àn [ châu fā́ ét gìsā̀ujejàu. ]
because hab [ thus some Agr-work:at:night ]
“as some people are wont to work at night”
In semantic terms, the sentence can be paraphrased as “since in most situa-
tions (i.e., most nights), there are some people who are working.” Importantly, the
DP paraphrased as some people is interpreted relative to the operator paraphrased
as most situations. That is, for each situation there were some people, but it could
have been different people each time.
Bear in mind, however, that DPs can be transparent even in the scope of the in-
tensional operator. In Alissa thinks Bill is handsome, Bill is interpreted inside the em-
bedded clause whether it is transparent or opaque. We cannot rely solely on scope to
derive the distinction. If the DP in (8) had been interpreted transparently, the people
would have been the same ones, and the sentence would be ascribing the property
of working at night to them. The transparent reading could be paraphrased as “there
are some people who in most situations are working.” Outside of a particular con-
text, the sentence is ambiguous. Typically, though, a given context will eliminate one.
While the context will generally eliminate one reading, a more reliable way is
to place the DP outside the scope of the operator. In the scope of the operator the
DP can be either transparent or opaque, but outside the scope, only a transparent
reading is allowed (Table 10.2).
Table 10.2
Possible interpretations of DPs based on scope
DP outside operator scope DP inside scope acceptable
operator transparent yes

operator opaque yes
transparent operator yes
opaque operator no
In English this effect can be forced by relativization: in the sentence There’s a book
that Alissa is looking for, the DP a book can only be interpreted transparently rela-
tive to look for. In Kiowa, simple dislocation suffices. In (10), the DP chát ([tsát]) is
not interpreted with respect to the habitual situation (the usual night). Instead, the
doors referred to are always the same ones.
(10) . . . chát àn [hábé pā́qùhè dáu dā́u.]

. . . door hab [sometimes lock+lie=without Agr–be]
“Our doors were sometimes unlocked.”
3.2 Emergent topicality
To summarize: Transparent DPs refer, while opaque DPs do not. This distinction
will help provide a motivation for the interpretation of transparent DPs as topics,
because the nebulous sense of topicality emerges from well-understood notions of
semantics and pragmatics. Dislocation has two effects:
1. It rules out an opaque reading, forcing a transparent or referential one.
The speaker is making sure that the DP is not interpreted opaquely.
2. Since the transparent DP could have been left in the postparticular
domain, its dislocation signals its prominence. The listener can infer that
the DP plays an important role in the discourse. In the absence of any
focus interpretation, the listener can then infer that the DP is likely to be
topical in some way, even if it carries no actual semantics of topichood.
Not only does this hypothesis explain the apparent topicality of dislocated
Kiowa DPs, it has the advantage of tying in to observations linking transparency
and topichood. Cross-linguistically, opaque DPs make poor topics. Under an ac-
count where topichood drives dislocation, this generalization needs additional
motivation. However, on the present account, nothing else is needed to prevent
opaque topics, because the apparent topicality of the DP emerges from its trans-
parent semantics, not the other way around.
This result leads to some other predictions, which can be tested with elicita-
tion. For one, dislocated DPs should not have quantificational variability effects,
as in (8), in which you get different people for each situation. Preliminary testing
indicates this is the case. For instance, in (16), speakers only get a transparent read-
ing for “child,” because it is not in the scope of the generic operator.
(16) Sân àn án cṑdép
ʃán an án–koːdép
child hab 3sD:3pnS–be bored.impf
[expected]: “A (specific) kid usually gets bored/There’s a kid who gets bored.”
If the DP is in the scope of the operator, an opaque reading is possible, as (17)

shows. In (17b), the meaning is paraphrased as “in most (classroom) situations, the
students (in those situations) get bored.” With this context, the speaker accepted
(17b) and disapproved of the dislocated DP in (17a).
Context:
A friend of yours is teaching Kiowa to some children. However, he’s not a very
exciting teacher, so oftentimes, the students stop paying attention.
(17) a. # Sân àn án cṑdép.

b. Àn sân án cṑdép.
So far, this account has only discussed the generic operator, introduced by the
habitual particle àn, but the hypothesis ought to apply to the particles that select
for negation, modality, and evidentiality. This warrants further testing, along the
same lines, but we already can see how it works by recalling example (11), repeated
below.
The lead-in sentence in (11) introduces a man and a woman, who serve as
topic for the follow-up. In the follow-up, the DP mā̀yī́ is contrasted against the
man. Speakers produced “the woman” in a preparticular domain, accepted it in a
postparticular one, and rejected it in a postverbal one.
(11) Qā́hî gàu māỳ ī́ è hḗbà, . . .
kʔjã́ːhi ː͂̂ ɡɔ maːyĩ́ː ẽ–héːba
man and woman 3d–enter.pf
“A man and a woman came in, . . .”
a. . . . nàu mā̀yī-́ héjáu háun kī́gâu.
nɔ͂ maːyĩ́ː hétɔ̀ hɔ͂́n Ø–khíː–ɡɔ̂ː
and.DS woman still not 3s–exit.neg
“and the woman still hasn’t left.”
b. . . . nàu héjáu háun mā̀yī ́ kī́gâu.

c. . . . *nàu héjáu háun kī ́gâu mā̀y–ī́ (AHW 2012:106)
The dislocation pattern reflects the pragmatic value of disambiguation—
moving “the woman” out of the scope of negation makes it impossible for it to be
interpreted relative to the negative (event) operator. Now, it may be the case that
this DP is a contrastive topic, which would bring particular meaning to the sen-
tence (Büring 1997). Even so, the topicality would not trigger dislocation. Instead,
it is the value the speaker gives to disambiguation that triggers it.
Besides explaining why DPs dislocate to the preparticular domain, this hy-
pothesis also explains why DPs can be placed between selective particles (14).
Again, the explanation is derived from the ordinary semantics of the DP and prag-
matics. The dislocation only needs to place the DP outside the scope of a particular
operator, relative to which the speaker finds it important to highlight the transpar-
ency of the DP. Standard notions of economy would preclude dislocating the DP
any further.
To summarize, these “topical” DPs are not inherently topical at all. They are
dislocated to highlight their transparent reading, and the dislocation lets the lis-
tener infer the DPs’ discourse prominence. This hypothesis also explains the facts
detailed in the previous section. There is no clear target for the dislocation because
the dislocation only needs to put the DP out of the scope of a particular operator.
The adverbials’ intensional operators set apparent domains through their scope-
bearing properties. The dislocation is preferred to leaving the DP in the postpar-
ticular domain because it eliminates a potential ambiguity. In addition, it accounts
for dislocation between particles, and it allows us to test for “topics” using ordinary
semantic fieldwork techniques.
4 Formal implementation
This section will work out the hypothesis and demonstrate the value of a particular
theory in developing semantic fieldwork. The formal framework assumed here is
a possibilistic situation semantics (Kratzer 1989, 2007). This is a semantics built
upon the well-known compositional semantics of Heim and Kratzer (1998), with
the addition of a class of semantic objects called situations. Readers familiar with
this model can skip ahead to section 4.3.
4.1 Situation semantics and contextual restriction
A situation is a part of a possible world.8 Situations can correspond to a wide va-

riety of events, spatiotemporal locations, and even individuals. They include the
room you are in right now, your stomach, the sack of Rome in 410, the 2012 Olym-
pics, and so on. Moreover, any two situations can be summed to form larger situ-
ations.9 As a result, most situations are composed of parts that are also situations.
Situations are used to restrict the evaluation of the truth of an utterance. For
instance, if you say It rained, you are not describing an entire world, but just a part
of it. That part of a world is a situation.10 The truth-value of the uttered proposition
is evaluated solely against that situation. If you assert It rained about the situation
corresponding to Montreal last night, the proposition is true if and only if it rained
in Montreal last night. The weather anywhere else last night, or in Montreal at any
other time, is irrelevant to the judgment.
8
Not necessarily a proper part, though. We can define a possible world as a situation that is not
part of any other situation. This allows us to maintain the semantic effects of possible worlds while
removing worlds from the ontology.
9
Logically speaking this is the case, although pragmatics can restrict the types of summing used
in actual natural language (von Fintel 2005).
10
Situations were first proposed by Barwise and Perry (1983), however, they did not characterize
situations as parts of a possible world, but rather as a set of properties of states of affairs.
4.2 Situations in the grammar
Situations are referred to by pronouns. We use the indexed variable si to denote

situations (i is any number), and they are of semantic type s. Situation pronouns
behave like any other pronouns—they can be anaphoric or bound, and they are
present in the syntax, even though they are silent (Cresswell 1990, Percus 2000).
The use of situations can be divided into two types: the first is the topic situation,
used with sentences (as above). The second use, the resource situation, is more
important for this chapter.
A resource situation is a pronoun inside a nominal expression that restricts
the domain of the determiner or quantifier. For instance, if you are in a café with
a friend, and you say “Everyone is looking at their cell phone,” you obviously don’t
intend for everyone to mean “everyone in the world.” Instead, you are referring to
some particular part of the world— a situation.11
Resource situations occur in the grammar as a silent pronoun inside the DP.
The DP [ Everyone (s3) ] picks out “everyone in s3”:
(18) [ Everyone (s3) ] is looking at their cell phone. →
∀x[ (x is a person & x is in s3) → x is looking at x’s cell phone ]
Resource situations also occur with definite determiners, preserving a presup-
position of uniqueness by restricting a determiner’s domain to a part of the world
with a unique object bearing the relevant property. For instance, in the sentence
The dog is chasing cars again, the DP the dog does not presuppose that there is only
one dog in the world. Instead, it contains a resource situation (let’s say s4) that
refers to a part of the world where there is a unique dog. If s4 refers to your family,
then [ the dog (s4) ] picks out the family dog.12
The above examples have anaphoric resource situations, but like any pro-
noun, resource situations can be bound by an operator. Through the binding of its
resource situation, a DP can be interpreted relative to an operator. For instance,
imagine you have a friend who is obsessed with the radio station at 97.5 FM. She
tells you:
(19) Every time I get in a car, I turn the radio to 97.5, baby!
11
As (Kratzer 2007) points out, there are surely other effects of domain restriction as well, such
as the exclusion of speech-act participants. The sentence here could be true even if you are in the café
and not looking at your cell phone. Whether effects like this are pragmatic or situation-based has yet
to be determined.
12
In theory, one could abuse this feature of resource situations to wreak pragmatic havoc. For
obvious reasons, we generally don’t, but we could. For instance, if you and a friend entered a library
full of desks, and you told your friend Sit at the desk, the friend would be puzzled because the context
would supply the library as the resource situation of the desk. Your request would only be felicitous if
there were a unique desk in the library. In this context that is a clearly false presupposition. Meanwhile,
you reply that you meant the desk in this particular spot you had in mind, having reduced the resource
situation to a spot where that desk was unique. It’s doubtful your friend would be amused.
The DP the radio doesn’t refer to a unique radio in theworld, or even a single
part of the world. It exhibitsquantificational variability: the radio could be differ-
ent everytime your friend gets in a different car. We can capture thisvariability by
binding the resource situation of the radio. For instance, assuming that every time
universally quantifies over situations, the sentence has the following meaning,
where the situation pronouns bound by the universal quantifier are in boldface:
∀s[ (s [ I get in a car ]) → ∃s′[ s′[ I turn on [ the radio (s) ] to 97.5 ] ] ]
(19′)
(19′) can be paraphrased: For all situations s such that s is a situation of me
getting in a car, there is a situation s′ of me turning on [ the unique radio in s ]
to 97.5.13
4.3 Formalizing the hypothesis
Given the use of resource situations, and the availability of binding them, we
can now formalize the way that dislocation rules out an opaque reading. Semi-
formally, an opaque DP is interpreted relative to some operator. We cannot derive
this interpretation by mere scope, because transparent DPs can be in the scope of
the operator. We can derive it with resource situations: A DP is opaque relative to
whichever operator binds its resource situation (20a). A transparent DP’s resource
situation is not bound by that operator (20b), whether it is in the scope of that
operator or not.
(20) a. opaque: [ OPs [ the DP (s) ] sentence ]
b. transparent: [ OPs [ the DP (s1) ] sentence ]
The use of resource situations can be applied to all the examples we have seen
so far, and also provides an explanation for an apparent counterexample. In the
following translation task, the speaker was given the lead-in sentence in (21a), and
tasked with translating the sentence in (21b).
Context: Two men are talking about a farm. One indicates the barren fields:
(21) a. [Speaker A:]
Óp háun étjé dáumgà gà dā́umâu.
óp hɔ͂́n éʔté dɔ͂́m-gja gja–dɔ͂́ː–mɔ͂̂ː
there not much earth-on 3pn–be–neg
“There isn’t much on the ground there.”
13
The expression in (19′) is simplified for exposition. Both the meaning and the paraphrase would
need to contain a matching function (Rothstein 1995) to ensure a one-to-one correspondence between
the situation of the restrictor and that of the nuclear scope, and the restrictor would require an expres-
sion of the minimality of the getting-into-a-car situation, to avoid also counting every situation that
includes it. (Heim 1990, Kratzer 2007)
b. [Speaker B:]
Chêgàu són àn ét cṓdófàutjàu.
tse͂̂:–ɡɔ sṍn ãn ét–koːdó+pɔʔ–tɔ
horse–inv grass hab 3iA:3pnO–much+eat.impf
“The horses eat a lot of grass.”
The follow-up clause (21a) introduces two dislocated DPs, neither of which
had been mentioned in the context. Thus, they should not be topics, and they were
clearly not focused. So why were they dislocated? An account based on discourse
role cannot explain the dislocation, but this account based on situation semantics
gives a clear explanation. The context makes it clear that the farm is under discus-
sion. The farm is referred to by a situation pronoun, let’s call it s7. The two DPs both
have the same resource situation: s7. Thus, (21b) can be translated as “[the horses
(s7)] are often eating [ the grass (s7)].” That is, the horses on the farm are eating the
grass that’s on the farm. The double dislocation highlights each DP’s transparent
reading, and crucially rules out any opaque reading relative to the habitual. Thus,
it rules out readings like “there are often horses eating a lot of grass.”
The hypothesis was further tested to see if the DPs could be left in a post-
particular domain. As predicted by the hypothesis, the speaker accepted one or
both DPs in the postparticular domain, but preferred dislocating them both, even
though neither is a topic.
The use of resource situations in opacity and the link to dislocation also leads
to an interesting prediction: donkey anaphors in Kiowa should not be able to
dislocate. Donkey anaphors have long posed problems to accounts of anaphora
reliant on c-command. Elbourne (2005) provides a solution based on resource
situations; essentially, the donkey anaphor is a covert definite description whose
resource situation is bound by the operator over situations.
(22) When a man has a donkey, he beats it
∀s[ s [ a man has a donkey ] → ∃s′[ s′ [ the man (s) ] beats [ the donkey
(s) ] ] ]
= “For all situations s such that a man has a donkey, there is a situation s′
such that the man in s beating the donkey in s.”14
A donkey anaphor is always interpreted relative to an aspectual operator;
thus, it cannot be dislocated outside of its scope. This prediction was tested with a
follow-up sentence, in which the DP was accepted in the postparticular domain,
but not the preparticular domain.
(23) Qā́hī̀ átháukàui án dā́uchḕ, . . .
kʔja͂́hĩː átʔɔ́kɔj án–dɔ́ː=tsẽː
man donkey 3sD:3sS–be=when.SS, . . .
“When a man has a donkey . . .”
14
This expression contains the same simplifications as (19′).
a. àn qāˊhī-́ gûgū̀
ʔj ́
an k a͂hĩː Ø–ɡûː–ɡu
hab man 3s–hit–impf
“the man hits it.”
b. *qāˊhī-́ àn gûgū̀
A full DP was used in (23) rather than a pronoun, because Kiowa lacks third-
person personal pronouns. Early attempts at elicitation triggered disapproval of the
donkey anaphor in the preparticular domain, as predicted. However, the speaker
hesitated to accept the postparticular domain. The confused judgment is likely due
to pragmatics. In addition to the lack of third-person personal pronouns, Kiowa
does not permit repeating full DPs very often (Watkins 1990). As a result, it may
simply be the case that any overt donkey anaphor is bad. Further testing is re-
quired, perhaps with clearer contexts, or speakers more willing to accept slightly
awkward expressions, but current indications are promising.
4.4 Summary
This section has presented a formalized account of the hypothesis, based on situa-
tion semantics. The transparency of the DP is actually determined by the transpar-
ency of its resource situation. Dislocation removes any possible opaque readings,
by taking the resource situation out of the scope of the operator. It thus highlights
the anaphoric nature of the resource situation, which in turn provides a referential
meaning for the DP.
5 Conclusion
This chapter has offered and tested a hypothesis about dislocated topical DPs
in Kiowa: that these DPs, thought to be dislocated because they are interpreted
as topics, are in fact interpreted as topics because they have been dislocated for
an independent pragmatic reason. The speaker wishes to disambiguate the DP
and highlight its transparency relative to some intensional operator. Disloca-
tion is not necessary for disambiguation, so the mere fact of dislocation allows
the listener to infer a discourse prominence for the DP. Instead of proposing
that the semantics of the DP is affected by the process, we can see that the
semantics of the DP is contributing to triggering the process. Thus, simple se-
mantic fieldwork can verify the semantics, and provide a plausible motivation
for the dislocation.
The hypothesis still requires more testing, but its promise has important
cross-linguistic implications. Many studies of dislocated DPs with apparent topi-
chood, or of scrambling, are at a loss to explain the motivation for the disloca-
tion, because of the difficulty in discovering what semantic or pragmatic effects
upon the DP result from topichood. This hypothesis provides a way out of that
labyrinth by making the semantics and pragmatics the motivation for the disloca-
tion, which leads to the topical interpretation. It explains the data, makes novel
confirmed predictions, and does not require the formulation of complicated no-
tions of topic or information structure. Lastly, it demonstrates the importance
and usefulness of semantic fieldwork even in areas traditionally thought to be out
of its reach.
References
Adger, David, Daniel Harbour, and Laurel Watkins. 2009. Mirrors and Microparameters:
Phrase Structure Beyond Free word Order. Cambridge University Press.
Barwise, Jon, and John, Perry. 1983. Situations and Attitudes. Stanford, CA: CSLI.
Büring, Daniel. 1997. The Meaning of Topic and Focus — the 59th Street Bridge Accent.
London: Routledge.
Cresswell, M. J. 1990. Entities and Indices. Dordrecht: Kluwer.
Elbourne, Paul 2005. Situations and Individuals. Cambridge, MA: MIT Press.
von Fintel , Kai. 2005. How to count situations (notes towards a user’s manual). Unpub-
lished manuscript, MIT. http://web.mit.edu/fintel/fintel-2005-counting.pdf.
Frascarelli, Mara, and Roland Hinterhölzl. 2007. Types of topics in German and Italian.
In On Information Structure, Meaning, and Form, K. Schwabe and S. Winkler (eds.),
87–116. Amsterdam/Philadelphia: John Benjamins.
Gundel, Jeannette K. 1988. The Role of Topic and Comment in Linguistic Theory. New York:
Garland.
Harbour, Daniel Laurel Watkins, and David, Adger. 2012. Information structure, discourse
structure, and noun phrase position in Kiowa. International Journal of American Lin-
guistics 78(1): 97–126.
Heim, Irene. 1990. E-type pronouns and donkey anaphora. Linguistics and Philosophy 13:
137–177.
Heim, Irene, and Angelika Kratzer. 1998. Semantics in Generative Grammar. Malden, MA:
Blackwell.
Kratzer, Angelika. 1989. An investigation of lumps of thought. Linguistics and Philosophy
12: 607–653.
Kratzer, Angelika. 2007. Situations in natural language semantics. In Stanford Encyclope-
dia of Philosophy, E. N. Zalta (ed.). Palo Alto, CA: CSLI. http://plato.stanford.edu/
archives/spr2014/.
Meadows, William, and Parker McKenzie. 2001. The Parker P. McKenzie Kiowa orthogra-
phy: How written Kiowa came into being. Plains Anthropologist 46: 233–248.
Percus, Orin. 2000. Constraints on some other variables in syntax. Natural Language Se-
mantics 8: 173–229.
Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosoph-
ica 27: 53–94.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of Grammar, Liliane
Haegemann (ed.), 281–337. Dordrecht: Kluwer.
Rothstein, Susan. 1995. Adverbial quantification over events. Natural Language Semantics
3: 1–32.
Watkins, Laurel. 1984. A Grammar of Kiowa. Lincoln: University of Nebraska Press.
Watkins, Laurel, and Daniel, Harbour. 2010. The linguistic genius of Parker McKenzie’s
Kiowa Alphabet. International Journal of American Linguistics 76(3): 309–333.
Watkins, Laurel J. 1990. Noun phrase versus zero in Kiowa discourse. International Journal
11
Reciprocity in Fieldwork and Theory

Sarah E. Murray
1 Introduction
Since 2006, during the summers I have worked with members of the Cheyenne
community on the Northern Cheyenne Indian Reservation in Montana.* We have
worked together on a variety of Cheyenne language projects, including several
semantics projects, with topics ranging from reflexivity and reciprocity to eviden-
tials and the marking of illocutionary mood. In this chapter, I discuss several of the
methods that I use in my semantic fieldwork, from eliciting judgments of felicity
and truth based on explicit contexts from existing texts, constructed discourses,
or images, to observing language use and attempting to learn the language myself.
In particular, I discuss an example from my fieldwork on reflexives and re-
ciprocals in Cheyenne. Cheyenne uses a single verbal suffix, -ahte, to express both
reflexivity and reciprocity, as in (1).
(1) Ka'ėškóne-ho é-vóom-ȧhtse-o'o.

child-pl.an 3-see-ahte-3pl.an
“The children saw each other.” / “The children saw themselves.”
The suffix -ahte is compatible with both plural subjects and singular sub-
jects, and further additions to the sentence can specify a reflexive or a reciprocal
*I would like to thank my Cheyenne consultants for their collaboration on and discussion of the
Cheyenne data. For detailed feedback, my thanks go to Ryan Bochnak, Lisa Matthewson, an anony-
mous reviewer, and the participants in the Workshop on Methodologies in Semantic Fieldwork at the
LSA Portland (January 2012). I am also grateful to Maria Bittner, Adrian Brasoveanu, Wayne Leman,
Richard Littlebear, Roger Schwarzschild, William Starr, Matthew Stone, Judith Tonhauser, and audi-
ences at Rutgers, Sinn und Bedeutung 12 (2007), and the Sixteenth Amsterdam Colloquium (2007) for
their comments and suggestions. This research was funded in part by a Phillips Fund Grant for Native
American Research from the American Philosophical Society and by a linguistic fieldwork grant from
the Endangered Language Fund. Any errors are my own. 287

donna donna donna
bobby laura bobby laura bobby laura
(a) Reflexive Scenario (b) (Weak) Reciprocal Scenario (c) Mixed Scenario
Figure 11.1 Simple Reflexive, Reciprocal, and Mixed Scenarios
interpretation. Given this data from Cheyenne, I developed an analysis of -ahte

where this marker is underspecified, not ambiguous (Murray 2007, 2008). This
formal analysis made strong predictions about other possible interpretations of
Cheyenne sentences with -ahte. Specifically, the analysis predicted that for subjects
involving three or more individuals, sentences with -ahte will allow a mixed con-
strual: one that is partly reflexive and partly reciprocal, as in Figure 11.1(c).
At the time of developing this analysis, I did not know if this was true or false
for Cheyenne, and it had not been discussed for other languages with “ambiguous”
markers of reflexivity and reciprocity. To test these predictions of the analysis, I de-
veloped various tasks for subsequent fieldwork. They involved Cheyenne sentences
with -ahte in a variety of scenarios, including mixed scenarios. Results from these
tasks show that mixed construals are available in Cheyenne. This provides strong
evidence that Cheyenne -ahte is underspecified for reflexivity and reciprocity, not
ambiguous. In fact, this turns out to be a robust pattern cross-linguistically for
what had been thought to be ambiguous forms (Murray 2007, 2008; Cable 2014).
Without the strong predictions from the formal analysis, I most likely would never
have asked about these possible interpretations.
In addition to discussing this example from my own fieldwork, this chapter
has two general goals. First, to illustrate that formal semantic training is valuable
training for fieldwork. Formal analyses, and their predictions, are one useful tool
for fieldworkers to be equipped with. In addition, semantic training can help in
documenting understudied languages, as the systematic investigation of the se-
mantic properties of a language should be part of the complete description of the
language. Second, I want to emphasize that there can be a reciprocal, mutually
beneficial relationship between fieldwork and theory (see also Rice 2006; Cover
and Tonhauser this volume; Gillon this volume; McKenzie this volume, among
others). Empirical findings from fieldwork inform our theories of natural language
semantics, and formally precise analyses can inform our fieldwork, making pre-
dictions that need to be tested in the field and providing novel questions to ask.
The next section of this chapter gives a brief background on the Cheyenne
language and the methodology used in the fieldwork discussed in this chapter. Sec-
tion 3 describes the marking of reciprocity and reflexivity in Cheyenne, contrasting
it with English. The underspecification analysis is presented in Section 4: I briefly
introduce the framework and then discuss the analysis of Cheyenne in detail, to
illustrate how the mixed construal was predicted. Section 5 discusses the ways that
these predictions were tested as well as the results. Section 6 is the conclusion.
Reciprocity in Fieldwork and Theory 289
2 Background on Cheyenne and Methodology
Cheyenne is a Plains Algonquian language spoken in Montana and Oklahoma.

