A Cognitive-Functional Approach To Nominalization in English

A Cognitive-Functional Approach
to Nominalization in English
WDE
G
Cognitive Linguistics Research
26
Editors
René Dirven
Ronald W. Langacker
John R. Taylor
Mouton de Gruyter
Berlin · New York
A Cognitive-Functional Approach
to Nominalization in English
by
Liesbet Heyvaert
Mouton de Gruyter
Berlin · New York 2003
M o u t o n de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter G m b H & Co. K G , Berlin
© Printed on acid-free paper

which falls within
the guidelines of the A N S I
to ensure permanence and durability.
ISBN 3 11 017809 5
Bibliographic information published by Die Deutsche Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data is available in the Internet at < h t t p : / / d n b . d d b . d e > .
© Copyright 2003 by Walter de Gruyter G m b H & Co. K G , D-10785 Berlin

All rights reserved, including those of translation into foreign languages. N o part of this book
may be reproduced or transmitted in any form or by any means, electronic or mechanical,
including photocopy, recording, or any information storage and retrieval system, without
permission in writing from the publisher.
Printed in Germany
To Wouter,
Judith and Oscar
Acknowledgements
The present study reflects a research project to which many people, each in
their own way, have contributed. I would like to thank those people here.
First and foremost I would like to thank Kristin Davidse. As my Ph.D.
supervisor, she was involved in my research on nominalization patterns
from its earliest stages onwards. I am deeply grateful for the continual en-
couragement she has given me, for the openness with which she has shared
her own learning with me, and for the precision with which she has read my
texts, time and again supplying extensive commentary. This book, in which
any remaining shortcomings are my own, in the first place represents the
dialogue I have been able to conduct with her.
I would also like to thank Hubert Cuyckens, Ad Foolen, Ronald Lan-
gacker and Willy Van Langendonck for their rigorous comments on an ear-
lier version of this book. In addition, I am grateful to René Dirven and
Ronald Langacker who, as editors of the Cognitive Linguistics Research
Series, guided me through the proposal and writing process and whose
careful comments on various chapters of the book I have greatly appreci-
ated. I thank the managing editor of the series, Birgit Sievert, for her pa-
tience and for her professional assistance throughout the publication proc-
ess. I thank Jürgen Benteyn for his swift and always good-humoured help
with the formatting of the manuscript, and Hubert Cuyckens, Kristin
Davidse and Lieven Vandelanotte for their general support. Thanks also to
Lieven Vandelanotte for his help with the drawings and to An Laffut and
Lieven Vandelanotte for proof-reading parts of the manuscript.
My thanks also go to a number of people who have in the past years
commented on my work and helped me with their suggestions and criti-
cisms. In addition to Hubert Cuyckens, Kristin Davidse, Ad Foolen, Ronald
Langacker and Willy Van Langendonck, these are: Eirian Davies, Renaat
Declerck, Dieter Kastovsky, An Laffut, Odo Leys, Lachlan Mackenzie, Bill
McGregor, Hans-Jörg Schmid, Lieven Vandelanotte, Frederike van der
Leek, Jean-Christophe Verstraete and Emma Vorlat. The University of
Leuven has contributed to the publication of this book by allowing me to
work on it as a postdoctoral fellow of the Research Council (postdoctoral
grant PDM/02/034).
I am most grateful to my parents and my parents-in-law for their unfail-
ing support. Finally, I would like to thank Wouter, without whose uncondi-
tional support, practical help and love I would never have been able to
bring this book to a good end.
Contents
Part I Towards a theoretical-descriptive approach to

nominalization
Chapter 1: Introduction 3
Chapter 2: Theoretical assumptions 9
1. Language: The relation between system and usage 9

1.1. A usage-based approach to language 9
1.2. Cognitive Grammar: The interface between system and 11
usage
1.3. The language system as network 13
2. The natural symbolic relation between lexicogrammar 15
and semantics
2.1. A natural description of language 15
2.2. The natural or non-arbitrary relation between grammar 17
and meaning
3. Constructions: Composition, classification and function 19
3.1. A functional approach to composition 20
3.2. Levels of functional analysis 21
3.3. A functional approach to classification 23
3.3.1. An internal functional analysis of constructions 24
3.3.2. Classes as complex categories: Schema and prototype 26
3.4. Composition and classification: Conclusion 28
4. Functional layers of organization 29
5. Constructions and the relations between them 33
5.1. Agnation and enation: A definition 35
5.2. Agnation and enation: Theoretical significance and heu- 37
ristic value
6. Theoretical assumptions: Conclusion 40
Chapter 3: Nominalization 41
1.
1.1. A usage-based account
Schematization of nominalization
and extension 42
43
1.2. Entrenchment 44
χ Contents
2. Nominalizations as non-arbitrary symbolic units 46

3. Nominalizations as composition and reclassification 47
3.1. Constructions as assemblies of symbolic units 48
3.2. Nominalization as functional reclassification 49
3.2.1. Rank shift and reclassification 49
3.2.2. Nominalization as functional reclassification 51
4. A multifunctional account of nominalization 52
5. The role of agnation in the analysis of nominalization 52
5.1. Networks of agnation 54
5.2. Agnation as a descriptive heuristic 55
5.3. The schematic nature of agnation 56
6. Towards a theoretical-descriptive approach to nominali- 58
zation: Conclusion
Chapter 4: The functional organization of nominal and clause 61
1. From noun type specification to nominal 61

2. From process type specification to clause 70
2.1. The clausal head 72
2.1.1. Finite vs. atemporal clausal heads 73
2.1.2. The clausal head, Objects and Complements 76
2.2. Instantiation and quantification of the clausal head 78
2.3. Clausal grounding 81
2.3.1. The finite element 82
2.3.2. Person deixis 85
2.3.3. Clausal grounding: Conclusion 86
2.4. The speech-functional role of the Subject 87
2.5. The functional organization of nominal and clause: Con- 92
elusion
3. Nominalization as functional reclassification 94
Part Π Deverbal -er nominalization

Chapter 5: Deverbal -er suffixation: Towards a descriptive 99
position
0. Introduction 99
1. -er nominalizations: Agent names or Subject names? 101
1.1. Agentivity and salience 102
Contents xi
1.1.1. -er nominalizations as natural symbolic units 104

1.1.2. -er suffixation as a complex conceptual category 106
1.1.3. The verbal base of deverbal-er nomináis 108
1.2. Non-agentives as external arguments 110
1.2.1. -er nomináis profiling oblique participants 115
1.2.2. -er nomináis based on unaccusatives 116
1.2.3. Middle-based-er nomináis 117
1.3. -er nominalizations as Subject names 118
2. Lexicalization versus ad hoc nominalization 120
3. Towards a coherent account of deverbal -er nominaliza- 123
tions
Chapter 6: The middle construction 125
1. Middle formation: A state of the art 126

2. The representational versatility of the middle formation 128
3. Towards a semantic typology of middle constructions 132
4. Middle constructions and modality 137
4.1. The modality of 'letting' 140
4.2. The middle as 'letting' construction 143
5. Conclusion 147
Chapter 7: A multifunctional approach to -er nominalization 149
0. Introduction 149
1. Representational semantics 151
1.1. Agentive-er nominalization 152
1.2. Non-agentive-er nominalization 153
2. A semantic typology of non-agentive -er nominalizations 154
3. The constructional properties of -er nominalization 158
3.1. Lexicalized -er nominalizations 159
3.1.1. Non-agentive -er nominalizations 159
3.1.2. Agentive -er nominalizations 161
3.1.3. A note on instrumental -er nominalization 163
3.2. Ad hoc nominalizations 165
3.2.1. Phorie ad hoc nominalizations 166
3.2.2. Non-phoric ad hoc nominalizations 168
3.2.3. Ad hoc-er nominalization: Conclusion 172
3.3. The constructional properties of -er nomináis: Summary 173
4. Deverbal -er suffixation as Subject-profiling 174
5. Conclusion 176
xii Contents
Part ΙΠ Factive nominalization
Chapter 8: Factive nominalization: Towards a descriptive po- 181

sition
0. Introduction 181
1. Factivity as truth presupposition 183
2. Factivity as embedded projection 185
2.1. Embedding vs. taxis 186
2.2. Projection vs. expansion 187
2.3. Facts as embedded projections 188
2.4. Conclusion 194
3. Delineation of the fact category 195
3.1. Ambiguity and vagueness 195
3.2. Gerundive nominalizations: Acts or facts? 198
4. Towards an internal, nominal analysis 200
4.1. Halliday's dependency analysis 201
4.2. Type vs. instance nominalization 203
Chapter 9: A functional analysis of factive nominalizations as 207

nominal constructions
1. ί/ζα/-facti ves: From finite clause to proper name strategy 207

1.1. Downranking as functional reclassification 208
1.2. The nominal paradigm 209
2. the fact íAútf-constructions: A case of apposition 211
2.1. The structural characteristics of the fact that- 213
constructions
2.2. The functional properties of the fact íAaí-constructions 217
2.3. the fact /to-structures: Conclusion 221
3. Gerundive facts: From atemporal clausal head to com- 221
mon noun or proper name strategy
3.1. From atemporal clausal head... 222
3.2. .. .to common noun or proper name strategy 228
3.2.1. Gerundive nominalizations: from transformation to 229
nominal constituent
3.2.2. Towards a systematic account of the nominal properties
of gerundive nominalizations 231
Contents xiii
3.2.3. The nominal functioning of gerundive nominalization:

Conclusion 239
3.3. Factive gerundive nominalization: Integrating the clausal
and nominal paradigms 241
3.4. Factive gerundive nominalization: Conclusion 244
4. The internal organization of factive nominalizations:
Conclusion 244
Chapter 10: Conclusion 249
Appendix 255
Notes 257
References 267
Index 281
Parti
Towards a theoretical-descriptive approach
to nominalization
Chapter 1
Introduction
The focus of the present study is on deverbal nominalization in English.

The analysis of nominalization that is presented in it has mainly been
prompted by two observations. First, the existing descriptive analyses of
nominalization fail to embed the peculiarities of specific nominalization
types in a coherent and systematic theoretical account of the fundamental
constructional mechanisms that underlie nominalization in general. Sec-
ondly, the description of nominalization systems tends to be biased towards
the clausal categories which they realize, to the neglect of their nominal
characteristics. Following Lees's (1960) influential analysis of nominaliza-
tion in English, nominalizations are still basically regarded as clause-like
constructions used in nominal units in an unfathomable way. The absence
of a coherent theoretical perspective on the process of nominalization has
led to descriptive fragmentation. Morphological processes of nominaliza-
tion such as -er or -ee suffixation, for instance, seem hardly relatable to
nominalization types which incorporate clause-like structures, as in [His
answering the phone] surprised me. Also, the one-sided, clause-based na-
ture of most descriptions has elucidated only part of the lexicogrammatical
and semantic properties of nominalizations.
These problems reflect deficiencies of theory as well as of description.
A central theoretical problem is, for instance: how can we, given the pre-
vailing emphasis on structural constituency, model nominalization, which
seems to elude and transcend ordinary constituency relations? A central
descriptive problem is: which features of clause and nominal enable the
reclassification from the one into the other that is generally held to be in-
volved in nominalization? Most attention so far seems to have gone to the
reclassification of verb into noun, and the 'reification' of a situational ref-
erent into an entitized one that seems to go with it. However, as a reclassi-
fication mechanism, this is too thin. There must be other factors, mainly
analogies between clause and nominal, which motivate the possibility of
reclassifying a clausal configuration into a nominal one.
To tackle the theoretical as well as the descriptive issues, this study is
composed of a general theoretical-descriptive part (Part I) and two case
studies (Parts Π and ΠΙ). Part I presents the theoretical and descriptive con-
structs required to model nominalization. Parts Π and ΙΠ present two case
studies and determine in how far the proposed theoretical-descriptive ap-
4 Introduction
proach can account for the lexicogrammatical properties and meanings of

specific nominalization systems. Part Π zooms in on the system of deverbal
-er nominalization (e.g., remind -» reminder) and in Part ΙΠ, three types of
factive nominalization are considered (viz. [That he killed her] surprises
me, [The fact that he killed her] surprises me and [His killing her] sur-
prises me).
The main theoretical motif in this study is that nominalization strongly
calls for a functional rather than purely structural approach. Crucially, the
reclassification from verbal/clausal starting point into nominal class is func-
tional in nature: nominalization involves the functional reclassification of a
particular level of assembly in the organization of the clause into a nominal
structure. A nominalization always adopts external nominal functions, but,
as will become clear, it may also acquire internal nominal functional or-
ganization. The reason why the identification of nominalizations as con-
structions of the nominal class requires a functional view on composition is
that nominalizations do not always have discernible structural components
for each function which they realize: as in ordinary nomináis, the internal
functions which nominalizations realize are not necessarily reflected iconi-
cally in their structure. It is therefore only a functional analysis which can
shed light on what nominalized structures have in common with the iconi-
cally and non-iconically coded members of the nominal class in general.
Moreover, it is only by adopting a functional viewpoint that one can relate
nominalizations which contain part of a clause or a full clause (e.g., her
signing the contract, that she signed the contract) to nominalizations at
word level (e.g., baker). Like nominalizations based on a simple verb stem
(e.g., bake —• baker), nominalizations containing a clausal unit reclassify
this clausal unit into a unit with nominal functional behaviour.
Because the analysis of nominalization requires a radically functional
approach, I will in Chapter 2 discuss two linguistic frameworks which have
shown a particular interest in the functional nature of language, i.e. Cogni-
tive Grammar (as developed in Langacker 1987a and 1991) and Systemic-
Functional Grammar (Halliday 1994). I will go more deeply into some of
their basic tenets, such as, firstly, the interaction between language usage
and the language system (Section 1); secondly, the natural symbolic rela-
tionship between lexicogrammar and semantics (Section 2); thirdly, the
'intrinsic' functional nature of linguistic structure and the need for a func-
tional approach to the notions of 'rank shift' and 'class' (Section 3);
fourthly, Halliday's multifunctional approach to language (Section 4); and,
finally, the need for an accurate analysis of not only the syntagmatic layer
of linguistic organization, but also of the paradigmatic, systemic aspect of
language (as revealed by the schematic networks or system networks of
Introduction 5
related structures and options in which both Cognitive Grammar and Sys-
temic-Functional Grammar situate a construction) (Section 5).
In Chapter 3,1 will show how these basic theoretical assumptions enable
us to identify some of the lacunae and weak points in the existing ap-
proaches to nominalization. Importantly, these principles also point out a
possible course of action in the description of nominalized constructions.
Going through the main theoretical divisions of Chapter 2 again, I will ar-
gue, firstly, that nominalization patterns have to be categorized both by
schematization and extension. It also has to be taken into account that
nominalizations may display varying degrees of entrenchment or automati-
zation: they may be lexicalized and function as fixed expressions in lan-
guage, or they may only be established as grammatical patterns or sche-
matic units, without themselves being included as conventional units in the
language system (Section 1). Secondly, I will stress the importance of
viewing nominalizations, like all constructions, as natural symbolic units of
which the meaning is encoded in the lexicogrammar (Section 2).
Thirdly, I will argue that nominalizations have to be viewed as func-
tional configurations which are situated on a particular level or rank of
functional organization (Section 3). As such, they find themselves in be-
tween a 'higher' rank in which the nominalization itself serves a function
and a 'lower' level that constitutes the nominalization's own internal or-
ganization. A full description of a nominalized structure requires an analy-
sis of its external, synthetic functioning, of its internal, analytic functional
outlook, and of the way in which the external and internal functions are
combined in the construction itself. Put differently, some nominalized con-
structions may represent an at first sight inexplicable mixture of external,
nominal behaviour and internal, clausal categories, but the nominalized
structure itself succeeds in combining these seemingly divergent functions
into one construction and therefore has to be accounted for. Moreover, for a
number of nominalizations, we need a concept of downranking, which al-
lows the possibility of functional configurations occurring at a level of or-
ganization which is not the one on which they normally function: clause-
like nominalizations such as íAaí-clauses can then be identified as having
been downranked from clausal to nominal level while preserving part of
their internal clausal outlook. By the same token, downranking entails its
own form of reclassification and involves external as well as internal re-
classification: by being reclassified, the clausal structure takes on the
grammatical and semantic properties inherent to the nominal unit. Failure
to recognize the reclassification involved in the rankshifted type of nomi-
nalization has been one of the main gaps in the description of nominaliza-
tion so far.
6 Introduction
Fourthly, nominalizations are best described from a multifunctional per-

spective. They encode representational semantics, they are related to the
ground and they serve textual functions. Their import cannot therefore be
reduced to their representational semantics (Section 4). Finally, I will also
argue that, if it is used systematically, the paradigmatic relationship of ag-
nation forms an important descriptive heuristic in the analysis of nominali-
zation patterns (Section 5). It helps to shed light on the clausal categories
that are present in nominalization, as well as on the similarities and differ-
ences between distinct nominalization types such as factive that-
nominalizations, the fact ^/-constructions and gerundive factives.
The theoretical principles discussed in Chapters 2 and 3 reveal that an
insightful description of the functional properties of nominal and clausal
organization is needed before one can tackle those of nominalized construc-
tions. In Chapter 4, therefore, I will complete the theoretical-descriptive
framework by elaborating on the internal functional organization of nomi-
náis and clauses. I will argue that nominalization or the functional reclassi-
fication from processual starting-point to nominal construction is possible
due to the fundamental correspondences that exist between the functional
organization of the nominal and that of the clause: clausal categories can be
mapped onto what is basically a nominal structure because the functional
organization of clauses and nomináis is to a large extent equivalent (Lan-
gacker 1991). Chapter 4 will be organized around the functional categories
which Langacker (1991) identifies for the nominal and the clause, i.e. those
of type specification, instantiation, quantification and grounding. I will first
demonstrate the relevance of these functional categories to the description
of clauses and nomináis (Sections 1 and 2). In a third section, then, I will
formulate a tentative analysis of the basic constructional mechanisms that
underlie nominalization. I will argue that the integration of clausal and
nominal properties in nominalized constructions develops along the lines of
the functional categories of type specification, instantiation, quantification
and grounding.
Parts Π and ΠΙ of this volume will then present two descriptive case
studies. The nominalizations which I will analyze in them broadly cover the
spectrum of clausal levels of assembly to which nominal reclassification
can apply: they include nominalizations derived from a simple verb stem,
i.e. -er nominalizations (e.g., baker, bestseller) (Part II); nominalizations
based on what I will call an atemporal clausal head (e.g., signing the con-
tract) and nominalizations which are centred on finite clauses (e.g., that she
signed the contract, the fact that she signed the contract) (Part ΙΠ). The
latter three structural types of nominalizations will be treated together be-
cause they have part of their semantics in common and can all be 'factive'.
Introduction 7
Each case study will start with a chapter in which I look at the main in-
sights thus far offered in the literature and establish my own descriptive
position. In Chapter 5 an overview will be given of the literature on dever-
bal -er nominalization; Chapter 8 will discuss the most interesting analyses
that have thus far been proposed for the system of factive nominalization.
In my own descriptive analysis of -er derivation and factive nominalization
(developed in Chapter 7 and Chapter 9 respectively), I will attempt to show
that it is possible to move beyond existing descriptions and provide an an-
swer to some of the long-standing moot points in the analysis of specific
nominalization types if the functionally-oriented approach which I set out
in Part I is systematically applied. Importantly, my analyses will focus not
on the external functioning of the nominalized constructions, but on their
internal functional organization, i.e. on the clausal and nominal categories
which they integrate.
Apart from the general functional perspective which I will take and the
central role which I will assign to the functional categories of the nominal
and the clause, two main motifs can be said to run through both my descrip-
tive case studies. First, my analysis will be based on the assumption that the
semantics of a construction is encoded in its lexicogrammatical properties
and can therefore be revealed through a careful analysis of those properties.
I will, in other words, not only present a detailed analysis of the lexico-
grammar of -er nominalizations and factive nominalizations; I will also try
to shed light on the basic meanings which these nominalization types en-
code.
Secondly, in each of these studies, the description of the syntagmatic in-
tegration of clausal and nominal properties in the internal organization of
the nominalized constructions is supplemented with observations about the
paradigmatic relations which the nominalized structures hold. In other
words, the syntagmatic properties of nominalizations will be systematically
considered against the background of the larger system network which they
form part of and which they select from. Not surprisingly, the paradigmatic
relations that will turn out to be most valuable to the elucidation of nomi-
nalizations are those with clausal structures and with nomináis. First, due to
the clausal nature of their starting point and the clausal categories which
they embody, nominalizations tend to be related to clausal structures. The
identification of related or 'agnate' clause types will be shown to play a
crucial role in the analysis of nominalized constructions. In my description
of -er nominalizations (Chapter 7), I will thus elaborate on the systematic
relationship that exists between -er suffixation (e.g., bestseller) and the
clausal middle construction (e.g., That book sells well) and I will show how
an accurate analysis of the lexicogrammatical properties of middle clauses
8 Introduction
(presented in Chapter 6) sheds a new light on the system of -er nominaliza-

tion.
Apart from being related to clausal structures, however, nominalizations
also connect with nominal constructions: by being made to function within
the nominal paradigm, nominalizations enter into relation with other nomi-
nal structures, nominalized as well as non-nominalized. The gerundive type
of factive nominalization, that-f&ctives and the fact ¿Ααί-constructions can
thus be considered as nominalized structures which are systematically re-
lated to each other and are linked up within the system of language. In my
analysis of factive nominalization, I will show that it is only by taking into
account the properties of each of them that one can come to a more accu-
rate description of the overall system of factive nominalization and its fac-
tive reading (Chapter 9). Importantly, the various types of factive nominali-
zations do not only relate to each other: as nominal constructions, they also
enter into relation with specific types of non-nominalized nominal struc-
tures. Once again, the elucidation of the precise paradigmatic relations
which each nominalization type has with 'ordinary' nominal constructions
will be argued to be vital to the elucidation of its properties. In the Conclu-
sion (Chapter 10), finally, I will attempt to answer the question as to what
constitutes a nominalization.
Chapter 2
Theoretical assumptions
Nominalization represents a theoretical challenge and has to be situated in

an overall theory of the language system. It is therefore essential to first
clarify the basic theoretical assumptions on which my analysis of nominali-
zation is based. This is even more important as my approach is eclectic and
some of the terminology which I use requires clarification. In this chapter I
will therefore zero in on the linguistic system and discuss the various 'lev-
els' that can be distinguished in it: in a first section, I elaborate on the lan-
guage system as such and its relation to language usage·, the second section
deals with the various strata that can be distinguished in language and fo-
cuses on the natural relation between lexicogrammar and semantics; in Sec-
tion 3,1 deal with the ranks or the levels of organization that can be distin-
guished within the grammar and I introduce two of the notions that are cru-
cial for the analysis of nominalization, viz. reclassification and rank shift.
Section 4 focuses on the multifunctional approach which I advocate for the
description of nominalization patterns; and, finally, Section 5 points out the
heuristic value of considering the relations that exist between different con-
structions.
1. Language: The relation between system and usage
1.1. A usage-based approach to language
The analysis of nominalization set out in this study is rooted in a view of

language which is at once functional and cognitive. It is functional in that it
builds on the assumption that the nature of the language system is "closely
related to the social and personal needs that language is required to serve"
(Halliday 1970b: 142): the language system is viewed as functional with
respect to the particular needs of the language user. More specifically, the
language system is claimed to be shaped by the functions which it serves,
i.e. they are built into it. An analysis that is functional in design can thus
"relate the internal patterns of language - its underlying options, and their
realization in structure - to the demands that are made on language in the
actual situations in which it is used" (Halliday 1970b: 165) (in Section 4 the
10 Theoretical assumptions
functional level of organization in language will be discussed in more de-

tail).
Apart from viewing language as functional, I also adhere to the claim
that the language system - and the constructions in it - is determined by
certain cognitive abilities of the language user: the language system is
viewed as an integral part of human cognition (Langacker 1987a). Lan-
guage is held to refer to "those aspects of cognitive organization in which
resides a speaker's grasp of established linguistic convention", and the lin-
guistic abilities of the language user are believed to "not necessarily consti-
tute an autonomous or well-delimited psychological entity" (Langacker
1987a: 57): there are a number of basic psychological phenomena involved
in language use which are essential to the language system, without being
limited to it (Langacker 1999). Because the cognitive processes which they
embody equally contribute to their make-up, an analysis of linguistic struc-
tures should articulate what we know of cognitive processing, i.e. it must
have "psychological reality" (Langacker 1987a: 42).
The language system is, however, not only shaped by the functions
which it serves and by the cognitive abilities of the ones who use it: it also
itself sanctions language use. "Usage events" (Langacker 1987a: 66) or
actual "instantiations" of language (Halliday 1992: 20; Langacker 1999:
99) are sanctioned by the restrictions and options of the system of language
underlying them: they are 'systemically motivated' (Langacker 1987a:
426). As Hjelmslev (1961: 39) puts it, an actual instance of language
(which he calls a 'process' or 'text') only "comes into existence by virtue
of a system's being present behind it, a system which governs and deter-
mines it in its possible development". This is not to say that language is a
generative or constructive system which generates well-formed, novel
structures as its output: it is not the language system which is responsible
for assembling novel expressions. Construing language is viewed as "some-
thing that speakers do" (Langacker 1987a: 65), which ties in well with
Hjelmslev's 'processual' interpretation of language use (Hjelmslev 1961:
39). Actual language use is "a problem-solving activity that demands a
constructive effort and occurs when linguistic convention is put to use in
specific circumstances" (Langacker 1987a: 65). In short, the language sys-
tem may provide the necessary "symbolic potential", but it is left for the
speaker to recognize this potential, i.e. to "exploit it in a fashion that re-
sponds to all the varied constraints inherent in the situation" (Langacker
1987a: 66).
It is the dynamic interaction between system and usage, or the confron-
tation of the conventionalized language units in human cognition with the
constructive effort of the language user which forms the "crucible of lin-
Language: The relation between system and usage 11
guistic structure" and the "source of language change" (Langacker 1987a:

70). If we are to elucidate particular constructions in the language system,
therefore, we must consider them from a usage-based perspective, i.e. we
must attach substantial importance to their instantiations in actual language
use. Hjelmslev (1961: 11) speaks of the "empirical principle", which states
that "a theory must be capable of yielding, in all its applications, results that
agree with so-called (actual or presumed) empirical data". A grammar
should be at once a grammar of the language system and a grammar of the
uses which it is put to (Halliday 1994: xxii); it must work bottom-up rather
than top-down (Langacker 1988: 132).
Langacker's cognitive model of grammar is aimed at modelling the lan-
guage system in a usage-based and psychologically plausible way (Lan-
gacker 1987a, 1988, 1990, 1991, 1999). It describes the interface between
system and usage, and the insight into the organization of the language sys-
tem gained from it in detail and with reference to the various psychological
processes that play a role in it. Because Cognitive Grammar posits a fun-
damental equivalence between, on the one hand, the interface between sys-
tem and usage, and, on the other hand, the organization of the language sys-
tem itself, - both of which are argued to be modelled on the same, essen-
tially cognitive principles - I feel that it goes a long way towards the for-
mation of a usage-based theory of language. I will therefore start with a
brief discussion of Langacker's analysis of the interaction between lan-
guage use and the language system, and of the model which he proposes to
account for the organization of the language system. A more detailed
analysis of particular aspects of the language system will follow in later
sections, together with a discussion of some basic, functional theoretical
constructs which I consider to be compatible with Langacker's model and
prerequisite for a truly usage-based account of language.
1.2. Cognitive Grammar: The interface between system and usage
Langacker considers the following psychological abilities of the language

user to be crucial for language use: symbolization, composition, compari-
son/ categorization and schematization.
Symbolization builds on the general ability of association, whereby
"one kind of experience is able to evoke another" (Langacker 1999: 94).
Symbolization in language concerns the "relation between a structure in
semantic space and one in phonological space" (Langacker 1987a: 77).
When an established symbolic relationship becomes automatized and ac-
quires 'unit' status, it forms a symbolic unit (e.g., [[OPENER]/[opener]],
which associates the semantic unit [OPENER] with the phonological unit
[opener]).
Composition refers to the ability to integrate two or more (component)
structures to form a composite structure (Langacker 1999: 94). Composi-
tion gives rise to relationships on the syntagmatic plane of language (Lan-
gacker 1987a: 75). An example of a composite structure is can opener,
which combines the symbolic units [[CAN]/[can]] and
[[OPENER]/[opener]], which in turn consists of [[OPEN]/[open]] and
[[-ER]/[-er]].
Apart from symbolization and composition, the abilities of comparison
and schematization are also relevant to language use. According to Lan-
gacker (1987a), the language user solves the problem of coding a detailed,
context-dependent conceptualization in terms of a linguistic expression by
making comparisons or categorizing judgements (Langacker 1987a: 65-
73): he/she assesses whether a specific expression can be categorized as a
member of the category defined by a conventional linguistic unit or a unit
that is widely shared by the language users of a speech community. A us-
age event is therefore always systemically motivated or "structured and
evaluated with reference to the conventional units of a grammar" (Lan-
gacker 1987a: 426).
Because structures which at first sight appear very different, may be
"quite comparable in a coarse-grained view" (Langacker 1999: 93), catego-
rization necessarily involves the ability of schematization: to determine
whether a usage event satisfies the specifications of specific conventional
units, one may have to abstract away from certain points of difference and
portray the distinct structures with lesser precision and specificity (Lan-
gacker 1999: 93). A usage event can then either be identified as elaborat-
ing/instantiating a category, or it can be said to extend it. When a linguistic
category is elaborated or instantiated by a usage event, the latter conforms
to the specifications of the category but is, schematically speaking, "charac-
terized in finer-grained detail" (Langacker 1999: 93). An elaborative or
instantiating novel usage is conventional or - in more traditional terminol-
ogy - grammatical·, it is conforming to the linguistic convention embodied
by the linguistic unit, i.e. it is fully sanctioned by it. When, on the other
hand, a discrepancy exists between the category that is defined by the lin-
guistic unit and the usage event which is related to it, the category is ex-
tended and the usage event is either ill-formed or non-conventional.
In general, a novel usage, irrespective of whether it merely elaborates or
extends a conventional unit, can become conventionalized or acquire unit
status via entrenchment (Langacker 1987a: 59). When the unit [[MOUSE]/
[mouse]], for instance, was first used with reference to a piece of computer
Language: The relation between system and usage 13
equipment, its meaning was extended. This extension has by now achieved
unit status because of its frequent use and the entrenchment resulting from
it (Langacker 1999: 108). Entrenchment or the automatization of certain
structures is a matter of degree and the boundary between units and non-
units fluctuates continually because "every use of a structure reinforces it
and entrenches it more deeply, whereas non-use has the opposite effect"
(Langacker 1999: 100).
1.3. The language system as network
Crucially, the relationships of symbolization and composition, comparison

and schematization, which the language user establishes when construing
language, are also claimed to determine the internal organization of the
language system itself. The terminology used to describe language use con-
sequently applies to the language system as well.
The language system consists of three types of units: firstly, semantic
(e.g., [PRINT]), phonological (e.g., [p]) and symbolic units (the latter asso-
ciate a semantic with a phonological unit, e.g., [[PRINT]/[print]]); sec-
ondly, schematic symbolic units or grammatical patterns (e.g., [[V/...]-
[[-ER]/[-er]]), and, thirdly, the categorizing relationships of elabora-
tion/instantiation, extension and mutual similarity (Langacker 1988: 134,
147). These units have acquired unit status because they are entrenched: the
language user can employ them "in largely automatic fashion" because they
constitute 'pre-packaged assemblies' (Langacker 1987a: 57). They are con-
ventional in that they are shared and recognized as being shared by a sub-
stantial number of language users (Langacker 1987a: 62).
The simplest kind of symbolic unit is the morpheme, "in which a se-
mantic and a phonological structure participate as unanalyzable wholes in a
symbolic relationship" (Langacker 1987a: 58; see also Halliday 1961). The
combination or (syntagmatic) integration of basic symbolic units leads to
the formation of a composite symbolic structure or a grammatical construc-
tion. The term 'grammatical construction' applies to the component struc-
tures which the composite structure consists of, to the relationship of inte-
gration that exists between them and to the resulting composite structure
(Langacker 1987a: 277).
In Cognitive Grammar, a construction is situated on the syntagmatic
plane of composition, as well as on a schematic plane, which is orthogonal
to that of the syntagm and distinguishes between schemata and specific
instantiations of schemata. The patterns for combining simpler symbolic
units into larger ones are in this perspective analyzed as constructional
schemata, which capture generalizations and serve as 'templates' for the

assembly of novel expressions (Langacker 1988: 148): the schema [[V/...]-
[[-ER]/[-er]] is thus instantiated by [[OPEN/open]-[-ER/-er]]. Because it is
believed that a speaker's knowledge of a construction is not restricted to a
single rule or schema standing in isolation, but also embraces the speaker's
"knowledge of how the rule is 'implemented' with respect to more specific
structures" (Langacker 1988: 151), the more specific instances of construc-
tional schemata are argued to be part of the system of language as well. As
Langacker puts it (1987a: 29), it is "gratuitous to assume that mastery of a
rule like Ν + -s, and masteiy of forms like beads that accord with this rule,
are mutually exclusive facets of a speaker's knowledge of his language; it
is perfectly plausible that the two might sometimes coexist". In addition to
being composite, a grammatical construction is therefore also complex: it
does not reside in a single structure, but rather in "a family of structures
connected by categorizing relationships" (Langacker 1988: 149).
The 'family of structures' that makes up a complex category is best
viewed as a schematic network or a network in which "linguistic structures
of any kind and any size are linked in pairwise fashion by categorizing rela-
tionships" (Langacker 1999: 103): the superordinate nodes in the network
are then schematic and the subordinate nodes are either subschemata or
more specific structures which elaborate/instantiate or extend the schema.
An example of a schematic network is given in Figure 1 (based on Lan-
gacker 1988: 131):
SCHEMA
THING PL
X -s
DOG PL TREE PL
dog -s tree -s
Figure 1. A schematic network
In including not only the categorizing relationship of extension, which is

focused on in prototype theory (e.g., Rosch 1975, 1978, cited in Langacker
1987a; Lakoff 1987), but also categorization based on schematization (i.e.
elaboration or instantiation), the network model distinguishes itself from
prototype theory. As Langacker (1988: 139) points out, it is necessary to in-
The relation between lexicogrammar and semantics 15
elude schematization among the categorizing relationships because a cate-

gory's extension from the prototype tends to be "accompanied by a certain
amount of upward growth, as schémas are extracted to generalize over a
more diverse array of category members".
The vision of the language system that emerges through Langacker's
model "is one of massive networks in which structures with varying de-
grees of entrenchment, and representing different levels of abstraction, are
linked together in relationships of categorization, composition, and sym-
bolization" (Langacker 1999: 95). The description of a construction in-
volves the elucidation of the relationships which the construction embodies
as well as the relations which it contracts with other constructions in the
network.
In the following sections, I will zoom in on specific aspects of these re-
lationships. I will start with the relationship of symbolization and briefly
discuss the natural or non-arbitrary nature of the symbolic relation be-
tween lexicogrammar and semantics (Section 2). In Section 3, I will con-
sider the syntagmatic relations that exist between the components of a con-
struction and lay out the main properties of a radically functional view of
composition and classification. Section 4 will go more deeply into the vari-
ous types of functional relationships that are at work in linguistic patterning
and it will discuss Halliday's tristratal functional interpretation of linguistic
structure. Finally, in Section 5, I will look at the various types of paradig-
matic relationships that obtain between specific constructions in the lan-
guage system and I will point to their value as a descriptive 'heuristic'. I
will argue that, apart from the categorizing relationships of elabora-
tion/instantiation, extension and mutual similarity which Langacker dis-
cusses, the relation of agnation, introduced in Gleason (1965), should be
included in the description of linguistic patterning as well.
2. The natural symbolic relation between lexicogrammar and

semantics
2.1. A natural description of language
The idea that a description of language, apart from being usage-based,

should also be natural is prominently present both in Langacker's cognitive
model of grammar and in Halliday's systemic-functional approach. In gen-
eral, the notion of 'naturalness' refers to the concern to develop a descrip-
tion of language which "deals with data in their own terms, with full regard
for the richness, subtlety, and complexity characteristic of linguistic phe-
nomena" (Langacker 1987a: 13). A natural model of language is a model

which respects the nature of language as a complex system. In Langacker
(1987a), a natural analysis of language is, for instance, argued to accom-
modate not only discrete, but also non-discrete aspects of language struc-
ture; it is said to carry the simplifications and idealizations which are some-
times needed to begin with the analysis of language not too far, so as not to
lose sight of the actual complexity of the subject matter; and, finally, a
natural description is claimed to operate with substantive constructs, or
constructs which are sufficiently clear to be elucidated "with a certain
amount of precision and detail" and which are real in that they correspond
to "some actual feature of linguistic organization" (Langacker 1987a: 22).
One of the unnatural distinctions which circulate in linguistics is that be-
tween lexis and grammar, and, related to it, the assumption that linguistic
theory needs the constructs of lexicon and syntax to describe it. Syntax is
thought of as the "domain of generality and regularity, of productive rules
with fully predictable outputs", and the lexicon is viewed as "the domain of
irregularity, idiosyncrasy, and lists" (Langacker 1987a: 26). The distinction
between lexicon and syntax has, for instance, been argued to underlie the
differences between nominalizations such as John's refusing the offer and
John's refusal of the offer (Chomsky 1970: 187): the latter, so-called 'de-
rived' type of nominal has been argued to form part of the lexicon, mainly
because the semantic relation which it has with the verb from which it is
derived tends to be idiosyncratic. Nomináis of the former, i.e. gerundive,
type, by contrast, are claimed to be part of the syntactic component of lan-
guage, being derived from clauses by means of 'syntactic transformations'
without fundamental changes to the clausal meaning (Lees 1960; Chomsky
1970): their meaning is argued to be related to that of the underlying clause
in a regular way.
Gerundive nomináis of the 'mixed' (Chomsky 1970: 215) or 'action'
type (Lees 1960), such as John's refusing of the offer, however, make clear
that the constructs of 'lexicon' and 'syntax' lack substance and cannot be
strictly distinguished between. Like gerundives of the type John's refusing
the offer, the meaning of action nominalizations seems regular, which sug-
gests that they are 'syntactically' derived; and yet, action gerundives are
not fully productive (e.g., *the being of a woman·, *his having of a car) and
they have the internal structure of a nominal, two features which are alleged
to classify them with lexically-based nomináis. The problematic use of the
constructs of 'lexicon' and 'syntax' in the analysis of nominalizations thus
illustrates that there is no reason to assume that the distinction between
lexical and grammatical issues is a sharp one (as also argued in Langacker
1987a: 18, 26). In Halliday's work, lexis is for that reason referred to as
The relation between lexicogrammar and semantics 17
"most delicate grammar" (Halliday 1961: 267), its relation with grammar is
argued to take the form of a cline or a continuum (an idea which is also
found in Langacker's work) and the term lexicogrammar is used instead of
grammar.
Apart from the artificial distinction between lexis and grammar, the re-
lationship that has most often been misrepresented is the fundamental rela-
tion between lexicogrammar and semantics: grammar and semantics have
often been conceived as autonomous entities or separate 'components' of
language (Langacker 1987a: 12; see also Halliday 1988). A description of
language which is truly natural, however, can only be arrived at when the
relation between grammar and semantics is recognized to be natural or
non-arbitrary.
2.2. The natural or non-arbitrary relation between grammar and meaning
Language is essentially symbolic in nature in that it consists of an open-

ended set of linguistic symbols or signs, "each of which associates a seman-
tic representation of some kind with a phonological representation" (Lan-
gacker 1987a: 11). In the case of most of the lexical items which are not
further analyzable into morphemes, the symbolic association between
meaning and form is arbitrary, i.e. there is no natural relation between the
meaning of the lexical item and the particular phonological form it is
mapped onto (Saussure 1972). Well-known exceptions to the arbitrariness
of unanalyzable lexical items are instances of onomatopoeia, such as hiss
and buzz.
One of the basic tenets of both cognitive and functional models of lan-
guage is that not only lexical items but also grammatical patterns are signs
or symbolic combinations of meaning and form. As Langacker (1987a: 12)
puts it, "morphological and syntactic structures themselves are inherently
symbolic" and this "above and beyond the symbolic relations embodied in
the lexical items they employ". A grammatical construction is thus sym-
bolic not only in that it contains symbolic elements: the patterns along
which these symbolic components are integrated is held to be symbolic as
well. Grammatical patterning itself is believed to associate a particular
meaning with a particular form.1 Most importantly, and unlike in most sim-
ple lexical items, the symbolic relationship established in grammatical pat-
terns is held to be non-arbitrary or natural, with the grammar 'encoding' or
'realizing' the semantics (Halliday 1994: xvii).
However, without further qualification, the Saussurean, bipolar interpre-
taton of the linguistic sign, which posits a link between meaning and pho-
nological form, seems unfit to capture the non-arbitrary nature of gram-

matical structure: "A semiology in which meaning is said to be directly as-
sociated with phonology cannot be sensibly linked to a theory of grammar"
(Davidse 1991: 1). As already hinted at in Saussure's work, it is necessary
to posit a third or intermediate stratum, in between the semantic and phono-
logical ones, to account for the non-arbitrary nature of grammatical struc-
ture. Halliday (1987, 1992) has used Hjelmslev's model of the linguistic
sign in an attempt to clarify the relation between lexicogrammar and se-
mantics.2
Hjelmslev (1961) takes over the Saussurean concepts of content and ex-
pression to talk about the linguistic sign and proposes that the content plane
of the linguistic sign has to be further analyzed into content-form and con-
tent-substance. Content-form, which Hjelmslev describes as language-
specific encoding (Hjelmslev 1961: 52), represents according to Halliday
(1987) the lexicogrammatical stratum of language. It is by passing through
this lexicogrammatical stratum that general communicative 'purport'
(Hjelmslev 1961: 51) or meaning which "can be 'translated' between lan-
guages" (Davidse 2000a: 33) is formed into content-substance, i.e. into
sign-specific meanings, which can then be mapped onto a particular phono-
logical form (Hjelmslev 1961: 52).
The linguistic sign is, in short, not an expression that points to a content
outside the sign. Rather, it is "an entity generated by the connexion be-
tween an expression and a content" (Hjelmslev 1961: 47), with the content-
level further divided into a lexicogrammatical and a semantic plane. Cru-
cially, it is only by virtue of content-form or lexicogrammar that content-
substance or semantics exists: content-substance appears by "the form's
being projected on to the purport, just as an open net casts its shadow down
on an undivided surface" (Hjelmslev 1961: 57). Or, as Langacker (1987a:
98) puts it, semantic structure is conceptual structure that is "shaped for
symbolic purposes according to the dictates of linguistic convention". The
relationship between content-substance and content-form is therefore natu-
ral or non-arbitrary·, the meaning of a linguistic sign is encoded or realized
via the lexicogrammar of the particular language it belongs to. Any attempt
to construe semantics and lexicogrammar as discrete components of the
language system is, consequently, at odds with the very nature of the lin-
guistic sign.
To establish the meaning of 'symbolically complex' units (Langacker
1987a: 82) - i.e. units which integrate two or more symbolic units into a
grammatical construction - , then, we have to analyze their lexicogrammar
in detail. Such an analysis mainly involves an elucidation of various 'con-
texts' (Langacker 1987a: 401). In what follows, I will concentrate on the
Composition, classification andfunction 19
syntagmatic context of the linguistic sign and the relations that are estab-
lished through composition. In Section 5,1 will discuss the systemic context
of linguistic signs or their position "within the schematic networks that col-
lectively constitute the grammar of a language" (Langacker 1987a: 401).
3. Constructions: Composition, classification and function3
... the components of a complex expression should not be thought of as

providing the material used to construct it - theirfunction is rather to cate-
gorize and motivate facets of the composite structure .... (Langacker 1991:
508)
Owing to the structural diversity of nomináis, a universally valid schematic

characterization must be couched in terms of meaning and semantic func-
tion rather than formal properties such as constituency. (Langacker 1991:
54)
In this section, a radically functional approach to the analysis of construc-

tions is described, which views them as functional configurations, the com-
ponents of which are essentially component functions and not necessarily
component structures. A radically functional analysis of composition thus
posits that it is functions rather than, for instance, constituents, which form
the overarching motivating principle in composition. The main properties
of a radically functional approach to composition will be sketched out as
follows: in a first part (Section 3.1), I will focus on the role of functions in
the analysis of composition and posit that the components of a construction
are best viewed as component functions.
In Section 3.2,1 argue in favour of integrating different levels of func-
tional organization in the description of a construction, levels which toge-
ther constitute a hierarchy or 'rank scale' (Halliday 1961) of functional
configurations, ranging from 'lower' to 'higher' levels of functional orga-
nization. In Section 3.3, classes of symbolic units are argued to be best es-
tablished on the basis of their specific functional behaviour, not only in
their external functioning, but also - and this is more controversial - with
respect to their own internal organization. Langacker's (1991) description
of the four main functions of nomináis and clauses will be argued to have
paved the way for such a 'radically functional' view of class.
3.1. A functional approach to composition
In Langacker's (1987a) model of language, composition is said to involve

components or component structures which are turned into a composite
structure or a grammatical construction. The notion of 'component' captu-
res the compositional relation by which components are integrated into a
composite whole. Like the terms 'construction' and 'assembly', it hints at
the active nature of the constructive efforts of the language user. 'Compo-
nent' is preferred to the notion of 'constituent', which, as pointed out by
McGregor (1997), is too much linked to one specific type of syntagmatic
relationship, viz. that of constituency (i.e. "the syntagmatic relationship
between a part and the whole to which it belongs", McGregor 1997: 21).
The term 'constituent' moreover suggests that the components of a con-
struction are building blocks out of which the composite structure is formed
(Langacker 1987a, 1999). A 'building-block' view on composition, how-
ever, fails to account for those constructions in which "components are
only partially discernible (or even indiscernible) within the composite
whole" (Langacker 1999: 152). Even though composite structures like
these are non-prototypical or 'marked', they form an essential part of the
nature of language, and should therefore be included in an account of com-
position that aims to be natural.
It is the description of constructions in which not all components are
discernible that illustrates best the assets of a functional approach to com-
position. Components which are only partially discernible or indiscernible
cannot be referred to as component structures. As Langacker's discussion
of non-canonical nomináis makes clear, such components are best descri-
bed in functional terms, i.e. as component functions which are not realized
by specific component structures. A functional analysis of such non-
canonically coded constructions not only sheds light on the constructions
themselves, but it also succeeds in revealing what they have in common
with other, canonically coded members of their class.
Let me illustrate this briefly (a more elaborate account of the functional
interpretation of nomináis given in Langacker 1991 follows later in Chapter
4). In a canonically coded nominal construction, Langacker argues, each
semantic function that is characteristic of nomináis is associated with a dis-
tinct element of structure (Langacker 1991: 54). An instance of a canonical
nominal is, for instance, the nominal those three black cats, in which
grounding (or the link with the speech event) is effected by the demonstra-
tive pronoun those, quantification by the numeral three and the type speci-
fication is provided by black cats.4 Non-canonically coded is a nominal like
she. Even though it contains no visible structural assembly, it can be argued
Composition, classification and function 21
to realize functions that are similar to those encoded in those three black
cats: it profiles a single instance of the schematically specified type 'human
female', is inherently definite, and its designatimi is identified as being dis-
tinct from the speech-act participants (Langacker 1991: 148). It is thus the
functional analysis of she which establishes it as a member of the class
'nomináis' and links it up with other nominal constructions. (From a
strictly structural perspective, she seems to have hardly anything in com-
mon with nomináis like those three black cats.)
The analysis of a construction thus essentially involves elucidating the
functions which it expresses (e.g., Haas 1954; Halliday 1994; McGregor
1997) and constructions can be interpreted as 'configurations of functions'
(Halliday 1994). In the following section, I will zoom in on the unique role
which functions play in the description and classification of constructions.
3.2. Levels of functional analysis
In Haas (1954), it is claimed that a functional analysis of composition is

necessarily two-directional: functional relations are said to characterize a
construction both analytically and synthetically. An analytic definition of a
construction involves the analysis of the functions served by its compo-
nents, whereas a synthetic definition refers to the functions which the com-
posite whole itself plays in a larger configuration (Haas 1954: 61). A con-
struction can, in this perspective, be described in functional terms exter-
nally (synthetically) as well as internally (analytically).
Apart from the external and internal functions of a construction, the
construction itself should be considered as well, as the level at which the
various external and internal functions are integrated.5 In this way, the con-
struction itself constitutes a third level of linguistic analysis. It is not "algo-
rithmically deducible" from its components, but must be treated in a non-
reductive way, "as a separate entity in its own right" (Langacker 1987a:
87).
The levels of analysis that are needed for the description of construc-
tions reflect a kind of hierarchy among the symbolic units of a language:
symbolic units seem to be situated among layers of 'lower' and 'higher'
levels of organization, whereby the lower levels define the construction
analytically or internally, and the higher levels situate the construction ex-
ternally or synthetically. To account for this hierarchy, Halliday (1961) in-
troduced the construct of the rank scale. In his original interpretation of the
notion of 'rank' - which is very much based on a constituency-oriented
model of language - a rank grammar is said to be "one which specifies and
labels a fixed number of layers in the hierarchy of constituents, such that

any constituent... can be assigned to one or other of the specified layers, or
ranks" (Halliday 1966b: 111). The various levels of organization that can
thus be distinguished correspond to specific, determinate stretches of struc-
ture, characterized by a distinction in 'size', i.e. the sentence, clause,
phrase, word and morpheme (Halliday 1961). These levels are claimed to
be related in such a way that, "going from top (largest) to bottom (small-
est), each 'consists of one, or more than one, of the unit next below" (Hal-
liday 1961: 251).6 The analysis of a construction, then, requires that it be
accounted for at all ranks. This is called the 'requirement of total account-
ability' in Halliday (1966b: 113). A clause, for instance, is analyzed into
phrases, which themselves consist of words, which, in turn, are made up of
morphemes. The rank scale allows for downward 'rank shift', whereby a
symbolic unit is used at a rank which is either equal to or lower than its
own rank. The use of a clause as nominal, as in [That Caesar was dead]
was obvious to all (Halliday 1994: 264), for instance, is considered to be a
case of downward rank shift.
The limitations of this "strong version" of the rank hypothesis
(McGregor 1991: 121) and the need for refining and restricting it have been
repeatedly pointed out (e.g., Matthews 1966; Huddleston 1988; McGregor
1991). Firstly, it has to be acknowledged that certain items, such as con-
junctions (e.g., and, but), interjections (e.g., hey, eh) and adjuncts (e.g.,
surely, probably, yet) cannot be included in the rank scale (Matthews 1966;
McGregor 1991). Secondly, the requirement of total accountability at all
ranks is arguably more illuminating in some cases than in others. As Mat-
thews (1966: 103) points out, an utterance like Yes! consists of a morpheme
assigned to sentence level, and analyzing it as 'one sentence, which is one
clause which is one phrase which is one word which is one morpheme'
does not seem to add anything to its analysis. Many criticisms are, more-
over, provoked by the 'structuralist' formulation of the rank scale in Halli-
day (1961), where the rank scale was presented as being a rank scale of
(canonically coded) structures.
Nonetheless, the idea that constructions occupy a particular position on
a rank scale of symbolic units is valuable to the description of construc-
tions, provided it is interpreted functionally and it is not detached from Hal-
liday's overall functional approach to linguistic organization. Halliday him-
self has stressed that ranks are basically a unit's "most generalized func-
tional environment" (Halliday 1966a: 66; [italics mine]) and that the identi-
fication of the rank of a unit "is a possible first step in the specification of
what Haas (1966: 125) calls 'functional relations'" (Halliday 1966a: 65).
Moreover, Halliday (1994: 22) links the constituency analysis implied in
the rank scale to so-called minimal or functional bracketing, that is "brack-

eting together only those sequences that have some function relative to a
larger unit". (Functional bracketing differs from 'immediate' constituent
analysis, which brackets together units that may not have any function at all
and which reveals "the order in which all the pieces are put together, pair
by pair", Halliday 1994: 23.)
From a functional perspective, the rank scale can thus be viewed as a hi-
erarchy of basically functional configurations and the analysis of a con-
struction can be said to involve a description of its functioning in higher-
level configurations, a description of the lower-level functions which it em-
bodies and the integration of these two into a coherent analysis of the rank
that lies in between, i.e. that of the construction itself.
3.3. A functional approach to classification
Symbolic units are typically assigned to particular classes because of their