It is an endangered language, with around 1,000 native speakers of Cheyenne in
Montana, most of whom are over 50 (Fisher et al. 2006). Though most children are
not learning the language at home, there are several ongoing language revitaliza-
tion efforts, including language classes in the local schools and summer language
camps for kids.
The data presented in this chapter are primarily from my own fieldwork,
building on a Cheyenne grammar (Leman 2011), texts (Leman 1980a, Leman
1987), a dictionary (Fisher et al. 2006), and a study of Cheyenne word order
(Leman 1999). In my fieldwork, I use a variety of methods, drawing on Newman
and Ratliff (2001) and Matthewson (2004), among others. These methods include
observation of language use and textual studies, including reading, reformatting,
glossing, and (re)translating texts.1 Much of my fieldwork involves direct elicita-
tions, often based on modified texts or constructed mini-discourses and stories.
A particularly useful strategy, I have found, is to build on naturally occurring ex-
amples by modifying the context or the target sentence, and then eliciting judg-
ments of felicity and truth. This can be useful because texts provide grammatical,
felicitous examples in a context, which can be a practical place to start exploring
a phenomenon. I am also learning the language, through my own study and lan-
guage classes, as well as volunteering at a summer language immersion camp for
kids. Learning the language has of course helped me in constructing examples, but
also in understanding the interaction of various components of the language and
the way the language is used naturally.
In elicitation sessions, I work individually with several consultants, all of whom
are native speakers of Cheyenne and English. I work on the same material with sev-
eral different people, presenting tasks in different orders, separated by other mate-
rial. When possible, tasks are confirmed with each consultant after an interval of
at least several days. Elicitation tasks include acceptability judgments about, and
corrections of, various kinds of examples in a given context as well as thinking of
contexts in which certain examples could be used. I use both naturally occurring
and constructed examples in constructed contexts or (modified) textual contexts.
In this chapter, I primarily discuss direct elicitations involving truth-value
judgments or felicity judgments of constructed examples in constructed contexts.
However, the constructed examples and contexts draw on uses found in Chey-
enne texts and the dictionary. In truth-value judgment tasks, I ask my consul-
tants if a sentence is true in an explicit context, provided in English or Cheyenne,
1
Many existing Cheyenne texts are translated on a word-for-word basis with the English directly
below the Cheyenne (e.g., Leman, 1987). In collaboration with language teachers, I have been reformat-
ting texts by separating the Cheyenne and English, formatting the Cheyenne into natural paragraphs,
and giving free translations in addition to interlinear glosses at the end.
sometimes supplemented by an image or drawing. For felicity judgment tasks, I

construct short discourses or dialogues in Cheyenne and ask if they are acceptable,
either in a context or out of the blue. Typically, sentences that are acceptable in a
context are both felicitous and true. However, sentences may be unacceptable in
a context because they are infelicitous or because they are false. These are impor-
tantly different, and truth-value judgment tasks can help tease them apart. Tasks
used in semantic fieldwork draw on those used in language acquisition research
(see, e.g., Crain and Thornton 1998; Eisele and Lust 1996). The use of these types of
tasks in semantic fieldwork is discussed in detail in Matthewson (2004).
3 Cheyenne Reflexives and Reciprocals
In some languages, including English, reflexivity and reciprocity are expressed by

means of distinct forms: themselves is a reflexive construction, as in (2), while each
other is reciprocal, as in (3).
(2) The children saw themselves.
(3) The children saw each other.
The English reflexive sentence (2) is true if each of the children saw himself/herself.
The English reciprocal sentence (3) is true if each child saw at least one other child,
and each child was seen by at least one other child. This reciprocal interpretation
is called weak reciprocity.2 For example, assume that there are three children, one
boy (bobby) and two girls (laura and donna). Figure 11.2 below depicts possible
reflexive and reciprocal scenarios for these three children, with lines indicating
what relations hold between the children.
The English reflexive sentence (2) is true in the reflexive scenario, Figure
11.2(a), but not in the reciprocal scenario, Figure 11.2(b). The English reciprocal
sentence (3) is true in the reciprocal scenario, Figure 11.2(b), but not in the reflexive
scenario, Figure 11.2(a). That is, the English constructions in sentences (2) and (3)
are specified for a reflexive meaning or a reciprocal meaning, respectively, and do
not overlap.3 However, the meanings do have something important in common:
donna donna
bobby laura bobby laura
(a) Reflexive Scenario (b) (Weak) Reciprocal Scenario

Figure 11.2 Simple Reflexive and Reciprocal Scenarios
2
Other reciprocal interpretations are possible for (3), including strong reciprocity, where each
child saw every other child (see, e.g., Langendoen 1978; Dalrymple et al. 1998).
3
Though there can be a collective interpretation of plural reflexives (e.g., The students helped
themselves, with the interpretation that the students as a group act to help that group).
they both require that the subject set (who performs the action, where the arrow
originates from) and the object set (who undergoes the action, where the arrow
points) be the same set. For example, in both scenarios in Figure 11.2, each child
sees some child, and each child is seen.
In contrast to English, many languages, including Spanish, French, and
German, use a single form to express both reflexivity and reciprocity (Langen-
doen and Magloire 2003; Maslova 2008; among others). Cheyenne is another such
language. The verbal suffix -ahte (phonologically realized as -ahtse or -estse) can
express both reflexivity and reciprocity (Leman 2011; Murray 2007, 2008). That is,
sentences with this suffix can be true in reflexive scenarios (e.g., Figure 11.2(a)) as
well as in reciprocal scenarios (e.g., Figure 11.2(b)). For example, Cheyenne (4),
which has a plural subject, allows both a reflexive interpretation, translated as
English (5), and a reciprocal interpretation, translated as English (6).
(4) [Several children were playing in the woods and got into some poison ivy.
Not long after, they were covered in itchy bumps.]
Ka'ėškóne-ho é-axeen-ȧhtse-o'o. (Murray 2008)
child-pl.an 3-scratch-ahte-3pl.an
(5) The children Scratched themselves. reflexive construal of (4)
(6) The children scratched each other. reciprocal construal of (4)
Cheyenne (4) can be true if each child scratched himself, as in the reflexive sce-
nario in Figure 11.2(a), or if each child was scratched and scratched at least one
other child, as the reciprocal scenario in Figure 11.2(b). This was established
through truth-value judgment tasks with contexts described in English as well as
with diagrams such as the ones in Figure 11.2. In addition, several texts show the
availability of each of these interpretations of -ahte.4
Cheyenne -ahte is also compatible with singular subjects, as illustrated in
(7). (In Cheyenne, singular agreement is the unmarked default on the noun and
verb.)
(7) Hetané-ka' ėškóne é-axeen-ahtse. (Murray 2008)

man-child 3-scratch-ahte
“The boy scratched himself.” reflexive construal only
Cheyenne (7) only has a reflexive interpretation. This is unsurprising, as there are
no individuals that the singular subject could bear the verbal relation to besides
himself. In a satisfactory analysis of the Cheyenne data, the reflexive interpretation
of (7) should follow from its having a singular subject.
4
See, e.g., The Bear, the Coyote, and the Skunk by Jeanette Howlingcrane in Leman (2011) for
a reciprocal interpretation of -ahte with a plural subject and Your Head is Covered (anonymous) in
Leman (1987) for a reflexive interpretation of -ahte with a plural subject. For an example with a singular
subject, see I’m Beading Moccasins by Jeanette Howlingcrane in Leman (2011).
A reciprocal construal with a plural subject can also be specified with the ad-
dition of a modifier nonámé'tó'e, as in (8).
(8) He'é-ka'ėškóne-ho noná-mé'tó'e é-axeen-ȧhtse-o'o. (Murray 2008)

woman-child-pl.an noná-other 3-scratch-ahte-3pl.an
“The girls scratched each other.” reciprocal construal only
With the addition of the modifier nonámé'tó'e, Cheyenne (8) is true only in re-
ciprocal scenarios, as in Figure 11.2(b). This modifier can also be used without the
reflexive/reciprocal suffix with a meaning of “one by one” or “in turns.” It is related
to the Cheyenne word mé'tó'e, which can occur independently with the meaning
“in turn” or “in exchange” (Fisher et al. 2006).
In summary, the Cheyenne reflexive/reciprocal suffix -ahte can have both re-
flexive and reciprocal interpretations. These meanings share a requirement that
the same set of individuals perform and undergo the verbal action, but they differ
in what relation is required to hold between the individuals (reflexive or recip-
rocal). The reflexive interpretation and the reciprocal interpretation can each be
specified, given certain changes or additions to the sentence.
This data suggests a unified analysis of the Cheyenne reflexive/reciprocal
morpheme, one where it is underspecified, not ambiguous. Because there is only
one morpheme, a unified analysis, if possible, is preferred. Given the overlap in
meaning between reflexive and reciprocal interpretations, it does not seem like a
true ambiguity (e.g., English bank). Furthermore, many languages express reflex-
ivity and reciprocity with a single form.
Given these facts, I set out to give a unified analysis of Cheyenne -ahte, ap-
pealing to underspecification. In general terms, the analysis of the Cheyenne
reflexive/reciprocal suffix -ahte requires that the subject and the object pick out the
same set of individuals. For (4), the truth conditions are that each child scratched
some child and that each child was scratched. However, it does not require any
particular relation to hold between these individuals. Thus, it is underspecified for
reflexivity and reciprocity. These relations can be further specified with additions
or changes to the sentence. In the next section, I discuss a formal analysis of Chey-
enne reflexives and reciprocals following this general line. I chose to implement
the analysis in a particular framework that provided a simple way of formalizing
this analysis, but it made strong, novel predictions.
4 Underspecification Analysis
The crucial point made in this section is that formalizing an underspecification

analysis led to a novel prediction. This prediction led to new questions to ask
during fieldwork, questions I would not have asked otherwise, and led to new
discoveries about the meaning of this construction in Cheyenne. Thus, formal
semantic training helped to uncover a pattern that may have otherwise gone
undocumented. What was essential was distinguishing between an ambiguity

analysis and underspecification analysis, and working out the predictions of the
analyses. This section discusses the details of the analysis in Murray (2007, 2008)
and how this predicted the mixed construal. However, other underspecification
analyses are possible (e.g., Cable 2014) and, while they may differ in other impor-
tant ways, also predict mixed construals. Those not interested in working through
the details of how the predictions were made could skip directly to section 5, where
the testing of these predictions and the results are discussed in detail.
A unified analysis of the Cheyenne reflexive/reciprocal suffix -ahte can be
given by appealing to underspecification: the suffix requires only that the subject
set and the object set are the same. There is no specification of a reflexive relation
or a reciprocal relation between the individuals in the set, as there is for English
themselves and each other.
The framework Dynamic Plural Logic (van den Berg 1996) allows a straight-
forward and simple analysis of this phenomena because of the way it represents
plurality. In Dynamic Plural Logic, there are two kinds of values for variables:
global values and dependent values. Dependent values are individuals while global
values are sets of individuals. Using this framework, Cheyenne -ahte can be ana-
lyzed as requiring only global identity, identity of the subject (x) and object (y) at
the set level. In Dynamic Plural Logic, this is written as in (9) below.
(9) Proposed translation: -ahte ↝ +[y = x] (Murray 2007, 2008)
In the remainder of this section, I give a brief overview of Dynamic Plural Logic
and then discuss in detail the analysis of Cheyenne given in (9), including the use with
singular and plural subjects, the reciprocal-only interpretation with nonámé'tó'e, and
the connection with English reflexives and reciprocals (adapted from Murray 2007,
2008). The proposed analysis can account for all of the data discussed in section 3,
but makes a novel prediction: that a sentence with Cheyenne -ahte can be true in a
mixed scenario, one that is partly reflexive and partly reciprocal.
4.1 Overview of the Framework: Dynamic Plural Logic
Dynamic Plural Logic (van den Berg 1996; henceforth DPlL) is an extension of
Dynamic Predicate Logic (Groenendijk and Stokhof 1991) developed to model
pluralities and the dependencies between them. As in Dynamic Predicate Logic,
a formula in DPlL denotes a relation between information states. In Dynamic
Predicate Logic, an information state is an assignment function. In DPlL, an in-
formation state is a set of assignment functions, each of which assigns at most one
individual to each variable. Such plural information states as a whole assign a set
to each variable, the collection of values assigned to that variable by the individual
functions in that information state.
For example, consider the sets {a, b}, {c, d}, and {e} assigned to the variables
x, y, and z, respectively. In the extension of Dynamic Predicate Logic to pluralities
in Kamp and Reyle (1993), {a, b} would be assigned to x, {c, d} to y, and {e} to z by
a single assignment function that assigns sets to each variable: g = {〈x, {a, b}〉, 〈y,
{c, d}〉, 〈z,{e}〉}. This information state is represented as the first matrix in (10),
below.
In DPlL, these same values would be assigned to these variables by a set of as-
signment functions, each of which assigns only a single individual to each variable.
One such information state is G = {{〈x, a〉, 〈y, c〉, 〈z, e〉}, {〈x, b〉, 〈y, d〉,〈z, e〉}}. This
information state can also be written as G = {g1, g2} where g1 = {〈x, a〉, 〈y, c〉, 〈z, e〉}
and g2 = {〈x, b〉, 〈y, d〉,〈z, e〉}. This information state is represented as the second
matrix in (10), below.
(10) Information states: assignment function versus set of assignment functions
G x y z
x y z
g1 a c e
g {a, b} {c, d} {e}
g2 b d e
Dynamic Predicate Logic
DPlL
Because of this structure, information states in DPlL allow a distinction be-

tween global values (columns) and dependent values (rows). The global value of
a variable is the set of values assigned to that variable by the entire information
state. For example, the global value of y in (10) is G (y) = {c, d}. A dependent
value of a variable is a subset of its global value, assigned to that variable by a sub-
state—the information state restricted to a particular value for another variable.
For example, there are two x sub-states in (10): G where x = a, written G|x = a, and
G where x = b, written G|x = b. Thus, there are two x-dependent y-values: G|x = a(y) =
{c}, and G|x = b(y) = {d}. DPlL information states can assign the same global values
to variables but differ on their dependent values, as illustrated in the information
states in (11), below.
(11) Same global values, different dependent values
G″ x y z
G x y z G′ x y z
g″1 a c e
g1 a c e g′1 a d e
g″2 b d e
g2 b d e g′2 b c e
g″3 b c e
The three information states in (11) agree on the global values for x, y, and z: they
each assign {a, b} to x, {c, d} to y, and {e} to z. However, the information states
assign different dependent values to the variables. Each of the information states
in (11) has two x sub-states, G|x = a and G|x = b, but what these sub-states are differs
from state to state. For example, the x = b sub-states assign different values to y in
each information state in (11): G|x = b(y) = {d} while G′|x = b(y) = {c} and G″|x = b(y) =
{c, d}.
These different dependent values represent different dependencies between
the variables. In G, b is related to d while in G′, b is related to c. G″ encodes the
same relation between b and d as G as well as an additional relation between b and

c. Thinking ahead to the analysis of reflexivity and reciprocity, these dependencies,
or rows, can be thought of as requirements on the verbal relation of the sentence.
(For example, state G might require that b saw d, state G′ would require that b saw
c, and state G″ would require both.)
The plural information states of DPlL can represent dependencies between
variables—relations between individual members of pluralities—as well as global
values. This distinction is crucial to the analysis of Cheyenne and English re-
flexives and reciprocals given in the next section: Cheyenne -ahte requires only
global identity, while English forms also specify certain dependencies between the
variables.
4.2 Analysis of Cheyenne Reflexives and Reciprocals
In DPlL, Cheyenne -ahte can be analyzed as requiring only global identity—

identity at the column level—with no requirements on the row relations. The pro-
posed translation is given in (12) below.5
(12) -ahte ↝ +[y = x] = (9)
This analysis allows sentences with Cheyenne -ahte and plural subjects, such as
(13), to be true in both reflexive and reciprocal scenarios.
(13) [Several children were playing in the woods and got into some poison ivy.
Ka'ėškóne-ho é-axeen-ȧhtse-o'o. =(4)
(i) “The children scratched themselves.” reflexive construal of (13)
(ii) “The children scratched each other.” reciprocal construal of (13)
The entire Cheyenne sentence (13) can be translated into DPlL as (14) below, where
C = child, PL = plural, and S = scratch. The εx introduces variable assignments to
the variable x (column), which one can think of as a discourse referent for x. The δx
is the distributivity operator over x, which requires each condition in its scope to
hold for each x at the individual (row) level. (See Murray 2007, 2008 for a full DPlL
fragment, with definitions and detailed explanations).
(14) εx ∧ δx (Cx) ∧ PLx ∧ δx (εy) ∧ δx(Sxy) ∧ + [y = x]
The first three conjuncts are contributed by the plural noun, the fourth and fifth
by the verb, and the sixth by the Cheyenne reflexive/reciprocal suffix. This requires
that the subject denote a plural set of children, each member of which scratched
5
The plus in +[y = x] makes this requirement a presupposition (see Murray 2007, 2008).
donna donna
bobby laura bobby laura
(a) Reflexive Scenario (b) (Weak) Reciprocal Scenario

Figure 11.3 Simple Reflexive and Reciprocal Scenarios
some child and was scratched by some child, allowing (13) to be true in both re-
flexive and reciprocal scenarios.
For example, again assume there are three children, bobby, donna, and laura.
Cheyenne (13) can be true in either a reflexive scenario like the one depicted in
Figure 11.3(a) or in a reciprocal scenario like the one depicted in Figure 11.3(b).
These facts are captured by the analysis of Cheyenne -ahte given in (12) be-
cause all that (12) requires is that the subject (x) and object (y) sets are identical.
This is true in both Figure 11.3(a) and Figure 11.3(b), where the subject and object
sets are both {b, d, l}, the set of bobby, donna, and laura. Likewise, several possible
information states could make (14) true. For example, consider the information
states in (15). The requirement imposed by -ahte, +[y = x], is true of each: G1 (x) =
G1(y) = {b, d, l} = G2(x) = G2(y). These information states correspond to the sce-
narios illustrated in Figure 11.3.
(15) Possible information states for (14)
G1 x y G2 x y
g 11 b b g 21 b l
g 12 d d g 22 l d
g 13 l l g 23 d b
(a) Reflexive (b) Reciprocal
The analysis of Cheyenne (13) in (14) introduces x values (εx) for a plural set of indi-
viduals (PLx), each of which are children (δx(Cx)), introduces y values (δx(εy)) that
are identical on a global level (+[y = x]), and requires that each child x scratched
at least one y (δx(Sxy)). In a model with three children, bobby, donna, and laura,
we get information states like those in (15): in (15a) each child scratched herself/
himself and in (15b) bobby scratched laura, laura scratched donna, and donna
scratched bobby. Many information states other than those in (15) could make (14)
true, but crucially (14) provides a unified analysis of Cheyenne (13) that allows it to
be true in both reflexive and reciprocal scenarios.
With the addition of Cheyenne nonámé'tó'e, only a reciprocal construal is
allowed, as in (16).
(16) Ka'ėškóne-ho noná-mé'tó'e é-axeen-ȧhtse-o'o.

child-pl.an noná-other 3-scratch-ahte-3pl.an
“The children scratched each other.” reciprocal construal only
This modifier can be analyzed as adding a requirement about the dependencies

between the individuals in the sets: it requires that each individual be related to at
least one other individual, written in DPlL as +[δy(y ⊘ x)]. This requirement can
be added to the translation of (13), given in (14), to produce the translation of (16),
given in (17).
(17) εx ∧ δx (Cx) ∧ PLx ∧ δx (εy) ∧ +[δy(y ⊘ x)] ∧ δx(Sxy) ∧ + [y = x]
This addition would rule out the information state G1 in (15a). Only information
states like G2 would satisfy this new requirement: each individual has to scratch at
least one other individual, yielding the reciprocal only interpretation.
The meaning contributed by Cheyenne nonámé'tó'e is a component of the
meaning of English each other. Following this analysis, this English reciprocal
marker can be analyzed as the combination of Cheyenne -ahte and nonámé'tó'e,
translated into DPlL as + [y = x] ∧ +[δy(y ⊘ x)]. English themselves can be analyzed
as the combination of Cheyenne -ahte, a plural requirement, and a requirement
that each individual is related at least to itself, translated as + [PLy] ∧ + [y = x] ∧
+[δy(y ○ x)]. (English himself would be similar, but with a singular requirement.)
That is, English each other is lexically specified for a reciprocal meaning, English
themselves is lexically specified for a reflexive meaning, and Cheyenne -ahte is un-
derspecified for reflexivity and reciprocity.
With singular subjects, as in as in (18), Cheyenne -ahte only has a reflexive
interpretation.
(18) Ka'ėškóne é-axeen-ahtse.

child 3-scratch-ahte
“The child scratched himself.” reflexive construal only
The analysis so far predicts this: because the subject is a single individual, the
object must be that same single individual. This results in the reflexive-only inter-
pretation with singular subjects. The translation of Cheyenne (18) is given in (19).
(19) εx ∧ δx (Cx) ∧ SGx ∧ δx (εy) ∧ δx(Sxy) ∧ + [y = x]
The only change between the translations of Cheyenne (13), with a plural subject,
and Cheyenne (18), with a singular subject, is the third conjunct, which now re-
quires the subject to be singular (SGx). This only allows assignments with identical
singleton sets assigned to x and y, as in (20), where G3(x) = G3(y) = {b}.
(20) Possible information state for (19)
G3 x y
g 31 b b
This underspecification analysis correctly predicts all of the data discussed

in section 3: both reflexive and reciprocal construals with plural subjects, only a
reflexive construal with a singular subject, and only a reciprocal construal with the
addition of nonámé'tó'e.
However, the analysis makes a novel prediction: for plural subjects of three
or more individuals, in addition to a reflexive and a reciprocal construal, a mixed
construal is predicted. A mixed construal is one that is partially reflexive and par-
tially reciprocal. For example, if there are three children, bobby, donna, and laura,
and the global values of x and y are the three children {b, d, l}, as in (15), then
another possible information state for Cheyenne (13), repeated below as (22), will
be (c) in (21). In (21c), just as in (21a) and (21b), the global values for x and y are
identical: G4(x) = G4(y) = {b, d, l}.
(21) Novel prediction: a mixed construal
G1 x y G2 x y G4 x y
g 11 b b g 21 b l g 41 b b Reflexive
g 12 d d g 22 l d g 42 d l
Reciprocal
g 13 l l g 23 d b g 43 l d
(a) Reflexive (b) Reciprocal (c) Mixed
Thus, Cheyenne (22) is predicted to have three kinds of construals: a reflexive

construal, a reciprocal construal, and a mixed construal—one that is partly reflex-
ive and partly reciprocal. Such an interpretation is hard to translate into English,
but an attempt is given in (22iii).
(22) [Three children were playing in the woods and got into some poison ivy.
Ka'ėškóne-ho é-axeen-ȧhtse-o'o. = (4)

(i)“The children scratched themselves.” reflexive construal of (22)
(ii)“The children scratched each other.” reciprocal construal of (22)
(iii)“The children were scratching.” mixed construal of (22)
These three construals are illustrated in Figure 11.4. For the mixed construal, the
reciprocal relation holds of some sub-group (here donna and laura) while the re-
flexive relation holds of the rest (here bobby).
This prediction of the proposed underspecification analysis of Cheyenne
-ahte is unavoidable: the properties of DPlL that allow a simple, unified analysis of
donna donna donna
bobby laura bobby laura bobby laura
(a) Reflexive Scenario (b) (Weak) Reciprocal Scenario (c) Mixed Scenario
Figure 11.4 Simple Reflexive, Reciprocal, and Mixed Scenarios
the reflexive and reciprocal construals of Cheyenne predict that a mixed construal
should be possible. In fact, any underspecification analysis with similar properties
would make the same prediction.
Prior to formalizing this underspecification analysis of Cheyenne, I had never
thought to ask if Cheyenne sentences like (22) were true in mixed scenarios like
Figure 11.4(c). Coming from an English perspective, with constructions specified
for reflexivity and reciprocity, mixed scenarios are somewhat strange. Such an
interpretation is certainly not predicted, or possible, if we think that Cheyenne
-ahte is ambiguous between English-like reflexive and reciprocal interpretations.
Thus, at the time of formalizing the analysis, I did not know if Cheyenne (22)
was true or false in a situation like Figure 11.4(c). If not for the strong predictions
made by this formal analysis, I might never have asked. Contra, for example, Evans
and Levinson (2009), abstracting to a formal analysis can free us from a language
bias, allowing us to examine patterns in typologically diverse languages from neu-
tral ground: “From the theoretical point of view all languages have equal status”
(Bittner 2008).
5 Testing the Predictions and Results
After realizing that an underspecification analysis of Cheyenne predicted that sen-

tences like (22) were true in mixed scenarios, as in Figure 11.4(c), I had to devise
ways to test this prediction during my next fieldwork trip. Several Cheyenne texts
(e.g., Leman 1980a, 1987) have instances of -ahte that are clearly reflexive and
ones that are clearly reciprocal (see note 4). These textual examples and a note
in the grammar (Leman 2011) were how I first became aware of this property of
Cheyenne, but no textual examples clearly illustrate a mixed interpretation. This
absence of evidence does not aid in determining whether construals are allowed—
semantic elicitation is needed.
To test whether or not mixed interpretations are available in Cheyenne, I devel-
oped three direct elicitation tasks, drawing on Matthewson (2004): (i) judgments
of sentences like Cheyenne (22) in situations described in English, (ii) judgments
of sentences like (22) with drawings of various situations, and (iii) constructing
short discourses that spell out the mixed construal of sentences like (22). The first
two of these tasks are truth-value judgment tasks. The third task is a combination
of a truth-value judgment and a felicity judgment task: I constructed short dis-
courses in Cheyenne that spelled out the mixed construal and asked if they were
acceptable, and if so, if they were true or false.
For the first task, I gave the contexts in English. All of my consultants speak
English fluently, and, though I am learning, I am not yet fluent in Cheyenne. In
addition, it is easier to ensure a reflexive or reciprocal interpretation in English,
since the English forms themselves and each other are specified for reflexivity and
reciprocity. All of the examples I constructed for this task were roughly of the form
given in (23). For the purpose of this chapter, I’ll use the same verb throughout,
though I tested with a variety of verbs.
In (23), I provided a context that specified a reflexive interpretation for the
whole group of children. Sentence (22), repeated in each example below, was
judged by my consultants to be true in this context, to be an accurate description
of the scenario.
(23) Reflexive [Three children were playing in the woods and got into some
poison ivy. Not long after, they were covered in itchy bumps. All of the
children got the poison ivy on their legs, so they each scratched their own
legs.]

“The children scratched themselves.” Judgment: True
In (24) I gave a scenario that specified a reciprocal scenario. Sentence (22) was
judged true in this scenario as well.
(24) Reciprocal [Three children were playing in the woods and got into some
children got the poison ivy on their backs, so they couldn’t scratch it
themselves. Each child scratched another’s back.]

“The children scratched each other.” Judgment: True
To make sure that identity of subject and object was required, I tried examples
where only part of the group was scratching, as in (25), and where the children
were being scratched by other people, as in (26). Sentence (22) was judged false in
both of these scenarios.
(25) Neither [Three children, a boy and two girls, were playing in the woods
and got into some poison ivy. Not long after, the boy had itchy bumps all
over his legs. He scratched his own legs while the girls checked themselves
for bumps.]
“The children scratched themselves/each other.” Judgment: False
(26) Neither [Three children were playing in the woods and got into some
children got the poison ivy on their backs, so they couldn’t scratch it
themselves. The children ran home and had their parents scratch their
backs for them.]
Ka'ėškóne-ho é-axeen-ȧhtse-o' o. =(22)

“The children scratched themselves/each other.” Judgment: False
So far these judgments support the analysis given in section 4: Cheyenne (22) is
true in both reflexive and reciprocal scenarios, and requires the subject and object
set to be the same. But what about the mixed construal? I constructed a parallel
scenario where part of the group of children were scratching each other and one
member of the group scratched himself, given in (26). Sentence (22) was judged
true in this scenario.
(27) Mixed [Three children, a boy and two girls, were playing in the woods and got
into some poison ivy. Not long after, they were covered in itchy bumps. The
boy had bumps all over his legs and the girls had bumps all over their backs.
The boy scratched his own legs and each girl scratched the other’s back.]

“The children were scratching.” Judgment: True
I also tried variants of these tasks giving specific names to each child and the re-
sults were the same.
The second task was judgments of sentences like Cheyenne (22) given pictures
that I drew. The drawings presented scenarios parallel to the ones described above
in English, and they produced parallel results. These were very useful for helping
to make clear the different relations. For the most part, they looked similar to the
illustrations in Figure 11.4. However, I drew these all on the fly, realizing only once
I was in the field how useful pictures could be. Since then (2006–2007), more at-
tention has come to using pictures and storyboards in elicitations (see, e.g., the
Totem Field Storyboards (http://totemfieldstoryboards.org/) and Burton and Mat-
thewson, this volume). In subsequent elicitations, I have used maps and images
to aid in elicitation, though they must also be supplemented with some verbal
contextual information.
For the third task, I constructed short Cheyenne discourses like (28), which
elaborate on the relations between the individuals, spelling out the mixed con-
strual. In (28), the first sentence, (28i), is Cheyenne (22). The second sentence,
(28ii), is a conjunction of Cheyenne (7) “the boy scratched himself ” and Cheyenne
(8) “the girls scratched each other.”
(28) Mixed [Three children, a boy and two girls, were playing in the woods and
got into some poison ivy. Not long after, they were covered in itchy bumps.
The boy had bumps all over his legs and the girls had bumps all over their
backs. The boy scratched his own legs and each girl scratched the other’s
back.]
i. Ka'ėskóne-ho é-axeen-ȧhtse-o'o. =(22)

child-pl.an 3-scratch.an-ahte-3pl.an
ii. Hetané-ka'ėskóne é-axeen-ahtse naa he'é-ka'ėškóne-ho

man-child 3-scratch.an-ahte and woman-child-pl.an
noná-mé'tó'e é-axeen-ȧhtse-o'o.
noná-non.id 3-scratch.an-ahte-3pl.an
Judgment: Felicitous and True
The mixed elaboration in Cheyenne (28ii) specifies a reflexive relation for the sub-
group of the boy and a reciprocal relation for the subgroup of the girls. This con-
tinuation of Cheyenne (28i) is felicitous—it is a good elaboration. In addition, (28)
is judged true in the context provided. That would not be possible if Cheyenne (28i)
were ambiguous between a reflexive and a reciprocal, as shown in (29) with English.
(29) a. # The children scratched themselves. The boy scratched himself and the
girls scratched each other.
b. # The children scratched each other. The boy scratched himself and the
girls scratched each other.
Thus, it is difficult to translate the Cheyenne discourse (28) into English. The least
awkward translation is (30), where Cheyenne (28i) is rendered as (30i), without
any object:
(30) i. The children were scratching.

ii. The boy scratched himself and the girls scratched each other.
In summary, results from various tasks show that mixed construals are al-
lowed in Cheyenne. Supporting evidence comes from each of the three tasks: sen-
tences like Cheyenne (22) were judged true in mixed scenarios (both visual and
described) and the mixed elaboration in (28) is grammatical, felicitous, and true
in mixed scenarios. This provides strong evidence that Cheyenne -ahte is under-
specified for reflexivity and reciprocity, not ambiguous. Mixed elaborations are
unavailable with English reflexives and reciprocals, so Cheyenne cannot just be
ambiguous between these two meanings.
6 Conclusions
In making a case for the indispensable role of direct elicitation methods in se-
mantic fieldwork, while (re)emphasizing the importance of texts, Matthewson
(2004) says “As pointed out by Mithun (2001) and many others, it is only by col-
lecting spontaneous speech that the researcher can be exposed to phenomena that
are outside the boundaries of his/her prior knowledge or imagination” (p. 376).
I strongly believe that collecting spontaneous speech and texts are invaluable tools
for fieldwork, for learning and studying a language. However, as demonstrated by
this paper, theoretical analyses can be another such tool. By formalizing analyses,
we make strong predictions that can also expose us to phenomena outside the
boundaries of our prior knowledge and imagination.
Mixed construals are available for Cheyenne -ahte. This is a fact that, coming
from an English perspective, might never have been discovered. The underspecifi-
cation analysis of this construction discussed in section 4 accounts for both sin-
gular and plural antecedents and the variety of construals. It also made the strong
prediction of a mixed construal. In fact, there being a mixed construal is a conse-
quence of the analysis, unavoidable. As it turns out, the prediction was borne out
by the Cheyenne data, but it might have been otherwise. If the prediction had been
falsified, it would have been evidence for an ambiguity analysis, and it would still
have revealed something new about Cheyenne.
To my knowledge, before Murray (2007), no one had explored the possibility
of mixed construals for any other languages with “ambiguous” reflexive/reciprocal
markers. However, many such languages exist. After working on the Cheyenne mixed
elaboration discourse in (28), I conducted an informal survey of other languages that
express reflexivity and reciprocity with a single form. Discourses parallel to Cheyenne
(28) suggest that mixed construals are available in many other languages: Polish (Maria
Bittner, personal communication), Romanian (Adrian Brasoveanu, personal commu-
nication), French (Viviane Déprez, personal communication), Spanish (Carlos Fasola,
personal communication), and German (Judith Tonhauser, personal communication).
So this pattern of mixed construals seems cross-linguistically quite robust, though not
before documented, and has been the topic of subsequent study (Cable 2014).
Two general conclusions are supported by the example discussed in this
chapter. First, formal semantic training is valuable for fieldwork. Formally precise
analyses make predictions that must be tested in the field. These predictions might
be things we as fieldworkers would not otherwise ask. Coming from my English-
speaking perspective, I might never have thought up such a scenario as (28), and
perhaps this is not the type of scenario we would ever find in naturally occurring
data. As Rice 2006 says, “enriched coverage comes about because linguistic theory
forces one to ask questions that likely would not have been asked otherwise.”
Second, there is a symbiotic relationship between fieldwork and theory. (Se-
mantic) fieldwork provides novel and varied data that pushes the limits of our
current theories, and making our (semantic) analyses formally precise can provide
new lines of inquiry for fieldwork.
References
van den Berg, Martin. 1996. Some aspects of the internal structure of discourse. PhD diss.,
ILLC, University of Amsterdam, Amsterdam.
Bittner, Maria. 2008. Aspectual universals of temporal anaphora. In Theoretical and Cross-
linguistic Approaches to the Semantics of Aspect, Susan Rothstein (ed.), 349–385. Am-
sterdam: John Benjamins Publishing Company.
Burton, Strang, and Lisa Matthewson. 2014. This volume. Targeted construction story-
boards in semantic fieldwork.
Cable, Seth. 2014. Reflexives, reciprocals and contrast. Journal of Semantics. 31(1): 1–41.
Cover, Rebecca, and Judith Tonhauser. 2014. This volume. Theories of meaning in the field:
Temporal and aspectual reference.
Crain, Stephen, and Rosalind Thornton. 1998. Investigations in Universal Grammar: A
Guide to Experiments on the Acquisition of Syntax and Semantics. Cambridge, MA:
The MIT Press.
Dalrymple, Mary, Makoto Kanazawa, Yookyung Kim, Sam Mchombo, and Stanley Peters.
1998. Reciprocal expressions and the concept of reciprocity. Linguistics and Philosophy
21(2): 159–210.
Eisele, Julie, and Barbara Lust. 1996. Knowledge about pronouns: A developmental study
using a truth-value judgment task. Child Development 67: 3086–3100.
Evans, Nicholas, and Stephen C. Levinson. 2009. The myth of language universals: Lan-
guage diversity and its importance for cognitive science. Behavioral and Brain Sciences
32:429–492.
Fisher, Louise, Wayne Leman, Leroy Pine Sr., and Marie Sanchez. 2006. Cheyenne Dictionary.
Lame Deer, MT: Chief Dull Knife College. http://www.cdkc.edu/cheyennedictionary/
index.html.
Gillon, Carrie. This volume. Investigating D in languages with and without articles.
Groenendijk, Jeroen, and Martin Stokhof. 1991. Dynamic predicate logic. Linguistics and
Philosophy 14(1): 39–100.
Kamp, Hans, and Uwe Reyle. 1993. From Discourse to Logic: Introduction to Modeltheoretic
Semantics in Natural Language, Formal Logic and Discourse Representation Theory.
Dordrecht: Kluwer Academic Publishers.
Langendoen, D. Terence. 1978. The logic of reciprocity. Linguistic Inquiry 9(2): 177–197.
Langendoen, D. Terence, and Joël Magloire. 2003. The logic of reflexivity and reciproc-
ity. In Anaphora: a Reference Guide, Andrew Barss (ed.), 237–263. Oxford: Blackwell
Publishing.
Leman, Elena M. 1999. Cheyenne Major Constituent Order. Dallas, TX: The Summer Insti-
tute of Linguistics.
Leman, Wayne (ed.). 1980a. Cheyenne Texts: An Introduction to Cheyenne Literature. Oc-
casional Publications in Anthropology, Series No. 6. Greeley, CO: Museum of Anthro-
pology, University of Northern Colorado.
Leman, Wayne. 1980b. A Reference Grammar of the Cheyenne Language. Occasional Publi-
cations in Anthropology, Series No. 5. Greely, CO: Museum of Anthropology, Univer-
sity of Northern Colorado.
Leman, Wayne (ed.). 1987. Náévȧhóó’ȯhtséme / We Are Going Back Home: Cheyenne History
and Stories Told by James Shoulderblad and Others. Memoir 4. Winnipeg, Canada: Al-
gonquian and Iroquoian Linguistics.
Leman, Wayne. 2011. A Reference Grammar of the Cheyenne Language. Raleigh, NC: Lulu
Press. Updated version of (Leman 1980b).
Maslova, Elena. 2008. Reflexive encoding of reciprocity. In Reciprocals and Reflexives, Ekke-
hard König and Volker Gast (eds.). Mouton de Gruyter.

McKenzie, Andrew. This volume. Deriving topic effects in Kiowa with semantics and
pragmatics.
Murray, Sarah E. 2007. Dynamics of reflexivity and reciprocity. In Proceedings of the Six-
teenth Amsterdam Colloquium, Maria Aloni, Paul Dekker, and Floris Roelofsen (eds.),
157–162. Amsterdam: ILLC/Department of Philosophy, University of Amsterdam.
Murray, Sarah E. 2008. Reflexivity and reciprocity with(out) underspecification. In Proceed-
ings of Sinn und Bedeutung 12 (2007), Alte Grønn (ed.), 455–469. Oslo, Norway: ILOS.
Newman, Paul, and Martha Ratliff (eds.). 2001. Linguistic Fieldwork. Cambridge: Cam-
grammars. In Catching Language: The Standing Challenge of Grammar Writing, Felix
K. Ameka, Alan Charles Dench, and Nicholas Evans (eds.), 235–268. Berlin: Mouton
de Gruyter.
12
Theories of Meaning in the Field: Temporal

and Aspectual Reference
Rebecca T. Cover and Judith Tonhauser
1 Introduction
Many linguists make it their goal to describe the grammars of under-studied lan-
guages that do not yet play a role in linguistic theorizing. Describing languages
is challenging for many reasons, including the following two.* First, describing
the grammar of an under-studied language is a complex, and therefore potentially
daunting, task. Second, the description needs to properly reflect the “genius” of
the language (Sapir 1921) without being excessively influenced by preconceived
notions about better-studied languages, but while still allowing for meaningful
comparisons to the grammars of other languages. Meeting this second challenge
is particularly important given the fast rate at which languages are dying and the
pressing need for bringing evidence from these languages to bear on theories of
meaning and on the study of language variation and universals. We argue in this
paper that these challenges can be met by the informed and cautious use of theory.
Benefits and pitfalls of involving theory in language description have been
widely discussed, mainly with an eye towards phonetics, phonology, morphology,
and syntax (see e.g., Dixon 1997, Gil 2001, Hyman 2001, Noonan 2005, Rice 2005,
Dryer 2006, Rice 2006, Haspelmath 2010a). In the study of meaning, however, the
*
We thank Ashwini Deo and Lev Michael for helpful discussion of the chapter in its preliminary
stages and Lynn Nichols for sharing her thoughts, years ago, on the relationship between theory and
data. Jürgen Bohnemeyer, as well as the editors of the volume, Ryan Bochnak and Lisa Matthewson,
gave detailed comments that led to important improvements. The chapter also benefited from feedback
by participants of Tonhauser’s course Semantic Fieldwork Methods at the 2013 Linguistic Institute in
Ann Arbor, MI. Of course, none of these individuals necessarily share the views we present here. We
also thank our Badiaranke consultants—Boubacar “Babacar” Niabaly, Bouly Niabaly, Kady Niabaly,
and Sountouré Niabaly—and our Guaraní consultants, especially Marité Maldonado. Tonhauser’s re-
search was supported by NSF award BCS-0952571. Cover’s was supported in part by NEH Grant #RZ-
306 50500-05 and by an ACLS New Faculty Fellowship.

Temporal and Aspectual Reference 307
interplay between theory and fieldwork-based description has received less atten-
tion. Matthewson (2004), the only paper outside this volume to discuss general
methodology for fieldwork on meaning, introduces readers to key semantic and
pragmatic notions and the intricacies involved in distinguishing them in fieldwork,
but does not explicitly address the interplay between theory and fieldwork-based
description. Tonhauser et al. (2013) develop diagnostics for exploring theoretically
informed properties of projective contents with theoretically untrained consul-
tants, but also do not discuss the relationship between theory and fieldwork.
In general, there remains a lamentable disconnect between theories of mean-
ing and fieldwork-based research on meaning, with several unfortunate conse-
quences. For one, many grammars do not discuss basic meaning properties of the
languages described, focusing instead on “good descriptions of the phonetics and
the phonology, as well as of the morphology and the syntax” (Noonan 2005:360;
see also Rice 2005 for a discussion of the contents of grammars). Other grammars
cover semantic topics but fail to define the terminology used, or use terminology
in ways that do not reflect the properties of the language being described. And in
some grammars, the data presented are intriguing, but the descriptions lack preci-
sion, making them unsuitable for cross-linguistic comparison and for assessing
theories of meaning.
Our goal in this chapter is to explicitly discuss the interplay between theory
and fieldwork in the study of meaning. Specifically, with respect to the two chal-
lenges of fieldwork-based research mentioned above, we argue that (i) theory can
guide fieldwork on meaning, and when it does, more comprehensive descriptions
of meaning result,1 and (ii) compared to linguistic fieldwork that is not theoreti-
cally informed, theoretically informed descriptive fieldwork has greater potential
for revealing the “genius” of the language under investigation (how it differs from
other languages), for improving theories, and also for increasing our knowledge of
language variation and universals.
We argue for these two points on the basis of fieldwork-based research on a
particular domain of meaning, namely temporal and aspectual reference. In so
doing, we also provide a theoretically informed guide to exploring and providing
meaning descriptions in this empirical domain. Meaning descriptions are state-
ments of empirical generalizations about the form-meaning mapping in a par-
ticular language. Such descriptions form the basis for, and are therefore distinct
from, formal semantic and pragmatic analyses, which rely on tools from set theory
and logic to formulate compositional models of the form-meaning mapping. We
also distinguish the description of a language from its documentation, which is
theory-independent and aims at “creat[ing] a record of a language in the sense of a
comprehensive corpus of primary data” (Himmelmann 2006:3).
We focus on this empirical domain for two reasons. First, no guide for con-
ducting theoretically informed fieldwork on temporal and aspectual reference
1
Murray (this volume) makes a related point.
is available (for an excellent cross-linguistic overview of this empirical domain,

see, e.g., Chung and Timberlake 1985). Second, both of us have explored temporal
and aspectual reference with theoretically untrained native speaker consultants of
Badiaranke (since 2004) and Paraguayan Guaraní (also since 2004), respectively.
Badiaranke is an Atlantic language, spoken by 10,000 to 15,000 people in Senegal,
Guinea, and Guinea-Bissau; Paraguayan Guaraní, a member of the Tupí-Guaraní
family, is spoken by over four million people in Paraguay and Argentina.
In section 2, after introducing a theoretical framework for exploring temporal
and aspectual reference, we discuss in detail how familiarity with this framework
can guide fieldwork on temporal and aspectual reference in sections 3 and 4, re-
spectively. Section 5 is devoted to discourse about the future, which can be real-
ized through temporal, aspectual, modal, and mood-based strategies. Throughout
these sections, we also discuss how assuming a theoretical framework can increase
the potential of the descriptive work to reveal the genius of the language and to
improve theory. The chapter concludes in section 6 with a discussion of five re-
quirements that theoretically informed descriptions must meet if they are to be
valuable for the study of meaning.
2 A Theoretical Framework for Describing Temporal and Aspectual Reference
The theoretical framework for describing temporal and aspectual reference that
we introduce in this section is a neo-Reichenbachian one. This framework as-
sumes that the temporal and aspectual reference of clauses can be described in
terms of temporal relations between three time intervals, which are introduced
in detail in section 2.1: the evaluation time, the topic time (sometimes also called
“reference time”), and the eventuality time (Kamp and Reyle 1993, Klein 1994).
The neo-Reichenbachian framework serves as a starting point for our discussions
of temporal and aspectual reference in sections 3 and 4, as it is the framework in
which we both work and in which much research on temporal and aspectual refer-
ence is couched.
The framework serves as a tool for accurate description, not as an ultimate,
irrevocable determinant of how every language will behave. If data discovered in
a language cannot be described within the framework, the data should not be dis-
carded, nor should they be twisted into conforming to the framework. Rather,
such a situation motivates modification of the framework. As Rice (2006:262) puts
it: in a good linguistic description “[t]he theory informs and shapes, but does not
control” the description.
2.1 Time Intervals in a neo-Reichenbachian framework
We assume that clauses describe eventualities, a cover term for events and
states (Bach 1986). The eventuality time (ET) of a clause is the time at which
the eventuality it describes is temporally located; for an event, this is the time at
which it occurs and, for a state, the time at which it holds. The utterance time
(UT) is the time at which a matrix clause is uttered; it is relative to this time that
the truth conditions of the clause are evaluated. But not all clauses are evaluated
relative to the utterance time: some subordinate clauses, for instance, may be
evaluated relative to the eventuality time of the matrix clause. The more general
evaluation time (EvT) is therefore used to refer to the time relative to which
a clause is evaluated. The third time interval assumed in neo-Reichenbachian
frameworks is the topic time (TT), which is the interval the uttered clause is
about. To illustrate the three time intervals, consider (1), adapted from Klein
(1994:3f.):
(1) Context: A judge (J) is interrogating a witness (W) in court.