(external) behaviour in larger configurations and, in some cases, because of
basic semantic features which they have. In functionally-oriented ap-
proaches to language, it is usually pointed out that classes of symbolic units
are best derived on the basis of the larger structural configurations in which
they can occur, i.e. on the basis of how they function in them (e.g., Halliday
1961; Quirk et al. 1985; McGregor 1997). The class of the noun phrase or
nominal, for instance, is thus characterized as a class of symbolic units
which "typically functions as subject, object, and complement of clauses
and as complement of prepositional phrases" (Quirk et al. 1985: 245). In
Croft (2000: 16), syntactic categories, including classes, are argued to be
"construction-specific" or derivative only "from the constructions that de-
fine them".
Apart from classification on the basis of external functioning, one also
finds - be it much less frequently - notional criteria to describe a number
of basic classes such as the noun and the verb, and certain of their sub-
classes, such as count and uncount or mass nouns. Probably the most elabo-
rate notional characterization of nouns and verbs is the conceptual or 'in-
trinsic' characterization given in Langacker (1987a, 1987b, 1991). It is nec-
essarily abstract or schematic because of the extremely heterogeneous na-
ture of nouns and verbs (Langacker 1987b: 58; 1991: 15). According to
Langacker's intrinsic characterization, a noun profiles a thing or "a region
in some domain, where a region is defined abstractly as a set of intercon-
nected entities" (Langacker 1991: 15). Verbs, on the other hand, are basi-
cally relational and profile 'interconnections' (Langacker 1991: 19). Within
the noun class, then, the distinction between the subclasses of count and
mass nouns mainly depends on "whether the profiled region is construed as
being bounded within the scope of predication" (Langacker 1991: 18):
count nouns (e.g., lake) include the boundaries of what they designate in
their scope of predication, whereas mass nouns (such as water) do not.
The external, grammatical functioning of a symbolic unit and the highly
schematic semantic characterizations of some basic classes have weighed
heavily on discussions about grammatical classes. Interestingly, internal
properties have not: internally, the various instantiations of a particular
class are typically considered to be too diverse to be generalized across.
Nomináis, which vary from having a common noun as head, to being a
proper name or a pronoun, seem to be no exception to this. The internal
structural diversity that is characteristic of many classes has, in fact, explic-
itly been said to render classification on the basis of internal properties im-
possible. Halliday (1961: 261) thus states that a class is "not a grouping of
members of a given unit which are alike in their own structure": he (1961:
261) holds that, in terms of the rank scale, classes are derived 'from above'
and not 'from below'.
In what follows, an alternative approach to grammatical classes will be
presented which does build in the perspective from below. In a first part, I
will argue that a radically functional approach to classification includes the
internal functional organization of symbolic units, irrespective of whether
the internal functions are symbolized separately or not. A functional ap-
proach to grammatical class which includes the internal functions in its
analysis manages to identify symbolic units which are structurally widely
divergent as members of one and the same class. In a second part, I will
argue that classes which are functionally derived allow for a finer-grained
description of what can be said to constitute a prototypical instantiation of a
class.
3.3.1. An internal functional analysis of constructions
A radically functional approach to classification identifies classes not only

in terms of their behaviour in larger functional configurations, but also by
the presence of a particular set of component functions. More particularly,
it views classes as combinations of certain external and internal functions.
The inclusion of internal functional properties in the classification of a
symbolic unit is in accordance with Haas's claim that "many constructions
can be adequately defined only by a combination of analytic and synthetic
criteria ..., and a classification of them will require a similar combination"
(Haas 1954: 70). Until recently, it was difficult to see how such an analysis
could be envisaged of the nominal, in which, in view of the topic of this
study, we are particularly interested. While generalizations that deal with
the nominal's external functioning are fairly readily arrived at (as, for in-
stance, in Quirk et al. 1985), internal functional generalizations are not.
This is not only because of the apparent disparity of the units that can func-
tion in the nominal slots of a construction, but also, and more fundamen-
tally, because the question of the internal functioning of the nominal has
tended to be approached from an exclusively structure-based perspective. It
is only Langacker's (1991) analysis of the internal functional organization
of the nominal that makes clear that the various instantiations of the class
can be generalized across also internally. Langacker's analysis is pathfind-
ing, both descriptively and theoretically, and adopts a view of the nominal
which is very much 'radically functional' in that it recognizes that "seman-
tic function (rather than constituency) is the critical factor for understand-
ing their internal organization" (Langacker 1991: 51).
The semantic functions which, according to Langacker, are characteris-
tically realized by nomináis are those of type specification, instantiation,
quantification and grounding (Langacker 1991). The "universally valid
schematic characterization" which Langacker proposes for the nominal is
that it profiles "a thing construed as an instance of some type and further
incorporates some specification of quantity and grounding" (Langacker
1991: 54). Simple nouns provide nothing more than a type specification:
they specify "the basis for identifying various entities as being representa-
tives of the same class", but are "not tied to any particular instance of that
class" (Langacker 1991: 53). Full nomináis such as the site, an excellent
site and two convention sites in the Midwest, on the other hand, presuppose
instantiation of the type in question and designate one or more instances. In
full nomináis, "information is furnished concerning both the number of in-
stances and their status vis-à-vis the speech-act participants" (Langacker
1991: 53).
These semantic functions, Langacker (1991: 54) argues, are realized in
all nomináis, be it in many different ways. If they are coded separately in
the nominal's structure, then the head noun, together with its adjectives and
other modifiers that render it more precise, provides the type specification
(e.g., excellent convention site in the Midwest); the quantifier is added as a
separate layer (e.g., three black cats), and a grounding element is appended
at the outermost layer (e.g., those three black cats). Many nomináis, how-
ever, fail to display this type of assembly, in which case "the semantic
functions in question are not uniquely associated with distinct levels of
constituency" (Langacker 1991: 54). Apart from pronominal nomináis
(such as she, which was discussed earlier), proper names probably repre-
sent one of the most extreme cases of departure from the iconic type of
coding: in proper names, Langacker (1991: 59) proposes, "type, instantia-
tion, quantity, and grounding are conflated in a single expression". In the
proper name Stan, for instance, a type specification is incorporated (that of
'male human'), and because it characterizes a specific person, it presup-
poses instantiation and quantification. Grounding is subsumed as well, "for
the nominal is definite and portrays the profiled individual as being
uniquely apparent to the speaker and hearer on the basis of this name
alone" (Langacker 1991: 59). Because these functions are fulfilled, the pro-
per name qualifies as a nominal, and this in spite of its non-canonical cod-
ing (Langacker 1991: 59). As Langacker (1991: 53) concludes, the func-
tions of type specification, instantiation, quantification and grounding thus
allow us to "profitably examine the organization of nomináis from the
standpoint of semantic function, abstracting away from any details of struc-
tural implementation".
Classification by internal functional properties should be distinguished
from intrinsic or notional classification. Notional classification makes ref-
erence to abstract cognitive events and, to make sure that all members of
the class (including the less prototypical ones) instantiate the class schema,
it is necessarily highly schematic (Langacker 1987b: 54).7 As Langacker
(1991: 52) himself points out, the characterization 'thing or region in some
domain' applies to nominal predications "as a broadly defined class": it is
too abstract to distinguish simple nouns like site from full nomináis like an
excellent convention site, "each of which profiles a thing and qualifies as a
noun in this inclusive sense of the term" (Langacker 1991: 52). The intrin-
sic classification of nomináis is especially meant to distinguish them from
relational predications. The functional description in terms of type specifi-
cation, instantiation, quantification and grounding, in contrast, applies to
the internal organization of nomináis and helps to identify the differences
between nominal constructions and simple nouns (Langacker 1991: 53).
Classification in general, then, is intrinsic as well as functional, and the
functional criteria involve both the external functions served by the sym-
bolic unit and the internal functions of its components.
3.3.2. Classes as complex categories: Schema and prototype
Langacker's functional description of nomináis shows that it is feasible to

include the internal functional organization of a construction into its classi-
fication: the functional characterization of nomináis as 'grounded instances
of some type', for instance, is applicable to all the members of the class
'nominal'. Put differently, all nomináis instantiate the same schema, which
"captures the pertinent generalization" (Langacker 1988: 130). All nomi-
náis are fully compatible with its specifications, but characterize them in
finer detail or "elaborate the schema in different ways along various pa-
rameters, to yield more precisely articulated notions" (Langacker 1987a:
68). By combining a schema with various instantiating structures, classes
form complex categories, or categories residing "in a family of structures
connected by categorizing relationships" (Langacker 1988: 149).
Among the categorizing relationships in a complex category, however,
Langacker distinguishes not only categorization by schema, but also cate-
gorization by prototype or extension. He (1987a: 371) views the two modes
of categorization as "inherently related and describable as aspects of a uni-
fied phenomenon", and his network model is a 'synthesis' of categorization
by prototypes and by schémas. Categorization by prototype offers a per-
spective on classification which is not so much alternative as complemen-
tary to that of categorization by schema. Both categorization by schema
and categorization by prototype can be said to involve comparing a usage
event or target structure with a conventionalized unit or standard. Instantia-
tion implies that all the specifications of the standard are satisfied by the
usage event. The standard is then viewed as a schema which only differs
from its instantiations in degree of schematicity or 'granularity'. This is
true of, for instance, the schematic characterization of nomináis as
'grounded instances of some type'. Extension of the standard, on the other
hand, implies that the target structure is only partially sanctioned by the
standard or deviates from it. The standard is then conceived as the proto-
type of the category.
It follows that a 'natural' approach to classification should not only look
for schematic generalizations, but should also take account of the occur-
rence of prototypical and less prototypical instances of a class (Langacker
1987a: 371). Instead of considering all members of a class as being fully
compatible with a schema, categorization by prototype judges class mem-
bership through a "perception of similarity" (Langacker 1987a: 69), which
is based on what is conceived to be a prototypical instance of the class. Pro-
totypicality thus results in "degrees of membership based on degrees of
similarity" (Langacker 1987a: 371; [italics mine]). The criteria used to set-
tle on the prototype of a particular class - and with them, the prototypes
themselves - are rather divergent. Semantic/intrinsic arguments can be dis-
tinguished from arguments that concern the structural implementation of
specific internal functions.
With regard to the general classes of language, Langacker argues that

they are "centered on a prototype that reflects a basic aspect of our every-
day experience" (Langacker 1991: 521). The archetypal conception of
'physical object' is thus generally held to constitute the semantic prototype
for nouns. In his functional analysis of nomináis, then, Langacker takes the
structural way of coding as starting-point: prototypical are held to be nomi-
náis "whose structure most directly mirrors semantic function" (Langacker
1991: 143), i.e. nomináis which reflect their component functions through
distinct and overt levels of constituency.8 It follows that, in Langacker's
view (as opposed to, for instance, that of Van Langendonck 1999), proto-
typical nomináis incorporate both a head noun and a determiner (Langacker
1991: 143). Examples of prototypical nomináis are this cup, some milk, a
big house, those three black cats. Non-prototypical are nomináis such as
three from Toledo, the poor, for Harold to resign·, they occur less fre-
quently or "only in special grammatical circumstances" (Langacker 1991:
143).
Interestingly, by linking prototypicality to iconicity, Langacker's analy-
sis of the internal organization of nomináis also allows one to distinguish
degrees of non-prototypicality: some nomináis, for instance, deviate from
the prototype in that the semantic functions which they realize are associ-
ated with the nominal as a whole (e.g., personal pronouns, proper names,
ίΑαί-nominalizations; Langacker 1991: 148), others consist of two nomi-
náis, related by means of apposition (e.g., the fact that whales are mam-
mals·, Langacker 1991: 149); still others lack a head noun because the
grounding predication itself functions as nominal (e.g., that, these, some,
any).
3.4. Composition and classification: Conclusion
In short, a radically functional approach to composition gives priority to

functions rather than to constituents and posits that an accurate description
and classification of symbolic units has to be based on a functional analysis
which includes the external and the internal functional organization of the
unit in question. These functional levels are situated on a rank scale of
functional configurations.
Yet another characteristic of a functional analysis of linguistic pattern-
ing is that it considers the components of a composite whole to serve sev-
eral functions at the same time. This kind of 'multifunctional' view on lan-
guage will constitute the focus of the next section.
Functional layers of organization 29
4. Functional layers of organization
The key to a functional interpretation of grammatical structure is the prin-

ciple that, in general, linguistic items are multifunctional. (Halliday 1994:
30)
Many of the functional theories of language are 'multifunctional' and posit

that constructions are characterized by different layers of functional organi-
zation (e.g., the Prague School; the functional schools of Dik and of Givón;
the systemic-functional school of Halliday). The multifunctional approach
to linguistic organization which I present here is that of Halliday's sys-
temic-functional model (with McGregor 1997 as most interesting recent
development in it), and this because of its emphasis on the semiotic princi-
ple or the correlation between symbolization and semantic value for each
layer of organization.
The systemic-functional interpretation of the multifunctional hypothesis
can be illustrated by means of the categories of Subject, Theme and Agent
in English (based on Halliday 1970b, 1988, 1994). If we take an English
clause like
(1) The duke gave my aunt this teapot.
the element considered as Subject is the duke. Its role in the clause can be
characterized in various ways: the duke can be analyzed as the Agent of the
process with which it is construed (i.e. gave); it can be said to be "the con-
cern of the message" (Halliday 1994: 30), and it is that of which something
is predicated. Not all English Subjects, however, realize these three fea-
tures. Consider the following clause:
(2) This teapot my aunt was given by the duke.
The Subject here is my aunt, but the Agent-role is realized in the fry-phrase
(i.e. by the duke) and the message concerns this teapot. The latter two fea-
tures - the role which a unit plays in the process designated by the main
verb and being the concern of the message - appear to be features which
can, but need not be, associated with the Subject: they can also be realized
by other components of the clause.9
What at first sight appear to be characteristic properties of a single unit
thus turn out to be basically functions on the clausal level. Rather than be-
ing properties or aspects of a specific unit (in this case, the Subject-unit),
they are distinct functions which are either mapped onto each other in a
single component (as in 1) or are realized by distinct components (as in 2).

In the case of a unit such as the duke in (1), Halliday has labelled its various
functions as Theme, Subject and Actor: as a Theme, the duke is the "point
of departure" of the message conveyed by the clause (Halliday 1994: 34);
the duke is also Actor or the "doer of the action" (Halliday 1994: 30); and,
finally, in its function of Subject, the duke is the one "on which rests the
truth of the argument" (Halliday 1994: 30). A unit such as the duke in (2) is
thus multifunctional, i.e. it serves several functions in the clause at a time.
To fully appreciate Halliday's multifunctional view on linguistic units,
three aspects of it have to be further looked into, viz. the so-called meta-
functional origin of the functions which it involves, its conflgurational na-
ture, and, thirdly, the significance that is attributed to the paradigmatic or
systemic choices that lie behind each metafunctional configuration. The
paradigmatic or systemic aspect of constructions will constitute the focus of
Section 5. The metafunctional and conflgurational nature of Halliday's
multifunctional approach to linguistic organization will be elaborated now.
Let us start with the notion of 'metafunction': what characterizes a
metafunction and how does this relate to the «íw/rifunctional view on lin-
guistic units? Language users are believed to select from among a large
number of interrelated options which together constitute as it were the
'meaning potential' of language (Halliday 1970b: 142). It is one of the most
fundamental claims of the systemic-functional approach to language that
the options which are embodied in this meaning potential and which are
reflected in the grammatical options of a language "combine into a very
few relatively independent 'networks'" (Halliday 1970b: 142). Within the
massive network of options that constitutes the grammar of a language, one
can thus distinguish certain clusters or sub-networks in which the "selec-
tions made by the speaker at one point tend to determine, and be deter-
mined by, the selections he makes at another" (Halliday 1979: 61). These
clusters of lexicogrammatical options have been interpreted by Halliday as
very general functions of language and they have therefore been called
'metafunctions' (Halliday 1994). Halliday distinguishes the ideational, the
interpersonal and the textual metafunction.
The ideational or representational metafunction covers all those options
in language which enable language users to express their "experience of the
real world, including the inner world of ... [their] own consciousness"
(Halliday 1970b: 143). These options either represent experience directly in
terms of processes, participants and circumstances (i.e. experientially), or
they represent experience indirectly, in terms of the fundamental logical
relations in language (i.e. logically) (Halliday 1979: 59). hi the clause, for
instance, the grammatical system of transitivity, which construes the world
Functional layers of organization 31
into a set of process types and participants, is basically experiential in func-

tion. The function of Actor served by the duke in example (1) is an experi-
ential function. Halliday argues that the logical metafunction determines
the relations that exist between the various clauses within a clause complex
(e.g., between the clauses in I would, if I could, but I can't; Halliday 1994:
218); but it is also at work in the nominal, where it imposes a head-modifier
structure (e.g., those two splendid old electric trains, with trains as head;
Halliday 1994: 191).
The interpersonal metafunction concerns the communicative function of
language or language as exchange: it incorporates the grammatical options
which a language user has to adopt a particular speech role and assign one
to others (e.g., making assertions, asking questions, giving orders). The in-
terpersonal metafunction also covers the linguistic options which the lan-
guage user has for relating what he says to the moment of encod-
ing/decoding, i.e. for grounding it in the speech event. The principal gram-
matical system for the interpersonal metafunction of language is that of
mood. The function of Subject is identified by Halliday as an interpersonal
function (Halliday 1970b, 1994; for a more elaborate discussion of the Sub-
ject, see Chapter 4). The ideational and the interpersonal metafiinctions to-
gether constitute the two main kinds of meaning in language. The third,
textual metafunction "breathes relevance into the other two" (Halliday
1994: xiii): it covers the options found in language "for making links with
itself' (Halliday 1970b: 143), i.e. for creating texts. The clause, for in-
stance, is organized as a message "by having a special status assigned to
one part of it", called the Theme. The latter then combines with the rest of
the clause into a message (Halliday 1994: 37).
In the metafunctional hypothesis lies the intrinsic character of Halli-
day's systemic-functional approach to language: the metafunctions link up
the two general purposes of language in life (the ideational and the 'active'
or interpersonal; Halliday 1994: xiii) with particular options that can be
chosen by the language user. Importantly, each of the metafunctions links
up with particular lexicogrammatical realizations: they represent form-
meaning, or better, form-function units (McGregor 1997) and cannot there-
fore be characterized as either 'semantic' or 'syntactic' functions (as are,
for instance, the functions in Dik 1980).10
A second main characteristic of Halliday's multifunctional hypothesis is
its configurational focus, which follows from the fact that the meanings of
particular functions necessarily derive from their relationship to the other
functions with which they are structurally associated: the function of Actor,
for instance, is interpretable only in its relation to other functions of the
same kind, i.e. other 'representational' functions such as Process and Goal
(Halliday 1994: 35). It is "the structure as a whole, the total configuration

of functions, that construes, or realizes, the meaning" (Halliday 1994: 34).
Each of the three metafunctions is therefore held to be reflected in a struc-
ture or configuration of functions (Halliday 1994: 30). The strands of
meaning which the three metafunctions form thus constitute three different
structural layers. Importantly, because not every element in a construction
discharges a function in every metafunction or layer of structure, construc-
tions cannot always be analyzed exhaustively in terms of all of the meta-
functions (Halliday 1994: 35). Some elements thus have a purely textual
function (e.g., conjunctions), while others serve a primarily interpersonal
function (e.g., modal Adjuncts such as probably, certainly, actually).
It is at the level of the clause that these metafunctional layers of organi-
zation are most clearly in evidence (Halliday 1979: 77). An example of a
three-layered metafunctional analysis of a clause is given in Table 1: the
first strand of analysis (consisting of the functions of Theme and Rheme) is
that of the textual metafunction; the second layer is the interpersonal one
(Complement-Subject-Finite-Predicator-Adjunct), and the third layer repre-
sents the experiential functions realized by the clause (Goal-Recipient-
Process-Actor).
Table 1. A multifunctional analysis of the clause
This teapot my aunt was given by the duke

Theme Rheme
Complement Subject Finite Predicator Adjunct
Goal Recipient Process: material Actor
It has repeatedly been pointed out by Halliday that this type of 'segmen-
tal' analysis of the clause does not "tell the whole story" (Halliday 1994:
36). For one thing, it captures the experiential metafunction more accu-
rately than it does the interpersonal and the textual metafunctions. In his
radically semiotic model of language, McGregor (1997) therefore argues
that syntagmatic relationships other than that of constituency underlie the
interpersonal and the textual metafunction. He proposes to view the inter-
personal metafunction, for instance, as a grammatical sign of which the
formal aspect is formed by the syntagmatic relationship of 'conjugation'.11
Conjugational relations are basically whole-whole relationships, which ei-
ther involve what McGregor (1997: 210) calls 'scoping' ("in which a unit
applies over a certain domain, leaving its mark on the entirety of this do-
main") or 'framing' ("in which a unit delineates the domain over which it
Constructions and the relations between them 33
applies, marking it off from everything else"). Μ the clause, the declarative
and interrogative moods are thus claimed to have 'scope' over the full
clause. An example of framing is formed by inter-clausal relationships
whereby one of the clauses expresses 'represented speech': like "a frame
around a picture, the representing clause delineates the represented clause
from its context... indicating that it is to be viewed or evaluated in a dif-
ferent way" (McGregor 1997: 253).
5. Constructions and the relations between them
[The] Language system involves regularly recurring patterns in sentence

structures. ... Such recurrent patterns within sentences, however, do not
exhaust the system of a language. There is an additional set of relation-
ships that grammar must describe, and this is at least as large and diverse.
These exist between pairs of sentences.... (Gleason 1965: 195)
Ultimately, the grammarian takes account in his statement of those rela-

tions which seem to fit into an interconnecting system ramifying far and
wide through the language. (Gleason 1965: 202)
It is a basic tenet of systemic-functional linguistics that functions cannot

only be defined by reference to the (syntagmatic) configuration of functions
which they belong to, but that they are primarily options or choices from a
paradigmatic system: a function in the system is held to be defined by the
total set of features from which it has been selected (e.g., 'past' is defined
by reference to 'present' and 'future'), just like a structural function is de-
fined by the total structural configuration it occurs in (e.g., 'modifier' by
reference to 'head') (Halliday 1970b: 61). The metafunctional layers of or-
ganization in the syntagm are, in other words, based on paradigmatic sys-
tems which, like them, are metafunctionally motivated. A systemic or para-
digmatically oriented description - i.e. a description of the systems that lie
behind particular functions in a syntagmatic configuration - is therefore
considered to necessarily complement a syntagmatic or structural descrip-
tion. Hence the term systemic-functional.
In systemic-functional grammar, a 'paradigmatic' description takes the
form of a system network, which maps out the various options of the lan-
guage user in hierarchically ordered systems. For each metafunctional layer
of organization in a construction, system networks of interrelated choices
can be set up. Any option in the system can then be linked to particular
wordings or instances of language use. An example of a system network
which accounts for part of the interpersonal choices made in the English
clause would be the one in Figure 2 (based on Halliday 1976: 93): it gives
an overview of the basic mood system of the clause and the various options
in it, each of which stands for a list of constructions. The option 'interroga-
tive, yes-no question', for instance, represents clauses such as Will you
close the door? and Did she climb that mountain?.
yes/no
• indicative
• interrogative
[
WH-
assertion
clause
• declarative
C
exclamation
jussive
• imperative
C optative
Figure 2. An example of a system network (Halliday 1976: 93)
In what follows, I will elaborate on the systemic or paradigmatic side of

constructions and, following Gleason (1965), I will point to the theoretical
and methodological significance of including into the analysis of particular
construction types the relations that exist between the constructions of the
language system. In terms of Halliday's system networks: the relations that
exist among the various options (i.e. constructions) of a system network are
highly relevant to the analysis of any construction that belongs to it. Or, in
terms of Langacker's network model of language: not only the relation be-
tween a construction and the schema which it instantiates/elaborates or that
between a construction and the prototype which it extends should be ana-
lyzed, but also the relation between the various instantiating constructions
themselves, as well as that between constructions which (seemingly) in-
stantiate distinct schemata. I will first zoom in on Gleason's notions of ena-
tion and agnation and discuss the inter-constructional relationships which
they represent (Section 5.1). In a second part, the theoretical significance of
agnation and enation, and their heuristic value will be pointed out and illus-
trated (Section 5.2).
5.1. Agnation and enation: A definition
What do the notions of 'agnation' and 'enation' stand for? Gleason (1965)
proposes the terms 'agnation' and 'enation' to describe two basic kinds of
relationships which he distinguishes between constructions. The terms are
derived from Latin enatus 'related on the mother's side' and agnatus 're-
lated on the father's side' and they are deemed appropriate to designate
"two contrasting types of relations, neither of which can exist without the
other" (Gleason 1965: 199fn2). The relationship of enation links up con-
structions with "identical structures", that is, "the elements ... at equivalent
places ... are of the same classes, and ... the constructions in which they
occur are the same" (Gleason 1965: 199). The relationship between the
clauses in (3a) and (3b) is one of enation:
(3) a. The dog bit the man.

b. The cat ate the canary.
Agnation is a relationship between two grammatical structures which

have the same major lexical items, but are different in structure, as in (4a)
and (4b). They are agnate because their "relation in structure is regular and
systematic", i.e. "it can be stated in terms of general rules" (Gleason 1965:
202).
(4) a. The dog bit the man.

b. The man was bitten by the dog.
Importantly, for a relation to be identifiable as agnation, it must involve

large numbers of constructions. Rather than being a "unique relation be-
tween isolated pairs" of constructions (Gleason 1965: 202), agnation re-
quires that groups of constructions are found which are agnate or system-
atically related to each other in a similar way. Agnation "is based on the
pervading patterns of the language", and it is, therefore, "always a recurrent
thing, involving large numbers of sentences" (Gleason 1965: 202). As the
examples in (5) and (6) make clear, the identification of groups of agnates
(linked up by means of <=>) always presupposes the identification of enate
constructions (indicated by I I I ) (based on Gleason 1965: 202):
(5) He saw it. <=> It was seen by him.
He heard it. <=> It was heard by him.
He felt it. « It was felt by him.
(6) He galloped the horse. <=> The horse galloped.