J: What did you notice when you looked into the room?
W: The light was on. [TT < UT; TT ⊆ ET]
The judge’s question fixes the topic time as the past interval when the witness looked
into the room. If the witness is a cooperative interlocutor who adheres to general
conversational principles (Grice 1975), in particular the principle of making her
utterance relevant to the current discourse goal, then her answer can be assumed
to elaborate on what the world was like at this topic time. The eventuality time of
the sentence W utters is the time at which the light was on. Competent speakers
of English understand W’s utterance to convey that the light was already on when
W looked into the room. That is, we understand W’s utterance to convey that the
eventuality time temporally includes the topic time, which precedes the evaluation
time (here, the utterance time).
This framework privileges two temporal relations between these three times:
the temporal relation between the evaluation time and the topic time of a clause
constitutes the clause’s temporal reference, whereas the temporal relation be-
tween the topic time and the eventuality time of the clause constitute its aspectual
reference.2 W’s utterance in (1) is annotated with an abbreviation of its temporal
reference (TT < UT) and its aspectual reference (TT ⊆ ET). (We provide such
annotations for select examples throughout the chapter to illustrate a variety of
temporal and aspectual references attested cross-linguistically.) This notation in-
dicates that the sentence uttered by W is only compatible with topic times that are
temporally located prior to the evaluation time, which is the utterance time in (1).
This restriction is due to the past tense form of the verb to be: past tense restricts
topic times to times prior to the evaluation time, as discussed in detail in section 3.
2
Cable (2013) argues that temporal remoteness markers in Gĩkũyũ temporally relate eventuality
times to evaluation times. This language thus provides evidence for a third privileged temporal relation,
namely between evaluation and eventuality times. See also Comrie (1985:ch.5) on temporal remoteness
markers.
The aspectual reference of the clause W utters is the inclusion relation, that is, the
eventuality time of the state of the light being on includes the topic time, as dis-
cussed in detail in section 4.
2.2 Distinguishing temporal/aspectual reference

from tense/aspect
A distinction that is crucial for the discussions in this paper is between the tem-
poral and aspectual reference of clauses, on the one hand, and tense and aspect,
on the other. We use these latter terms exclusively to refer to natural language
expressions with certain properties. In particular, a tense is an expression that
forms part of a grammatical paradigm and that constrains the temporal reference
of the clause in which it occurs;3 a (grammatical) aspect is an expression that
forms part of a paradigm and constrains the aspectual reference of the clause in
which it occurs.4
The distinction between temporal/aspectual reference as properties of clauses
and tense/aspect as expressions that constrain temporal/aspectual reference is es-
sential for exploring the form-meaning mapping across languages. For example,
whereas every language can realize clauses with past temporal reference (where
the topic time precedes the evaluation time), not all languages have past-tense
morphemes that constrain the topic time to a past time. Likewise, every language
can convey progressive aspectual reference (where the eventuality time temporally
includes the topic time), but not every language has a distinguished progressive
aspect marker to convey this aspectual meaning.
Keeping temporal and aspectual reference separate from tense and aspect
allows precise statements about similarities and differences in the form-meaning
mapping, as illustrated for English and Paraguayan Guaraní (henceforth Guaraní)
in (2). In the context given, both the English example in (2a) and the Guaraní
example in (2b) convey that the eventuality of the speaker bathing is ongoing at
the topic time, the time when the door bell rang, which temporally precedes the
utterance time. But whereas this meaning is conveyed in English by uttering a
sentence with a past-tense finite verb (was) and a progressive aspect construction
3
This definition differs from Comrie ’s (1985) much-cited definition of tense as “grammaticalized
expression of location in time” (p. 9), which does not specify what is located in time, for example,
the eventuality time, the topic time, or something else altogether. It also differs from the definition of
tense in Chung and Timberlake (1985:203) as locating “the event in time by comparing the position
of the frame [the eventuality time, RC/JT] with respect to the tense locus.” This definition of tense, as
temporally relating the eventuality time and the utterance time, is also found in, for example, Zagona
(1990) and Stowell (1996).
4
A linguistic paradigm is generally defined as a set of forms that are derived from the same
base form and that contrast with one another semantically and morphosyntactically (see, e.g., Beard
1995:254). Some paradigms, including the English tense paradigm, are impoverished, consisting
only of an overt morpheme and an unmarked form. English modals, in contrast, are not para-
digmatic by this definition, since the English modal verbs (can, must, might, etc.) do not share a
common root.
(be V-ing), it is conveyed in Guaraní by a sentence that only consists of the verb
stem –jahu “bathe” inflected for first-person singular.5
(2) Context: I tell my mother that yesterday my doorbell rang at a very
inopportune moment. My mother asks me: What were you doing when the
doorbell rang?
a. I was bathing. English [TT < UT; TT ⊂ ET]
b. A-jahu. Guaraní [TT < UT; TT ⊂ ET]
A1sg-bathe
“I was bathing.”
Thus, past temporal progressive aspectual reference is conveyed by both the English
and the Guaraní utterances in (2). But only in English is this meaning conveyed as
part of the meanings of the morphemes that comprise the uttered sentence; uttered
in other contexts, the Guaraní sentence is compatible with other types of temporal
and aspectual reference. If we used, for example, “tense” to refer both to the tempo-
ral reference of a clause and to an expression that constrains temporal reference, as
some authors do, we would need to say that both the English and the Guaraní exam-
ples have past tense, thereby obscuring the differences between the two languages.
The examples in (2) also allow us to make another point that is key for our dis-
cussions in this paper, namely the importance of considering the temporal and aspec-
tual reference of utterances, that is, sentences in context,6 rather than of sentences in
isolation. The context in (2) plays a very important role: it fixes a particular temporal
and aspectual reference for the utterances. The fact that the utterances in (2a) and
(2b) are judged to be acceptable by English and Guaraní speakers, respectively, in
this context is evidence that they are both compatible with past temporal progressive
aspectual reference, despite differing in form. Crucially, sentences presented without
a context, or translations of sentences into the contact language do not constitute
data that can be used to evaluate hypotheses about the form-meaning mapping (as
discussed in Matthewson 2004:§3 and Cover’s and Deal’s chapters in this volume).
2.3 Theoretically informed meaning description
The neo-Reichenbachian framework just introduced takes a particular perspective

on the temporal and aspectual reference of utterances: the framework privileges
5
Glosses in this chapter use the following abbreviations: 1, 2, 3 = 1st, 2nd, 3rd person; 1sg.3sg = 1st
person acting on 3rd person (etc.); A = set A cross-reference marker; aff.decl = affirmative declarative
clause marker; compl = complementizer; d3 = deictic particle; def = definite; det = determiner; fem =
feminine; fut = future; imp = imperative; impf = imperfective; inc = incompletive; incl = inclusive;
loc = locative; masc = masculine; nom = nominative; nmlz = nominalizer; npst = non-past; p =
preposition; par = participial; pfv = perfective; pl = plural; pres = present; pret = preterite; prosp =
prospective; q = question; sg = singular; term = terminative; top = topic.
6
Context is taken to be the utterance context, the context in which the utterance is made, which
is a body of information held in common by the interlocutors in the discourse, including information
from the utterance situation, the linguistic context in which the utterance is made, as well as the infor-
mation structure of the preceding discourse (e.g., Roberts 2004:197f.)
three time intervals (the evaluation, topic, and eventuality times) and two tempo-
ral relations between them (the temporal reference relation between the evalua-
tion and the topic times, and the aspectual reference relation between the topic
and eventuality times). Inherent to the framework is thus a very strong hypothesis
regarding the form-meaning mapping in the domain of temporal and aspectual
reference in any language, namely that this mapping can be described by refer-
ence to these three time intervals and two temporal relations. In particular, for
any given linguistic expression of the language, the framework demands that the
researcher ask whether the meaning of the expression constrains either of the two
relations, and, if so, how. And for any given temporal or aspectual reference re-
lation expressible in the framework, the researcher is held to explore how that
particular relation is expressed in the language. This strong hypothesis, and the
questions resulting from it, can guide fieldwork on temporal and aspectual refer-
ence in a particular language.
Another feature of the framework is that the relevant categories, such as
“temporal reference” and “tense”, are (at least partially) defined on the basis of the
three privileged time intervals, not only on the basis of particular structures or
morphological forms. As a consequence, the definitions of the categories provide
guidance about the positive and negative evidence the researcher has to provide
in support of a particular hypothesis about the form-meaning mapping. To argue,
for instance, that a particular form restricts temporal reference, the researcher
has to provide evidence that clauses with this form are compatible with some but
not all of the possible temporal relations between the evaluation time and the
topic time.
The framework can lead to more comprehensive meaning descriptions since,
simply put, the language description is only complete once the contributions to
temporal or aspectual reference of all the forms of the language have been identi-
fied, and once the linguistic realizations of all possible temporal and aspectual
reference relations have been explored. Awareness of how temporal and aspectual
reference is realized in other languages allows the researcher to identify which
properties of the form-meaning mapping in the language under investigation are
cross-linguistically novel, and which ones are attested already. Descriptive research
on temporal and aspectual reference within this neo-Reichenbachian framework
also has the potential to lead to revisions of theories of temporal and aspectual ref-
erence, namely whenever an empirical generalization established for a particular
language cannot be captured in the framework.
Despite such advantages of theoretically informed descriptive fieldwork,
some researchers argue that bringing theory into the field is problematic since, as
Haspelmath puts it, theoretical frameworks may “set up expectations about what
languages should, can and cannot have, and once a framework has been adopted,
it is hard to free oneself from the perspective and the constraints imposed by it”
(Haspelmath 2010a:303). More specifically, the worry is that working within a par-
ticular theory will lead the researcher to impose theoretically motivated categories
on the language to be described, without recognizing that “all languages have dif-
ferent categories” (Haspelmath 2010a:302). Thus, such researchers think it best
not to use theory in language description, or to only use “atheoretical” frame-
works, such as General Comparative Grammar (Lehmann 1989), Canonical Ty-
pology (Corbett 2005, Corbett 2007), Basic Linguistic Theory (Dixon 2010), and
Framework-Free Description (Haspelmath 2010a, b). We argue that the aforemen-
tioned qualities of theoretically informed descriptive fieldwork far outweigh any
potential dangers.
In fact, every description is theory-dependent to some extent. Even purport-
edly atheoretical descriptions make some assumptions about how language is
structured and how it can be described. For instance, one case study that Haspel-
math uses to illustrate Framework-Free Description, a description of Tagalog
syntax by Schachter and Otanes (1972), uses terms like “core” and “topic”. In order
for a description that employs these terms to be explicit, these terms must be de-
fined in the context of some theoretical framework, even if these are language-spe-
cific theoretical terms. Rather than asking researchers to abandon their theoretical
assumptions when conducting fieldwork, it seems more productive to encourage
an explicit discussion of the theoretical assumptions that were made as the re-
search was conducted. We therefore advocate for the informed and cautious use of
theory in conducting descriptive fieldwork, coupled with a willingness to abandon
or modify the theory in light of relevant empirical findings.
Some resistance against using a theoretical framework in language descrip-
tion seems to stem from the perception that such frameworks generally assume
language universals in structure and the form-meaning mapping, that is, that such
frameworks generally subscribe to the Chomskyan Universal Grammar paradigm.
Evans and Levinson (2009), for instance, maintain that there is “[a] widespread
assumption among cognitive scientists, growing out of the generative tradition in
linguistics, . . . that all languages are English-like but with different sound systems
and vocabularies” (p. 429). Theoretical frameworks that have built-in assumptions
about language structures and the form-meaning mapping may indeed be par-
ticularly likely to unduly affect descriptions of meaning in a particular language.
For example, if a framework for the description of temporal and aspectual refer-
ence assumes that every sentence in any language realizes a Tense Projection (TP)
with a T(ense) inflectional head, or that the topic time is introduced by a tense
morpheme, there is a danger that descriptions in this framework assume such
structures or morphemes without providing empirical evidence for them. But
theory-informed descriptions need not make any assumptions about linguistic
universals: theory-informedness merely involves doing description in a way that
acknowledges the existence of, and uses as a descriptive tool, some theoretical
framework. In particular, the neo-Reichenbachian framework we use in this chap-
ter (and in our research) makes no assumptions about the universality of particu-
lar syntactic structures or morpheme inventories. Rather, the framework provides
the researcher with a general hypothesis about a particular domain of meaning,
namely temporal and aspectual reference, and a means for providing theoretically
informed meaning descriptions that employ well-defined terminology.
3 Exploring Temporal Reference
In the previous section, the temporal reference of a clause was defined as the tem-
poral relation between the topic time and the evaluation time. In other words, the
temporal reference of a clause is the set of topic times with which it is compatible.
We speak of a clause being compatible with past temporal reference if the set of
topic times the clause is compatible with includes ones that temporally precede
the evaluation time; likewise, we speak of a clause being compatible with pres-
ent temporal reference or future temporal reference if the set of topic times the
clause is compatible with includes ones that overlap with or follow the evaluation
time, respectively.
In exploring temporal reference in a language, one may have the goal of
describing the entire language, or the more modest goal of discussing the con-
tributions of particular expressions to temporal reference. In the context of the
theoretical framework introduced in the last section, the first goal amounts to ex-
ploring hypotheses about how temporal reference is constrained in the language,
and the second one involves exploring hypotheses about how certain expressions
constrain temporal reference.
Cross-linguistic research has revealed that the temporal reference of clauses
may be constrained by tenses, temporal adverbials, embedding constructions, and
context. Section 3.1 illustrates how temporal reference restrictions can be explored
in matrix clauses, where the evaluation time is the utterance time. Section 3.2 then
turns to the temporal reference of subordinate clauses. Our discussion in this sec-
tion, as well as in sections 4 and 5, focuses on the kind of data that support hypoth-
eses about temporal and aspectual reference rather than on methods for obtaining
such data. One such method is judgment elicitation, but data may also be obtained
from (spoken and written) corpora and questionnaires (such as that used in Dahl
1985). For discussions of methods see Matthewson (2004), Krifka (2011), and the
chapters in this volume, especially that by Bohnemeyer.
3.1 Temporal reference in matrix clauses
The examples in (1) and (2) already illustrated that the temporal reference of
matrix clauses is constrained by context. The examples in (3) show that tempo-
ral adverbs also constrain temporal reference: whereas both (3a) and (3b) have
past temporal reference, as they address a contextually given question about a past
topic time, the temporal adverb yesterday in (3a) constrains the temporal refer-
ence of the clause to the daylong interval that precedes the day that contains the
utterance time, and the temporal adverb last year in (3b) constrains the temporal
reference of the clause to a yearlong interval that precedes the year that contains
the utterance time. We thus understand Mario’s writing to be temporally located
within the interval denoted by yesterday in (3a) and within the interval denoted
by last year in (3b).
(3) Context: When did Mario write an obituary? [TT < UT; ET ⊆ TT]
a. Mario wrote an obituary yesterday. [TT < UT; ET ⊆ TT; TT is yesterday]
b. Mario wrote an obituary last year. [TT < UT; ET ⊆ TT; TT is last year]
In some languages, temporal reference is also constrained by tenses. The past-

tense verb wrote in (3a) constrains the topic time of the clause in which it occurs
to a time prior to the utterance time. This constraint on the temporal location of
the topic time introduced by the tense is compatible with the constraints on the
temporal location of the topic time introduced by context and the temporal adverb
yesterday, rendering this an acceptable utterance. The example in (4), by contrast,
is judged to be unacceptable.7 We hypothesize that the unacceptability judgments
are due to conflicting constraints on the temporal location of the topic time: the
non-past tensed verb writes constrains the topic time to a time at or in the future
of the utterance time, which is incompatible with the constraints on the topic time
introduced by context and the temporal adverb.
(4) Context: What did Mario write?

#Yesterday, Mario writes an obituary.
In the following, we illustrate two diagnostics that are typically used to estab-
lish temporal reference restrictions in research with theoretically untrained native
speakers.
Diagnostic #1: Co-occurrence restrictions with temporally locating ad-

verbs The first diagnostic rests on the assumption that temporally locating
adverbs constrain the temporal location of the topic time, as illustrated with the
examples in (3). If the expression under investigation can co-occur with a particu-
lar temporally locating adverb, this can be taken as evidence that the expression is
compatible with the temporal reference restrictions imposed on the clause by the
adverb. Inability to co-occur with such an adverb can support the hypothesis that
the expression is incompatible with the temporal reference restrictions introduced
by the adverb (assuming that one has excluded other possible reasons for the un-
acceptability judgment).
We illustrate this diagnostic with the so-called present-tense verb form of
Standard High German. As illustrated by the example in (5), this verb form may
co-occur with the temporally locating adverbs im Augenblick “right now” and
7
We use # to indicate that the unacceptability of the utterance is hypothesized to be due to
semantic/pragmatic (rather than syntactic) reasons.
morgen “tomorrow”, but not with gestern “yesterday”. These data are compatible
with the hypothesis that the “present” tense verb form of Standard High German is
compatible with present and future temporal reference, but not with past temporal
reference (and is therefore better referred to as a non-past tense).
(5) Im Augenblick / Morgen /#Gestern arbeite ich an meiner

at.the moment tomorrow yesterday work.npst I at my
Dissertation.
thesis
“Right now I am / Tomorrow I will be / #Yesterday I was working on my thesis.”
The diagnostic can also be applied to explore temporal reference restrictions

of adverbs, such as Guaraní kuri. In the examples in (6), kuri co-occurs with the
verb a-jahu “A1sg-bathe”, which consists of a verb stem inflected only for person/
number information, and various temporally locating adverbs. The fact that kuri
can co-occur with kuehe “yesterday” in (6a), but not with ko’ãga “now” or ko’ẽro
“tomorrow” in (6b,c), provides support for the hypothesis that kuri restricts the
temporal reference of the clause in which it occurs to topic times that precede the
utterance time.
(6) a. Kuehe a-jahu kuri.

yesterday A1sg-bathe past
“Yesterday I bathed/was bathing.”
b. #Ko’ãga a-jahu kuri.
now A1sg-bathe past
(Intended: “I am bathing right now.”)
c. #Ko’ẽro a-jahu kuri.
tomorrow A1sg-bathe past
(Intended: “Tomorrow I am going to bathe.”)
However, as mentioned above, it is important to establish not only that a sentence
in which particular forms co-occur is judged to be unacceptable, but also why it
was judged to be so. Note, for instance, that (7), which differs from (6c) only in the
omission of kuri, is also judged to be unacceptable (in contrast, (6b) without kuri
is judged to be acceptable).
(7) #Ko’ẽro a-jahu.

tomorrow A1sg-bathe
(Intended: “Tomorrow I will bathe.”)
(6c) thus does not provide conclusive evidence that kuri is incompatible with
future temporal reference, since the unacceptability of that example may instead
be due to the incompatibility of ko’ẽro “tomorrow” with a-jahu “A1sg-bathe”.
Diagnostic #2: Contextually constrained temporal reference The second di-

agnostic relies on the assumption that the context in which a clause is uttered
constrains its temporal reference, as illustrated with the example in (2). Thus, if
an uttered clause is judged to be acceptable in a context that constrains the tem-
poral reference of the utterance to a particular time, this provides evidence that
the clause is compatible with that particular temporal reference. Unacceptability
of the utterance in such a context can support the hypothesis that the utterance is
incompatible with that particular temporal reference.
We can use this diagnostic to provide further evidence for the hypothesis that
the Standard High German “present”-tense verb restricts the temporal reference of
the clause to a non-past time. One way of contextually constraining the temporal
reference of an utterance is through a question that utterance is intended to answer
(as discussed in section 2.1), but assertions can also be used to constrain topic times
(see, e.g., example (10)). The answer utterance is thus contextually restricted to
present or future temporal reference in (8a), or to past temporal reference in (8b).
(8) a. Context: What are you doing right now/tomorrow morning?

Ich arbeite an meiner Dissertation.
I work.npst at my thesis
“I am / will be working on my thesis.”
b. Context: What did you do yesterday morning?
#Ich arbeite an meiner Dissertation.
I work.npst at my thesis
(Intended: “I worked on my thesis.”)
The observation that the answers in (8a) are judged to be acceptable, but not that
in (8b), further supports the hypothesis that the “present” tense form restricts the
temporal reference of the clause in which it occurs to non-past times.
We can also use this diagnostic to explore whether utterances of Guaraní
clauses with kuri are compatible with future temporal reference. The context in
(9a) restricts the temporal reference of the answer to times that precede the utter-
ance time (i.e., past temporal reference), whereas the context in (9b) restricts it to
future temporal reference. The answer utterance in (9c) is judged to be acceptable
in response to (9b), but not in response to (9a). These observations support the
hypothesis that (9c) is compatible with future temporal reference,8 but not past
temporal reference. The utterance in (9d) differs from that in (9c) only in the ad-
dition of kuri. This utterance is judged to be acceptable in response to (9a) but not
8
These data are also compatible, however, with the hypothesis that ko’ẽro “tomorrow” temporally
locates the eventuality time (see the first pitfall discussed below). In accordance with this hypothesis,
Tonhauser 2011a, b) provides evidence that –ta “prosp” is not a future tense, which temporally locates
the topic time in the future of the evaluation time, but a prospective aspect/modal, which temporally
locates the eventuality time in the future of the topic time.
in response to (9b). These findings suggest that (9d) is not compatible with future
temporal reference, but only with past temporal reference.
(9) a. Context 1: A music festival will take place tomorrow in the next town
over. Yesterday, I ran into Raul, who was very grumpy. Today, I ask his
mother: Why was Raul so grumpy yesterday?
b. Context 2: A music festival will take place tomorrow in the next town
over. I discuss with my friend how the organizers had to cancel a lot of
acts for financial reasons. I ask my friend: Who will still perform?
c. Raul o-purahéi-ta farra-há-pe ko’ẽro.

Raul A3-sing-prosp party-loc-at tomorrow
“Raul will sing at the festival tomorrow.”
d. Raul o-purahéi-ta kuri farra-há-pe ko’ẽro.
Raul A3-sing-prosp past party-loc-at tomorrow
“Raul was going to sing at the festival tomorrow.”
[TT < UT; TT > ET, ET within tomorrow]
Note that, to make this argument about (9d), it was crucial to establish not only
that this example is judged to be unacceptable in the context of (9b), and differs
from (9c) in this regard, but also that the example is judged to be acceptable in
some context (i.e., is grammatical). What (9d) conveys in the context of (9a) is that,
at the past topic time, Raul had the intention of singing at the festival; the utterance
implicates that the singing event will not take place. These examples thus provide
evidence that clauses with kuri are compatible only with past temporal reference.9
Potential pitfalls One potential pitfall of diagnostic #1 is that temporally
locating adverbs may temporally locate not only the topic time of the clause but
also its eventuality time. In (6c), for instance, ko’ẽro “tomorrow” might also specify
the time at which the event of bathing will take place (and this is the analysis ad-
vocated for in Tonhauser 2011a, b). Likewise, the temporal adverb on Sunday in
(10) temporally locates the event of Rick submitting his homework, whereas the
topic time is constrained by the preceding adverbial clause when I talked to him
on Monday:
(10) Rick had homework assignment yesterday (Wednesday). However, when I
talked to him on Monday, he told me that he had submitted it on Sunday
already.
Thus, co-occurrence patterns with temporally locating adverbs should ideally be
complemented with evidence from contextual reference restrictions.
9
Although kuri “past” restricts the temporal reference of the clauses it occurs in, it is not a tense,
given our definition of tense, since it is not paradigmatic: Guaraní clauses do not necessarily realize a
temporal adverb (see Tonhauser 2010, 2011a for discussion).
Another potential pitfall is a too-limited application of the diagnostics, re-

stricted to familiar temporal adverbs and adverbial constructions. Given exam-
ples like (7), for instance, one might presume that Guaraní verbs inflected only
for person/number information are incompatible with adverbs denoting future
times. But exploration of a wider range of adverbs and adverbial constructions re-
veals that this conclusion is premature, as discussed in Tonhauser (2011b). In (11),
for example, the matrix clause contains the verb a-juka “I kill” and the temporal
adverb ko ka’ aru “this afternoon”, which in the given context refers to the upcom-
ing afternoon:
(11) Context: It’s morning and the speaker is talking about a goose walking past
her and the addressee.
Ja’ú-ta-re ko gánso ko’ẽro, a-juka ko

a1pl.incl-eat-prosp-for this goose tomorrow a1sg-kill this
ka’arú-pe.
afternoon-at
“Since we are going to eat this goose tomorrow, I will kill it this afternoon.”
(Tonhauser 2011b:260)
In sum, empirically sound generalizations about temporal reference restrictions
emerge from consideration of a wide variety of examples in which context or tem-
poral adverbs constrain temporal reference.
3.2 Temporal reference of subordinate clauses
Subordinate clauses also have temporal reference (for discussion, see, e.g., Enç
1987; Ogihara 1995; Abusch 1997; Gennari 2003; Kusumoto 2005; Kubota et al.
2009; Smirnova 2009). And since expressions that constrain temporal reference
may also occur in subordinate clauses, a description of temporal reference in a
language is incomplete if only matrix clauses are considered. One of the ways in
which tenses across languages differ is brought out in subordinate clauses, in-
cluding clauses embedded under propositional attitude verbs, relative clauses,
and temporal adjunct clauses. Some tenses are absolute, which means that the
evaluation time relative to which they constrain the temporal location of the
topic time is the utterance time in both matrix and subordinate clauses. Other
tenses are relative, which means that the evaluation time relative to which they
constrain the topic time is the utterance time in matrix clauses but a time other
than the utterance time in subordinate clauses, typically the matrix clause even-
tuality time.
The contrast between absolute and relative tenses is illustrated with the
pair of examples in (12): both the English example in (12a) and the Japanese
example in (12b) convey, in the given context, that Anna was sick at the past
time of Ken’s report. In the English example, the verb in the subordinate clause
is marked with a past tense, whereas it is marked with a non-past tense in the
Japanese example.
(12) Adapted from Kubota et al. (2009:310)
Context: Anna was sick yesterday when Ken visited her. Immediately after
his visit, Ken told Sandra: “Anna is sick.” Earlier today, Sandra told her
mother:
a. Ken said that Anna was sick.
b. Ken-wa Anna-ga byooki da to it-ta. Japanese

Ken-top Anna-nom sick be.npst compl say-past
“Ken said that Anna was sick.”
This observation suggests either that the tenses in the two languages do not have
the same meanings, or that the tenses have the same meanings but that the two
languages differ in their syntax-semantics interfaces (see the references above for
proponents of the two proposals). On the former proposal, one might argue that
the English past tense in (12a) is an absolute tense, whereas the Japanese non-past
tense in (12b) is a relative tense.
The reading of (12a) is sometimes referred to as a temporally “overlapping”
reading. This reading is in contrast to a temporally “back-shifted” reading of
the same example in a different context, which conveys that Anna was sick at a
time prior to Ken’s saying. In Japanese, the backshifted reading is realized with
a (relative) past-tensed complement clause. The observation that, in English, a
past-tensed complement clause embedded under a past-tensed matrix clause
can lead to a temporally overlapping and a back-shifted reading has been taken
by some as a motivation for saying that English past tense is relative (like in
Japanese) and that the two languages differ at the syntax-semantics interface;
specifically, with respect to whether a “Sequence of Tense” rule is available (see
references above).
To determine whether a tense is relative or absolute, it is necessary to identify
whether its evaluation time in subordinate clauses is the utterance time, as can be
argued for the English example in (12a), or whether it may be some other time,
such as the matrix clause eventuality time, as in the Japanese example in (12b). For
a discussion of how different propositional attitude verbs affect the temporal refer-
ence of the subordinate clause, see Smirnova (2009).
Another example of a subordinating construction that may affect temporal
reference is temporal adjunct clauses with after and before, exemplified in (13).
These two expressions convey a temporal ordering between the eventualities de-
scribed by the matrix and the subordinate clauses.
(13) (Beaver and Condoravdi 2003:41)

a. Mozart died after he finished the Requiem.
b. Mozart died before he finished the Requiem.
Tenses also show intriguing behaviors in the antecedents of conditionals. For

example, as discussed in Kaufmann (2005), the example in (14a) with the present-
tensed verb submits, is “only felicitous under a special reading which includes an
element of ‘certainty’ or ‘scheduling’ ” (p. 232f.), whereas “[t]his connotation is
absent” when this clause realizes the antecedent of a conditional, as in (14b). Con-
ditional antecedents realized with past-tensed verbs give rise to counterfactual in-
terpretations; as illustrated in (15), the past-tensed antecedent clause need not have
past temporal reference (see, e.g., Iatridou 2000, Ippolito 2003 for discussion).
(14) (adapted from Kaufmann 2005:232)

a. Sam submits his paper to a journal.
b. If Sam submits his paper to a journal, we won’t include it in our book.
(15) If you bought it tomorrow, you would get a discount.
The interpretation of tenses in conditionals shows that explorations of temporal

reference quickly encroach on questions of modality and mood. We return to this
matter in section 5.
3.3 Tenseless languages and temporal reference in discourse
The observation that some languages make do without tense morphemes is entirely
compatible with what we have said above about the role tenses play in constrain-
ing temporal reference: in tenseless languages, temporal reference is constrained
only by optional temporal adverbials, context, and embedding constructions. In
addition to Guaraní, languages that have been described as tenseless and that have
received tenseless analyses include Yukatek (Mayan, Bohnemeyer 2002, 2009) and
Kalaallisut (Eskimo-Aleut, Bittner 2005, 2011). See Matthewson’s (2006) analysis
of St’át’imcets (Salish) as a “superficially tenseless” language.
As already mentioned in section 2.3, some syntactic and semantic frameworks
conceive of all languages as tensed, regardless of whether the language has tenses
and a tensed analysis is empirically motivated. In some Chomskyan frameworks,
for example, the T(ense) node is obligatory since the realization of subject noun
phrases is intimately tied to the specifier position of the Tense Phrase that T proj-
ects (e.g., Chomsky 1995). Likewise, under the assumption that the meaning of
a tense is needed in order for a sentence to denote a proposition (e.g., May 1991,
Partee 1992, as discussed in von Fintel and Matthewson 2008:157), a tenseless lan-
guage necessarily receives a tensed analysis.
Even outside the confines of such syntactic and semantic frameworks, it has
been proposed that all languages receive tensed analyses. An argument in favor of
such proposals is made on the basis of observations about how utterances in tensed
and tenseless languages are interpreted in discourse. Consider the discourse in
(16), which conveys that the events of Juan getting up, bathing, and eating break-
fast temporally occur in sequence, one after the other.
(16) Juan got up. He bathed and he ate breakfast.

The temporal interpretation of the discourse in (16) is generally assumed to be due
to the topic time of a clause being anaphorically dependent on the topic time of
preceding discourse (see, e.g., Partee 1984; Hinrichs 1986; Kamp and Reyle 1993).
Some authors attribute the context-dependency of English utterances to (the past)
tense, which is said to be anaphoric to the contextually salient topic time (e.g.,
Partee 1984). Other authors assume that the topic time is introduced by the func-
tional category T (Stowell 1996; Kratzer 1998). The observation that narrative dis-
courses in tenseless languages also exhibit temporal progression interpretations,
as illustrated for Guaraní in (17), is then taken to provide evidence that such lan-
guages have a T node that realizes (phonologically empty) tenses (e.g., Matthew-
son 2002) or introduces the topic time.
(17) Context: What did Juan do last Sunday?
O-pu’a, o-jahu ha o-rambosa.

A3-get.up A3-bathe and A3-breakfast
“He got up, bathed and ate breakfast.” (Tonhauser 2011b:264)
But, as discussed in Ritter and Wiltschko (2004), it is not necessary to assume that
the topic time is introduced by T. For example, Shaer (2003) argues that it is intro-
duced by the verb itself. For Bohnemeyer (2009), the topic time of an utterance
constitutes part of the interpretation of the utterance, regardless of whether what
was uttered was a tensed or a tenseless sentence. Thus, depending on the theoreti-
cal framework assumed, it is possible to provide tenseless (and T-less) analyses of
tenseless languages.
3.4 Interim summary
In the neo-Reichenbachian framework, theoretical terms such as “tense” and

“temporal reference” receive precise definitions, and the framework can thereby
serve as a guide to identifying evidence for a) whether a language has tenses, b) if it
has tenses, what kind of tenses it has, c) temporal reference in matrix and subordi-
nate clauses, and d) temporal reference in discourse. Thus, within this framework,
theoretically informed meaning descriptions, such as “The so-called German
‘present’ tense is a non-past tense,” that are useful for cross-linguistic comparison
can be established.
4 Exploring Aspectual Reference
As with temporal reference, there are two questions a fieldworker might attempt to
answer about aspectual reference in the target language. First, how can a particular
aspectual reference be expressed in the language? And second, what constraints do

particular expressions in the language impose on aspectual reference?
In section 3, we discussed four ways of constraining temporal reference:
context, adverbials, tenses, and subordinating constructions. Similarly, aspectual
reference can be constrained by context and by adverbial expressions, as well as
by paradigmatic forms whose purpose is to convey aspectual reference—that is,
grammatical aspects. Additionally, aspectual properties of the lexical content itself
can play a role in restricting aspectual reference. We first consider the effect of
grammatical aspects.
4.1 Effects of grammatical aspects on aspectual reference
Grammatical aspects are grammaticalized ways of encoding different relation-

ships between the topic time (TT) and the eventuality time (ET). In the discussion
below, we follow Comrie (1976) in capitalizing names of grammatical aspects in
particular languages (while leaving types of aspectual reference in lowercase).
An aspectual distinction that is commonly encoded morphologically is
that between perfective and imperfective aspectual reference. For the former, a
common definition is that the topic time temporally includes the eventuality time:
(18) Perfective aspectual reference: ET ⊆ TT10
Pulaar (Atlantic) is one language that has grammaticalized this contrast. Example
(19) illustrates a Perfective-marked Pulaar clause.11
(19) Haŋki Mariam def- ii. [TT < UT; ET ⊆ TT; TT is yesterday]
yesterday Mariam cook- pfv
“Yesterday Mariam cooked.”
To establish that -ii is a perfective aspect, we need two kinds of evidence. The first
is that clauses with -ii can be felicitously and truthfully used in contexts where
the eventuality time is included in the topic time. Sentence (19), for instance, can
be truthfully uttered in a context in which sometime during the preceding day,
Mariam cooked. The eventuality time is the time spanned by the event of Mariam
cooking, while the topic time is delimited by the adverb haŋki “yesterday”. Thus
the eventuality time is indeed included within the topic time.
Compatibility with a context where the topic time includes the eventuality
time is a necessary condition for perfective aspectual reference, but not a sufficient
10
Klein (1994) proposes a less restrictive version of perfectivity: TSit at TT (where Klein’s TSit is
equivalent to our ET and at represents temporal overlap). On that analysis, it is possible for only the
tail end (but not, according to Klein, only the front end) of the eventuality time to be included in the
topic time.
11
The Pulaar sentences were modified from examples in Fagerberg-Diallo (1983:248), then judged
by three native speakers as being truthful in the given contexts.
one. To ensure that -ii entails perfective aspectual reference, we need negative evi-
dence: namely, we must establish that clauses with -ii are not judged true in con-
texts where the eventuality time is not included in the topic time. Indeed, Mariam
def-ii is not a possible response to a question like “When you walked into the house,
what was Mariam doing?” In that context, the topic time is the moment at which
the addressee entered the house, and the questioner is asking for a description
of an eventuality whose time includes the topic time. Because it can only be used
when the eventuality time is included in the topic time, -ii is a perfective aspect.
If, in contrast, a certain form requires that the eventuality time include the
topic time, then that form entails imperfective aspectual reference. (See Deo 2009
for a more sophisticated, unified analysis of imperfective aspectual reference.)
(20) Imperfective aspectual reference: TT ⊂ ET
Imperfective aspectual reference is commonly divided into two subtypes: pro-

gressive and habitual (e.g., Bybee et al. 1994:151). To establish that a form is an
imperfective aspect, then, we need to determine that it can have both progressive
meaning and habitual meaning, depending on context.
In Pulaar, a single Imperfective form can indeed realize progressive or ha-
bitual aspectual reference. In (21), the initial clause sets up the topic time as the
time at which the speaker saw Mariam. The sentence can be truthfully uttered if at
that topic time, Mariam had begun cooking but had not yet finished—that is, the
cooking event was in progress at the topic time.
(21) Mi yiy- ii Mariam omo def- a. [TT < UT; TT ⊂ ET]

1sg see-pfv Mariam 3sg.impf cook impf
“I saw Mariam cooking.” (Literally: “I saw Mariam [while] she was cooking.”)
One might hypothesize that an expression that realizes progressive aspectual ref-
erence entails imperfective aspectual reference. To determine whether the hypo
thesis is empirically adequate, one needs a context that demands habitual aspectual
reference—meaning, roughly, that the eventuality time spans not a single event,
but a set of multiple events of the same type, which together temporally subsume
the topic time. If the hypothesized imperfective form is acceptable in this context,
then the hypothesis is sustainable; if not, the form is more likely only a progressive
aspect. Example (22) provides a relevant example for Pulaar.
(22) Context: The speaker is talking about a woman named Mariam. Because
she is the only woman of working age in the household, she is responsible
for cooking lunch and dinner every day.
Omo def- a ñande fof.[UT ⊆ TT; TT ⊂ ET]

3sg.impf cook- impf day every
“She cooks every day.”
In this example, the topic time is not specified; instead, it is implicitly assumed to
be some interval approximately equivalent to the time of utterance. (The adverb
“every day” does not denote the topic time, but instead specifies the frequency of
individual cooking events.) Thus there is a span of time during which a series of
events occurs, each one consisting of Mariam cooking something; this span of
time is the eventuality time, and it includes the topic time. Since the same con-
struction that realized progressive aspectual reference in (21) realizes habitual as-
pectual reference in (22), this construction appears to be an imperfective aspect.
Beyond inclusion, another temporal relationship between the topic time and
the eventuality time is precedence. Aspects that entail that the eventuality time fol-
lows the topic time (TT < ET) are called prospective aspects, and those that entail
that the eventuality time precedes the topic time (ET < TT) are sometimes called
perfect aspects. In English, prospective aspectual reference is conveyed by the be
going to construction, illustrated in (23).
(23) Max was going to watch a DVD tonight, but his DVD player broke.
Example (23) is true in a particular context if and only if the time at which the
DVD viewing had been planned to take place follows the past topic time, perhaps
the time at which Max decided on his program for the evening. Thus, the eventu-
ality time follows the topic time; this relationship between the topic time and the
eventuality time defines prospective aspectual reference.
(24) Prospective aspectual reference: TT < ET

Despite occasional assumptions to the contrary (e.g., Bybee and Dahl 1989, Dahl
and Velupillai 2011), be going to is not a simple future tense, as attested by its ability
to combine with auxiliaries entailing at least past and present temporal reference,
as in Louisa was going to eat/is going to eat. (See Schroeder 2011 for additional
arguments to this effect.)
Like the Prospective, the English Perfect—which consists of a tensed have
auxiliary followed by a participle—is compatible with multiple kinds of temporal
reference (e.g., Louisa had (already) eaten/ has (already) eaten/ will have (already)
eaten). Example (25) exemplifies the Perfect aspect with past tense.
(25) By the time I got to the station, my train had already left. [TT < UT; ET < TT]
Some researchers, including Klein (1994), define perfect aspectual reference as in
(26).
(26) Perfect aspectual reference: ET < TT

To establish how a language conveys the meaning in (26), one needs to set up a
context in which the eventuality time ends before the topic time begins. Sentence
(25), for instance, is felicitous and true in a context where the speaker arrived at the
train station at 3:00 p.m., but her train departed at 2:58 p.m. The initial clause of
(26) sets up the time of the speaker’s arrival as the topic time; the time of the train’s
departure is the eventuality time. In this context, where the eventuality time fully
precedes topic time, English uses the Perfect aspect.
The analysis in (26) works well enough for (25), which illustrates the exis-
tential reading of the Perfect: by topic time, there has been an event of the train’s
leaving. This analysis does not properly capture other instances of the Perfect,
however. In (27), for example, the eventuality time of the speaker living in Colum-
bus does not precede the topic time—which A’s question sets up as a time interval
surrounding the utterance time—but includes it.12 This reading of Perfect clauses
is called the universal reading. (For discussions of different readings of the Eng-
lish Perfect, see Kiparsky 2002 and Portner 2003.)
(27) A: Where are you living these days?
B: I live in Columbus. Actually, I’ve lived there since 2001.
[UT ⊆ TT; TT ⊆ ET]
Another problematic prediction of the analysis in (26) is that perfects should

be unable to combine with other aspects that entail different relationships between
the eventuality time and the topic time. This prediction is wrong, as illustrated in
(28), which combines the Perfect with the Progressive.
(28) The cat has been playing with that string for the last three hours.
To address these problems with the Neo-Reichenbachian analysis, a number of
alternative approaches to the Perfect have been proposed. These include current rel-
evance theories (e.g., Inoue 1979, Moens and Steedman 1988, Binnick 1991:100-104,
Bybee et al. 1994:61, and Smith 1997:107) and Extended Now theories (e.g., McCoard
1978, Bennett and Partee 1978, Dowty 1979, Iatridou et al. 2003, Portner 2003).
The analysis in (26) seems to be a better fit for what Bohnemeyer (2002) calls
terminative aspect in Yukatek, which contributes the information that the relevant
eventuality is terminated at the topic time. In example (29), the clause that is marked
with the terminative aspect ts’o’k “term” implies that at the topic time, which is es-
tablished by the initial clause, the event of the child’s dying has already terminated.
(29) K-u k’uch-ul-o’b-e’, ts’o’k u kim-il le chàampal-e’.
impf-A.3 arrive-inc-top term A.3 die-inc def small:child-D3
“(By the time) they arrived, the baby had already died.”
(Bohnemeyer 2002:284)
12
Note that in (27), we allow for the possibility of non-proper inclusion of the topic time by the
eventuality time—that is, the possibility that the topic time may be coextensive with the eventuality
time. This prediction is validated by the following twist on (25):
(i) On his seventh birthday, Michael moved to Michigan. Up to that time, he had lived in Mississippi.
Here, the topic time for the second sentence is set up as Michael’s entire lifetime from birth until his
seventh birthday. The sentence asserts that the state of Michael’s living in Mississippi endured from the
time of his birth until he moved to Michigan on his seventh birthday. Thus, the time of the Michael-
living-in-Mississippi eventuality is coextensive with the topic time.
In sum, a grammatical aspect is better analyzed as contributing terminative as-

pectual reference than perfect aspectual reference if it does not allow for universal
readings.
The grammatical aspects discussed thus far map to five different kinds of as-
pectual reference: perfective, imperfective, prospective, perfect, and terminative.
Not every clause in every language is marked morphosyntactically with a particu-
lar aspect. Nonetheless, as was illustrated in example (2), and as we will discuss
further in section 4.2 below, every clause uttered in a given context has some as-
pectual reference—just as finite clauses without tense marking have a temporal
reference (see section 3.3 above). A given language may have more or fewer (gram-
matical) aspects than the five discussed above. It might, for instance, have separate
progressive and habitual aspects, rather than a general imperfective. On the other
hand, for a particular aspectual reference in a particular language, there may or
may not be a grammatical aspect that entails that aspectual reference.
Cross-linguistically, it is common for an expression to constrain not only as-
pectual reference, but also modal and/or temporal reference—that is, to refer to
possibility/probability/necessity, or to the relationship between the topic time and
the utterance time (or some other evaluation time). We now illustrate some ex-
pressions that restrict a combination of aspectual, temporal, and modal reference.
Aspectual/modal forms
Any time that an affirmative declarative clause that contains a certain expres-
sion does not entail full realization of the eventuality it describes, that expres-
sion must be suspected to describe possible, rather than actual, eventualities. The
most famous case of this is arguably the imperfective paradox. Many authors,
starting with Dowty (1977, 1979), have noted that with some sentence pairs, as in
(30), the Past Progressive (a subtype of imperfective) entails Simple Past (with
a perfective reading), but with others, like those in (31), this entailment relation
does not hold.
(30) a. Harry was flying on his broomstick.