Ill Ill
He marched the soldiers. Ό The soldiers marched.
Ill
He walked the dog. <=> The dog walked.
It is only because the clauses He heard it and He felt it, which are enates of
He saw it display a relation of agnation with It was heard by him and It was
felt by him, which are enates of It was seen by him, that the relationship
between He saw it and It was seen by him in (5) can be analyzed as being
one of agnation.
Likewise, relations cannot be identified as enate if they do not have
identical sets of agnates. Constructions such as The man saw a stranger and
The man seemed a stranger, for instance, are not enate because they belong
to different agnation networks: A stranger was seen by the man is possible,
while A stranger was seemed by the man is not. The man seemed to be a
stranger, on the other hand, is acceptable, while The man saw to be a
stranger is not (Gleason 1965: 203). The identification of agnate structures
is, in other words, an ideal heuristic tool for disambiguating constructions
which are seemingly identical (Davidse 1998a: 283). Because it is possible
that constructions share some agnates, but behave differently with respect
to others, Gleason also recognizes partial enation or enation between struc-
tures which are only partially structurally identical (which is elaborately
illustrated in Laffut 2000). As pointed out by Davidse (1998a: 283), it
seems consistent with Gleason's way of thinking to speak of a relationship
of non-enation when constructions are structurally non-identical. The rela-
tionship between The man saw a stranger and The man seemed a stranger
can then be labelled as being one of 'non-enation'.
In short, relationships of agnation and enation cannot be stated clearly
without one another: they are mutually defining notions that make up a
two-dimensional set to analyze constructions (Gleason 1965: 201). Most
importantly, they show that "formal evidence is not restricted to 'observ-
able characteristics' of the syntagm in question, but ... also includes the
syntagm's systematic relation to its agnates" (Davidse 1998a: 284). In more
mainstream terminology, one would say that formal evidence includes the
syntagm's behaviour with regard to syntactic tests (Davidse 1998a: 284).
5.2. Agnation and enation: Theoretical significance and heuristic value
Gleason's claim that agnate and enate relationships form an integral part of
the analysis of a construction is theoretically significant in that it assigns a
central role to the paradigmatic or systemic aspect of language use. As
pointed out by Davidse (1998a), it has had a great impact on Halliday's
thinking, and even though Halliday has never theorized much about agna-
tion (Davidse 1998a: 287), his identification of system networks and,
linked to it, of the various metafimctions, is based on inter-constructional
relations. For one thing, the categories in his system networks essentially
represent the proportionality of relations between constructions. As Halli-
day puts it, a category's label is "no more than the name of a proportional
relation" between constructions (Halliday 1994: xxxii).
Proportional relations and, consequently, linguistic categories are de-
fined by agnation and enation (Davidse 1998a: 287): more precisely, they
are established as well as semantically interpreted by means of the two-
dimensional set of agnation/enation. To give an example, the categories in
the system network in Figure 2 represent constructions of the following
types:
(7) a. You opened the door, (declarative: assertion)

b. Who opened the door? (interrogative: WH-question)
c. Did you open the door? (interrogative: yes-no question)
d. Open the door! (imperative: jussive)
Between constructions such as those in (7a) and (7c), situated on the hori-
zontal level of (8), there exists a proportional relation of agnation, which is
confirmed by the enate constructions on the vertical axis:
(8) You opened the door. :: Did you open the door?
She climbed the mountain. Did she climb the moun-
tain?
He washed the car. Did he wash the car?
The proportionality which these constructions show (indicated by '::') cen-

tres on their Subject-Finite unit: more particularly, in the paradigm in the
left-hand column, the Subject precedes the finite verb, while in the para-
digm on the right-hand side, it is the Finite that precedes the Subject. It is
this proportionality which is captured by the categories labelled as 'declara-
tive' and 'interrogative'. Importantly, these categories also derive their
'meaning' from it: linked to the declarative order of the mood element is
the interpersonal speech function of 'giving information', while the inter-
rogative involves the demand of information (Halliday 1994). On a inore
general level, the Subject-Finite unit can thus be said to serve a central role
in what is called the interpersonal function of language.
Particularly interesting, then, is that the inter-constructional relations
that underlie the categories in Halliday's system networks differ depending
on the metafunctional layer of language which they represent (Davidse
1998a): while interpersonal networks such as the one in Figure 2 map out
agnate structures (as shown in 7) and consist of categories "dealing with
clausal variants that are, by and large, applicable to any one clausal syn-
tagm" (Davidse 1998a: 290), agnation has a different use in system net-
works of the ideational domain. There, agnation is mainly used to distin-
guish non-enate structures: the categories of 'intransitive', 'transitive' and
'ditransitive' clauses, for instance, do not label proportionalities of agnate
structures, but of non-enate constructions (Davidse 1998a). Consider the
following ideational network (based on Halliday 1968):
s - intransitive (Actor - process)

e.g., John is running
clause
goal-directed (Actor - process - Goal)
e.g., John hit the ball
transitive
descriptive (Initiator - process - Actor)
e.g., John marched the prisoners
Figure 3. An example of an ideational system network
The non-enate status of the constructions John hit the ball and John
marched the prisoners, and, consequently, of the categories 'goal-directed'
and 'descriptive', is revealed through the different sets of agnates which
they can be related to (Davidse 1998a: 293). The construction John
marched the prisoners is part of the paradigm illustrated in (9):
(9) a. John marched the prisoners.

b. John made the prisoners march.
c. The prisoners marched.
d. What the prisoners did was march.

e. What John made the prisoners do was march.
As shown in (10), the construction John hit the ball does not belong to that
same paradigm:
(10) a. John hit the ball.

b. *John made the ball hit.
c. *The ball hit.
d. *What the ball did was hit.
e. *What John did to the ball was hit.
The categories in the ideational network in Figure 3, in short, represent

non-enate constructions, identified on the basis of the distinct agnation
paradigms 'behind' them. Importantly, these agnation paradigms also serve
"as a heuristic to interpret the semantic difference" between John hit the
ball and John marched the prisoners (Davidse 1998a: 293): the agnates of
John hit the ball show that the ball is the 'undergoer' or Goal of the proc-
ess, whereas those of John marched the prisoners reveal that the prisoners
is Actor.
Crucially, the relations of agnation and non-enation, which seem to un-
derlie Halliday's metafunctional distinction between interpersonal and
ideational categories, can be said to tie in with the overall import of the
metafunctions. The interpersonal layer of organization in language is
mainly concerned with fitting representational content into the communica-
tive event of the language exchange. It represents different ways of constru-
ing one and the same experiential meaning. It seems natural, then, that the
various construals which the interpersonal function of language can give
rise to, are linked up by means of relationships of agnation, which relate
constructions of which the lexical items are identical, but the structure is
different. The representational metafunction of language, in contrast, de-
scribes and categorizes the language user's experience of the world and
seems to do that primarily by ordering or classifying the processes and par-
ticipants in it into distinct categories. Not surprisingly, these categories are
not agnates of each other (representational functions do not operate with
identical lexical items organized in different patterns). Rather, they are non-
enate, i.e. they are structurally, nor lexically identical.
In short, the relations of agnation, enation and non-enation form an im-
portant undercurrent of Halliday's Systemic-Functional Grammar: they
motivate the categories or options that are distinguished in his system net-
works and thus help to draw up the paradigmatic side of a construction,
and, ultimately, the system of language. When used as a heuristic for the
description of particular constructions, they seem to serve one of two pur-
poses: either they disambiguate structures which seem identical, but are
non-enate and they help to interpret the semantic differences between them
(e.g., John hit the ball vs. John marched the prisoners)·, or they reveal the
various construal types which a construction can occur in and indicate the
semantic impact of each construal type (e.g., He opened the door vs. Did he
open the door?). In any case, the relations which constructions hold with
other constructions seem to form an essential part of the 'family of struc-
tures' that constitutes a grammatical construction (Langacker 1988: 149).
6. Theoretical assumptions: Conclusion
In this chapter, I have set out the overall cognitive-functional view of the
language system in which my analysis of nominalization has to be situated.
Among the most prominent features of this cognitive-functional perspec-
tive, I have argued, are: its usage-based character (Section 1.1); the atten-
tion that is devoted in it to both schematization and extension or prototypi-
cality (Sections 1.2 and 1.3); its view of grammatical constructions as natu-
ral symbolic units which encode meaning (Section 2.2); its multifunctional
analysis of constructions (Section 4); and, finally, the interest which it
shows in the systemic or paradigmatic side of constructions and the impor-
tance which it attaches to inter-constructional relationships (Section 5). I
have also argued in favour of a radically functional approach to composi-
tion and to the construct of 'class' (Sections 3.1 and 3.3) and I have pointed
to the existence of a functional hierarchy or rank scale on which construc-
tions are situated and which identifies them as combinations of specific
external and internal functions (Section 3.2).
In the next chapter I will confront these theoretical insights with analy-
ses that have been suggested for nominalization. I will show how the cogni-
tive-functional framework set out in Chapter 2 not only enables us to iden-
tify the lacunae and weak points in the literature but also gives an indica-
tion as to how to move beyond them and set up a more coherent and sys-
tematic account of the constructional mechanisms that underlie nominaliza-
tion.
Chapter 3
Nominalization
Nominalization is not only pervasive but theoretically significant ....

(Langacker 1991: 22)
The focus of this study is on deverbal nominalizations, i.e. on nominal con-

structions which, at some layer of their assembly, incorporate a reclassified
verbal component. The nominalizations included in the discussion differ as
to the level of assembly at which the nominalization process takes place:
firstly, they may be structures that have been transcategorized from verb to
noun class, i.e. nominalizations at the level of the word and with a simple
nominal head (examples 1 and 2); secondly, they may nominalize structures
that are intermediate between a verb and a full clause and thus form com-
plex nominal heads (illustrated in 3); thirdly, they may consist of clauses
functioning as full nomináis (as in 4) (unless otherwise mentioned, the
source of the examples is the COBUILD corpus, henceforth CB):
(1) computer, teacher, bestseller

(2) ... removal of references to the monarch is an act of sedition and a
breaking of the oath of office taken by politicians. (CB)
(3) Since old Nanapush had saved her from death, since he was her only
friend on the reservation, her visiting the man stood to reason. (CB)
(4) I am not a father who regrets that he didn 't see them enough because
of work. (CB)
My interest in nominalization has been prompted by the observation

that, in spite of the fact that nominalizations have been extensively dis-
cussed in the literature, no systematic and coherent theoretical-descriptive
approach of nominalization has been provided yet. Too often, nominaliza-
tions are regarded as 'empoverished' clauses, rather than as constructions in
their own right. Moreover, descriptive accounts of their lexicogrammatical
properties tend to be biased towards representational categories (such as the
process-participant relations which they express), to the neglect of other
functional categories. And finally, few attempts have been made to connect
the lexicogrammar of the various nominalization types to the meanings
which they realize. General problems such as these are not simply descrip-
tive but can be related to theoretical problems such as the absence of the
notions of rank and rank shift, the lack of attention that is paid to form-
42 Nominalizaîion
meaning correlations and the absence of a multifunctional perspective in

the description of nominalization patterns. The theoretical-descriptive
model which I will attempt to develop presupposes a coherent theoretical
view of the language system. It is in the dialectic between theory and de-
scription that the answers to many of the moot points in the analysis of
nominalization lie. The analysis of nominalization itself then turns out to be
"theoretically significant" (Langacker 1991: 22) and contributes insights to
issues such as form-function relations, the multifunctional analysis of lin-
guistic organization, and rank and class.
In what follows, I will present a first confrontation between theory and
the phenomenon of nominalization. I will follow the order in which I set
out the theoretical model in Chapter 2.1 will start by pointing out some of
the properties of nominalizations that reflect the usage-based character of
the system that lies behind them (Section 1). I will then reflect on the status
of nominalized constructions as symbolic units characterized by a natural
relation between semantics and lexicogrammar (Section 2). In Section 3,
issues related to the composite nature of nominalizations will be discussed,
together with the systemic-functional notions of 'rank shift' (Halliday
1961) and 'reclassification' (McGregor 1997). The advantages of a multi-
functional interpretation of nominalization will be treated in Section 4, fol-
lowed by an analysis of the role of agnate relationships in the analysis of
nominalization patterns (Section 5).
1. A usage-based account of nominalization
Nominalization constitutes a domain that is characterized by the frequent

occurrence of novel formations and extensions and by varying degrees of
productivity. Not surprisingly, therefore, the analysis of nominalization can
benefit greatly from a usage-based approach or an approach which views
nominalizations as 'systemically motivated' constructions which are struc-
tured with reference to the conventional units of language, and linked to
them in terms of schematization and extension (Section 1.1); an approach,
also, which recognizes that, as a result of the constant interaction between
language use and the language system, nominalizations are characterized
by varying degrees of entrenchment, which not only explain (diachronic)
changes in productivity but also account for the distinction between 'lexi-
calized' and 'ad hoc' nominalizations (Section 1.2).
A usage-based account of nominalization 43
1.1. Schematization and extension
First, the analysis of a specific nominalized construction involves a recon-

struction of the various categorizing judgements involved in its construal. It
entails an attempt to reconstrue the constructional schemata which the
nominalization instantiates or extends and tries to map out the family of
structures which the nominalization, as a complex category, is related to.
Two claims of Cognitive Grammar relating to the nature of complex cate-
gories are particularly relevant to the analysis of nominalization (Langacker
1987a, 1988, 1991, 1999): firstly, categorization by prototype and categori-
zation by schema are claimed to represent complementary modes of catego-
rization; and, secondly, categorization by schema is said to involve both
highly abstract and low-level schemata, the latter being probably more im-
portant for the computation of novel expressions than the former (Lan-
gacker 1988: 133).
Langacker has repeatedly pointed out that "extension tends to be ac-
companied by schematization", i.e. "that the 'outward' growth of a network
by extensions from a prototype tends to induce its 'upward' growth via the
extraction of higher-level schémas" (Langacker 1999: 102). When, in other
words, a construction extends a schema, the language user still perceives
some similarity between the extension and the schema which it extends,
and this similarity - representing the commonality between the basic
schema and the extension - constitutes "a schema having the two for in-
stantiations" (Langacker 1988: 139). Extension is thus inextricably linked
to schematization and leads to the extraction of a schema that generalizes
over a more diverse group of category members (Langacker 1988: 140).
Langacker also stresses the importance of recognizing both high-level
schemata which generalize across all instances of a construction type and
low-level schemata which hold for only a limited set of data and explain the
variation that exists among them. Low-level schemata cannot be consid-
ered in isolation from the high-level schema which they instantiate because
they primarily specify which of the options that are sanctioned by the high-
level schema "are in fact conventionally exploited" (Langacker 1991: 47).
How does this view on categorization tie in with the analysis of nomi-
nalizations? When a number of nominalized constructions are parallel in
formation and a pattern is discernible among them, the language user ex-
tracts a constructional schema that represents their commonality (Lan-
gacker 1991: 47). The various instantiations of that schema are then catego-
rized as members of one particular type of nominalization (e.g., deverbal
-er nominalization; gerundive nominalization), and their structure is con-
sidered to be "motivated rather than arbitrary" (Langacker 1991: 47). In
44 Nominalization
addition to this highly abstract schema, various low-level schemata can

then be distinguished which account for variations within the system and
which, along with the high-level schema, constitute the backbone of the
schematic network that characterizes the nominalization type. Research into
nominalizations, however, has tended to be primarily oriented towards low-
level schemata, and it often refrains from trying to identify the high-level,
generalizing schema that lies behind them. A case in point is the analysis of
deverbal -er nominalization. Descriptions of deverbal -er nomináis mostly
consist of elaborate enumerations of the wide range of entities which the
-er suffix can profile (e.g., human Agents as in teacher and hunter, inani-
mate Agents as in eye-opener and reminder, Instruments as in eraser and
poker, Locations, as in diner and sleeper, etc.). Because -er nominaliza-
tions are so diverse in profile, few attempts are made to link up the low-
level schemata which they instantiate with the high-level schema that gen-
eralizes across them. Yet, schematic networks consist of high-level and
low-level schemata. If it aims at being accurate, an account of a particular
nominalization type therefore necessarily involves a description of both.
1.2. Entrenchment
Schematic networks are not only characterized by different degrees of

specificity (ranging from concrete expressions to low- and high-level
schemata); the units that belong to them also display varying degrees of
entrenchment or automatization. Constructions fall "along a continuous
scale of entrenchment in cognitive organization" (Langacker 1987a: 59)
and differ in the degree to which they form "well-rehearsed routines" (Lan-
gacker 1999: 93). Two phenomena related to the degree of entrenchment of
a unit are particularly relevant to the analysis of nominalization patterns,
viz. productivity and lexicalization. First, the likelihood that a structure is
activated and leads to the construal of novel units - i.e. its productivity -
seems to correlate positively with both entrenchment and specificity (Lan-
gacker 1991: 48). Secondly, along with specificity, the degree of entrench-
ment of a unit also turns out to play an important role in the lexicalization
of it.
As far as lexicalization is concerned, the parameters of specificity and
entrenchment are equally determining: structures come to form part of the
"set of fixed expressions in a language" (Langacker 1991: 45) if, firstly,
they are high in specificity and represent individual expressions rather than
schemata, and, secondly, if they are highly entrenched, i.e. have achieved
the status of conventional units (Langacker 1991: 45). Note that symbolic
A usage-based account of nominalization 45
complexity is not relevant to lexicalization: lexical items can take the form
of morphemes, words, phrases and even longer expressions (Langacker
1991: 45; see also Halliday 1961: 274). Lexicalized constructions are to be
distinguished from grammatical patterns which are sufficiently entrenched
to be established as schematic units, but not as specific expressions. Gram-
matical patterns are either situated at word level (traditionally called 'mor-
phology') or at a level higher than the word (i.e. 'syntax').12 Lexicalizations
are thus not always simple words, and non-lexicalized items are not neces-
sarily larger than the word. Both for that reason, and because the degree of
entrenchment which determines the lexicalization of a unit fluctuates con-
tinually, the boundary between lexicon and grammar is not a sharp one, but
rather forms a continuum.
The distinction between lexicalized constructions and constructions
which are schematic rather than specific is highly relevant to the discussion
of nominalization patterns such as those in (5) and (6):
(5) Alex Gough, the Welsh destroyer of Del Harris's hopes on Tuesday
....(CB)
(6) ... a passenger steamer, which crossed the Channel escorted by a de-
stroyer. (CB)
The nominalizations in (5) and (6) show that, in a schematic network that
elaborates one and the same abstract schema (in this case, that of [[V/.. .]-
[[-ER]/[-er]]), some instantiations are lexicalized, while others are not.
Whereas destroyer in (6) forms a lexicalized nominal and has itself
achieved the status of a conventional unit, destroyer in (5) is what I will
call an 'ad hoc' nominalization: it has been formed according to the same
high-level schematic symbolic unit [[V/...]-[[-ER]/[-er]] as has the lexicali-
zation destroyer, but it is not included in the language system as a particu-
lar, fixed expression.13
As to the productivity of particular schematic units - i.e. the likelihood
of their being activated for the construal of novel units - Langacker (1991:
48) remarks that low-level schemata are more likely to be invoked than
more abstract schemata. He (1991: 48) argues that evidence in favour of
this claim comes from the fact that instantiations of a schema are normally
concentrated in certain regions which correspond to low-level schemata,
rather than being "distributed randomly through the space of possibilities
defined by the highest-level schema". Langacker's claim that it is low-level
rather than high-level schemata which determine the production of novel
expressions is comparable to what is traditionally referred to as analogy or
the "direct modelling of novel expressions on the basis of familiar ones"
46 Nominalization
(Langacker 1999: 144). It differs from analogy in its emphasis on the role
of schematization: "even the learning of specific expressions (required as
the basis for analogy) involves abstraction and schematization from actual
usage events" (Langacker 1999: 144).
The low-level nature of productivity which Langacker describes can be
clearly observed among the agentive and the non-agentive instances of -er
nominalization. Both are characterized by certain 'regions' of productivity
or low-level schemata which give rise to instantiations that are sufficiently
entrenched to be established as units. Agentive regions are, for instance,
those of agent-like instrumental -er nomináis (e.g., computer, printer,
parser), 'second-order' Agents (as inpageturner 'a book that almost forces
you to turn the pages' and yawner 'an event that makes you yawn') and
'professional' Agents (such as programmer and fire-fighter). Non-agentive
regions of productivity are especially those that designate non-agentive In-
struments (e.g., front-loader, stroller, stepper, walker) and items of food
(e.g., broiler 'a chicken suitable for broiling', cooker 'an apple for cook-
ing', sipper 'a drink that you have to sip', dipper 'vegetable/fruit/other
snack that has to be dipped before being eaten').
2. Nominalizations as non-arbitrary symbolic units
In spite of the fact that nominalizations are symbolic units which encode
their meaning, research on nominalization patterns often seems to consider
the analysis of lexicogrammatical properties as an end in itself and only
rarely attempts to link it up with the semantics of the construction. Certain
lexicogrammatical features moreover tend to be - erroneously - considered
as 'meaningless' and are therefore frequently left unanalyzed.
The tendency to leave things at a detailed account of a construction's
lexicogrammatical properties only is perhaps best illustrated by the follow-
ing nominalizations:
(7) And it's the third anniversary of the opening of East Germany's bor-
ders. (CB)
(8) Opening a book is like opening the door into a wonderful new world.
(CB)
(9) I still regret opening the door so suddenly, (personal example)
In Lees' (1960) account of nominalization, the nominalization in (7) is

labelled 'action nominal', the nominalization in (8) is described as a 'ge-
rundive nominal referring to actions', and that in (9) is said to be a 'gerun-
Nominalization as composition and reclassification 47
dive nominal referring to facts'. In spite of the fact that he gives an excel-
lent description of the lexicogrammatical characteristics of each of these
types of nominalizations, Lees does not succeed in tracing the origin of the
nominalizations' semantic differences in them. Since Lees (1960), numer-
ous analyses have been devoted to these nominalization types, some of
which significantly elucidate certain of their properties and the meanings
associated with them (e.g., Kiparsky and Kiparsky 1971; Schachter 1976;
Halliday 1985, 1994; Langacker 1991; Davidse 1991). Yet, it remains un-
clear why, for instance, the nominalization opening the door into a wonder-
ful new world in (8) designates what Lees calls an 'action', while opening
the door so suddenly in (9) is factive in meaning. Likewise - and in spite of
the fact that their lexicogrammatical differences have been extensively de-
scribed - the differences in meaning between the 'action' nominal in (7)
and the 'gerundive nominal' in (8) remain to be clarified.
One of the reasons why the semantic distinction between nominaliza-
tions like the ones in (7) and (8) has not yet been fully accounted for, is that
the semantic import of the grammatical marker of in (7) has tended to be
ignored (Langacker 1991: 35): along with the 's of Zelda 's in a nominaliza-
tion like Zelda's signing of the contract, o/has often been assumed to be
'empty' and inserted for purely grammatical purposes (Langacker 1991:
35). And even if grammatical markers like's and of are not explicitly ar-
gued to be meaningless, they are not taken into account in the analysis of
the nominalization in which they function. The analysis of nominalization
offered in Halliday (1994) and Halliday and Matthiessen (1999), for in-
stance, ignores the presence of periphrastic markers by relating nominaliza-
tions such as the writing of business programs and writing business pro-
grams to the same clausal agnate, i.e. to (people) write business programs.
However, grammatical markers such as the and of form part of the lexico-
grammatical properties of nominalizations and motivate a facet of their
composite structure. As Langacker (1991: 35) points out, they "invariably
have some kind of conceptual import, which may be quite abstract but is
nonetheless essential to their function". A natural account of nominaliza-
tions has to take account of all lexicogrammatical properties, including
those the import of which at first sight seems negligible.
3. Nominalization as composition and reclassification
In addition to relationships of categorization and symbolization, a construc-

tion also establishes compositional relationships by integrating two or more
component symbolic units into a complex unit or a symbolic assembly.
48 Nominalization
Two aspects pertaining to the assembly of symbolic structures are particu-

larly important for nominalization: first, the analysis of a construction
should not only deal with the construction as product, but should also con-
sider the order of assembly of its component parts, i.e. the construction as
process (this will be dealt with in Section 3.1). Secondly, both within and
above assemblies of symbolic structures, different levels of organization
can be distinguished which are functionally defined and which together
form a hierarchy or rank scale. One aspect of the analysis of a construction
is to situate it relative to this rank scale. I will show that, in the case of
nominalizations, a distinction can be made between, on the one hand,
nominalizations which reclassify a verbal unit of the rank of the word into a
nominal unit of the same rank, and, on the other hand, reclassification
which involves a clausal unit being 'downranked' to nominal rank (Section
3.2).
3.1. Constructions as assemblies of symbolic units
The importance of looking at the order of assembly of the component parts

of nominalized constructions is illustrated by the analysis which Langacker
(1991: 32) gives of the following nominalizations:
(10) Sam's washing of the windows was meticulous.