b. Harry flew on his broomstick.
(31) a. Petunia was knitting a sweater for Dudley.
b. Petunia knit a sweater for Dudley.
The imperfective paradox can be explored in other languages by pairing an imper-

fective or progressive sentence with its perfective equivalent, then testing whether
the first entails the second.
This puzzle involves a property of clauses that we discuss in section 4.3
below—namely lexical aspect. Specifically, a Past Progressive entails its Simple
Past counterpart with activities—which describe only a process with no logical
endpoint—but not with accomplishments—which describe a process leading up
to a logical endpoint (e.g., the existence of a sweater in (31)).
To solve the paradox, many authors have argued for a modal approach: Eng-
lish Progressive sentences entail that, at the topic time, an eventuality has begun,
and that, at a time in the future of the topic time, this eventuality is likely to de-
velop into one of the type described by the sentence. However, if the eventuality is
an accomplishment, there is no guarantee that its logical endpoint is ever reached
in the actual world (so Petunia, for instance, may never complete the knitting of
the sweater). For analyses along these lines, the reader is referred to Dowty (1977,
1979), Landman (1992), and Portner (1998).
Prospective aspectual reference, too, is commonly intertwined with modal
reference. The modal properties of the English Prospective be going to, in particu-
lar, are clearest with past temporal reference, where the construction implicates
non-realization of the eventuality (much like the Guaraní construction in (9d) in
section 3.1). Use of (32a), for instance, implicates that the speaker’s singing plans
have been changed. Evidence that the non-realization of the speaker’s singing is
only an implicature, not an entailment, comes from the fact that this inference can
be canceled, as illustrated by the continuation in (32b).
(32) a. I was going to sing at the opera yesterday.
b. I was going to sing at the opera yesterday, and, in fact, I did!
The interpretation of (32b) and its Guaraní counterpart in (9d) provide fur-
ther evidence that the English be going to construction and the Guaraní suffix –ta
“prosp” encode a modal meaning in addition to prospective aspectual reference
(TT < ET). After all, if, for example, (32a) only conveyed past temporal prospective
aspectual reference (where the topic time is temporally located prior to the utter-
ance time, and the eventuality time temporally follows the topic time but is tem-
porally located within the denotation of yesterday), then (32a) would entail that the
speaker sang yesterday. But despite the possible continuation in (32b), (32a) does
not entail event realization, as illustrated by the fact that it can be continued with
. . . but I couldn’t because I was sick (and likewise for the Guaraní counterpart, as
discussed in Tonhauser 2011a).
Aspectual/temporal forms
The (so-called) English bare Present tense restricts temporal reference to non-past
topic times. But the form’s meaning is not only temporal (contrary to what its
name might suggest): in addition, it imposes habitual aspectual reference.13 (33),
13
This verb form is compatible with other temporal and aspectual references in so-called histori-
cal present or sportscaster’s present uses (see Comrie 1985:37 for discussion) and in news headlines that
assert the occurrence of an event in the recent past:
(i) McCain Says Decree by Egypt’s Mursi Is “Unacceptable.” (Bloomberg News, November 27, 2012).
Context: Headline of a story about a statement that Senator John McCain made in reaction to
a decree by President Mursi of Egypt. http://www.bloomberg.com/news/2012-11-25/mccain-says-
decree-by-egypt-s-mursi-is-unacceptable-.html.
for example, would be an appropriate way of talking about Eleanor’s habitual be-
havior toward Brussels sprouts.
(33) Eleanor eats Brussels sprouts.
Evidence that this verb form does not entail imperfective aspectual reference is
that it is incompatible with a progressive interpretation: (33) is unacceptable in a
context in which Eleanor is gingerly nibbling at the first Brussels sprout she has
ever tasted. Thus, expressions that constrain both temporal and aspectual refer-
ence constrain both the relation between the topic time and the eventuality time,
and the relation between the topic time and the evaluation time.
Expressions can also simultaneously constrain temporal, aspectual, and
modal reference. For discussion, see, for example, Cipria and Roberts (2000),
Ippolito (2004), and Hacquard (2006).
4.2 Effects of context and adverbials on aspectual reference
Context can also constrain aspectual reference, either in addition to or in place of

overt aspect markers. We can see this effect with the English simple past (“Pret-
erite”), which lacks aspectual marking and can receive both a perfective and an
imperfective interpretation. In the context of A’s utterance in (34a), B’s statement
receives a perfective interpretation: B biked to school, and the riding-to-school
event was fully contained within the day of utterance. In the context of A’s ut-
terance in (34b), however, the same statement receives a habitual interpretation:
riding-to-school events recurred regularly.
(34) a. A: What did you do for exercise today?
B: I biked to school.
b. A: I got my first car when I turned 16. After that, I drove to school every
day.
B: I never had a car in high school. I biked to school.
Aspectual reference can also be constrained by adverbials. In English, differ-
ent types of adverbial clauses (e.g., a time measure phrase preceded by for versus
in) restrict aspectual reference in different ways. Hence Ellen played the sonata
receives a perfective interpretation in (35a)—since the sonata is completed within
the hour-long topic time—but an imperfective interpretation in (35b).
(35) a. Ellen rehearsed the sonata in an hour.
Interpretation: Within a time interval lasting one hour, Ellen played the
entire sonata from beginning to end. The sonata cannot be more than
one hour long (at Ellen’s pace).
b. Ellen rehearsed the sonata for an hour.
Interpretation: Within a time interval lasting one hour, Ellen played
part(s) of the sonata. The entire sonata may or may not be more than
one hour long.
If context or an adverbial constrains the temporal relation between the topic

time and the eventuality time for a particular clause, and an aspectual marker is
acceptable in that clause, then we can conclude that that aspect is compatible with
the aspectual reference required by the context or the adverbial. In the Badiaranke
example in (36), for example, context and the adverbial clause together require a
perfective interpretation.
(36) Context: Aamadu and Binta began writing a letter at 1:00 and finished
writing the letter at 2:00.
birĩ 1:00 haː 2:00 safiŋә- b� de leːtar.
since 1:00 until 2:00 write- 3pl aff.decl letter
“Between 1:00 and 2:00, they wrote a letter.”
The sentence in (36) contains a verbal stem (safiŋ- “write”),14 followed by a subject
agreement suffix (-bə̃ ‘3pl’), followed by a particle, de, that appears in affirmative
declarative clauses. From the acceptability of the construction in the given con-
text, we can conclude that the construction is compatible with perfective aspectual
reference. Sentence (36) is not acceptable, however, in a context where the letter-
writing began before 1:00 and ended after 2:00, so the construction is not compat-
ible with imperfective aspectual reference.
4.3 Effects of lexical aspect on aspectual reference
In (36), only a (past) perfective interpretation is available for safiŋə-bə̃ de leːtar

“they wrote a letter”. The same is true of the examples in (37)–(38), which employ
the same construction (verb + subject agreement + de): example (37) is not ap-
propriate if Mari is still laughing, and (38) cannot be uttered truthfully until the
stick-breaking is done.
(37) Context: Mari, a child, has just laughed after catching sight of the
household cat. She is no longer laughing.
Mari das- ə̃ de.
Mari laugh- 3sg aff.decl
“Mari laughed.”
(38) Context: The speaker has just seen a video in which a person snaps a
wooden stick in half.
kutt- � doko sẽ de.
break- 3sg stick det aff.decl
“He broke the stick.”
14
The [ə] after [safiŋ] in (36) is epenthetic.
In (37)–(38), as in (36), the construction in question describes events completed

during a topic time in the immediate past. In (39)–(40), however, the same con-
struction receives a present imperfective interpretation: the states described are
ongoing at the topic time, which here overlaps with the utterance time. Example
(39) is a claim about the way most Africans look in general, including at the utter-
ance time; (40) is an exchange about B’s current state of health.
(39) baː Afrik bajə- bə̃ de.

people.of Africa be.black- 3pl aff.decl
“Africans are black.” (Cover 2010:67)
(40) Context: The speakers are exchanging greetings.
A: kẽdan- iː de ba? B: haː kẽdan- � de.
be.healthy- 2sg aff.decl q yes be.healthy- 1sg aff.decl
“Are you healthy?” “Yes, I’m healthy.”
The data in (36)–(40) pose a puzzle: Is there any semantic property shared by
(36)–(38), but not by (39)–(40), that correlates with the observed divide in aspec-
tual reference?
The key difference is that the sentences in (36)–(38) describe events, while
those in (39)–(40) describe states. States are eventualities that are essentially un-
changing. For linguistic (not evolutionary) purposes, the skin color of Africans
is a stable property; while the state in (40) is more ephemeral; B’s physical con-
dition is fundamentally the same at any point during the time that she or he is
healthy. Events, in contrast, evolve over time: different sounds are emitted during
the laughing event, the stick changes from unbroken to broken through the break-
ing event, and so on.
The state–event distinction belongs to the domain of lexical aspect, also
known as Aktionsart, propositional aspect (Verkuyl 1993), or situation type (Smith
1997). None of the commonly used terms are quite ideal: Aktionsart overempha-
sizes “actions,” to the exclusion of other eventuality types; lexical aspect wrongly
suggests that the property in question pertains only to individual lexical items,
and cannot emerge compositionally; propositional aspect goes to the other extreme
by implying that only full propositions have the property; situation type uses the
term situation, which means something different in formal semantics, to refer to
eventualities. We will use the term “lexical aspect,” while noting that the domain
of this property is not limited to individual lexical items. Rather, the word “lexical”
here means that the aspectual properties in question arise from lexical items (in
the absence of tenses and grammatical aspects) and the ways they combine—that
is, lexical aspect is compositional. “Lexical aspect” stands in contrast to the “gram-
matical aspects” discussed above (perfective, imperfective, etc.).
The idea of lexical aspect is that the lexical content of a sentence bears some
inherent aspectual properties, distinct from those introduced by grammatical
aspect. The lexical content, sometimes called the “untensed proposition” of a sen-
tence, is the semantic content contributed by the lexical items themselves, what is
left when the tense and grammatical aspect are stripped away (Klein 1994:1–12). As
with lexical aspect, lexical content is a property not only of isolated lexemes, but
also of their combination. Thus in a sentence like Jim squished a bug, the lexical
content describes a Jim-squishing-a-bug eventuality; the grammatical component
contributes the temporal and aspectual information. Similarly, in (41a.i) below,
the -s on loves is not part of the lexical content, since it marks present temporal
reference; the lexical content of the sentence is <Mary loves cupcakes>, which de-
scribes a Mary-loving-cupcakes eventuality.15
Five lexical aspect categories are typically recognized, namely states, activi-
ties, accomplishments, achievements, and semelfactives. (All the types that are
not states—i.e. activities, accomplishments, achievements, and semelfactive—are
events.) English sentences exemplifying each are given in (41).
(41) Sentences illustrating different kinds of lexical aspect
a. States:
i. Mary loves cupcakes.
ii. Millicent was happy.
b. Activities:
i. The monkey danced.
ii. Mildred and George were chatting animatedly on the phone.
c. Accomplishments:
i. The children are building a sandcastle at the beach.
ii. This bridge freezes over in the winter.
d. Achievements:
i. My train will arrive at 2:13.
ii. Wilfred’s pet tarantula died.
e. Semelfactives:
i. He knocked once, loudly.
ii. Nancy blinked innocently.
See Bar-el (this volume) for an in-depth discussion of lexical aspectual classes and
their cross-linguistic properties.
As we saw in (36)–(40), lexical aspect sometimes constrains aspectual refer-
ence. So does grammatical aspect, as discussed in section 4.1. These two influences
are not independent of one another: in English, as in Badiaranke, the meaning
and even acceptability of a given grammatical aspect is often dependent on lexi-
cal aspect. The English Progressive, for instance, is frequently claimed to be in-
compatible with stative lexical contents (e.g., #Kim is knowing French; see, e.g.,
15
We follow Klein (1994) in using angled brackets to distinguish lexical content from fully in-
flected sentences.
Vendler 1957, Comrie 1976). To evaluate this claim, we need to create, and assess
the meaning of, Progressive-marked sentences in which the lexical content is
stative. The examples in (42) meet this description.
(42) a. You’re being ridiculous.
b. I’m loving this veggie burger.
The two sentences in (42) are judged grammatical by native speakers. Sentence
(42a) can be aptly addressed to someone who is exhibiting ridiculous behavior,
whether or not she or he is ridiculous in general, and (42b) can be said while con-
suming a delicious burger. It is not quite right, then, that stative lexical contents
disallow Progressive aspect. A better characterization, suggested by the data in
(42), is that the Progressive coerces statives into the reading that the states are
time-delimited and/or manifested in some specific behavior (cf. Michaelis 2003).
Similarly, aspectual coercion occurs when the Progressive occurs with a se-
melfactive predicate, forcing the lexical aspect to shift to an activity made up of
multiple occurrences of the semelfactive. In (43), the only reading is that at the
moment when Kelly woke up, a series of knocks was ongoing.
(43) Kelly woke up with a start. Someone was knocking loudly at the door.
While in Badiaranke the distinction between stativity and eventivity is key
for aspectual interpretation, Bohnemeyer and Swift (2004) argue that in some lan-
guages, a similar divide exists between telic and atelic sentences. Telicity is the
property of having an inherent logical endpoint (e.g., the logical endpoint of the
lexical content of Owen run three miles is the three-mile mark; the logical endpoint
for The mirror shatter suddenly is a transition from an intact mirror to a broken
one). According to Bohnemeyer and Swift (2004), in German, Russian, and In-
uktitut, clauses unmarked for aspect receive a perfective interpretation by default
when their compositionally derived lexical content is telic, but an imperfective
interpretation when it is atelic. In the Inuktitut data in (44), taken from Bohne-
meyer and Swift (2004:267), neither clause is marked for aspect or tense; the telic
sentence in (44a) receives a past perfective reading, while the atelic one in (44b)
receives a present imperfective reading (the walking is ongoing at a topic time
including the utterance time) in an out-of-the-blue context.
(44) a. ani- juq.

go.out- par.3sg
“S/he went out.”16 [ET ⊆ TT; TT < UT]
b. pisuk- juq.
walk- par.3sg
“S/he is walking.” [TT ⊂ ET; UT ⊂ TT]
16
The “participial” is “the standard indicative mood in [this dialect of Inuktitut]” (Bohnemeyer
and Swift 2004:267).
Due to the ability of lexical aspect to affect aspectual reference, one cannot
generalize about the aspectual reference of particular forms on the basis of the
meaning of a small, random selection of sentences; instead, exploring the aspec-
tual reference of some construction necessitates consideration of a systematic
array of sentences representing all types of lexical aspect. (This systematic array
need not be presented in a boring manner; see Louie, this volume.)
In seeking to explore the full range of lexical aspects, language-specific di-
agnostics for lexical aspect must be applied—language-specific, because certain
classic tests for English, such as the for an hour/ in an hour test for telicity (Vendler
1967, e.g.; see (35) above), simply fail in other languages (including Badiaranke,
which lacks the prepositions to express this distinction). Meanwhile, diagnostics
that work for other languages might well not work for English. Badiaranke is just
one of many African languages in which a single construction gets, by default, a
past perfective interpretation when the lexical content is eventive, but a present
imperfective interpretation when the lexical content is stative (as we saw in (36)–
(40)). Welmers (1973) calls such constructions “factatives”. In such languages, a
lexical content whose stativity is unclear can be tested by placing it in the construc-
tion in question, then checking whether or not the clause can be understood as a
statement about an eventuality that is ongoing at speech time.
Crucially, when a sentence is translated to another language, lexical aspect
might well change. To figure out the aspectual properties of a particular lexical
content in the target language, we can begin with reasonable working assump-
tions based on what we know about other languages, while remaining open to
the possibility that these assumptions are wrong. These working assumptions will
help us develop language-specific diagnostics for lexical aspect; at the same time,
the language-specific diagnostics help us to test the working assumptions. In Ba-
diaranke, for instance, the above generalization about factative meaning emerges
from a consideration of a large number of simple sentences with different predi-
cates. In simple sentences including only the predicate, subject agreement, and de,
predicates including those in (45) all behave like those in (39)–(40) with respect to
their temporal and aspectual interpretation:17
(45) ka-dʒinn-e “be red”, ka-niŋanaː-e “be happy”, ka-kab-e “know”,
ka-lafiŋieːn-e “want”, ka-roːm-e “be short” . . .
In contrast, predicates including those in (46) trigger a past perfective reading in
the same construction:
(46) ka-kam-e “dance”, ka-seːt-e “talk”, ka-waj-n-e velo “repair a bike”, ka-rədd-e
paːdə “build a room”, ka-nij-e kaːs “break a glass”, ka-wub-e “cough”, ka-
peredʒ-e “blink” . . .
17
The Badiaranke predicates are given here in the infinitival form; there is no morphological
equivalent to English bare stems.
All the predicates in (45) denote properties that remain stable for some non-
momentary interval—that is, states; those in (46), in contrast, denote eventualities
that inherently involve change, that is, events. With exposure only to the predi-
cates in (45)–(46), then, we could hypothesize that the difference in temporal and
aspectual interpretation correlates with a difference between stative and eventive
lexical contents. (This hypothesis would be disproven if, for instance, other predi-
cates turned out to meet the semantic requirements for eventivity, yet received a
present imperfective interpretation in this construction.)
Armed with this working hypothesis, we could use the construction in ques-
tion to determine the aspectual class of a lexical content whose categorization is
uncertain. For instance, it is not immediately obvious how Badiaranke will catego-
rize a predicate like ka-datta:-e, which means “sleep.” In English, the verbal form
(as opposed to the adjectival form asleep) is treated as an activity: (47a), but not
(47b), can felicitously and truthfully be used to assert that at the utterance time,
the children are not awake.
(47) a. The children are sleeping.
b. #The children sleep.
The question is, when it is placed into the factative construction, as in (48), can
the Badiaranke counterpart of sleep receive a present imperfective interpretation,
or not?
(48) bepoːse pã dattaː- b� de.

children det sleep- 3pl aff.decl
“The children are asleep.”
The answer is “yes”: (48) can be felicitously and truthfully used to assert that a
group of children is asleep at the utterance time. Thus we can conclude that the
Badiaranke verb dattaː- “sleep” is stative, unlike its English verbal counterpart.
This process of making hypotheses about how different lexical aspectual
classes behave and figuring out which aspectual class a particular lexical content
exemplifies is not a linear one. Instead, there is a constant feedback loop: both the
hypotheses about diagnostics for different aspectual classes and the conclusions
about the aspectual class of a particular lexical content inform one another and are
subject to change.18
Thus far, we have treated lexical aspect as a property of lexical content. Sepa-
rating the lexical content from the sentence in which it appears allows us to talk
about the effect of grammatical aspects on different types of lexical content. The
effect of Progressive aspect on statives, as in (42) above, and the imperfective para-
dox illustrated in (30)–(31) are cases in point.
18
We thank Lisa Matthewson for helping to clarify this point.
Lexical content is not, however, the only level at which aspectual classes can be
defined. As pointed out by Dowty (1986), it is equally important to consider the aspec-
tual properties of entire sentences (fully inflected with tense and aspect). Dowty uses
the same terminology—states, activities, and accomplishments/achievements19—as
do treatments based on lexical content, but defines them at the sentence level. He
defines aspectual classes of sentences on the basis of the subinterval property: if a
sentence is true at an interval i, it is a state if it is also true at all subintervals of i, an
activity if it is true at subintervals of i down to a certain size, and an accomplishment/
achievement if is true at no subinterval of i (Dowty 1986:42).
From the perspective of Dowty (1986), Progressive sentences are inherently
stative (also Smith 1997): all Progressive-marked sentences adhere to the strictest
version of the subinterval property (i.e., they are true at any instant during the
interval at which they are true).20 Thus (41b.ii) and (41c.i), which were classified
above as containing activity- and accomplishment-type lexical content, respec-
tively, are both stative at the sentential level.
This analysis accounts for certain similarities between Progressive sentences
(derived statives) and sentences whose lexical content is stative (lexical statives).
Neither Progressive sentences nor non-Progressive stative sentences move the
topic time forward in time, unlike eventive sentences with perfective aspectual
reference (see the discussion in section 3.3). In (49a), for instance, the topic time
for He shuddered is understood to be later than that of A snake fell down on him, so
that the shuddering event is understood to follow the snake-falling event. In (49b),
the second clause is a lexical stative and, in (49c), it is a derived (progressive)
stative. In both of these examples, the shuddering events are understood to tem-
porally overlap with the eventualities described in the preceding clauses, rather
than following them.
(49) a. Indiana Jones opened his eyes. A snake fell down on him. He shuddered.
b. Indiana Jones opened his eyes. The pit he was in was full of snakes. He
shuddered.
c. Indiana Jones opened his eyes. Snakes were falling down on him. He
shuddered.
A similar stative property holds of Perfect sentences. In (50), Indiana Jones is still
awake (in the result state of having woken up) when he shudders.
19
Dowty (1986:42–43) treats accomplishments and achievements as a single aspectual class of
“telic” sentences.
20
Jürgen Bohnemeyer (personal communication) brings to our attention the dilemma of whether
imperfectives and progressives are inherently stative, or whether instead statives are inherently imper-
fective. There certainly are such things as perfectives of statives, as in I was happy today from 2 p.m. to
4 p.m., but depressed the rest of the time—but the inherent stativity of imperfectives or imperfectivity
of statives is a thorny issue. The solution to this chicken-and-egg problem will depend on a number of
theoretical assumptions about aspect—including the extent to which grammatical and lexical aspect
should be separated at all—and we will not be able to resolve the dilemma here. The reader is referred
to Sasse (2002) for a very thorough discussion of the divergent perspectives on this issue.
(50) Indiana Jones opened his eyes. He had awoken from a nightmare about
snakes. He shuddered.
Thus, depending on one’s research focus, it can be important to look both at
the aspectual properties of the lexical content itself, and of the aspectual class of
the fully inflected sentence.
4.4 Summary
We have seen in this section that aspectual reference may be constrained by con-
text or by adverbials. It is also possible for aspectual reference to arise by default
from lexical content. And of course, languages have dedicated aspects, and/or
combined tense/aspect/modal forms. Note that although we have attempted to
isolate these factors for clarity of exposition, in many examples aspectual reference
will be constrained by some combination thereof.
5 The Linguistic Realization of Future Discourse
This section is concerned with ways in which languages allow their speakers to talk
about the future. We refer to this as future discourse, that is, discourse about even-
tualities that are temporally located in the future of the utterance time (UT < ET).
Future discourse may be realized through future temporal reference (UT < TT), as
discussed below, but is distinct from it.
It is well established that across languages past discourse (ET < UT), that is,
discourse about eventualities that are temporally located in the past of the utter-
ance time, can be realized through past temporal reference, as discussed in section
3; through (present or past temporal) perfect aspectual reference; as discussed in
4, or through a combination thereof (e.g., past temporal perfect aspectual refer-
ence). In contrast to past discourse, future discourse is intimately connected with
modality since the future, in contrast to the past, is necessarily unknown and in-
determinate. As a consequence, future discourse is inherently modal, in the sense
of being about possible worlds that characterize ways in which the world may
develop. We have to be careful, however, to distinguish the inherent modality of
future discourse from the meanings coded by natural language expressions used
to realize future discourse: despite the modal nature of future discourse, linguistic
expressions used to realize future discourse need not themselves have a modal
component to their semantics.
5.1 Grammatical Aspects That Realize Future Discourse
As discussed above, prospective aspect markers in combination with present tem-

poral reference realize future discourse by temporally locating the eventuality time
in the future of the (present) topic time (TT< ET). The English and Guaraní pro-
spective aspect markers have a modal meaning component in addition to the as-
pectual one, since eventuality realization is not entailed even when the eventuality
time is temporally located prior to the utterance time, as in examples (9) and (32b).
Another example of a prospective aspect is Scottish Gaelic a’ dol a, accord-
ing to Reed (2012). As shown in (51), a’ dol a is compatible with past, present, and
future tense, which supports the hypothesis that it does not constrain the temporal
relation between the topic and the utterance time (i.e., is not a tense). That a’ dol
a requires the topic time to precede the eventuality time (making it a prospective
aspect) is illustrated in (51b), where the only acceptable reading is that “noon” de-
limits the topic time, placing it in the past of the utterance time.
(51) (Reed 2012:6)

a. Bha/tha/bithidh Calum a dol a phòsadh Mairi.
be.past/be.pres/be.fut Calum a’ dol a marry.par Mairi
“Calum was/is/will be going to marry Mairi.”
b. Aig meadhon-latha bha Calum a dol a phòsadh Mairi, ach aig

at mid-day be.past Calum a’ dol a marry.par Mairi but at
uair gabh e an t-eagal.
hour take.past 3sg.masc the.sg.masc fear
“At noon Calum was going to marry Mairi, but at 1 he got scared.”
(Reed 2012: 16)
The prospective aspect is not the only aspect that can be used to realize future
discourse cross-linguistically. In Badiaranke, statements about future times re-
quire the same aspectual/modal construction that is used to express imperfec-
tive aspectual reference (progressive and habitual); Cover (2010, 2011) calls this
the Imperfective construction. Badiaranke Imperfective clauses consist, at mini-
mum, of a verb stem preceded by an aspectually-conditioned subject agreement
prefix. In (52), the (bold-faced) Imperfective-marked clause asserts that the sing-
ing eventuality is ongoing at the topic time (progressive aspectual reference); in
(53), raining events recur regularly throughout the topic time (habitual aspectual
reference). But in (54), the singing eventuality is predicted to occur within a future
topic time (perfective aspectual reference, future temporal reference). Note that
although this construction is called Imperfective, the aspectual reference of (54) is
in fact perfective (the running event will not take up the entire topic time denoted
by “tomorrow”).
(52) Context: A child walked home from school, singing all the way. The
speaker caught sight of the child as she walked along.
dʒeːnә- mãnã de mpə- tʃimə (pәraŋe).
see- 1sg.3sg aff.decl 3sg.impf- sing going
“I saw her singing (as she went along).”
(53) Context: The conversation takes place in 2006. In 2004, the speaker visited
the north of Senegal, known as the Fuuta. During her stay, it rained once a
week.
paːrə peːs u- jak- akõ fe Fuuta wẽ loːkũ woː

two.years.ago nmlz.sg- be- past.1sg p Fuuta det week every
mpi- dʒafə de pakkã.
3sg.impf- rain aff.decl one
“Two years ago when I was in the Fuuta, every week it used to rain once.”
(54) Context: A certain individual is planning to run two miles the next day.
kũpia kilomeːtr maːe mpə- kər.

tomorrow kilometer two 3sg.impf- run
“Tomorrow it’s two kilometers he’ll run.”
Use of a single form—optionally or obligatorily—for imperfective aspectual

reference and future temporal reference is also attested in other West African
languages, including Pulaar, Wolof (Nussbaum et al. 1970:360), Balanta (Fude-
man 1999:94), Kisi (Tucker Childs, personal communication), and Mani (Tucker
Childs, personal communication). Indeed, as observed by, for example, Dowty
(1977) and Copley (2002), even the English Progressive has a futurate use:
(55) Context: The speaker has just been asked to go to a movie later in the
evening and is now scrambling for an excuse to decline.
Sorry, I can’t—I’m washing my hair tonight.
See, e.g., Bohnemeyer (2002:ch. 6.2.2) and Bittner (2005) for a discussion of
aspect/modal markers in Yukatek and Kalaallisut, respectively, that can realize
future discourse.
5.2 Mood and Modality Can Realize Future Discourse
As discussed in Bittner (2005), one strategy for realizing future discourse in Ka-
laallisut is through mood-marking expressions that convey that the speech act is
a request or wish:
(56) Qimmi-t nirukkar-niar-tigik.

dog-pl feed-please-imp.1pl.3pl
“Let us feed the dogs, OK?” (Bittner 2005:353)
Expressions that encode modal meanings can also convey future discourse,
as illustrated with the Guaraní examples in (57) with the existential modal-ne
“might”. The time of the raining eventuality described in (57a) is temporally
located within the denotation of ko’ ẽro “tomorrow”, that is, in the future of the
utterance time. The example conveys that this eventuality is an epistemic possibil-
ity from the perspective of the utterance time. Sentences with –ne “might” are
merely compatible with an interpretation in which the eventuality time tempo-
rally follows the utterance time. That they do not entail such an interpretation is
illustrated with (57b), in which the eventuality time temporally overlaps with the
utterance time.
(57) Adapted from Tonhauser (2011a:210)

a. Ko’ẽro o-ky-ne.
tomorrow A3-rain-might
“It might rain tomorrow.”
b. Context: A family is discussing who might be disrespectful to them. The
father says to the daughter:
Nde rei-kuaa-ne, che memby!
you A2sg-know-might my child
“You might know, my child!”
For discussions of the interaction between modality and temporality see, for ex-
ample, Thomason (1984, 2007), Condoravdi (2002), Kaufmann (2005), Kaufmann
et al. (2006), Hacquard (2010) and references therein.
5.3 Future tense
In language description, an expression that is used to realize future discourse is

often called a “future tense” or, simply, a “future.” For instance, in the Guaraní
grammar tradition (e.g., Gregores and Suárez 1967, Liuzzi and Kirtchuk 1989; Zar-
ratea 2002), the suffix –ta “prosp” is typically considered a future (tense). But, as
the discussions in the preceding sections have made clear, not every expression
that realizes future discourse is a future tense, that is, an expression that restricts
the temporal location of the topic time to times that follow the utterance time.
To determine whether a linguistic form is a future tense, it needs to be estab-
lished whether use of that form entails that the topic time follows the utterance
time. To distinguish, for instance, between a future tense (UT < TT) and a pro-
spective aspect (TT < ET), the acceptability of the form in a context in which one
of these conditions is met but the other one is not needs to be explored. We expect
that a future tense should be acceptable in contexts in which the topic time fol-
lows the utterance time, even if the eventuality time does not follow the topic time
(but instead includes it, or is included in it). On the other hand, we expect that a
prospective aspect is acceptable in contexts in which the eventuality time follows
the topic time, even if the topic time precedes or includes the utterance time. With
respect to Guaraní –ta, examples like (9d) in the context of (9b) provide evidence
that this linguistic form is a prospective aspect, since (9d) is acceptable even when
the topic time precedes the utterance time, as long as the eventuality time tempo-
rally follows the topic time. If (9d) had not been grammatical, or had not been ac-
ceptable in the context of (9b), –ta would remain as a candidate for a future tense.
While it is possible to show that a form is not a future tense, but instead is, for
example, a grammatical aspect, it is more challenging to be certain that a form is
a future tense. One difficulty in identifying whether there are future tenses is that
it is still an open question how to characterize this meaning category (see also
Comrie 1989). Whereas Yavaş (1982), for instance, maintains that a future tense
should be able to occur in all clauses that have future temporal reference, Bohne-
meyer (2000) points out that many languages have designated modal markers that
convey that a future eventuality is epistemically possible, an obligation, a desire, a
wish, or a prediction. He argues that it may be implausible to expect a future tense
to occur in clauses that realize future discourse with these modal attitudes: the
existence of designated modal expressions in the language may simply block the
occurrence of the future tense in such clauses.
Another point of contention concerns the question of whether a future tense
should only constrain the temporal relation between the topic time and the ut-
terance time, or whether it may also convey a modal meaning (and which modal
meanings it may convey). Some researchers argue that future temporal refer-
ence is a necessary but not a sufficient condition for future tenses (e.g., Enç 1996;
Kaufmann 2005). Given that future discourse, and hence also future temporal
reference, is inherently non-factual, and given that non-factual assertions are ac-
companied by a modal attitude, it is unclear whether “pure” future tenses exist, or
whether a future tense should always be expected to contribute a modal meaning
as well. We agree with Comrie (1985) that whether there are languages with pure
future tenses can only be answered “on the basis of the investigation of grammati-
cal expressions of future time reference across a number of languages” (p. 44). That
this is not a trivial task is evidenced by the fact that no consensus has been reached
even for the English auxiliary will, which has been analyzed as a future tense that
also encodes modality (e.g., Enç 1996; Kaufmann 2005), as a future tense that does
not encode modality (e.g., Kissine 2008), and as a modal marker that does not
(necessarily) entail future temporal reference (e.g., Werner 2006, Klecha 2014).
These authors’ papers provide a plethora of pointers to the kind of evidence that
can be brought to bear on the question of whether a particular linguistic form in
the language under investigation is a future tense, or at least whether and how it
differs from English will.
6 Requirements for Theoretically Informed Meaning Descriptions
In this chapter, we have argued that theory can guide fieldwork on meaning,
that theoretically informed fieldwork can result in more comprehensive mean-
ing descriptions, and that theoretically informed descriptions are instrumental to
improving theories of meaning and to uncovering language universals and varia-

tion. By taking into account what we already know about how linguistic forms
map to meanings and vice versa, theoretically informed descriptive work makes it
possible to see how a newly described language expands that knowledge.
We end the chapter by pointing to specific qualities that enable theoretically
informed meaning descriptions to contribute to the assessment and development
of theories of meaning and to the study of cross-linguistic universals and varia-
tion. In the spirit of Bird and Simons (2003), who recommend best practices for
maximizing the long-term usability of language documentation, we present five
requirements for meaning descriptions. These requirements must be met in order
to maximize a description’s potential for contributing to the assessment of theories
of meaning, and to our understanding of the cross-linguistic mapping between
form and meaning.
(58) Requirements for fieldwork-based, theoretically informed descriptions
of meaning
a. The description should be based on well-defined meaning categories.
b. The description should be based on semantic data.
c. The description should be based on positive and negative evidence.
d. The description should maximize replicability.
e. The description should maximize generalizability.
In the following, we discuss each requirement in turn.
(58a) The description should be based on well-defined meaning categories.
A meaning category (such as “tense” or “aspect”) is well defined if the definition
makes predictions about how an expression that encodes that meaning is distrib-
uted in the language (e.g., which other expressions it can(not) co-occur with) and
which contributions the expression makes to the meanings of sentences in which
it occurs. An example of an insufficiently precise definition is Comrie’s (1985:9)
definition of tense as “grammaticalized expression of location in time” (see foot-
note 3). See Tonhauser (2008) for a discussion of the importance of basing descrip-
tions on well-defined meaning categories.
Descriptions that use terms such as “past tense” or “imperfective aspect” with-
out providing definitions for the terms are unsatisfying, since such terms can be
defined in a variety of ways. If, for instance, a description discusses a “perfec-
tive aspect” without defining it, readers will, at best, be unsure about the meaning
of the perfective in this language and, at worst, they will assume the meaning of
some better-known perfective. The need for clear, language-specific explanation
of category labels is especially strong if the meaning of the form for which a label
is used differs from the meaning for which the label is typically used (as noted by
Haspelmath 2010b.) It is for this reason that in section 5.1 we took care to state
how the meaning of the Badiaranke Imperfective differs from that of more familiar
imperfective aspects.
(58b) The description should be based on semantic data.

Data relevant to semantic/pragmatic theorizing consist of sentences uttered in a
specific context and a judgment of the acceptability/truth of the uttered sentence
in that context by a native speaker (see also Matthewson 2004:371). These data
may be collected through a variety of methods: judgment elicitation is suitable
for obtaining both positive and negative evidence (cf. 58c), but positive evidence
can also be obtained from written or spoken corpora (which can be assumed to
contain only sentences judged to be acceptable; see Cover this volume for discus-
sion), questionnaires, and production experiments (for a discussion of methods,
see, e.g., Krifka 2011 and Bohnemeyer this volume). Translations do not constitute
data (though they may provide clues about the meanings of sentences of the object
language; see Matthewson 2004 and Deal this volume).
(58c) The description should be based on positive and negative evidence.
As illustrated with examples (8) to (11), positive evidence identifies particular
meanings with which an utterance is compatible, whereas negative evidence iden-
tifies particular meanings with which an utterance is incompatible. A meaning
description thus only identifies the truth and felicity conditions of a particular
utterance if both positive and negative evidence are provided.
(58d) The description should maximize replicability.
By this we mean that the information provided about how the empirical gener-
alization was established should be precise enough to allow other researchers to
replicate the findings in the same language or to explore comparable meanings in
other languages. The level of detail should be on par with the level of detail pro-
vided in “method” and “analysis” sections of experimental papers. In particular,
the discussion should include information about how the researcher interacted
with the consultants (one-on-one elicitation, group interviews, participant obser-
vation, etc.); the number of consultants that participated in the research; relevant
characteristics of the consultants (sex, age, linguistic background, etc.); the range
of examples elicited; the prompts given to the consultants in elicitation; and even,
ideally, the kinds of responses by consultants that led to positing that the consul-
tant judged the utterance to be (un)acceptable (including verbal and nonverbal
cues, such as puzzled looks, laughing, and shaking of heads). This last piece of
information is particularly important to provide to ensure cross-consultant and
cross-language comparability. A meaning description that provides this level of
detail about the methodology by which the data were obtained not only maxi-
mizes replicability, but also allows for subsequent explorations of the same topic to
improve on the methodology, as discussed in Tonhauser et al. (2013).
(58e) The description should maximize generalizability.
The more consultants participate in a descriptive research project and the wider
the range of utterances that are judged, the more likely it is that the empirical
generalizations that constitute the meaning description will generalize to the

grammar of the entire language and to the wider population of speakers. Further-
more, the more consultants are involved and the more utterance types are judged,
the easier is it to identify patterns about the language that are particular to a subset
of consultants due to, for instance, sociological or dialectal variation. In practice,
it is often not feasible to conduct fieldwork-based research with many consultants
(e.g., due to limited speaker availability and/or fieldwork time constraints). To
identify the extent to which the meaning description is generalizable, a descriptive
research project should be explicit about how many consultants participated, the
range of data used to establish a particular empirical generalization, and for which
data and empirical generalizations there were disagreements among the consul-
tants’ judgments.
References
Abusch, Dorit. 1997. The now parameter in future contexts. In Context Dependency in the
Analysis of Linguistic Meaning, Barbara Partee and Hans Kamp (eds.), 1–20. Stuttgart:
IMS Working Papers.
Bach, Emmon. 1986. The algebra of events. Linguistics and Philosophy 9: 5–16.
Bar-el, Leora. 2014. This volume. Documenting and classifying aspectual classes across
languages.
Beard, Robert. 1995. Lexeme-Morpheme Base Morphology. Albany, NY: SUNY Press.
Beaver, David, and Cleo Condoravdi. 2003. A uniform analysis of before and after. In Se-
mantics and Linguistic Theory (SALT) XIII, Rob Young and Yuping Zhou (eds.), 37–54.
Ithaca, NY: Cornell University.
Bennett, Michael R., and Barbara H. Partee. 1978. Toward the Logic of Tense and Aspect in
English. Bloomington, Indiana University Linguistics Club.
Binnick, Robert. 1991. Time and the Verb. Oxford: Oxford University Press.
Bird, Steven, and Gary Simons. 2003. Seven dimensions of portability for language docu-
mentation and description. Language 79: 557–582.
Bittner, Maria. 2005. Future discourse in a tenseless language. Journal of Semantics 22:
339–387.
Bittner, Maria. 2011. Time and modality without tense or modals. In Tense Across Languages,
Renate Musan and Monika Rathert (eds.), 147–188. Tübingen: Niemeyer.
Bohnemeyer, Jürgen, 2000. Time reference across languages. Handout from LOT summer
school, Tilburg, June 19–30, 2000.
Bohnemeyer, Jürgen. 2002. The Grammar of Time Reference in Yukatek Maya. Munich:
Lincom.
Bohnemeyer, Jürgen. 2009. Temporal anaphora in a tenseless language. In The Expression of
Time in Language, W. Klein and P. Li (eds.), 83–128. Berlin: Mouton de Gruyter.
Bohnemeyer, Jürgen. 2014. This volume. A practical epistemology for semantic elicitation
in the field and elsewhere.
Bohnemeyer, Jürgen, and Mary Swift. 2004. Event realization and default aspect. Linguistics
and Philosophy 27: 263–296.
Bybee, Joan, and Östen Dahl. 1989. The creation of tense and aspect systems in the lan-
guages of the world. Studies in Language 13: 51–103.
Bybee, Joan, Revere Perkins, and William Pagliuca. 1994. The Evolution of Grammar: Tense,
Aspect, and Modality in the Languages of the World. Chicago: University of Chicago
Press.
Cable, Seth, 2013. Beyond the past, present and future: Towards the semantics of ‘graded
tense’ in Gĩkũyũ Natural Language Semantics. Volume 21, pages 219–276.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
Chung, Sandra, and Alan Timberlake. 1985. Tense, aspect and mood. In Language Typol-
ogy and Syntactic Description, Timothy Shopen (ed.), 202–258. Cambridge: Cambridge
University Press.
Cipria, Alicia, and Craige Roberts. 2000. Spanish imperfecto and preterito: Truth condi-
tions and aktionsart effects in situation semantics. Natural Language Semantics 8:
297–347.
Comrie, Bernard. 1976. Aspect. Cambridge, England: Cambridge University Press.
Comrie, Bernard. 1985. Tense. Cambridge, England: Cambridge University Press.
Comrie, Bernard. 1989. On identifying future tense. In Tempus, Aspekt, Modus: DIE lexika-
lischen und grammatischen Formen in den germanischen Sprachen, Werner Abraham
and T. Janssen (eds.), 51–63. Tübingen: Niemeyer.
Condoravdi, Cleo. 2002. Temporal interpretation of modals: Modals for the present and for
the past. In The Construction of Meaning, David Beaver, Luis Casillas Martínez, Brady
Clark, and Stefan Kaufmann (eds.), 59–87. Stanford, CA: CSLI Publications.
Copley, Bridget Lynn, 2002. The Semantics of the Future. PhD diss., MIT.
Corbett, Greville G. 2005. The canonical approach in typology. In Linguistic Diversity and
Language Theories, Zygmunt Frajzyngier, David Rood, and Adam Hodges (eds.), 25–49.
Amsterdam: John Benjamins.
Corbett, Greville G. 2007. Canonical typology, suppletion, and possible words. Language
83: 8–42.
Cover, Rebecca. 2010. Aspect, Modality, and Tense in Badiaranke. PhD diss., Berkeley.
Cover, Rebecca. 2011. Modal aspects of Badiaranke aspect. Lingua 121: 1315–1343.
Cover, Rebecca. This volume. Semantic fieldwork on TAM.
Dahl, Östen, and Viveka Velupillai. 2011. The future tense. In The World Atlas of Lan-
guage Structures Online, Matthew S. Dryer and Martin Haspelmath (eds.), chapter 67.
Munich: Max Planck Digital Library. http://wals.info/chapter/67.
Deal, Amy Rose. This volume. Reasoning about equivalence in semantic fieldwork.
Deo, Ashwini. 2009. Unifying the imperfective and the progressive: Partitions as quantifi-
cational domains. Linguistics & Philosophy 32: 475–521.
Dixon, R. M. W. 1997. The Rise and Fall of Languages. Cambridge: Cambridge University
Press.
Dixon, R. M. W. 2010. Basic Linguistic Theory, Vols. 1 and 2. Oxford: Oxford University Press.
Dowty, David. 1977. Towards a semantic analysis of verb aspect and the English ‘imperfec-
tive progressive.’ Linguistics and Philosophy 1: 45–78.
Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Dowty, David. 1986. The effects of aspectual class on the temporal structure of discourse:
Semantics or pragmatics? Linguistics and Philosophy 9: 37–61.
Dryer, Matthew. 2006. Descriptive theories, explanatory theories, and Basic Linguistic
Theory. In Catching Language: The Standing Challenge of Grammar Writing, Felix
K. Ameka, Alan Dench, and Nicholas Evans (eds.), 207–234. Berlin: Mouton de
Gruyter.
Enç, Mürvet. 1987. Anchoring conditions for tense. Linguistic Inquiry 18: 633–657.
Enç, Mürvet. 1996. Tense and modality. In The Handbook of Contemporary Semantic Theory,
Shalom Lappin (ed.), 345–358. London: Blackwell.
Evans, Nicholas, and Stephen C. Levinson. 2009. The myth of language universals: Lan-
guage diversity and its importance for cognitive science. Behavioral and Brain Sciences
32: 429–492.
Fagerberg-Diallo, Sonja. 1983. Introduction to Pulaar. Senegal: American Lutheran Church
in Senegal.
25: 139–201.
Fudeman, Kirsten Anne. 1999. Topics in the Morphology and Syntax of Balanta, an Atlantic
Language of Senegal. PhD diss., Cornell University.
Gennari, Silvia. 2003. Tense meaning and temporal interpretation. Journal of Semantics 20:
35–71.
Gil, David. 2001. Escaping Eurocentrism: Fieldwork as a process of unlearning. In Linguistic
Fieldwork, Paul Newman and Martha Ratliff (eds.), 102–132. Cambridge: Cambridge
University Press.
Gregores, Emma, and Jorge A. Suárez. 1967. A Description of Colloquial Guaraní. The Hague:
Mouton de Gruyter.
Grice, Paul. 1975. Logic and conversation. In Syntax and Semantics 3: Speech Acts, P. Cole
and J. Morgan (eds.), 64–75. New York: Academic Press.
Hacquard, Valentine. 2006. Aspects of Modality. PhD diss., MIT.
Hacquard, Valentine. 2010. On the event-relativity of modal auxiliaries. Natural Language
Semantics 18: 79–114.
Haspelmath, Martin. 2010a. Comparative concepts and descriptive categories in cross-
linguistic studies. Language 86: 663–687.
Haspelmath, Martin. 2010b. Framework-free grammatical theory. In The Oxford Hand-
book of Grammatical Analysis, Bernd Heine and Heiko Narog (eds.), 341–365. Oxford:
Himmelmann, Nikolaus. 2006. Language documentation: What is it and what is it good
for? In Essentials of Language Documentation, Jost Gippert, Nikolaus P. Himmelmann,
and Ulrike Mosel (eds.), 1–30. Berlin and New York: Mouton de Gruyter.
Hinrichs, Erhard. 1986. Temporal anaphora in discourses of English. Linguistics and Phi-
losophy 9: 63–82.
Hyman, Larry. 2001. Fieldwork as a state of mind. In Linguistic Fieldwork, Paul Newman
and Martha Ratliff (eds.), 15–33. Cambridge: Cambridge University Press.
Iatridou, Sabine. 2000. The grammatical ingredients of counterfactuality. Linguistic Inquiry
31: 231–270.
Iatridou, Sabine, Elena Anagnostopoulou, and Roumyana Izvorski. 2003. Observations
about the form and meaning of the Perfect. In Ken Hale: A Life in Language, M. Ken-
stowicz (ed.), 189–238. Cambridge, MA: MIT Press.
Inoue, Kyoko. 1979. An analysis of the English present perfect. Linguistics 17: 561–589.
Ippolito, Michela. 2003. Presuppositions and implicatures in counterfactuals. Natural Lan-