(11) Sam s washing the windows was a shock to everybody.
Langacker shows that these nominalizations essentially differ as to the level

of organization at which the nominalization process takes place: in the ac-
tion nominalization in (10), the verb stem wash has first been nominalized
and turned into a noun, i.e. washing, which then functions like any other
noun as nominal head. In the gerundive nominal in (11), on the other hand,
the nominalization process applies to a "higher-level structure" or "a proc-
essual expression that has all the ingredients of a finite clause except an
explicit subject and a grounding predication" (Langacker 1991: 32). When
this higher-level structure is nominalized, it comes to function as a whole as
nominal head, which can then be premodified by Sam's. The order in which
the component parts of the nominalizations in (10) and (11) have been as-
sembled thus determines their structural outlook.14
3.2. Nominalization as functional reclassification
Nominalization types differ according to the level of organization at which

the nominalization process takes place (see also Langacker 1991). As
pointed out in the introduction to this chapter, three types of nominaliza-
tions can be distinguished: nominalizations at the level of the word (e.g.,
teacher, Sam's washing of the windows), nominalizations which nominalize
a structure that lies in between a verb and a full clause (e.g., Sam's washing
the windows) and, finally, nominalizations consisting of full clauses (e.g.,
that Sam washed the windows). The latter two types deviate from the 'nor-
mal' rank scale of units in that they represent nomináis or phrases which
consist of clausal or clause-like structures. They have therefore often been
regarded as problematic, and it has even been claimed that /AaZ-structures
are not nominalizations (e.g., Dik 1997; McGregor 1997).
In what follows, I argue that structures with a clause-like component
(washing the windows as in Sam's washing the windows) and even con-
structions which have all the characteristics of a full clause (that Sam
washed the windows) can be said to be nominalized and 'reclassify' a
clausal unit into a nominal one, and this both externally and internally. A
notion that is crucial for the analysis of this kind of nominalization, how-
ever, is that of 'rank shift', which was introduced by Halliday (1961) as a
constituency-bound mechanism, and which I have reinterpreted in a radi-
cally functional way in Chapter 2 (Section 3.2). McGregor (1997) has
added a further capital clarification by pointing out that rank shift entails
reclassification. I will discuss this point and its relevance to nominalization
in Section 3.2.1. In a second part (Section 3.2.2), I will briefly introduce the
idea that all nominalizations - i.e. those which imply rank shift as well as
those which do not - involve a functional reclassification which is both ex-
ternal and internal.
3.2.1. Rank shift and reclassification
Halliday speaks of downward rank shift, when "an item normally having
the function of (entering into the paradigmatic and syntagmatic relations
associated with) rank χ characteristically 'loses' these functions on taking
over those of rank y" (Halliday 1966b: 114). A clause which is rankshifted
to phrase level (Halliday 1961: 253 speaks of 'group' level) thus "cannot
enter into direct syntagmatic relations with clauses outside the structure of
that group" (Halliday 1966b: 114). As Halliday defines it, rank shift basi-
cally involves a change in the external functions of a symbolic unit: a
50 Nominalization
clause which is rankshifted to phrase level, for instance, acquires the exter-
nal functioning potential of phrases (as is shown, in, for instance, [That he
committed the crime] surprised us, where the that-clause has come to func-
tion as Subject of the verb surprised). Halliday (1961: 261) concludes that
"by reference to the rank scale, classes are derived 'from above' (or
'downwards') and not 'from below' (or 'upwards')".
Because rank shift involves "a change in part-of-speech membership"
(McGregor 1997: 128), McGregor defines downward rank shift as reclassi-
fication. Most importantly, and against Halliday, he holds that rank shift
involves external as well as internal reclassification. He (1997: 127) de-
fines rank shift as "the process whereby a unit of a given rank is as it were
demoted in size, and reclassified as a unit of lower rank, as a result of
which it takes on the grammatical and semantic properties inherent to the
lower ranking unit". He (1997: 128) further points out that, semantically
speaking, the reclassification of, for instance, a clause into a nominal unit
implies that the clausal meaning is 're-presented' "as a thing - a second-
order entity - rather than an ongoing phenomenon located at some point in
space and time". In addition, McGregor emphasizes that there must be
some formal registration of the rankshifted status of a unit and he argues
that rankshifted clauses are characterized by a reduced number of gram-
matical options that are available. Among the options that are no longer
available are "those which permit the anchoring of the situation to a spe-
cific occurrence in the referent world" (McGregor 1997: 130). From this
premise, McGregor (1997: 131) draws the conclusion - wrongly in my
opinion - that finite clauses cannot be rankshifted or nominalized, whereas
non-finite clauses can. (A similar view is expressed in Dik 1997.) Finally,
McGregor (1997: 130) states that, apart from having a reduced structural
potential, the embedded clause may also acquire morpho-syntactic proper-
ties that are normally associated with nomináis.
In my opinion, McGregor rightly points to the importance of including
the internal properties of rankshifted units into their description and classi-
fication. Yet, he fails to clarify their significance to the analysis of rank-
shifted units, because he only considers overtly realized structural proper-
ties, instead of taking in a functional perspective. By focusing on the
clausal structural properties of the downranked clause only, McGregor fails
to observe and account for its nominal functioning. This results in inexpli-
cable discrepancies between the external (nominal-like) functioning of par-
ticular rankshifted units (e.g., as Subject of a clause as in [That he was the
one that killed her] surprised us most) and the internal status of these units,
which McGregor describes as non-rankshifted. McGregor's analysis also
fails to clarify the correspondences in meaning and in external functioning
of finite and non-finite nominalizations (such as [That he killed her] sur-

prised us and [His killing her] surprised us). In McGregor's view, gerun-
dive (non-finite) nominalizations are rankshifted and discharge a function
in the clause; that-structures, by contrast, are not rankshifted and serve no
such function. As Langacker (1991: 35, 449) points out, however, when a
finite clause comes to function as a constituent in another clause, the finite
grounding which it establishes as a clause may retain its processual charac-
ter, but it is viewed through an additional viewing frame which construes it
holistically, as a unitary entity. I will moreover show in my discussion of
factive nominalizations in Chapter 9 that full clauses and intermediate
clausal expressions which are downranked are fitted into the nominal para-
digm and connect with other, non-nominalized nomináis.
3.2.2. Nominalization as functional reclassification
I will show that all nominalizations - irrespective of the level of assembly

at which the nominalization process has taken place - are essentially nomi-
nal constructions and involve the functional reclassification of a non-
nominal into a nominal unit. Each nominalized construction thereby adopts
external but also internal functions of the target class. In those cases where
a simple verbal root is nominalized, typically by means of a nominalizing
suffix, the nominal status of the reclassified unit is not problematic. How-
ever, if nominalizations contain clear units of clausal patterning, then the
nominal character of the construction has eluded linguists more - or it has
not been pursued much in linguistic analysis. In fact, a 'constructional'
analysis of these nominalizations (which integrates their clausal and nomi-
nal characteristics) has not been achieved. Even though the theoretical con-
cepts of reclassification and rank shift are available, a descriptive break-
through with respect to these clause-like nominalizations is needed. This
descriptive breakthrough, I will show, can be realized by means of the
functional categories which Langacker (1991) has proposed for the clause
and the nominal and which I will zoom in on in Chapter 4. In particular, all
nominalizations will be argued to realize the highly abstract schema which
Langacker (1991) identifies for the nominal and according to which a
nominal realizes a 'grounded instance of some type'. I will show, however,
that the implementation of these functions ranges from being discernible
(e.g., in the case of action nomináis such as the opening of the door) to be-
ing non-prototypical (e.g., in gerundive nominalizations such as [Opening
the door] was not a good idea).
52 Nominalization
4. A multifunctional account of nominalization
The account of nominalization which I will attempt to develop is also built

on the assumption that constructions should be given a multifunctional
analysis. Nominalizations encode ideational, interpersonal and textual
meaning. Their semantic import cannot be reduced to, for instance, their
representational semantics and certainly not to the state of affairs that is
represented by the verbal component. What is encoded of their relation to
the speech event (i.e. their interpersonal meaning) and to the immediate
textual context (their textual functioning) has to be brought into the picture
as well. In Chapter 4,1 will argue that Langacker's description of the nomi-
nal in terms of type specification, instantiation, quantification and ground-
ing allows me to integrate these functional strands.
Research on nominalization patterns has thus far focused primarily on
ideational or representational functions: nominalizations have mainly been
analyzed in terms of the kind of experience which they designate. In Mat-
thiessen (1995) and Halliday and Matthiessen (1999), for instance, nomi-
nalizations are primarily seen as a resource for reconstruing the speaker's
experiences in a nominal way. However, by describing an action nominali-
zation such as Tom's cleaning of the kitchen and a gerundive nominal like
Tom's cleaning the kitchen only in terms of the kind of process that under-
lies them and the participants that accompany it, one cannot shed light on
the important constructional differences that exist between the two nomi-
nalizations. Another example of a nominalization type that has mainly been
analyzed in ideational terms is -er nominalization: suffixation with -er has
been described as a system that is basically agentive in profile, but allows
for non-agentive formations as well (e.g., among others, Marchand 1969;
Levin and Rappaport 1988; Ryder 1991, 1999b). The processes that are
being reclassified by it have been argued to be typically transitive (Lem-
mens 1998). However important it is for the analysis of -er nominalization,
a purely experiential description of -er suffixation, however, is unable to
identify a high-level constructional schema that generalizes across all in-
stances of -er nominalization, irrespective of which type of process they
derive from and of whether they are agentive or non-agentive.
5. The role of agnation in the analysis of nominalization
The analysis of nominalization can also benefit from the identification of

agnate constructions, or structures which nominalizations are systemati-
cally related to. Gleason (1965: 211) himself has pointed out that agnation
Agnation and nominalization 53
can relate "two constructions which are not on the same grammatical le-
vel", as is the case with the clause the boy runs and the nominalization the
boy's running (Gleason 1965: 212). In fact, relationships of agnation, ena-
tion and non-enation have always figured in descriptive accounts of nomi-
nalization patterns (and, indeed, of structural patterns in general: the central
idea underlying the construct of 'transformation' in Transformational
Grammar was that of capturing systematic relationships between different
structures). In Lees's (1960) Grammar of English Nominalizations, for in-
stance, it is thus relationships of agnation which underlie the transforma-
tional relation between nominalized structures and 'deep structure' clauses
(Lees [1960] 1968: 32, 65). Nominalizations and the clauses they are said
to be transformationally derived from are agnates in that they have the
same major lexical items, but differ in structure:
(12) his rapid drawing of the picture <=> He drew the picture rapidly.
(13) his bringing up the box » He brought up the box.
(14) John is the signer of the check. o John signs the check.
Notice that the 'tests' which Lees uses to identify the differences in struc-
tural behaviour of action nomináis (e.g., his drawing of the picture) and
gerundive nomináis (e.g., his drawing the picture) are likewise based on the
assumption that constructions can only be analyzed in more detail if their
systemic or paradigmatic relations are considered: in the examples in (15)
and (16), for instance, the paradigmatic relations of Aw and of drawing as
they function in his drawing of the picture and his drawing the picture are
considered (Lees 1968: 65-66):
(15) a. his drawing of the picture « the drawing of the picture

b. his drawing the picture <=> *the drawing the picture
(16) a. his drawing of the picture <=> *his having drawn of the
picture
b. his drawing the picture o his having drawn the picture
Agnation may have been used before in the analysis of nominalization,

but it has never been fully exploited. Often relationships of agnation fail to
be interpreted and linked up with the semantics of the nominalized con-
struction in question (e.g., Lees 1960). Or they are used in a way which
deviates from Gleason's (1965) understanding of them in that, rather than
being viewed as a network or grouping of structures, agnation is restricted
to one-to-one correspondences between nominalizations and clausal struc-
54 Nominalization
tures (as happens in Lees's transformational account of nominalization, but

also in Halliday 1994 and Halliday and Matthiessen 1999).
In my analysis of nominalization, I will attempt to use agnates more sys-
tematically. Three aspects of agnation seem to me important to the descrip-
tion of nominalization patterns. First, relationships of agnation are always
part of elaborate networks which characterize a specific construction. Sec-
ondly, agnation can be used as a heuristic not only for the disambiguation
of structures (as discussed in Chapter 2, Section 5), but also for the identifi-
cation (and description) of the components of a construction. And, finally,
relationships of agnation are best captured in schematic terms. They enable
one to derive high-level schemata capturing the general meaning of a con-
struction type.
5.1. Networks of agnation
As I pointed out in Chapter 2 (Section 5.1), Gleason does not view agnation
as "a unique relation between isolated pairs of sentences", but argues that it
is based on the 'pervading patterns' of language and that it is "therefore
always a recurrent thing, involving large numbers of sentences" (Gleason
1965: 202). Gleason has also pointed out that one agnate structure high-
lights a particular grammatico-semantic feature of a construction, but does
not uniquely identify the structure which it agnates with. A detailed picture
of the grammatico-semantic choices that are realized by a particular struc-
ture can thus only be provided by agnation networks, in which each agnate
structure sheds light on one or more of the structure's properties.
Gleason's understanding of agnation uncovers one of the central weak-
nesses of the transformational-generative and the systemic-functional ap-
proaches to nominalization, viz. their being focused on one - necessarily
clausal - agnate.15 Clausal agnates are very important to the analysis of
nominalized structures, but other, non-clausal agnates may contribute to the
identification of the lexicogrammatical properties of a nominalization as
well. The focus on one agnate structure which is clausal in nature tends to
lead to incomplete or even inaccurate analyses. For instance, in Halliday
and Matthiessen's (1999) account of nominalization, both a structure such
as Tom's cleaning the kitchen and Tom's cleaning of the kitchen would be
related to the clause Tom cleans the kitchen. This, however, fails to identify
the relevant features of the nominalizations it is related to and needs to be
accompanied by a set of other agnates to distinguish the gerundive nominal
Tom's cleaning the kitchen from the action nominalization Tom's cleaning
of the kitchen}6
The central role which is assigned to clausal agnates moreover tends to

reduce nominalizations to clause-like constructions, rather than viewing
them as unique - nominal - constructions in their own right. This is espe-
cially the case in the transformational-generative approach, which views
nominalizations as constituting the 'surface level' of language, while their
'source sentences' (Lees 1968: 65) are conceived of as the 'deep' or more
fundamental level of linguistic organization. The distinction between deep
and surface structures is rejected in Cognitive Grammar, which claims "that
grammatical structure is almost entirely overt: things really are what they
appear to be, provided we know how to interpret them properly" (Lan-
gacker 1987a: 27). Systemic-functional linguistics likewise rejects the dis-
tinction between deep and surface levels of organization, but by focusing
solely on clausal agnates nonetheless fails to analyze nominalizations as
constructions in their own right.
5.2. Agnation as a descriptive heuristic
As noted in Chapter 2 (section 5.1.), Gleason (1965: 203) shows that ag-
nates can crucially be used to disambiguate apparently identical structures.
This also applies to the disambiguation of nominalizations. Consider, for
instance, the nominalized structures in (17) and (18):
(17) a. I still regret not going to school.

(18) a. Not going to school is not a very sensible thing to do.
At first sight, the structure not going to school in (17a) is identical to that in
(18a). Yet, if one considers the set of agnates that is related to not going to
school in (17a) and in (18a) (illustrated in 17d, 17e, 18d and 18e), as well
as the paradigmatic options of their component functions (illustrated in 17b,
17c, 18b and 18c), it becomes clear that they are different structures:
(17) b. I still regret not having gone to school.

c. I still regret their not going to school.
d. I still regret that I did not go to school.
e. I still regret the fact that I did not go to school.
(18) b. *Not having gone to school is not a very sensible thing to do.
c. *Their not going to school is not a very sensible thing to do.
d. *That you do not go to school is not a very sensible tiling to do.
e. *The fact that you do not go to school is not a very sensible thing to
do.
56 Nominalization
In my opinion, yet another function of agnation lies in the identification

of the component parts of a construction. A good example here is the
analysis of the nominalization in the following clause:
(19) Sam's washing the windows was a shock to everybody. (Langacker

1991: 32)
I have pointed out in Section 3.1 that Langacker's analysis of the order of
assembly of this nominalization is essential to distinguish it from the action
nominalization Sam's washing of the windows. Langacker (1991: 32) has
pointed out that the gerundive nominalization in (19) nominalizes a
"higher-level structure" or "a processual expression that has all the ingredi-
ents of a finite clause except an explicit subject and a grounding predica-
tion". The fact that this 'higher-level' structure combines washing and the
windows into one component unit of the assembly Sam's washing the win-
dows, which is then premodified by Sam's, is confirmed by agnation. Was-
hing the windows agnates, for instance, with wash the windows as it is used
in the constructions in (20), (21) and (22):
(20) You will wash the windows, won't you?

(21) What will you do? Wash the windows.
(22) I will wash the windows. Will you? Yes, I will.
The clauses in (20), (21) and (22) illustrate that, also within clauses, a verb
(minus the grounding elements of tense and modality) and its Object(s)
combine into a separate component unit (Halliday 1994). The clause-level
agnates thus provide evidence in favour of the claim that the order of as-
sembly described in Langacker (1991) is the correct one and that washing
the windows constitutes a component unit in the assembly of Sam's was-
hing the windows.
5.3. The schematic nature of agnation
In this final section, I put forward my view that agnation is best thought of
as involving degrees of schematicity·. relationships of agnation between
specific examples of constructions always involve relations between the
schemata which they instantiate. I also argue that agnate relationships bet-
ween nominalizations and clauses reveal the importance of clausal catego-
ries to the analysis of nominalized constructions.
Both claims can be illustrated by a critique of the transformational-

generative approach to nominalizations which was formulated by Dik
(1967). One of the points which Dik argues for is that nominalizations and
the clauses they are related to should be viewed as two distinct construc-
tions: often "the 'derived' complex word contains a semantic aspect not
present in the proposed underlying structure", and "the proposed underly-
ing structure contains semantic aspects not present in the 'derived' complex
word" (Dik 1967: 378). Dik gives the example of John is the signer of the
check, which in Lees (1968: 32) is identified as a transformation of the
clause John signs the check. While the clause contains an explicit indica-
tion of tense, the nominalization is said to be "neutral in relation to time,
but [it] can be referentially used in different ways" (Dik 1967: 378): it can,
for instance, be interpreted as designating a past event, as in John signed
the check. Clauses, Dik concludes, "necessarily contain an aspect of tense,
which is not present in the final derived structure" (Dik 1967: 378). Dik's
analysis of the relationship between John is the signer of the check and
John signs/signed the check illustrates two characteristics of agnation
which seem to me essential for the description of nominalization. First, it
shows that relationships of agnation are basically schematic in nature and
that, in some cases, a relationship of agnation is captured more accurately
in schematic terms than as a relation between examples. Secondly, Dik's
description of the possible temporal meanings of John is the signer of the
check and of their (schematic) link with the clausal tense system provides a
first indication of the precise nature of the schematic relationship between
nominalized constructions and their clausal agnates. I now develop these
two points.
I have thus far described agnation as a relationship between specific ex-
amples. Each of these examples, however, instantiates a constructional
schema. Agnation can therefore be said to essentially involve a relationship
between schemata. In fact, in some cases the agnate relationship is perhaps
better captured in schematic terms than by means of relations between spe-
cific examples. A good example is Dik's analysis of what I describe as an
'ad hoc' nominalization, i.e. signer of the check. Dik - in my view, cor-
rectly - argues that it is 'neutral' in time and adopts a particular tense only
in the context in which it is used. Various specific clausal constructions
can, in other words, be said to agnate with signer of the check: e.g., he signs
the check, he signed the check or perhaps even future he will sign the check.
One way of generalizing across the different temporal interpretations is by
viewing them schematically. Ad hoc -er nominalizations can then be said to
relate to the overall temporal mode of grounding in clauses. The relation-
ship of agnation between ad hoc -er nominalizations and temporally groun-
58 Nominalizatìon
ded clauses can, in turn, be captured by means of high-level schémas of

which both are instantiations. This abstract schema could, for instance, be
argued to be that of 'temporal grounding', or, on an even more abstract
level, that of clausal grounding in general.
This brings me to a second aspect of the schematic nature of agnation
which is relevant to the analysis of nominalizatìon: the schematic relation-
ship of agnation that exists between nominalizations and clausal construc-
tions alerts us to the role of schematic clausal categories in nominalized
constructions. The clausal category of grounding, for instance, somehow
appears to be relevant to the analysis of the system of -er nominalizatìon.
Agnation thus does not only help to disambiguate seemingly identical
nominalizations and identify the component units of a nominalized struc-
ture; the clausal agnates of a nominalized construction also provide infor-
mation concerning the internal functional behaviour of nominalizations and
the clausal categories that play a role in it. I will argue that this information
is crucial for the description of specific nominalizatìon types and for the
identification of the basic mechanisms that underlie deverbal nominaliza-
tions.
6. Towards a theoretical-descriptive approach to nominalizatìon:

Conclusion
I have argued in this chapter against the tendency to view nominalizations

as 'empoverished clauses' whose main purpose is to periphrastically realize
processes and their participants and of which certain lexicogrammatical
properties (such as the inclusion of, for instance, of and '5) can be ignored.
Nominalizations, I have suggested, should be viewed as constructions in
their own right: they are reclassifications of non-nominal into nominal units
or units which have adopted nominal functions both in their external and in
their internal functioning. An accurate description of these nominal units
involves a description of, firstly, their symbolic status (as units which en-
code a particular meaning); secondly, of the compositional relationships
which they realize (in particular, of their component functions)·, and,
thirdly, of the (paradigmatic) relationships which they enter into as complex
categories (i.e. with the high- and low-level schemata which they instanti-
ate or extend and with agnate constructions).
At the same time, however, nominalizations are reclassifications of
various levels of assembly of the clause and it is clauses which they most
typically agnate with. A natural account of the component functions of
nominalizations and of the meanings which they encode can therefore only
Conclusion 59
be arrived at when nominalizations are analyzed not only in terms of the

nominal but also in terms of the clausal functional categories that they real-
ize. More specifically, analyses of nominalization patterns should be fo-
cused on the way in which nominalizations integrate nominal and clausal
categories. In the next chapter (Chapter 4), I will therefore zoom in on the
functional categories which, in my opinion, are best qualified to elucidate
the integration of clausal and nominal functions in nominalizations, viz. the
functions of type specification, instantiation, quantification and grounding
(Langacker 1991). Crucially, Langacker has argued that these functions
figure in nominal structures as well as in clauses and that they thus reveal a
fundamental parallellism between the functional organization of clauses
and ordinary nomináis. After I have dealt with the functional organization
of nomináis and clauses (Sections 1 and 2), I will set out the main princi-
ples of what a functionally-oriented theoretical-descriptive approach to
nominalization may look like (Section 3).
Chapter 4
The functional organization of nominal and clause
The key to a descriptive analysis of nominalizations which does not lose

sight of the various theoretical concerns raised in Chapter 3 lies in an accu-
rate account of their functional organization. This, in turn, presupposes an
insightful account of the functional resources of both the nominal and the
clause: only when the parallelisms between the functional organization of
nomináis and clauses are clear can the process of nominalization itself be
elucidated and the mechanisms that lie behind specific nominalization types
be identified.
I indicated earlier that Langacker's description of nomináis and clauses,
combined with elements of Halliday's multifunctional approach to linguis-
tic organization, offers a very productive perspective on the functional
make-up of the nominal and the clause. This chapter will therefore first
zoom in on the functional categories which Langacker (1991) and Halliday
(1994) distinguish. Langacker's analysis of nomináis and finite clauses as
'grounded, quantified instances of some type' will form the leitmotif of the
chapter and will be elaborated in detail; Halliday's metafunctional hypothe-
sis and, in particular, his interpersonal analysis of linguistic organization
will be shown to throw a complementary light on the assembly of clauses,
from their innermost layer of organization (i.e. the process type specifica-
tion) to the outermost layer or grounding elements. I will first discuss the
assembly of the nominal (Section 1) and then elaborate in detail on the par-
allels and differences that exist between the organization of the nominal
and that of the clause (Section 2). In a final section, I will turn to nominal-
ized constructions and sketch the main principles that underlie the func-
tional reclassification which they realize (Section 3).
1. From noun type specification to nominal
The overall function of a nominal is to refer to a thing and make it "a mo-
mentary focus of attention" within the speech event (Langacker 1991: 53).
To single out the thing one wants to talk about - to the exclusion of other
things - one can basically employ two strategies. Either one provides the
thing with its own, unique label and uses a proper name·, or one follows a
strategy based on type specifications which one then specifies in terms of
62 The functional organization of nominal and clause
quantity, as well as in terms of their relation to the speech event and its par-
ticipants (the ground) (Langacker 1991: 53). It is by being grounded in the
speech event that the instance of the type specification that one wants to
talk about becomes uniquely apparent to the speech participants. However
different the latter, common noun strategy may seem from that based on
proper names, both strategies give rise to nomináis that conform to the
schematic characterization of the nominal as a 'grounded, quantified in-
stance of some type'. They only differ in the "structural implementation of
these semantic functions" (Langacker 1991: 148): unlike proper names,
nomináis with a common noun as head tend to reflect the semantic func-
tions which they realize iconically, with the type specification as the in-
nermost layer of organization, the grounding predication as the outermost
layer and intermediate between the type specification (or simple noun) and
the grounded type (or nominal), the functions of instantiation and quantifi-
cation (Langacker 1991: 54).
Which is the contribution of each of the semantic functions which Lan-
gacker (1991) distinguishes to the ultimate grounded status of the thing des-
ignated by the nominal? And how are these functions encoded in the struc-
ture of the nominal? Both formally and semantically, the type specification
constitutes the innermost functional layer. Semantically speaking, it makes
"an initial delimitation among the potential objects of thought" and "speci-
fies the basis for identifying various entities as being representatives of the
same class" (Langacker 1991: 53). In the nominal my office, for instance,
the noun office specifies the type or class of entities that is being talked
about and thus forms the representational core of the nominal (Halliday
1994: 189). Structurally speaking, it is the head noun which provides the
type specification, along with any adjectives or other modifiers which ren-
der the type specification more precise and add certain refinements: in the
nominal an excellent convention site, the head noun convention site is thus
rendered more specific by the inclusion of excellent. Postmodifying struc-
tures can also add to the type specification, as illustrated by in the Midwest
in the nominal an excellent convention site in the Midwest (Langacker
1991: 52). Together with the head noun, these modifiers create a 'higher-
order type specification' which forms a kind of nucleus within the nominal
and which typically "supplies a vast amount of detailed conceptual content"
A nominal type specification is instantiated when it is thought of as
"having a particular location in the domain of instantiation" (Langacker
1991: 57). For nomináis, the typical domain of instantiation is space. The
instantiation of a type specification implies that one conceives the domain
of instantiation as being able to support the simultaneous manifestation of
From noun type specification to nominal 63
multiple entities. It involves anchoring the type specification at a particular

spot which is different from other possible locations (Langacker 1991: 57).
Only very rarely, a noun type specification is not instantiated (and
grounded) and functions as a type specification in the clause: Langacker
(1991: 69) gives the French examples of Alain est professeur 'Alain is a
professor' and Pierre est médecin 'Pierre is a doctor'. An example of an
English non-instantiated noun type may be Mary became Queen (Davidse
p.c.).
What counts as an instance of a type specification depends on the kind
of noun type that is instantiated. Common nouns can, in this respect, be
divided into three categories: there is, firstly, the category of the singular
count noun which designates a discrete entity (e.g., pebble). An instance of
the singular count noun pebble is then a single small stone. There is the
category of plural nouns (e.g., pebbles), which Langacker classifies among
the mass nouns and which designates a 'replicate mass' in that "consider-
able prominence is accorded to the discrete entities out of which the mass is
constituted" (Langacker 1991: 78).17 An instance of the plural noun pebbles
is necessarily a set consisting of two or more stones. A third category of
common nouns, then, is formed by non-plural mass nouns such as gravel,
which designate a non-replicate mass and focus on the continuous nature of
the profiled mass. An instance of the mass noun gravel must be perceived
as internally uniform rather than particulate (Langacker 1991: 78). Instan-
tiation of a type specification is an important prerequisite to both quantifi-
cation and grounding, but, unlike the latter, it cannot receive separate struc-
tural symbolization: prototypically, instantiation is indicated by determiners
which also serve the function of grounding the instantiated type (Langacker
1991:58).
If a nominal is to serve its function of singling out a particular instance
of a specified type, it must also include some indication of the size or quan-
tity of the profiled instance (Langacker 1991: 53). Importantly, quantifica-
tion of an instance does not specify the number of instances which the
nominal profiles: every nominal profiles a single instance of a type, and this
irrespective of whether the type specification is singular or plural (Lan-
gacker 1991: 81). The nomináis the pebble, the pebbles and the gravel thus
each designate a single instance of the type specifications pebble, pebbles
and gravel. Even when a quantifier is added, as in the seven pebbles, the
nominal profiles only one instance. Rather than specifying the number of
instances, quantification merely gives more information concerning the size
of the designated instance (Langacker 1991: 81).18
Nomináis select from a variety of quantifiers to give a finer-grained in-
dication of the size of the instance which they designate. Langacker distin-
guishes between absolute or true quantifiers, on the one hand, and relative
quantifiers, on the other. Absolute quantifiers "offer a direct description of
magnitude" (Langacker 1991: 83). Examples are seven, many, few, little.
Relative quantifiers specify a quantity in relation to a 'reference mass',
which in the default case consists of the "maximal instantiation of the per-
tinent category" (Langacker 1991: 82). A relative quantifier such as all in
all dogs, for instance, profiles a quantity that is identical to the reference
mass. In most dogs, in contrast, the profiled mass only constitutes a subpart
of the reference mass, be it one "that comes reasonably close to exhausting
it" (Langacker 1991: 82). Because they situate the instances which they
quantify relative to a definite reference mass, relative quantifiers also serve
a grounding function. This explains why they cannot be used with other
grounding predications (e.g., *the all dogs, *those most dogs). Absolute
quantifiers do not imply a reference mass: the nomináis many dogs, several
dogs and seven dogs give a (vague or precise) indication as to the quantity
of dogs involved, but they do not situate this quantity relative to a reference
mass. Absolute quantifiers do not therefore serve a grounding function and
can be preceded by a grounding predication (e.g., those three dogs, the
many cats I have owned) (Langacker 1991: 82-83).
Langacker argues that every nominal contains some indication of quan-
tity, either in absolute terms (e.g., three cats) or proportionally (e.g., most
cats). He does not, however, elaborate on the quantificational information
provided by those nomináis which leave the size of the profiled instance
implicit or characterize it only 'schematically' (Langacker 1991: 89), such
as proper names and nomináis with articles, zero-determiners, demonstra-
tives and possessives. It is interesting in this respect to consider the analysis
which Milsark (1976, 1977) gives of determiners according to the type of
quantification that they can express (Davidse 1999: 216-217). Milsark's
analysis is to be situated in the context of a systematic distributional analy-
sis which he makes of the use of determiners in the existent NP of existen-
tial constructions, more precisely in the unmarked 'cardinal' type of exis-
tential such as There are five weak spots in the human body (for an exten-
sive discussion of existentials, see Davidse 1999). Milsark distinguishes,
firstly, 'cardinality words' which "express the size of the set of entities de-
noted by the nominal with which they are construed" (Milsark 1977: 23).
Cardinal determiners can be used in the existent NP of cardinal existentials
and they include, apart from the quantifiers which Langacker (1991) classi-
fies as absolute, also the indefinite article a, the zero-determiner with plural
and mass nouns in non-generic reading and, finally, cardinal numbers. If
determiners do not give a cardinal value themselves, Milsark argues, they
"must always be understood with reference to a set of some cardinality"
(Milsark 1977: 23), i.e. they are what have been called 'relative' quantifi-
ers. The relative group of determiners comprises the definite article, bare
plural and mass nouns with generic reading, pronouns, proper names, pos-
sessive and demonstrative determiners, genitives and various relative quan-
tifiers, such as all, every, both, most (Davidse 1999: 217). The latter group
of determiners is excluded from the cardinal existential construction.
Davidse (1999) argues that Milsark's analysis of the quantifying proper-
ties of determiners can be integrated with Langacker's analysis of the
nominal as 'quantified instance of some type'. The distinction between ab-
solute (or cardinal) and relative or proportional quantification can, in other
words, be extended to all nomináis, irrespective of whether they contain an
explicit quantifier. Some of the analyses which are proposed in Davidse
(1999) are the following. With regard to non-generically used bare mass
nouns and plural count nouns (as in 1 and 2), Davidse argues that they real-
ize absolute quantification (Davidse 1999: 222), because they alternate sys-
tematically with nomináis introduced by absolute quantifiers such as some,
much and many (illustrated in 3 and 4):
(1) There's milk on the table.

(2) There are mice in the shed.
(3) There is some/ much milk on the table.
(4) There are some/ many/ three mice in the shed.
Generic bare plurals and mass nouns (as in 5 and 6), in contrast, are
claimed to receive a relative quantificational reading (Davidse 1999). They
resemble proper names in that they designate unique instances. Or, as Carl-
son (1978) puts it, they are proper names of kinds conceived of as individu-
als. Unlike proper names, however, the instances which they designate are
instances in type space and not in physical space: the nomináis koalas and
laurel do not refer to specific, spatially-instantiated animals or plants.
Rather, they have to be understood with reference to a type universe, in
which they constitute unique individuals (for a more detailed account, see
Davidse 1999: 209-215). They realize relative quantification because they
"quantify universally, as it were, over the one instance contained by the
reference mass in type space" (Davidse 1999: 225).
(5) Koalas carry their young in pouches.

(6) Laurel has red berries.
Finally, nomináis which include a possessive determiner (see 7) or a

genitive (as in 8) typically realize relative quantification: they invoke the
concept of a reference mass (all the instances of a type specification in the

current discourse space), and present the instances of the type specification
which they profile as coinciding with either all or part of this reference
mass (Davidse 1999: 223-224):
(7) My students protested loudly.

(8) the teacher's books
With the size of the profiled instance specified, it remains for the
speech-act participants to establish contact with the instance: the objective
of the speaker in using a nominal is to refer to an instance of a type which
the hearer can identify. The speaker has to make the profiled instance into
something that can be exchanged in communication, something that can be
'talked about'. As Langacker (1991: 91) puts it, "the speaker (S) and hearer
(H), who jointly form the ground (G), face the task of coordinating their
mental reference to some instance t¡ of type Τ drawn from the reference
mass RT". Within the system of identification, the basic options are definite
vs. indefinite identification. Indefinite identification presents instances
which are not presumed known to the hearer; definite identification marks
an instance as 'presumed known'. Definite identification occurs when "full
coordination of reference" between speaker and hearer is achieved (Lan-
gacker 1991: 91).
As pointed out in Davidse (2001: 8), the functions of identification and
quantification have not always been properly distinguished: it is thus un-
clear how the definite-indefinite distinction applies to the various determin-
ers that are found in the nominal, and how this distinction relates to that
between relative and absolute quantifiers. Davidse (2001) suggests viewing
all definite reference as implying relative quantification. She (2001: 9) ar-
gues that "speaker and hearer can be thought of as coordinating their men-
tal reference to P, the actually designated mass, which, through the opera-
tion of definite reference, is delineated as an identifiable set or mass".
Unlike indefinite reference, definite reference delineates or delimits the
portion of the reference mass which the hearer should make mental contact
with. Note that this definition of definite reference is compatible with that
by Langacker (1991) in terms of 'full coordination of reference to some
instance t, of type Τ drawn from the reference mass Rf '; it only differs from
it in that the mechanism of relative quantification is brought into the defini-
tion and is identified as a prerequisite for definite identification. Davidse
(1991: 9) thus takes over the schematic representation of definite reference
in Langacker (1991: 92) and only changes the profiled mass t¡ into a Ρ re-
sulting from a presupposed operation of relative quantification:
Figure 4. Definite identification (see Langacker 1991: 92)
Domain of instantiation
Figure 5. Indefinite identification (see Langacker 1991: 92)
In Davidse's (2001: 10) approach, indefinite reference "does not delimit

a mass or a set of instances" but only involves "the hearer recognizing in-
stances as a t¡ of type T". As Langacker (1991: 104) puts it, indefinite refer-
enee puts an instantiation on the scene but is "not sufficient to put the
hearer in mental contact with a uniquely determined instance of the cate-
gory". In the example Hand me a wrench (Langacker 1991: 104), for in-
stance, the speaker tells the hearer 'what sort of thing' he wants, but a
wrench does not delimit a set of one specific wrench. In that the hearer has
to be able to recognize an instance corresponding to the type specification
used by the speaker, indefinite reference presupposes 'type-identifiability'
(Gundel et al. 1993: 275, as cited in Davidse 2001: 13). Besides represent-
ing the instantiation relation between t¡ and T, Figure 5 also indicates that
speaker and hearer both have to make mental contact with the type T.
It is thus not only determiners that contribute to the reference act: the
type specification makes an essential contribution to it as well: as Davidse
(2001: 12) points out, "successful reference is dependent as much on a suf-
ficiently informative and contextually well-positioned 'categorization' as
on the appropriate choice of determiner". In fact, the type specification
turns out to play a different role with definite and indefinite identification.
In nomináis which are introduced by a definite determiner, Davidse (2000b,
2001) argues, the type description defines a 'reference mass' of 'all the in-
stances corresponding to that type in the discourse context'. The following
examples are given to illustrate this point:
(9) When God asked why they were wearing clothes, it was the man who
spoke first.
(10) Here are the exams that still need correcting.
(11) If you take care of your exams that still need correcting, I'll take care
of mine.
In (10) and (11), the contextually relevant reference mass is 'all exams that
still need correcting'. While the definite article in (10) refers to all of those,
the possessive your in (11) picks out only some instances. In (9), the con-
textually relevant reference mass (in the context of the Garden of Eden)
consists of only one instance of the type 'man'. The use of the definite arti-
cle to refer to 'man' is therefore correct.
Indefinite nomináis, then, merely designate "'some' instantiation of a
type, without indicating whether there are - or aren't - other instances of
the type in the discourse context" (Davidse 2000b: 1113). The type specifi-
cation has a purely "classificational" function (Davidse 2000b: 1113). The
designated instance may be presented as indefinite, but specific (as in 12):
the instance is then being introduced into the discourse for the hearer's
benefit. Or no specific instance may correspond to the indefinite nominal
for either speaker or hearer (as in 13) (Davidse 2000b: 1113):
(12) I had only just arrived when a man suddenly entered the restaurant
and started shooting.
(13) A tie is always an appropriate present for a man who is rather con-
servative.
I have discussed the semantic functions of type specification, instantia-

tion, quantification and grounding and I have shown how they contribute to
the functional organization of the nominal. These functions can be realized
by separate words or phrases, ordered in a layered kind of organization, as
in those three black cats,19 or they may not be reflected iconically. There
are, for instance, nomináis in which the semantic functions are associated
with the nominal as a whole. Langacker (1991: 148) gives the examples of
proper names, personal pronouns and nominalized finite clauses. Proper
names (e.g., John, Belgium) carry with them a type specification (e.g., 'hu-
man male', 'nation'). Van Langendonck (1999: 109) refers to it as the
'categorical presupposition' of the proper name and argues that it forms its
basic level meaning. Proper names are moreover semantically singular, i.e.
they always designate a single instance of the type specification (also when
the latter is plural, as in the Alps, the Philippines, Van Langendonck 1999:
109-110). Finally, they are grounded in a definite way. As pointed out in
Van Langendonck (1999: 111), the definite nature of proper names is either
expressed by a fixed definite article (as in the Nile) or it is incorporated in
the structure of the proper name itself (as in John).
Like proper names, personal pronouns such as I and she profile a single,
definitely grounded instance of a schematically specified type (e.g., 'hu-
man' in the case o f / ; 'human female' in the case of she). The distinguish-
ing property of pronouns is that their profile is entirely determined by the
speech event and the participants in it: they are either first- and second-
person pronouns referring to the speaker and hearer respectively, or, if an
entity is profiled that is distinct from the speech participants, they are in the
third person (Langacker 1991: 148).
The nominalization of a finite structure, then, also constitutes a nominal,
"as witnessed by its ability to occur alone as subject or direct object" (Lan-
gacker 1991: 148):
(14) That prices will continue to rise is obvious.
The type specification, Langacker (1991: 149) argues, is in this case pro-
vided by the proposition conveyed by the nominalized clause. Because the
clause is finite and uniquely related to the time and the participants of the
speech event, the type specification which it implies has only a single in-
stance which is definitely grounded.
Other nomináis consist of two structures which are themselves nomináis
and which are in apposition: the two components have "equal claim to the
status of profile determinant" and the resulting, composite structure is a
"doubly-characterized, doubly-grounded thing" (Langacker 1991: 149).
Examples of such appositional structures are the fact that whales are mam-
mals·, all those kittens and my daughter the lawyer (Langacker 1991: 149).
Still other nomináis consist of a grounding predication only (e.g., that,
these, some, any, all). They can be said to incorporate a type specification,
in spite of the fact that they lack a head noun. A nominal such as these thus
schematically characterizes the entity which it grounds as being a replicate
mass or plural; each, on the other hand, incorporates the type specification
of a discrete entity (Langacker 1991: 151).
In short, the semantic functions of type specification, instantiation,
quantification and grounding can be encoded in many different ways and
they do not always receive individual symbolization. Yet, they are realized
in every nominal and together, they serve to turn a 'thing' into something
which is related to the speech event and its participants, something which
one can zoom in on and talk about. Crucial for the analysis of nominaliza-
tions is Langacker's claim that the "relation between a simple noun and a
nominal is directly analogous to that between a simple verb and a finite
clause" (Langacker 1991: 191). In the next section, I will go more deeply
into the functional organization of the clause. Again, the analysis that is
presented is based on Langacker (1991). More than in the description of the
nominal, however, a number of complementary or alternative analyses that
have been proposed for certain aspects of clausal assembly are included as
well. The latter are especially based on Halliday (1994), Davidse (1992b,
1997), Heyvaert (2000) and Verstraete (2002).
2. From process type specification to clause
The pivotal element in the clause is the verb. It may stand on its own (as in
15a), or it may form part of a more elaborate kind of verb group with a
varying number of auxiliaries (as in 15b and 15c):
(15) a. She cleaned the room.

b. She may clean the room.
c. She must have been cleaning the room.
From process type specification to clause 71
Langacker (1991) argues that, on the basis of semantic function and gram-
matical behaviour, the verb group falls apart into two parts, viz. into a
grounding predication and a type specification. Importantly, this distinction
does not coincide with that between the main verb and its auxiliaries.
Rather, Langacker suggests "that the specification of tense and modality be
analyzed as the grounding predication, with the remainder of the verb
group (other auxiliaries and the main verb) regarded as a complex clausal
head analogous to a head noun" (Langacker 1991: 195). For the verb
groups in (15), this leads to the following analysis (with the grounding
predications in italics):
(16) a. She cleaned the room.

b. She may clean the room.
c. She must have been cleaning the room.
Semantically speaking, tense and modality seem to stand apart from the
rest of the verb group in that they are "the only auxiliary elements that spe-
cifically invoke the ground as a reference point" (Langacker 1991: 195):
tense relates the process to the time of the speech event, while the presence
of a modal explicitly situates the profiled relationship with respect to the
judgement of the speaker. A structural argument in favour of the proposed
division comes from the fact that an infinitive (marked by to) can be
formed on the main verb together with any combination of auxiliaries, mi-
nus the markers of tense and modality. It is thus possible to have to clean,
to be cleaned, to have cleaned, to have been cleaning. It is not possible to
do the same with a verb group which is marked for tense or modality, e.g.,
*to should wash, *to had been washing (Langacker 1991: 195). There is, in
other words, a distinction that cuts right through the auxiliaries of the verb
group and distinguishes those auxiliaries that serve to ground the process
type from all the others. Together with the main verb, the latter, non-
grounding auxiliaries make up the clausal head. In what follows, I will suc-
cessively discuss the properties of the clausal head (Section 2.1), the way in
which it is instantiated and quantified (Section 2.2), and the means that ex-
ist of grounding it (Section 2.3). Both in the instantiation and in the ground-
ing of a process type specification, the Subject will be argued to play an
important role. A third function of the clausal Subject, viz. that of 'speech
functionally responsible' element in the clause (Halliday 1994), will be de-
scribed in Section 2.4.
2.1. The clausal head
Langacker argues that the head-part of the verb group, which provides the
process type specification of the clause, may be quite complex internally:
like the head in nomináis, the essential semantic content provided by the
main verb (e.g., clean) may be elaborated with any desired degree of preci-
sion by the non-grounding auxiliaries with which it is combined (Lan-
gacker 1991: 196). Each auxiliary element is said to impose its profile on
the structure it combines with, "thereby deriving a distinct higher-order
type specification" (Langacker 1991: 197). It is, however, only "the left-
most verb in the sequence" (Langacker 1991: 196) which determines the
profile of the entire verb group. If a verb group is to function as clausal
head and provide a clausal type specification, Langacker (1991: 199) ar-
gues, its leftmost verb must profile a 'process', or a series of component
states which are distributed continuously through time and scanned in a
sequential way (Langacker 1991: 21, 246). The function of the auxiliaries
have (present perfect), be (progressive aspect) and be (passive voice) is
precisely to impose a processual profile on the atemporal relation that is
formed by the affixal/inflectional pair members -ed (perfect aspect), -ing
(progressive aspect) and -ed (passive voice) (Langacker 1991: 199). More
particularly, the "members of a given pair must co-occur if the resulting
expression is to be processual and thus able to serve as clausal head" (Lan-
gacker 1991: 200). Examples of clausal heads which illustrate the use of
specific auxiliary pairs are the following:
(17) a. They must have known it all the time, (perfect aspect)
b. She will be singing in several operas this year, (progressive as-
pect)
c. You will be notified in due course, (passive voice)
The auxiliary pairs can also be combined:

(18) a. They have argued that the Inkatha leader has been deliberately
orchestrating the recent violence. (CB) (perfect and progressive)
b. Samples from 90 horses are being tested at the Australian Ani-
mal Health Laboratory, in Geelong. (CB) (progressive and passive)
c. A computer program which diagnoses schizophrenia has been
developed by a University of Queensland research team. (CB) (per-
fect and passive)
d. And will the Queensland Government agree? Well, the papers
have been being put into the Cabinet in-basket and taken out of the
Cabinet in-basket since some time in March. (CB) (perfect, pro-

gressive and passive)
Two aspects of Langacker's notion of 'clausal head' deserve further at-

tention. First, there is the fact that clausal heads are presented as being ori-
ented towards being grounded, in finite clauses. Secondly, it remains to be
seen whether the verb group minus the grounding predication exhausts the
type specification of the clause or whether other clausal elements form part
of the clausal head as well. I will start with a discussion of the processual
nature of the clausal head and formulate an analysis of the clausal head in
English which basically distinguishes two options, i.e. a finite and an atem-
poral one (Section 2.1.1). In a second part, I will present arguments in fa-
vour of viewing a clause's Objects and Complements as part of the clausal
head or type specification (based on Halliday 1994 and Davidse 1997)
(Section 2.1.2).
2.1.1. Finite vs. atemporal clausal heads
Langacker's discussion of the clausal head focuses on finite clauses, i.e. on

clausal heads which are grounded in the speech event in terms of tense or
modality. I would like to broaden this picture by proposing that the gram-
matical system of clausal heads offers not one, but two options: verb groups
with a processual profile cannot only be grounded; they can also be atem-
poralized (see also Heyvaert 2000). Atemporal clausal heads constitute an
option in the functional organization of the clause which should be ac-
counted for.
Which evidence can be adduced in favour of distinguishing a systematic
atemporal alternative to grounding the clausal head? I mentioned earlier
that one of the main structural arguments which Langacker gives in favour
of distinguishing the grounding predications of tense and modality from the
rest of the verb group is that it is possible to add a to-infinitive to any verb
group minus the markers of tense and modality (e.g., to clean, to have been
cleaning, but *to had been cleaning, *to must clean) (see Langacker 1991:
195). It seems to me, however, that this does not only illustrate that it is the
non-grounding auxiliaries (like have and be), rather than the grounding
predications, which form a strong unit with the main verb. These structures
also show that the combination of main verb and non-grounding auxiliaries
which Langacker calls 'clausal head' cannot only be turned into a finite
expression, but also into a /o-infinitive. The infinitival form can, in fact, be
attached to an equally wide range of auxiliaries and auxiliary combinations

as the finite forms. The to-infinitive can, moreover, be replaced by -mg:20
(19) washes/washed washing/ to wash

has washed having washed/ to have washed
has been washing having been washing/ to have been
washing
Similar to grounding predications, the -ing suffix and the to-infinitive can
be used with both so-called 'perfective' and 'imperfective' verbs. Imperfec-
tive verbs profile processes with identical component states (e.g., know —>
knowing/ to know). Perfective verbs profile some kind of change due to the
differences that exist among the component states (e.g., hit -» hitting/ to
hit) (Langacker 1987a: 254-267).
I propose that the to-infinitive and the -ing forms in (19) represent
clausal heads that have been atemporalized. They constitute an alternative
to grounding the clausal head by tense or modality. Significantly, ground-
ing a clausal head and atemporalizing it are mutually exclusive choices.
Atemporal markers cannot therefore be added to a grounded structure, nor
can an atemporalized structure be grounded:
(20) a. *to was sitting on that bench (Langacker 1991: 421)