guage Semantics 11: 145–186.
Ippolito, Michela. 2004. Imperfect modality. In The Syntax of Time, Jacqueline Guéron and
Jacqueline Lecarme (eds.), 359–387. Cambridge, MA: The MIT Press.
Kamp, Hans, and Uwe Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer.
Kaufmann, Stefan. 2005. Conditional truth and future reference. Journal of Semantics 22:
231–280.
Kaufmann, Stefan, Cleo Condoravdi, and Valentina Harizanov. 2006. Formal approaches
to modality. In The Expression of Modality, W. Frawley (ed.), 71–106. Berlin and New
York: Mouton de Gruyter.
Kiparsky, Paul. 2002. Event structure and the perfect. In The Construction of Meaning,
David Beaver, Luis Casillas Martínez, Brady Clark, and Stefan Kaufmann (eds.), 113–
133. Stanford, CA: CSLI Publications.
Kissine, Mikhail. 2008. Why will is not a modal. Natural Language Semantics 16: 129–155.
Klecha, Peter, 2014. Diagnosing modality in predictive expressions. Journal of Semantics
31(3): 443–455.
Klein, Wolfgang. 1994. Time in Language. New York: Routledge.
Kratzer, Angelika. 1998. More structural analogies between pronouns and tenses. In Pro-
ceedings from Semantics and Linguistic Theory (SALT) VIII, 92–110. Ithaca, NY: CLC
Publications.
Krifka, Manfred. 2011. Varieties of semantic evidence. In Semantics: An International Hand-
book of Natural Language Meaning, Claudia Maienborn, Klaus von Heusinger, and
Paul Portner (eds.), vol. 1, 242–267. Berlin: Mouton de Gruyter.
Kubota, Yusuke, Jungmee Lee, Anastasia Smirnova, and Judith Tonhauser. 2009. The cross-
linguistic interpretation of embedded tenses. In Proceedings of Sinn und Bedeutung 13,
Arndt Riester and Torgrim Solstad (eds.), 307–320.
Kusumoto, Kiyomi. 2005. On the quantification over times in natural language. Natural
Language Semantics 13: 317–357.
Landman, Fred. 1992. The progressive. Natural Language Semantics 1: 1–32.
Lehmann, Christian. 1989. Language description and general comparative grammar. In
Reference Grammars and Modern Linguistic Theory, Gottfried Graustein and Gerhard
Leitner (eds.), 133–162. Tübingen: Niemeyer.
Liuzzi, Silvio, and Pablo Kirtchuk. 1989. Tiempo y aspecto en Guaraní. Amerindia 14: 9–42.
Louie, Meagan. This volume. The problem with no-nonsense elicitation plans (for semantic
fieldwork).
Matthewson, Lisa. 2002. Tense in St’át’imcets and in Universal Grammar. In Papers for the
37th International Conference on Salish and Neighbouring Languages, Carrie Gillon,
Naomi Sawai, and Rachel Wojdak (eds.), vol. 9 of UBC Working Papers in Linguistics.
Vancouver: UBC.
Matthewson, Lisa. 2006. Temporal semantics in a superficially tenseless language. Linguis-
tics and Philosophy 29: 673–713.
May, Robert. 1991. Syntax, semantics, and logical form. In The Chomskyan Turn, Asa Kasher
(ed.), 334–359. Oxford: Oxford University Press.
McCoard, Robert. 1978. The English Perfect: Tense-Choice and Pragmatic Inferences. Amster-
dam: North-Holland.
Michaelis, Laura A. 2003. Word meaning, sentence meaning, and constructional meaning.
In Cognitive Perspectives on Lexical Semantics, H. Cuyckens, R. Dirvin, and J. Taylor
(eds.), 163–210. Amsterdam: Mouton de Gruyter.
Moens, Marc, and Mark Steedman. 1988. Temporal ontology and temporal reference. Com-
putational Linguistics 14: 15–28.
Murray, Sarah E. This volume. Reciprocity in fieldwork and theory.
Noonan, Michael. 2005. Grammar writing for a grammar-reading audience. Studies in Lan-
guage 30: 351–365.
Nordlinger, Rachel, and Louisa Sadler. 2008. When is a temporal marker not a tense? Reply
to Tonhauser (2007). Language 84: 325–331.
Nussbaum, Loren, William W. Gage, and Daniel Varre. 1970. Dakar Wolof: A Basic Course.
Washington, DC: Center for Applied Linguistics.
Ogihara, Toshiyuki. 1995. The semantics of tense in embedded clauses. Linguistic Inquiry
26: 663–679.
Partee, Barbara H. 1984. Nominal and temporal anaphora. Linguistics and Philosophy 7:
243–286.
Partee, Barbara H. 1992. Syntactic categories and semantic type. In Computational Linguis-
tics and Formal Semantics, Michael Rosner and Roderick Johnson (eds.), 97–126. Cam-
bridge: Cambridge University Press.
Portner, Paul. 1998. The progressive in modal semantics. Language 74: 760–787.
Portner, Paul. 2003. The (temporal) semantics and (modal) pragmatics of the perfect. Lin-
guistics and Philosophy 26: 459–510.
Reed, Sylvia L. 2012. Evaluating prospectivity in a neo-Reichenbachian aspectual system.
In Coyote Papers, Dan Brenner (ed.), vol. 19. Tucson, AZ: University of Arizona Lin-
guistics Circle.
Reichenbach, Hans. 1947. Elements of Symbolic Logic. Berkeley: University of California
Press.
Rice, Keren. 2005. A typology of good grammars. Studies in Language 30: 385–415.
grammars. In Catching Language: The Standing Challenge of Grammar Writing, Felix K.
Ameka, Alan Dench, and Nicholas Evans (eds.), 235–268. Berlin: Mouton de Gruyter.
Ritter, Elizabeth, and Martina Wiltschko. 2004. The lack of tense as a syntactic category:
Evidence from Blackfoot and Halkomelem. In UBC Working Papers in Linguistics, J.
Brown and T. Peterson (eds.), vol. 14.
Roberts, Craige. 2004. Context in dynamic interpretation. In Handbook of Contempo-
rary Pragmatic Theory, Laurence Horn and Gregory Ward (eds.), 197–220. Oxford:
Blackwell.
Sapir, Edward. 1921. Language: An Introduction to the Study of Speech. New York: Harcourt.
Sasse, Hans-Jürgen. 2002. Recent activity in the theory of aspect: Accomplishments,
achievements, or just non-progressive state? Linguistic Typology 6: 199–271.
Schachter, Paul, and Fé T. Otanes. 1972. Tagalog Reference Grammar. Berkeley: University
of California Press.
Schroeder, Sara Elizabeth. 2011. A Case for be going to as Prospective Aspect. M.A. Thesis,
University of Montana.
Shaer, Benjamin. 2003. Toward the tenseless analysis of a tenseless language. In Semantics of
Under-Represented Languages (SULA) II, Jan Anderssen, Paula Menéndez-Benito, and
Adam Werle (eds.), 139–156. Amherst, MA: GLSA.
Smirnova, Anastasia. 2009. A case against ‘defective’ tense in the Bulgarian subjunctive. In
Proceedings of Formal Approaches to Slavic Languages (FASL) 17: The Yale Meeting. Ann
Arbor: Michigan Slavic Publications.
Smith, Carlota S. 1997. The Parameter of Aspect. Dordrecht: Kluwer, 2nd edition.
Stowell, Tim. 1996. The phrase structure of tense. In Phrase Structure and the Lexicon, Johan
Rooryck and Laurie Zaring (eds.), 277–291. Dordrecht: Kluwer.
Thomason, Richmond H. 1984. Combinations of tense and modality. In Handbook of Philo-
sophical Logic: Extensions of Classical Logic, Dov Gabbay and F. Guenthner, 135–165
Dordrecht: Reidel.
Thomason, Richmond H. 2007. Three interactions between context and epistemic locu-
tions. In Modeling and Using Context, 467–481: Berlin: Springer.
Tonhauser, Judith. 2008. Defining cross-linguistic categories: The case of nominal tense.
Reply to Nordlinger and Sadler (2008). Language 84: 332–342.
Tonhauser, Judith. 2010. Is Paraguayan Guaraní a tenseless language? In Proceedings of Se-
mantics of Under-Represented Languages of the Americas (SULA) 5, 227–242. Amherst,
MA: GLSA.
Tonhauser, Judith. 2011a. The future marker –ta of Paraguayan Guaraní: Formal seman-
tics and cross-linguistic comparison. In Tense Across Languages, Renate Musan and
Monika Rathert (eds.), 207–231. Tübingen: Niemeyer.
Tonhauser, Judith. 2011b. Temporal reference in Paraguayan Guaraní, a tenseless language.
Linguistics & Philosophy 34: 257–303.
onomy of projective content. Language 89: 66–109.
Vendler, Zeno. 1957. Verbs and times. The Philosophical Review 66: 143–160.
Vendler, Zeno. 1967. Verbs and times. In Linguistics and Philosophy, 97–121. Ithaca, NY: Cor-
nell University Press.
Verkuyl, Henk. 1993. A Theory of Aspectuality. Cambridge: Cambridge University Press.
Welmers, William E. 1973. African Language Structures. Berkeley and Los Angeles: Univer-
sity of California Press.
Werner, Tom. 2006. Future and non-future modal sentences. Natural Language Semantics
14: 235–255.
Yavaş, Feryal. 1982. Future reference in Turkish. Linguistics 20: 411–429.
Zagona, Karen, 1990. Times as temporal argument structure. Unpublished manuscript,
University of Washington, Seattle.
Zarratea, Tadeo. 2002. Gramática Elemental de la Lengua Guaraní. Asunción, Paraguay:
Marben.
Index
accomplishment, 37, 78–9, 93–101, 105, 327–8, bare nouns, 175–6, 179–83, 185–6, 190–200
332, 336 Blackfoot, 48–9, 51–2, 56–8, 63, 66–8, 141, 193
achievement, 76, 79, 81, 91–3, 103–4, 105, 243, BowPed (stimulus), 17–18, 27–8
332, 336
act-out tasks, 15, 22, 23, 40 Cheyenne, 287–303
activity (aspectual class), 24, 37, 57, 76, 78–9, 81, comparatives, 110–11, 118–23, 125, 127–32
87–90, 98, 105, 327, 332–3, 335–6 conjoined, 118–23
adverbial, temporal, 37, 64, 78, 98–9, 235, 241, conditionals, 59, 61–8, 163, 167, 215, 237, 321
247, 260, 275, 314, 318–19, 321, 329–30 subjunctive, 59, 63–8
Akan, 92–3 unreal, 63–8
Algonquian, 48, 141, 191, 289, 304 context, 2–3, 22–7, 28, 48–56, 58–70, 96–7,
ambiguity, 31, 159, 163–4, 230, 253, 270, 276–80, 99–100, 118–23, 125–31, 138–9, 142–3, 145–7,
288, 292, 299, 302–3 159–60, 166, 170–2, 179, 181–3, 185–7, 207–30,
Arabic, Egyptian, 92 241–5, 248–51, 277, 280, 283–4, 289, 299, 311,
articles, covert, 180–2, 184–5, 190–1, 193, 195–6, 317–20, 329–30, see also discourse contexts
198–200 context-dependence, 36, 68, 111–12, 322
aspect, 25–7, 37–8, 57, 75–107, 234, 247, 254, 258, contradiction, 17, 35, 94–5, see also law of
277, 283, 308–13, 322–37 contradiction
comparative, 124 control (by an agent), 94–6
grammatical, 75, 80, 84, 254, 327, 332 counterfactuals, 145
imperfective, 88, 94, 237–8, 240–1, 253–5, crisp judgments, 121–3
259–62, 264–6, 271, 323–5, 327, 329–30, 333–5, culmination, 90, 93–9
338–9, 342, see also imperfective paradox Cut & Break Clips (stimulus), 23–5
lexical, 75–107, 330–7
perfect, 255–6, 324–7, 337 definiteness, 25, 175, 177–83, 186–7, 190–1, 193,
perfective, 38, 75, 82–3, 85, 88–90, 93–8, 102, 195, 199, 277, 281
234–5, 238, 247, 251, 253–5, 258, 323–34, 336, deixis, 176, 179, 188, 190, 199
338, 342 demonstration, 15, 22–3, 40–1
prospective, 150–2, 325, 328, 338, 340 demonstratives, 175–6, 178, 193
aspectual classes Dëne Sųłiné, 78–9, 89–90
accomplishment, 37, 78–9, 93–101, 105, 327–8, discontinuous past, 237–8, 254–5, 262–3, 266
332, 336 discourse contexts, novel, 180, 182, 186, 191, 196,
achievement, 76, 79, 81, 91–3, 103–4, 105, 243, 199
332, 336 discourse-sensitive phenomena, 137
activity, 24, 37, 57, 76, 78–9, 81, 87–90, 98, 105, dislocation, 269–70, 271, 274–6, 278–80, 283–4
327, 332–3, 335–6 domain restriction, 179, 185, 186, 189, 190, 196,
semelfactive, 76, 83–4, 87, 90, 105, 332–3 199, 281
state/stative, 40, 76–7, 80–2, 84, 86–7, 91–2,
101–7, 228, 234–5, 238, 247, 253–6, 262, 271, ecological validity, 31–2
308–9, 326, 331–6 elicitation, 13–15, 20–3, 28, 32, 34, 42, 47–8, 61–8,
Assiniboine, 91 96, 106, 120–3, 128–31, 137–9, 142, 153–4,
association tasks, 19, 23, 40 246–57, 301–2
attitude reports, 214–23 paradigmatic, 7, 47–66, 69, 241, 246–7, 255
elicitation frame, 28
Badiaranke, 235, 237–57, 259–67, 308, 330–5, 338, Elicitation, Golden Rule of, 15, 42
340 empirical methods, 15–20
Ball & Chair (stimulus), 29–32 Empiricism, Constructive, 19
352 Index
English, 24, 26, 79, 84–5, 89, 93, 98, 168, 175, 177, language description, 248, 306, 312, 340
179, 181–90, 199, 224–5, 290, 297, 302, 325–6, law of contradiction, 185–6, 188, 194–6,
328–9, 332, 339 200
entailment, 17, 34–5, 87 lexical competition, 123–31
culmination, see culmination Lithuanian, 175–6, 180, 186, 190–200
termination, 87, 89–90
equivalence, 157–73 Mandarin, 97
equivalent implicatures hypothesis, 169 Maori, 181, 184, 193
equivalent judgments hypothesis, 159 metalinguistic utterance/knowledge, 4, 14, 20,
equivalent translations hypothesis, 158 22–3, 34, 36, 75, 106, 212–13
event, 87–91, 94–8, 105, 238, 331–2 modality, 3, 6, 8, 25, 135–8, 157, 162–71, 230, 234,
prior, 104 237–8, 247, 279, 317, 321, 327–9, 337–41
eventuality, see event circumstantial, 147, 149–51
deontic, 147, 238
familiarity, 180–2, 186–7, 190–1, 194, 196, epistemic, 27, 136–9, 141, 147, 168–71, 238, 240,
199 264–6, 340–1
future, 26–7, 63–6, 235, 240, 314, 317, 325, strength of, 149–50, 151–2, 230
337–41 modification (modifiers), 88, 98–9, 102, 111–20,
125–6, 131, 292, 297
Gitksan, 135, 138, 141, 149–50 monotone, upward, 165–7
gradable predicates, 111–15, 122, 124–5 motion verbs (stimulus), 38–9, 251
absolute versus relative standard, 112–13
dimensional, 113–14 narratives, 32, 60–2, 68–70, 137, 142–3, 145–6,
evaluative, 113–14 153–4, 211, 254, 259, 264, 321
positive and negative polarity, 114, 117 naturalness of elicited speech, 8, 71, 86, 139,
gradability, 110–32 142, 153–4, 213, 220, 223, 247, 262–3,
266
hermeneutic methods, 16, 41–2 Navajo, 7, 111, 116–17, 124–32
Hindi, 95–6 Nez Perce, 8, 158–69, 230
homogeneity (of events), 87–8, 102–3, 105 Niger–Congo, 8, 234–5
norm–relatedness, 7, 111, 114–32
imperfective paradox, 37, 327
implicature, 18, 89–90, 93–7, 118, 165–7, 169, 171, objectivism, 19–20
187, 239, 247–8, 255 observation of speakers, 8, 15, 17, 20–1, 32, 41, 43,
anti-perfect, 239, 255 234, 246, 263–7, 289
scalar, 167, 171 opaque readings, 276–8, 282–4
inceptive, 80, 88, 91–2, 98–9, 101–2 Original Word Game, 14, 16
inchoative, 24, 102–4 overarching storyline elicitation plan, 7, 47,
indefinites, 177–8, 180–5, 195 60–70
Innu-aimun, 175–6, 186, 190–200
interpretation (of a stimulus or task), 14–16, 27, Paraguayan Guaraní, 9, 308, 310–11, 316–19,
33–4, 42 321–2, 328, 338–40
Inuttut, 175–6, 186, 190–200 paraphrase, 15, 22–3, 227–9, 240, 277
participant observation, see ‘observation of
Japanese, 153, 319–20 speakers’
judgment, 20, 23, 28, 34–9, 48, 58–9, 62–4, 68, pluractionality, 8, 208, 223–30, 251–2, 258
71, 96–7, 157–9, 169–72, 207–9, 248–9, 290–1, plurality, 56, 183–4, 187, 189, 196–8, 224, 226, 229,
299–302 245, 287, 290–1, 293, 295–8, 303
felicity, 23, 47–8, 64, 68, 90, 98, 157, 207, 208, predictions, 1, 5, 17, 42, 79, 115, 131, 136, 172–3, 256,
221, 289, 299 270, 278, 283–5, 288, 292–3, 297–9, 303, 326,
truth-value, 37–8, 47–8, 58–9, 207–9, 219, 226, 338, 341
228–9, 289–91, 299 Prototype Theory, 18–19, 21
Kaqchikel, 208, 212–13, 223–9 questionnaire, 7, 25–6, 76–7, 83–6, 111, 246, 314,
Kiowa, 271–6, 278–81, 283–5 343
Index 353
reciprocal, 8–9, 287–8, 290–5, 295–303 tense

reference absolute, 27, 319–20
aspectual, 9, 233, 236, 307–14, 322–5, 327–39 past, 8, 59, 88, 141, 167, 223, 236, 237–8, 240,
temporal, 9, 59, 212, 234–5, 307–23, 325, 327–9, 246–7, 255, 262, 266, 309–11, 315, 318–22,
332, 337–41 325, 327, 338, 342
referential communication task, 27, 29–32, 34 relative, 27, 234, 319–20
referential indeterminacy, 19 tenseless, 235, 321–2
reflexive, 8, 103, 287–8, 290–3, 295–303 termination, 87, 89–90, 105
relativist agnosticism, 14 texts, 2, 6, 8, 25, 49, 60, 106, 142, 182, 196, 233,
replicability, 3, 5, 142, 210, 230, 342–3 235–7, 245–8, 252–64, 266, 287, 289, 291, 299,
resource situations, 281–4 302–3
response, 7, 14–17, 20–8, 31, 34–6, 38–40, 42, 85, Thai, 92
137, 139, 158, 160, 162, 172, 220–2, 241, 245, time
265, 317, 343 evaluation, 308–12, 314, 317, 319–20, 327,
329
Salish, 7–8, 78, 94–6, 149, 151, 176, 178–9, 186, eventuality, 26, 98, 234, 238, 255, 265–6, 308–11,
199, 321 317–18, 320, 323–32, 337–41
Samoan, 179 topic, 26, 98, 308–41
satiation, 36 topic/topichood, 215–23, 269–85, 313
scale structure, 111–18, 124, 126–8, 131 Topological Relations Picture Series (stimulus),
scope, 6, 150, 163, 181–90, 193–4, 200, 224, 226, 17, 27–8
238, 273, 277–80, 282–5 translation, 2, 8, 20, 22–3, 25–6, 48, 57, 59, 68,
semantic transfer, 18–19 80, 82, 85–6, 90, 93–4, 96–7, 115–17, 137, 139,
Sənčáθən, 95, 102 153–4, 157–8, 160–73, 209, 213, 215–16, 221–3,
situations/situation semantics, 63–8, 83–8, 101, 227, 233, 241–3, 260–1, 265, 267, 282, 302,
105, 123, 160, 276–84 311, 343
Skwxwú7mesh, 7–8, 78, 88–90, 94–6, 98–9, transparent readings, 276–80, 283
101–2, 175, 178–9, 186–90, 192–3, 199 truth-value, 37–8, 47–8, 58–9, 207, 210, 219, 226,
Spanish, 18, 103–4, 226–7 228–9, 289–91, 299
specificity, 40, 48, 179, 187, 276, 278
staged discourse, 21, 32, 258 underspecification, 19, 260–1, 288–9, 292–3,
St’át’imcets, 8, 94, 149–51, 167, 171, 176, 181, 188, 297–9, 303
190, 193, 199, 223, 321 uniqueness, 180–1, 183, 186–7, 189, 191,
stimulus, 6, 15–16, 19–23, 25–8, 31–2, 34–8, 42, 86, 281
121–3, 129, 137–8
non-visual, 7, 123, 125, 128–32, 211 vagueness, 19, 112, 121–2, 132, 168
visual, 6–7, 28–34, 38, 111, 118, 120–4, 128–32, Vendler classes, 37, 75–82, 87, 104–6, 333
139, 210, 227, 302
Washo, 7, 110–22, 128–9, 131–3
Tatar, 101
telicity, 37, 78–80, 82, 84, 88–9, 94, 96, 103, 105–6, Yucatec (Yucatec Maya), 20–1, 26–7, 38–40, 207,
238, 333–4, 336 214–23

Bochnak, M. R., & Matthewson, L. (Eds.) - 2015. Methodologies in Semantic Fieldwork.

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

Bochnak, M. R., & Matthewson, L. (Eds.) - 2015. Methodologies in Semantic Fieldwork.

Uploaded by

Copyright:

Available Formats

Introduction

M. Ryan Bochnak and Lisa Matthewson

Methodologies in Semantic Fieldwork. M. Ryan Bochnak and Lisa Matthewson.

Conferences such as Semantics of Underrepresented Languages in the Americas

2 Experiments Large and Small

It is important to emphasize that our defense of the small-scale experiments

3 Coverage of the Volume

Bohnemeyer’s chapter discusses methods of linguistic data collection for reveal-

to make and test hypotheses about modals in Gitksan (Tsimshianic), St’át’imcets

Haider, Hubert 2007. As a matter of facts—comments on Featherston’s sticks and carrots.

A Practical Epistemology for Semantic Elicitation

The aim of this chapter is to sketch a classification of elicitation methods in seman-

Methodologies in Semantic Fieldwork. M. Ryan Bochnak and Lisa Matthewson.

directly observable—so how else would it be accessible if not through interpreta-

(1) Which observable properties of communicative behavior can be exploited

2 The Empirical Basis of (Field) Semantics

Figure 1.1 An elaborate update of Brown’s (1958) “Original Word Game”

The richest and most basic source of observational evidence in semantics is

(4) The x is under the y.

Stimulus: contact language utterance

Recording of spontaneous Recording of staged speech

Linguistic behavior (“response”) + + +

conditions. From this perspective, psycholinguistic experiments involve the elici-

Target Type I — Type II — Type V — Type VI — Type VII —

a linguistic representation used as stimulus is a context description employed in

4 Type I: From Target Language Utterance to Target Language Utterance

Responses to theme prompt (7) Responses to instrument prompt (8)

Responses to theme prompt (7) Responses to instrument prompt (8)

because “separation in material integrity” occurs in them in a manner similar to

5 Type II: From Contact Language Utterance to Target Language Utterance

A translation task directs a speaker to translate a stimulus utterance in the contact

(10) Q: Ba’x a=tukul-ik

A: Chéen k’uch-uk-o’n wal=e’,

6 Type IV: From Non-Linguistic Representation to Target Language Utterance

The elicitation of descriptions of non-linguistic stimuli has become the method

However, when I ran BowPed with Yucatec speakers, I had an experience I

Figure 1.6 Ball & Chair picture 2.5

(11) a. Estée, u séegere-e-e . . . chan fòotoa’,

b. u frèente e sìiyao’, tu tohile don Jorgeo’,

c. ti’ yàani’. Tu’x ku nakta’ máako’,

d. estée, ta frèente súutu’.

f. ti’ yàan ump’ée bòolai’.

g. Ta xts’íi - ta xts’íihk’abil [unintel.], ti’ yàan

h. kàasi tu tohil u yòok yàan ti’.

Line g is a correction of line e. The individual propositions of this description can

7 Type V: From Target Language Utterance to Judgment

Judgments are metalinguistic utterances that may comment on a variety of prop-

I have encountered the phenomenon alluded to in the first point on numerous

There can be a variety of reasons for why a decontextualized expression appears to

The templates in (12a–d) might be used to test the well-formedness of an ut-

(13) a. Floyd was pushing a cart

(14) Pedro=e’ táan u=k’àay,

káa=t-u=k’at-ah u=báah Pablo.

Pedro=e’ t-u=p’at-ah u=k’àay.

Be’òora=a’ ts’o’k=wáah u=k’àay Pedro?

(15) H-na’k le=chan kanìika y=óok’ol le=tàabla=o’

8 Type VI: From Target Language Utterance to Linguistic Representation

A very powerful strategy for elucidating the meanings of linguistic expressions

(16) Le=chan tàabla=o’ h=péek-nah-ih, káa=h-na’k

le=chan kanìika y=éetel che’ te’l y=óokol=o’.

As soon as it is made explicit that it was the ground—the plank—that moved,

9 Type VII: From Target Language Utterance to Non-Linguistic

identifying the 20 most frequently mentioned types of themes—humans; various

(13) P1: A Paradigmatic Elicitation Plan (ML/BB Elicitation 10-25-2011)