*shoulding have sat on that bench (Langacker 1991: 421)
b. *should being killed
*will to play in the garden
However, if the atemporal -ing suffix is elaborated into a structure with be,
thus forming the auxiliary pair be V-ing, a new processual structure or
clausal head is formed and grounding becomes possible again: when, for
instance, being dropped is turned into be being dropped, it can be grounded
in the speech event, e.g., is being dropped.
Can atemporalized clausal heads effectively figure as head of the clause,
and which is their semantic import? Like clausal heads that have been
grounded in the speech event, atemporalized clausal heads can serve the
function of head of the clause. As with finite clausal heads, a varying num-
ber of other clausal components (e.g., Subject, Object, Complement) tends
to accompany them. In the non-finite clause in (21), for instance, the apples
functions as Subject, and the ¿y-phrase realizes the Agent of the process. In
(22), then, the atemporal clausal head is accompanied by an adverbial {con-
scientiously) and an Object (the woods he loved).
(21) Those who ate the apples ran off wild and mad into the forest - the
apples having been poisoned by a woman who had loved Merlin ....
(CB)
(22) There his father ... lived the life of a recluse, conscientiously tending
the woods he loved. (CB)
Observe that the clauses in which atemporal clausal heads function are nec-
essarily subordinate, i.e. they cannot stand alone as independent clauses. I
will come back to this briefly in Section 2.3.3.
As to their semantic import, I hold that atemporal clausal heads basi-
cally represent a different way of looking at the process type specification.
They do not ground it or instantiate it, but preserve its status as type speci-
fication. The change of processual structures such as play or have been
playing into the atemporal structures playing or having been playing, for
instance, does not instantiate them.21 It does, however, change the way in
which the component states of the verb type are scanned: rather than the
step-by-step, serial way of scanning which is typical of processual struc-
tures (such as kill, have killed, be killing) and finite clauses, atemporalized
clausal heads offer a holistic view and activate the component states of a
verb cumulatively, through summary scanning (as defined in Langacker
1991: 554). They thus present the component states of a verb as "a single
gestalt" (Langacker 1991: 21). This is true for both atemporal clausal heads
in -ing and for those that take a io-infinitive.
It is beyond the scope of the present study to examine the precise se-
mantic differences that exist between -ing and to-infinitives as they func-
tion in clausal heads. It is important, though, to point out that the -ing of an
atemporal clausal head should not be confused with progressive -ing. Pro-
gressive -ing has been characterized as "restricting its profile to a series of
component states that does not include the initial and final states" and as
rendering a verb "imperfective" in that its component states "are construed
at a level of schematicity that neutralizes their differences" (Langacker
1991: 209). Verbs that are clearly imperfective cannot therefore take pro-
gressive -ing. As pointed out before, however, the -ing of atemporal clausal
heads can be used with perfective as well as imperfective verbs (e.g., know-
ing the truth, having been there) and the clausal head which it forms can
itself contain a progressive (e.g., having been washing; having been playing
in the garden). As Langacker has moreover pointed out to me, an atemporal
clausal head such as eating that whole chocolate cake in [Eating that whole
chocolate cake] was a big mistake does not take an internal perspective on
the atemporalized process (i.e. one that excludes its endpoints). Rather, the
atemporal clausal head designates the entire event of eating the cake. In
that respect, the -ing of atemporal clausal heads also differs from construc-
tions in -ing that follow verbs of perception (e.g., We saw [the ship sink-
ing], We heard [him yelling]) in which -ing can be attributed the same
value that it has in the progressive (i.e. it excludes the endpoints of the
event and offers an 'inside perspective') (Langacker 1991: 443).
To conclude, then, the process type specification which forms the core
of the English clause is either grounded in the speech event, or it is ex-
cluded from grounding by being explicitly atemporalized. In the latter case,
the process type specification remains ungrounded. Atemporal or un-
grounded clausal heads need not, however, remain uninstantiated: in what
follows, I will argue that atemporal clausal heads, like all clausal heads, are
instantiated by being combined with a Subject. Crucially, the role of the
clausal Subject should be distinguished from that of the Object, which is
primarily type-specifying: the clausal Object forms part of the clausal head,
which the Subject serves to instantiate. I will first elaborate on the type-
specifying role of the clausal Object (Section 2.1.2), and will only in Sec-
tion 2.2. turn to the instantiating properties of the Subject.
2.1.2. The clausal head, Objects and Complements
In Langacker's functional analysis of the clause, it is the verb group minus

the markers of tense and modality which is presented as clausal type speci-
fication or head. The inclusion of the participants that are involved in the
process and realized as Subject or Object of the clause is argued to turn the
process type specification into an instance (Langacker 1991: 34). Davidse
(1997: 420) has proposed an alternative view according to which the type
specification of the clause also includes "any Objects, Complements, and
circumstantial Adjuncts that may be present in the clause". Instead of serv-
ing the role of instantiator, the Objects, Complement and Adjuncts merely
'subcategorize' the process type expressed by the verb stem "into more
specific process types" (Davidse 1997: 422).
This claim builds on Halliday's (1994) observation that a fundamental
structural split seems to exist between the Subject and the finite verb (and
polarity), on the one hand, and the rest of the clause, on the other. Halliday
views this split as being essentially interpersonally motivated, i.e. it divides
the clause into two main parts according to the function which they serve in
turning the representational content of the clause into a message that can be
exchanged in communication. In Halliday (1994)'s interpersonal analysis
of the clause, the Subject-Finite complex is referred to as the 'Mood ele-
ment', and the rest of the clause is labelled the 'Residue'. Semantically
speaking, the Mood is argued to relate what is expressed in the Residue to

the context of the here-and-now of the speech event (Halliday 1994).22
Davidse (1997, p.c.) brings together Halliday's construct of 'Residue' with
Langacker's functional analysis of the clause and views the Residue as real-
izing the clausal process type specification: the Objects, Complement and
circumstantial Adjuncts that belong to the Residue, she argues, serve to
'subcategorize' the general process type expressed by the verb group minus
the grounding elements. Or, as Davies (1979: 64) puts it, a clause minus its
grounding devices is a "complex name" or a description, to which the
grounding devices can be attached. In his study of the Subject-Predicate
relation, Strawson (1974) observes that the Subject serves to 'individuate'
the non-particulars that are expressed by the Predicate: the Subject gives
"an individually distinguishing specification of a particular situation of a
certain general type" (Strawson 1974: 95). In Davidse's view, the Residue
(or clausal head) thus contains not only the main verb and its non-
grounding auxiliaries (as argued in Langacker 1991), but also the Objects,
Complements and Adjuncts that function in the clause. In the examples in
(23) and (24), the clausal head is in italics:
(23) She cleaned the room.

(24) She must have gone to school.
The fundamental structural and semantic distinction which appears to be at

work within the structure of the clause between the Subject and the ground-
ing predications of tense and modality, on the one hand, and the non-
grounding auxiliaries, the main verb and the Objects and Complements that
follow it, on the other hand can be represented as follows (the analysis that
is given is based on the clauses I am writing you a letter and You 're not
being a pest, are you?)·.
Mood element Residue

Subject Finite Polarity Predicator 10 DO Complement
TYPE SPECIFICATION
st , ... you a
/ 1 P·, be writing ^
pres.tense
2 p., , , . a pest,
You not be being ,
pres.tense (are you?)
Figure 6. The Residue as clause type specification

We have thus far identified the semantic function of the Residue as be-
ing that of clausal head, providing the type specification within the clause.
In the next section, I will go more deeply into the question of how instan-
tiation and quantification of the clausal head are realized.
2.2. Instantiation and quantification of the clausal head
Broadly speaking, the instantiation of a process type is in many ways simi-

lar to instantiation in the nominal, as described in Langacker (1991). Like
the instantiation of a noun type specification, instantiation of a process type
implies that one conceives the domain of instantiation (in this case time) as
having sufficient expanse to support the manifestation of multiple instances
and that the profiled processual instance is situated at a specific location
which is distinct from other possible locations. As in the nominal, instantia-
tion of a process type always results in one instance, the size of which can
be further specified through quantification, while its relation to the speech
event can be established through grounding. As pointed out by Langacker
(1991), the basic functional schema which clauses realize thus does not dif-
fer fundamentally from that of the nominal.
With Davidse (1992a, 1997), I hold that it is by being tied to a particular
Subject that the type specification expressed in the clausal head or Residue
is turned into the conception of an instance.23 Because typically, instances
are also grounded in the speech event, however, it is difficult to find evi-
dence in favour of the claim that it is actually the Subject which fulfils the
function of instantiator of the clause. As pointed out in Davidse (1991:
364), the best place to observe the precise function of the Subject in Eng-
lish is in non-finite, nominalized clauses, where the grounding effect of the
Finite is cancelled. Consider the following examples:
(25) Writing a letter is something that can be done in half an hour.

(26) Your having written us a letter made all the difference.
The non-finite structure in (25) merely represents a type specification,

which is not tied to a particular instance. The structure in (26), in contrast,
has been linked up with a Subject (which is periphrastically realized by
your) and represents an instance of the type specification. In other words,
by adding a Subject to an otherwise ungrounded clausal type specification,
the latter is turned into the conception of an instance. Notice that the instan-
tiated nature of the clausal type specification in (26) is not only made clear
by the presence of a Subject: it is also reflected in the 'secondary' tense
having written which is marked on the clausal type itself. This type of sec-
ondary tense in the non-finite verb group is not deictic, i.e. it does not itself
anchor the clausal type in the here-and-now of the speech event (Halliday
1994: 204). It does, however, imply grounding (or primary tense), which, in
turn, presupposes that the clausal type specification is viewed as being in-
stantiated.
It can be observed here that the claim that the Subject has special status
with respect to the instantiation of the clause fits in with the Cognitive
Grammar analysis of the Subject as 'primary focal participant' (Langacker
1991: 321) and that of 'initial reference point accessed in arriving at the
processual conception' (Langacker p.c.). As pointed out to me by Lan-
gacker, it also meshes nicely with other ideas that have emerged from re-
search in Cognitive Grammar. It reveals, for instance, an interesting parallel
between the instantiation of the clausal process type and that of a nominal
type with possessor: in both of them instantiation is effected by the element
that is characterized as the primary reference point (i.e. the Subject and the
possessor respectively) (see also Langacker 1993). In my opinion, the Sub-
ject in that respect differs fundamentally from Objects or 'secondary' focal
participants, which, I argued before (Section 2.1.2), cannot themselves in-
stantiate a type specification (though they can serve to specify in more de-
tail the instance that has been established by the clausal type specification
and the Subject).
How, then, can the size of the single clausal instance which the Subject
establishes through instantiation be further specified, i.e. how are clauses
quantified!? Langacker (1991: 197, 421) proposes that it is the non-
grounding auxiliaries of aspect (perfect and progressive) that serve to quan-
tify the verb's processual profile: they "indicate, with respect to some tem-
poral reference point, whether the process designated by the content verb
has been completed or is still in progress (hence only partially accom-
plished)" (Langacker 1991: 197). At the same time, however, Langacker
argues that these non-grounding auxiliaries serve to derive a higher-order
type specification, and he compares their function to that of number in the
nominal, where the plural of a noun like pebbles creates a type specification
that is distinct from that of pebble·, while pebbles designates a replicate
mass type (which belongs to the category of mass nouns), pebble desig-
nates a discrete entity and represents a count noun (Langacker 1991: 78).
The claim that the non-grounding auxiliaries of aspect derive a higher-order
type specification, however, seems to conflict with the idea that they ex-
press quantification, which presupposes instantiation (Davidse 1997: 423).
On this point, I will thus follow Davidse (1997), who argues that it is
not the non-grounding auxiliaries which quantify over a clause's processual
instance, but the system of polarity. Davidse (1997) refers to Davies

(1979), who argues that polarity expresses 'occurrence value' ('Ov') and
that "One or other Ov (positive/negative) is built into every description
which involves a predication" (Davies 1979: 67). When the speaker uses
negative polarity, as in My son didn 't sleep yesterday, he or she states that
the occurrence value of the instance is 'nil' (Davidse 1997: 423). By using
positive polarity, in contrast, the occurrence value of the instance is claimed
to be 'at least one', as in My mother called yesterday. As pointed out in
Halliday (1994: 88), polarity in between positive and negative is expressed
by modality. Epistemic modality of probability realizes the polarity of'"ei-
ther yes or no', i.e. maybe yes, maybe no, with different degrees of likeli-
hood attached" (Halliday 1994: 89). Epistemic modality of usuality is ar-
gued to be equivalent to 'both yes and no', i.e. "sometimes yes, sometimes
no, with different degrees of oftenness attached" (Halliday 1994: 89). De-
ontic modality of obligation, then, realizes 'do' and 'don't' and intermedi-
ate degrees; that of willingness expresses 'will' and 'won't' and intermedi-
ate degrees. Notice that the polarity that is linked to occurrence value is,
like nominal quantification, essentially associated with instantiation and
grounding. It should be distinguished from what has traditionally been
called 'internal' negation or negation that follows a modal auxiliary and
negates the type specification rather than the modality itself (Davidse 1997:
424), as in:
(27) You guys must / not have closed the door all the way. (CB)
(28) Your friend must/ not have very good eyes. (CB)
(29) You may/ not have read in the Constitution about Election Day for the
Supreme Court.... (CB)
(30) The researchers say the findings suggest women may/ not have
equivalent access to these procedures (CB)
Importantly, Davidse's proposal to assign a quantifying role to polarity

is in line with the claim that instantiation is situated in the Subject and
grounding is realized by both Subject and Finite (see Section 2.3). Quanti-
fication presupposes instantiation (only the size of an instantiated process
type can be further specified) and it is tied up with grounding (only quanti-
fied instances can be grounded). Not surprisingly, therefore, quantification
is situated within the Subject-Finite or Mood-element of the clause. The
relation between the functional analysis of the clause as a 'grounded quanti-
fied instance of some type' and the systemic-functional interpersonal analy-
sis of clause structure (as presented earlier in Figure 6) can now be refined
as follows:

Predi- Comple-
Subject Finite Polarity IO DO
cator ment
INSTAN- QUANTIFI-
TYPE SPECIFICATION
TIATION CATION
ist
1 p.,
be writ- you a
/ present
ing letter.
tense
2 p., a pest,
be be-
You present not (are
ing
tense you?)
Figure 7. Clausal instantiation and quantification
In this view, the clause closely resembles nomináis with common noun
head, realizing the functions which it encompasses iconically, with the
process type specification as the innermost layer of organization and the
functions of instantiation and quantification added to it. The clause can
leave it at that and profile a quantified instance of a type which is not
grounded in the speech event, witness the occurrence of non-finite clauses
with atemporal clausal heads. In Section 2.3,1 will look into quantified in-
stances of a process type that are grounded and I will give an overview of
the types of grounding which they incorporate.
2.3. Clausal grounding
In many functional descriptions (Hengeveld 1989; Langacker 1991; Halli-

day 1994), it is held that the clause is grounded or related to the here-and-
now of the speech event by the finite element of the verb group. Typically,
the finite element is said to ground the clause either through tense, by refer-
ence to the 'temporal zero-point' or the time of encoding and/or decoding
of the utterance (Declerck 1991a), or by means of modality, i.e. by refer-
ence to the judgement of the speaker as to how 'probable' or how 'obliga-
tory' occurrence of the process instance is. In what follows, I will discuss
two analyses which refine this basic model of clausal grounding. In Section
2.3.1,1 elaborate on the idea that all tensed finite clausal heads ground the
clause in what is basically a modal, epistemic way of grounding (Verstraete
2002). In Section 2.3.2, I discuss a proposal by Davidse (1997) to analyze
the clause as being grounded not only in terms of tense/modality, but also
in terms of grammatical person or person deixis. The Subject is argued to
contribute to this type of clausal grounding by serving as the objective or

explicit realization of grammatical person. It is thus identified in its second
main function, in addition to that of instantiator of the clausal process type.
2.3.1. The finite element
In his discussion of the grounding predications in the English verb group,

Langacker (1991) distinguishes between a schematic characterization of
their properties and one in terms of more specific values which the ground-
ing predications may have. At the schematic level, he (1991: 244) argues,
"time per se is not a factor" and "the system is purely one of modality"
(Langacker 1991: 244). Clauses which have a tensed verbal group but lack
a modal auxiliary are, in this perspective, viewed as realizing a schematic
type of modality.
Langacker (1991: 242fnl) classifies the schematic modality of finite full
verbs as being 'epistemic' in nature or related "to what a particular person
knows (or thinks he knows)". A similar claim is made by Davies (1979,
2001) and Verstraete (2002). Both argue that the indicative mood (or the
verbal group that contains a finite full verb rather than a modal auxiliary)
realizes epistemic modality, or encodes - as do the epistemic modal auxil-
iaries - the speaker's position with respect to the plausibility of the proposi-
tional content of the clause. Unlike epistemic modals, however, the indica-
tive does not express probability or possibility. Rather, it expresses the
speaker's "commitment to the truth of the proposition" (Verstraete 2002:
58), i.e. the proposition is 'known to be true' (Davies 2001: 230).
In an attempt to integrate this schematic epistemic modality of the in-
dicative into the overall distinction between epistemic and deontic modal-
ity, Verstraete (2002), basing himself on Halliday (1970a) and Davies
(1979), redraws the traditional distinction between epistemic and deontic
modals and posits that a fundamental distinction can be made between, on
the one hand, various types of epistemic grounding (including the indica-
tive, 'objective' deontic modality and 'dynamic' modality), and, on the
other hand, the 'subjective' type of deontic modality, together with the im-
perative.
What precisely does the notion of subjective deontic modality stand for
and what distinguishes the subjective type of deontic modality from objec-
tive modality? The notions of subjective and objective modality go back to
a distinction that was first made in Halliday (1970a). Halliday argues that
two main subtypes can be distinguished in the overall system of modality.
A first type of modality represents a "form of participation by the speaker
in the speech event" in that the speaker "associates with the thesis an indi-
cation of its status and validity in his own judgment" (Halliday 1970a:
335). This type of modality is called 'interpersonal' modality or 'modaliza-
tion'. It comprises the epistemic modal assessments of probability/ possibil-
ity/ certainty, as well as those of usuality/ typicality. A second type of mo-
dality is called 'modulation' and is argued to be "part of the thesis", i.e.
"part of the ideational meaning of the clause" (Halliday 1970a: 336). Under
modulation, Halliday includes the modals of volition and ability (grouped
together as 'dynamic' modals by Palmer 1990: 2, as cited in Verstraete
2002: 20), as well as instances of deontic modality to do with obligation
and permission. In Lyons (1977), the terms 'subjective' and 'objective'
were introduced to refer to the interpersonal and ideational types of modal-
ity respectively: subjective or interpersonal modality thus encodes the
speaker's position with regard to the propositional content of the clause.
Objective or ideational modality, in contrast, is internal to the proposition
(Verstraete 2001: 1507).24
Halliday (1970a) and Verstraete (2001, 2002) suggest that a subjective
type of modality exists in addition to that of epistemic modality: they argue
in favour of recognizing a subjective type of deontic modality as well. Ex-
amples of subjective deontic modality are given in (31) and (32). The mo-
dals in (33) and (34) illustrate objective deontic modality (based on Ver-
straete 2002: 38):
(31) Almost daily, I was told No, Clarkson, you are a fool and you may not
wear training shoes for school. (CB)
(32) For taking my sister's life, and all the other lives, too, she must be
locked up forever. (CB)
(33) But Ramadan means more than just physical deprivation. It has spiri-
tual and moral obligations, too. Followers must refrain from bad
thoughts, words and actions .... (CB)
(34) As you know, Mr. Karmal, a king must always return a gift with
something of similar value. (CB)
In the clauses in (31) and (32), the modal serves to encode a particular posi-
tion of the speaker with regard to the propositional content. The only dif-
ference between this type of deontic modality and epistemic modality con-
cerns "the relevant domain for the position", i.e. "deontic modality encodes
positions concerning the desirability of actions, whereas epistemic modality
encodes positions about the plausibility of propositions (Verstraete 2002:
40). The examples in (33) and (34), in contrast, do not encode a particular
position of the speaker, but they "report on the existence of a particular ob-
ligation without necessarily committing the speaker to it" (Verstraete 2002:

38). The speaker merely describes the obligation, but is not necessarily the
source of it. As such, the modals in these examples are objective or idea-
tional in nature and closely related to the dynamic modalities of willingness
and ability. With the existence of a subjective type of deontic modality es-
tablished, the system of modality in English can be schematized as follows:
Table 2. The system of modality in English
subjective or interpersonal modality/modalization

-epistemic modality -probability, possibility, certainty
-usuality, typicality
-deontic modality -obligation, permission25
objective or ideational modality/ modulation

-dynamic modality -willingness, ability
-deontic modality -obligation, permission26
When the various types of subjective and objective modals are analyzed
in terms of whether or not they establish epistemic grounding, another pic-
ture emerges and the main distinction turns out to be that between subjec-
tive deontic modals (and imperatives), on the one hand, and all the other
types of modality and the indicative, on the other:
Table 3. Epistemic vs. non-epistemic grounding
Epistemic grounding: -indicative mood

-epistemic modal auxiliaries
-'objective' deontic modality
-'dynamic' modality
Non-epistemic grounding: -imperative
-'subjective' deontic modality
Epistemic grounding turns out to be linked to tensed verbal groups, to all

those modality types that can realize tense (i.e. those of objective deontic
modality and dynamic modality), and, of course, to epistemic modal auxil-
iaries. Non-epistemically grounded are the imperative and the subjective
type of deontic modality, both of which are tenseless. It can be observed
that Verstraete (2002) thus analyzes clauses with objective modality (of the
deontic or the dynamic type) as establishing epistemic grounding.27 An im-
portant indication of the presence of epistemic modality in clauses with
objective modality is the presence of tense marking in them. Unlike subjec-
tive deontic modals, objective deontic modality does not encode the
speaker's commitment to the obligation, but merely describes the existence
of such an obligation. And this can be located in past, present or future, just
like any other proposition (Verstraete 2002: 63):
(35) In Bangalore, computer companies have to/ had to/ will have to do
privately what they cannot depend on the government to do.
The same counts for clauses with dynamic modality, which are also tensed
(Verstraete 2002: 65):
(36) He could write, he could play any instrument.... (CB)
Significantly, the presence of tense in clauses such as those in (35) and (36)
distinguishes them from clauses with subjective deontic modality: subjec-
tive deontic modals apply to tenseless or virtual domains. The state of af-
fairs that combines with subjective deontic modals is desired and, conse-
quently, purely virtual.
To conclude, the finite element does not simply ground the clause either
by tense or by means of modality: at a schematic level, all clauses that real-
ize tense can be analyzed as being modal in nature. They establish epis-
temic modality and express the speaker's position with regard to the plausi-
bility of the prepositional content of the clause. As Davies (2001) argues,
the semantics of indicative mood and of epistemic modality can thus be
treated as an integrated continuum. Interestingly, the objective type of de-
ontic modality and the dynamic modalities of ability and volition also seem
to establish epistemic modal grounding and thus form part of the epistemic
continuum as well. It is only the subjective type of deontic modality and the
imperative which do not imply any epistemic position on the part of the
speaker about the plausibility of the proposition. Instead, they express the
speaker's desire or wish and therefore deal with states of affairs which are
purely virtual and tenseless.
2.3.2. Person deixis
Davidse (1997) proposes that, in addition to being grounded by the Finite,

the clause is also grounded by 'person deixis': the process instance is not
only located with respect to the ground by being related to the speech event
in terms of modality and tense, but also by being tied to 'grammatical per-
son', which may be either the speaker/encoder, the hearer/decoder, or some
third person (Davidse 1997: 425). It is obvious that first and second person
refer to the ground: they refer to the speech participants who are primary
constituents of the ground. With regard to third person, note the unique ref-
erential status of the Subject which is pointed out by Keenan as a 'truly
universal' property of Subjects: the Subject's reference "must be determin-
able by the addressee at the moment of utterance. It cannot be made to de-
pend on the reference of other NPs which follow it" (Keenan 1976: 313).
The structure * He-self admires John cannot therefore replace John admires
himself. A third-person Subject is thus the non-speaker/hearer referent that
is most immediately accessible for identification by speaker and hearer.
How is the system of person deixis encoded in the clause? Davidse ar-
gues that it is coded subjectively or implicitly (Langacker 1991: 93) by the
Finite, i.e. through person and number marking, and objectively or explic-
itly by the Subject.28 The Subject thus makes explicit the grammatical per-
son which is marked subjectively on the Finite.29 Notice that there is no
person and number marking on non-finite or atemporal clausal heads: it
seems logical that atemporal clausal heads, which do not ground the proc-
ess type temporally or modally, also fail to realize the person deixis that is
inextricably linked to it. As to the Subject function in non-finite clauses,
then, I suggest that it is reduced to its instantiating role. It is only when a
Subject functions in a finite clause that it co-operates in the grounding of
the clause by establishing objective person deixis. Whereas the nominal
may be grounded either in terms of spatial proximity (by means of demon-
stratives) or in terms of person deixis (via possessives), the clause is thus
grounded both in terms of temporal/modal proximity and person deixis
(Davidse 1997: 424).
2.3.3. Clausal grounding: Conclusion
Summarizing, in my description of the various systems that operate in lan-

guage to relate quantified instances of a process type to the ground, I have
highlighted two issues: first, every finite clause encodes, in one way or an-
other, the speaker's position towards the proposition expressed by the in-
stantiated process type. Every grounded clause thus realizes a schematic
modal feature, and this may or may not be combined with an indication of
tense. Secondly, grounding is more than the expression of temporal/modal
proximity: finite clauses also identify the instance which they ground with
respect to the speaker and hearer of the speech event, and they encode this
by means of person deixis or grammatical person, marked on the finite verb
and the Subject. The overview of the functional organization of the clause
can now be completed, with grounding located both in the Subject and in
the finite element of the clause:

Predica- IO Comple-
Subject Finite Polarity
tor DO ment
INSTANTIA- QUANTI-
TYPE SPECIFICATION
TION FICATION
GROUNDING
1 p., be writ- you a
/
pres.tense ing letter.
a pest,
2 p., be be-
You not (are
pres.tense ing
you?)
Figure 8. The role of the Subject and the Finite in clausal grounding
The Subject-Finite or Mood-element of the clause does not only serve to

ground the clausal type specification. According to Halliday (1994), the
Mood-element also defines the speech functional value of the clause. It ex-
presses the "illocutionary force of an utterance" or "its function as a speech
act" (McGregor 1997: 213). This illocutionary force is closely tied up with
the grounding function of the Mood-element: non-finite clauses cannot
therefore occur as independent speech acts in the discourse. Verstraete
(2002) points out that this is due to the fact that non-finites do not have
grounding that is expressive of the position of the speaker:
"Speaker/interlocutor-positioning is a functional prerequisite for a proposi-
tional content to be put forward as a speech act in discourse: what functions
in discourse is not bare descriptions of SoAs, but speakers' and interlocu-
tors' positions with respect to these SoAs" (Verstraete 2002: 79). The
speech functional value of finite clauses identifies a third function of the
Subject, viz. that of 'speech functionally responsible' element (Halliday
1994). In the following and last section, I will briefly elaborate on the
speech functional role of the Subject.
2.4. The speech-functional role of the Subject
I have thus far shown that a clause is always grounded by person deixis and
in terms of modality: when the clause realizes epistemic modality, it en-
88 Thefiinctionalorganization of nominal and clause
codes the speaker's position with regard to the plausibility of the proposi-
tion; subjective deontic modality, in contrast, expresses the speaker's posi-
tion with respect to the desirability of the proposition. For a clause to func-
tion as a fixll speech act, however, it has to provide an indication as to who
takes responsibility for the modal position that is expressed in it. This indi-
cation is provided by the basic clause types declarative, interrogative and
imperative: together with the modal positions that can be taken in the
clause, these clause types serve to define the speech function of the clause
(Verstraete 2002: 83). In what follows, I will first briefly discuss these
speech functions and show how they indicate whether it is the speaker or
the hearer that is responsible for the modal position that is implied in the
clause. My discussion of the speech functions is based on Davies (1979)
and, in particular, on Verstraete (2002: 83-106). In a second part, I will
zoom in on the speech functional role of the Subject. My analysis of the
role of the Subject in the clause as speech act builds on Halliday (1994) and
Davidse (1997).
Declarative and interrogative structures serve to encode alternative as-
signments of responsibility for the modal position which they imply: while
the declarative assigns responsibility to the speaker in his or her own turn,
the interrogative transfers responsibility to the hearer in the next turn. As
Davies (1979: 51) puts it, in an interrogative, the speaker assigns to the ad-
dressee "the role of subsequent teller in so far as the speaker's use of this
construction incorporates a demand for reply". In the declarative examples
in (37) and (39), the speaker thus takes the responsibility in his or her own
turn and is committed to the assessment of possibility realized by might and
the assessment of certainty implied in the indicative in (39). The interroga-
tives in (38) and (40), on the other hand, signal that the speaker transfers
the responsibility to the hearer in the next turn. Verstraete (2002: 85) ob-
serves that this does not imply that the hearer will take this position:
"rather, transfer of responsibility implies that it is up to the interlocutor to
accept, reject or modify the modal position which the speaker has chosen as
the starting point for the exchange" (the examples are based on Verstraete
2002: 85):
(37) Gene therapy might very well become a major new revolution in
medicine. (CB)
(38) Might it be of some value? (CB)
(39) Then he got us involved in Iran-Contra. (CB)
(40) Were the police involved in that? (CB)
Note that in imperatives, as in declaratives, it is the speaker that takes re-

sponsibility for the modal position: the speaker expresses the desirability of
a certain action and takes responsibility for it:
(41) Leave this room at once!
The same difference between declarative and interrogative can be ob-

served when the position that is encoded in the clause is one of subjective
deontic modality. The declarative in (42) thus signals that the speaker takes
responsibility for the position of permission that is encoded by may,
whereas the interrogative in (43) signals transfer of the responsibility to the
hearer. Note that, again, this transfer does not imply that the hearer will
take the position. It only implies that the hearer will decide about the issue
of permission raised by the speaker:
(42) You may return to your duties, Colonel Pople. (CB)

(43) May we sit down? (CB)
What about the role of the Subject in declaratives, interrogatives and

imperatives? Halliday (1994: 76) views the Subject as the 'modally respon-
sible element' or the element that is "so to speak, being held responsible -
responsible for the functioning of the clause as an interactive event". Now
that we have distinguished between two basic modal positions (i.e. that re-
lated to the plausibility or truth of the proposition and that related to the
desirability of the proposition), Halliday's analysis of the Subject as
'modally responsible' can be understood as follows. The Subject is not re-
sponsible in the way that the speaker or hearer is. It is not responsible for
the modal position of plausibility/truth/desirability that is implied in the
speech act: it is the speaker who assumes a particular position and the
hearer who is assigned a position. Instead, the Subject is the entity which
the speaker holds responsible for what the modal position involves, i.e. the
Subject is presented as being 'responsible' for the truth or the plausibility
of an epistemically grounded clause, and for the success of the obliga-
tion/desire in a subjective deontic clause and an imperative.
The precise nature of the speech functional role of the Subject thus var-
ies according to the type of schematic modality that is realized: in clauses
that realize epistemic modality, the Subject is the entity in which the plau-
sibility or truth of the proposition is vested:
(44) a. He gave the book to Mary,

b. Mary was given the book.
c. The book was given to Mary.

(45) a. Did he give the book to Maiy?
b. Was Mary given the book?
c. Was the book given to Mary?
In the declaratives in (44), it is the speaker who assumes responsibility for

the epistemic modal position of truth, but it is the Subject in which the truth
of the proposition is vested: whether or not the proposition is actually true
is presented as being dependent on the Ae-person in (44a), on Mary in (44b)
and on the book in (44c). If one wants to find out whether the propositions
in these examples are true, the Subjects indicate who or what to turn to: one
can ask the he in (44a), one can turn to Maty and check with her in (44b),
and one can try to locate the position of the book in (44c). The interroga-
tives in (45) are in this respect similar to the declaratives: they also present
their Subjects as the entities on which the truth of the proposition depends.
Different from the declaratives in (44), however, they transfer the responsi-
bility for the modal position of truth to the hearer, and thereby also ask the
hearer to decide on whether the Subjects support the truth of the proposi-
tion.
In clauses with objective deontic modality (which imply epistemic mo-
dality and belong to the epistemic continuum), the Subject is responsible
for the existence of the obligation. Consider, for instance, the following
examples (based on Davidse 1997):
(46) a. He has to tell mother the truth.

b. The truth has to be told to mother.
c. Mother has to be told the truth.
In (46a), the responsibility for the existence of the obligation is attributed to

the Ae-Subject: the Subject is not only the Agent who will have to do the
'telling', the Subject is also held responsible for the fact that an obligation
exists in the first place. The Ae-Subject may, for instance, be identified as
the one who has not told the truth to mother yet, as a result of which he is
now responsible for the fact that the obligation of telling her exists. In (46b)
and (46c), by contrast, the obligation is made to depend on characteristics
of the truth and of mother. There may, for instance, be "some moral im-
perative associated with 'the truth'", or the speaker's respect for mother
may explain the existence of the obligation in (46c) (Davidse 1997: 427).
In clauses with subjective deontic modality, then, as well as in impera-
tives, the modal position which is assumed by the speaker relates to the de-
sirability of the proposition. Because no epistemic modality is involved, the
Subject is not so much held responsible for the existence of the obligation,
but for carrying out the desired process: it depends on the Subject whether
the command will actually be carried out and will thus be successful. In
active clauses, the Subject is held responsible for carrying out the desired
process (as shown in 47a and 47b):
(47) a. (You) be quiet!

b. You must leave at once!
Halliday's notion of responsibility can, however, also be recognized in

passive imperatives and with passive subjective deontic modality. In (48a),
the Subject has been dissociated from the Agent of the process (i.e. the one
who invites), but is still responsible for the success of the command (Halli-
day 1994: 76). In (48b), the Subject is responsible for the success of the
obligation as well, though she may not willingly co-operate in it. It is,
nonetheless, the Subject that is assigned a central role in the obligation: it
does not matter who does the locking up, as long as it leads to her being
locked up: the success of the desired process is made to depend on the Sub-
ject ( Ί will not rest until she has been effectively locked up')·
(48) a. (you) Get yourself invited to their meeting, will you?

b. She must be locked up forever for what she did to me.
Observe the difference between (48b) and its objective deontic alternative
She has to be locked up forever·, in the latter, the sAe-Subject is held re-
sponsible for the existence of the obligation: she has committed a crime,
and therefore the obligation of locking her up exists.
Like the analysis of the Subject as clausal 'instantiator', its characteriza-
tion as 'modally responsible' element in the clause is reconcilable with the
Cognitive Grammar analysis of Subjects as 'primary focal participants'
(Langacker p.c.). Modal responsibility is assigned to the Subject-entity by
the speaker and therefore resides in the speaker's conceptualization of the
scene (Langacker 1990b). Like the notion of 'primary focal participant',
that of 'modally responsible element' is thus essentially a subjective cate-
gory. It manages to specify the precise nature of the speaker's conceptuali-
zation in more detail by linking it to the modal position (of plausibility,
truth or desirability) which the speaker takes in the clause. Importantly, the
subjective characterization of Subjects in terms of 'responsibility' is sche-
matic and generalizes across the range of Subjects that can be found in the
clause. Subjects can be said to differ, however, as to whether they are also
'objectively responsible', i.e. as to whether there is an objective basis for
selecting them as modally responsible elements (Langacker p.c.): while in

the examples in (44) to (48), some objective motivation for conferring focal
prominence or modal responsibility to the Subject-entity was identified,
this is not possible for an example such as It is obvious that Belgium is a
great place for beer drinkers·, it cannot be said to objectively motivate the
speaker's choice of construing it as Subject. I therefore agree with Lan-
gacker (p.c.) when he argues that the objective situation described may of-
fer varying degrees of motivation for conferring modal responsibility to an
element, but that it is ultimately the speaker who chooses which element
will function as Subject.
In short, the Subject-function serves to instantiate the process type ex-
pressed by the clausal head; it co-operates in grounding instantiated types
by relating them to the system of person; and, finally, the Subject is speech
functionally responsible: it is the entity which the speaker picks out as be-
ing responsible for the plausibility/truth of the proposition or for the suc-
cess of the obligation.
2.5. The functional organization of nominal and clause: Conclusion
The description of the internal functional organization of nomináis and

clauses in terms of Langacker's functions of type specification, instantia-
tion, quantification and grounding reveals that, at a highly schematic level,
nomináis and finite clauses realize the same functional schema: both repre-
sent grounded instances of a type (as argued in Langacker 1991). Impor-
tantly, these functions enable us to generalize over all instances of the class
of the nominal, as well as over all finite clauses, irrespective of whether the
functions are discernible (explicitly coded) or not. They thus show that the
classes of the nominal and the clause can be defined not only in terms of
their external functions and their intrinsic properties, but on the basis of
their internal organization as well.
Interestingly, when Halliday's metafunctions are brought into the pic-
ture, some of the differences between the nominal and the clause show up.
Firstly, in nomináis the ideational or representational content is located in
the nominal head, which "supplies a vast amount of conceptual content"
(Langacker 1991: 143). Whereas the noun type specification and its modi-
fiers serve to characterize the designated entity, the grounding predications
in nomináis are "wholly extrinsic" and do not pertain to "properties inher-
ent in the entity itself' (Langacker 1991: 143). Clauses, by contrast, do not
show compartmentalization of the representational content which they ex-
press, but rather have their representational content as a strand of meaning
that is realized throughout the clause (Halliday 1994: 179). The representa-
tional functions in the clause are thus not located in the clausal head only
(even though the clausal head does provide the basic categorization of ex-
perience in the clause): representational functions are mapped onto the Sub-
ject as well.
Halliday (1994: 35) argues that the textual metafunction in the clause,
which relates the clause to the larger discourse context, is expressed by
means of "peaks of prominence located at beginnings and endings": the
textual component in the clause expresses the textual status of parts of the
discourse by assigning them to the boundaries of the clause. There is thus a
special significance attached to 'coming first' and 'coming last' (Halliday
1979: 69). In the structure of the nominal (which is less flexible in terms of
word order), by contrast, textual functions are primarily located in the de-
terminer: determiners can thus relate the nominal to the discourse context
or to an immediate referent "within the text" (Halliday and Hasan 1976:
33). Determiners, in other words, establish a direct link with the speech
event and its participants or they have an immediate referent in the dis-
course or text.
Finally, as to the interpersonal metafunction, it was shown that Halli-
day's (interpersonal) Mood-Residue analysis of the clause matches Lan-
gacker's functional description of the clause quite closely. While the Resi-
due expresses the clausal type specification, the Mood-element turned out
to realize instantiation, quantification and grounding. Not surprisingly,
therefore, the interpersonal differences between the clause and the nominal
- which also realizes a 'grounded instance of a type' - are smaller than are
those in terms of the representational and the textual metafunctions. Inter-
personally speaking, both the nominal and the clause consist of a compo-
nent that is ungrounded and combines with grounding elements that relate it
to the speech event, thus turning it into something that can be exchanged in
communication. In the clause, it is the Mood-element consisting of the Sub-
ject-Finite link that is most essential to grounding and realizes instantiation
and quantification; in the nominal, it is determiners (in the broad sense) that
assume the functions of instantiation, quantification and grounding.
Now that we have gained insight into the functional parallels and differ-
ences that exist between nomináis and clauses, we can turn to nominaliza-
tion again and try to establish the impact of the functional organization of
the nominal and the clause on the process of nominalization.
3. Nominalization as functional reclassification
I argued in Chapter 3 (Section 3.2) that one of the main problems related to
the traditional analyses of nominalization is that they fail to see nominaliza-
tions as constructions in their own right, i.e. as constructions which com-
bine clausal and nominal categories in a unique way but which are essen-
tially nominal in status and function as such in larger configurations. The
nominal character of nominalizations derived from a verb stem is not con-
troversial. That of nominalizations containing clause-like units, as in [Being
able to play the piano] has always been a dream of mine has eluded lin-
guists more and has not been accurately described yet. The nominal fea-
tures of the latter type of nominalization have often failed to be recognized
because they tend to be encoded non-prototypically, i.e. as non-discernible
functional components. I argued in Chapter 3 that for the internal analysis
of nominalizations, a functional account of the nominal class is needed
which is sufficiently schematic to cover all nominal constructions, proto-
typical as well as non-prototypical. As I have shown in Section 1 of this
chapter, the functional analysis of the nominal in terms of type specifica-
tion, instantiation, quantification and grounding can provide such an ac-
count and thus makes an analysis of the nominal status of the various types
of nominalized constructions - including those that contain clause-like
units - possible.
In addition to the nominal characteristics of nominalized constructions,
the clausal categories which accompany the clausal level of assembly that
forms the starting point of nominalizations have to be accounted for as
well. The schematic characterization of the clause which I have set out in
this chapter (Section 2) allows us to define the distinct levels of assembly
which nominalizations can start from more accurately (as first suggested in
Langacker 1991): nominalization either applies to a simple process type
specification or verb stem, to what I have defined as an atemporal clausal
head or to a fully grounded (or finite) clause. I also pointed out in Chapter 3
(Section 5) that nominalizations should not be reduced to being clause-like
constructions, but should be analyzed as being systematically related to
clausal constructions through agnation. More specifically, I argued that re-
lationships of agnation between nominalizations and clausal constructions
can shed light on the clausal categories that play a role in nominalization.
The functional model of the clause and, in particular, the description of the
types of grounding that are found in it and of the various functions served
by the Subject, I will show, form an excellent tool for the description of the
clausal agnates of nominalized structures.
Nominalization as functional reclassification 95
Finally, I posited in Chapter 3 (Section 2) that the literature on nomi-

nalization tends to lose sight of the natural symbolic nature of nominalized
constructions: nominalizations encode their meaning in their lexicogram-
matical properties, and only a careful analysis of these lexicogrammatical
features can shed light on the semantics of the nominalized constructions. I
have thus far argued that nominalizations are combinations of nominal and
clausal functional categories. Crucially, the lexicogrammar of nominaliza-
tions is not made up of a random combination of clausal and nominal cate-
gories. Rather, nominalizations reclassify a clausal level of assembly into a
nominal construction and thereby integrate the functional values attached
to the clausal starting point with the functional organization of the nominal
structure in which it comes to figure. Importantly, this integration cannot be
described from a representational perspective only: it has to be accounted
for in terms of a multifunctional analysis which also considers, for instance,
its interpersonal properties or the role which each component plays in link-
ing the nominalization to the ground (as argued in Chapter 3, Section 4).
Once again, the functions of type specification, instantiation, quantification
and grounding prove to be pivotal to such an analysis: one of the most at-
tractive properties of these functions for the description of nominalizations
is precisely that they schematically characterize both the clausal and the
nominal mode of organization and thus make it possible to come to an inte-
grated analysis of the nominal and clausal contributions to nominalized
constructions, and this in terms of one and the same functional framework.
I will show that there exist interesting correlations between the clausal level
of assembly which a nominalization derives from, the type of reclassifica-
tion that applies to it (i.e. involving rank shift or not) and the nominal strat-
egy that is adopted (i.e. proper name or common noun strategy).
In short, the model of functional organization of the nominal and the
clause that has been presented in this chapter provides the descriptive
breakthrough that was needed to move beyond the existing analyses of
nominalization. It helps to identify the main principles that lie behind the
process of nominalization and, together with the theoretical concerns which
I raised in Chapters 2 and 3, it provides a descriptive-theoretical framework
for the analysis of specific nominalization systems.
To shed light on the functional reclassification mechanism which under-
lies nominalizations and to find out more about the precise nature of the
integration which nominalizations establish between nominal and clausal
categories, I will now embark on the description of specific nominalization
types. I will discuss nominalizations that are based on a simple verb or
process type specification (as in teacher and the signing of the contract),
nominalizations that derive from an atemporalized clausal head (as in I
don 7 regret [having waited for him]), and nominalizations that take a full
finite clause as starting point (as in [That he left her] surprised me). In spite
of the fact that they derive from different levels in the functional assembly
of the clause, all of these structures are nominalizations, i.e. they have been
reclassified and have come to function as nomináis. My analysis is aimed at
elucidating the reclassification process that leads from processual starting
point to nominalization.
In Part Π, I will look at deverbal -er nominalizations. In Part ΙΠ, I will
turn to the system of factive nominalization, which includes three lexico-
grammatical subtypes: gerundive nomináis (e.g., his being fired), that-
nominalizations (e.g., that he was fired) and the fact /Aai-nominalizations
(e.g., the fact that he was fired). When discussing gerundive nominaliza-
tions, I will also mention gerundive nomináis with an 'action' meaning
(e.g., [Firing someone] is a terrible thing to do) and action nominalizations
(e.g., the writing of that book).
Part II
Deverbal -er nominalization
Chapter 5
Deverbal -er suffixation: Towards a descriptive
position
0. Introduction
In this chapter, I will discuss the literature on deverbal -er nominalization

and single out the two main descriptive themes of my own analysis (pre-
sented in Chapter 7). First, as regards the semantic profile of -er nomináis,
the question will be raised whether the system can (still) be characterized as
agentive. Secondly, the distinction between lexicalized -er nomináis (e.g.,
destroyer 'warship') and 'ad hoc' nominalizations (e.g., the destroyer of
our hopes) will be introduced, together with the question of how to inte-
grate the differences between these types of -er nominalizations in the
overall analysis of -er suffixation.
As to the profile of deverbal -er nominalizations, it is widely agreed
that, diachronically speaking, the -er suffix (including its orthographic vari-
ants in -or and -or) is agentive·, the Old English suffix -er(e) primarily
served to derive Agents, and then especially personal or human Agents
(e.g., bœcere 'baker' from bacan 'to bake', bëâtere 'beater, boxer' from
bëâtan 'to beat', cwellere 'killer' from cwellan 'to kill', for more examples,
see Marchand 1969; Kastovsky 1971, 1985). The few non-agentive coin-
ings that exist in Old English, Kastovsky (1971: 295) points out, "are more
or less restricted to translations from Latin, and thus probably did not be-
long to the general vocabulary": examples of Latin-based, patientive -er
nomináis are scëawere 'buffoon, actor' (from scëawian 'to look at, see',
recorded as a gloss for Latin scurra) and scëawere 'mirror' (from scëawian
'to look at, see', a translation of Latin speculea). Designating the Location
rather than the Patient of the process of looking is scëawere in the meaning
of 'watch-tower' (Kastovsky 1971: 295). The instrumental type of -er
nominal, which is so productive in Modern English (e.g., eraser, toaster,
screwdriver, nutcracker), is hardly found among Old English -er nomináis:
the only example of an Old English instrumental -er nominalization found
by Kastovsky is pünere 'pestle' from pünian 'to pound' (Kastovsky 1971:
295).
Nowadays, suffixation of a verb with -er still prototypically results in a
nominalization of which the referent is the Agent of the base process
(hence the label nomina agentis; see, among others, Jespersen 1914-1929,
100 Deverbal -er suffixation: Towards a descriptive position
6; Lees 1960; Marchand 1969; Quirk et al. 1985). Apart from human
Agents, -er nomináis also designate animate Agents or animals (e.g., re-
triever, wood-pecker) and immaterial Agents (e.g., pointer, reminder). In-
strumental nomináis like scraper, scratcher, toaster, transmitter have been
argued to come close to being material Agents (see, for instance, the -er
entry in the Oxford English Dictionary and Marchand 1969). However,
Modern English has also seen the emergence of many different types of
non-agentive deverbal -er nominalizations. The non-agentives that are most
often identified in the literature are those based on verbs of cooking (as in
1), those denoting items of clothing (as in 2), locative nouns (as in 3), and,
finally, nomináis designating a variety of other non-agentive entities (as in
4):30
(1) broiler 'a young chicken suitable for broiling'

roaster 'a kind of meat suitable for roasting'
steamer 'an edible clam'
(2) jumper 'a type of pullover'
sneakers 'casual shoes with rubber soles'
slippers 'a kind of shoe that can be slipped into'
(3) sleeper 'a train you can sleep in'; 'a bed or sofa in a train, in
which you can sleep'
diner 'a place where you can dine'
(4) bestseller 'a book or other item that sells well'
dipper 'a vegetable/ fruit/ other snack that has to be dipped be-
fore being eaten'
scratcher 'a lottery ticket that has to be scratched to reveal the po-
tentially winning patterns'
The issue that needs to be tackled here is whether the Modern English
-er suffix can still justifiably be analyzed as agentive, or whether another
characterization has to be looked for which describes the present distribu-
tional properties of the suffix more accurately. Put differently, is agentivity
still the highest-level schema which deverbal -er nomináis instantiate, or do
we have to look for another schema? And if so, what does this alternative
high-level schema look like?
Not only the occurrence of non-agentive formations has to be accounted
for. Deverbal -er nominalizations also differ according to the degree to
which they are entrenched·. I referred earlier to the occurrence of 'ad hoc'
nominalizations (as in 5) and lexicalized -er nomináis (as in 6):31
-er nominalizations: Agent names or Subject names ? 101
(5) Alex Gough, the Welsh destroyer of Del Harris's hopes on Tuesday
....(CB)
(6) destroyer 'a small, heavily armed warship'
Lexicalizations are included in the language system as 'fixed expressions'

(Langacker 1991: 45) and name what are basically types of things or peo-
ple. Ad hoc nominalizations, in contrast, represent -er formations which are
derived on the basis of the constructional schema of -er nominalization to
serve a specific discourse purpose. These two types are characterized by
differences in degree of entrenchment and specificity, as well as by a num-
ber of semantic and structural differences. It has, for instance, been ob-
served that ad hoc nominalizations tend to be postmodified by an o/-phrase
(i.e. of Del Harris's hopes) realizing what would clausally have been a
participant in the process. Lexicalizations typically do not take this kind of
o/-phrase.
In what follows, I will go through the analyses of -er nominalization
that have been proposed thus far. I will first look at descriptions of the se-
mantic profile of -er nomináis (Section 1) and then at analyses of the dis-
tinction between lexicalizations and ad hoc nominalizations (Section 2).
Characteristic of the analyses of the profile of -er nominalization is that
they view all -er nomináis as designating what is basically a semantic role
(Ryder 1991, 1999a, 1999b), an argument structure configuration (Levin
and Rappaport 1988; Mackenzie 1990; Rappaport Hovav and Levin 1992)
or a complex conceptual category (Panther and Thornburg 2002). The label
of 'Subject name', I will show, has thus far been brought into the discus-
sion of -er nominalization only occasionally and without reflection on the
actual value and use of the Subject in clausal contexts (e.g., Marchand
1969; Kastovsky 1971; Bauer 1983; Booij 1986, 1992). As for the differ-
ences between lexicalized and ad hoc -er nominalizations, I will point out
that they have been repeatedly discussed in the literature (e.g., Jespersen
1914-1929, 6; Strang 1968, 1969; Levin and Rappaport 1988; Mackenzie
1990; Rappaport Hovav and Levin 1992), but they have not yet been
brought together into a coherent analysis, nor has the distinction between
lexicalized and ad hoc nominalizations been integrated in the analysis of
deverbal -er suffixation in general.
1. -er nominalizations: Agent names or Subject names?
Research into the semantic profile of deverbal -er nominalization has al-
ways struggled with the system's tolerance of non-agentive -er nomináis.
Some of the early accounts of the suffix simply ignore the existence of non-
agentive -er nomináis and refer to -er as an agentive suffix (e.g., Oxford
English Dictionary; Lees 1960; Quirk et al. 1985). Others acknowledge that
-er can also derive non-agentive nomináis, but do not attempt to integrate
the non-agentive cases in the system of -er suffixation in general: their
account of -er nominalization is restricted to a mere list of possible forma-
tions (e.g., Jespersen 1914-1929, 6; Quirk et al. 1972; Kastovsky 1985).
Still others point to the disparity between agentive and non-agentive -er and
conclude that the non-agentive type is not part of the system of -er nomi-
nalization, that it is idiosyncratic or non-systematic in nature (e.g., Keyser
andRoeper 1984; for Dutch: Booij 1986; Taeldeman 1990).
Linguists who have tried to come to terms with non-agentive -er nomi-
nalization are, firstly, Ryder (1991, 1999a, 1999b), who offers an analysis
of the system of -er nominalization which focuses on the prototypically
agentive status of the designated entity and its salience within the event in
which it figures; secondly, Panther and Thornburg (2002), who suggest that
-er nominalization forms a complex conceptual category which is centred
on an agentive prototype to which a large number of other meanings are
linked through metaphor and metonymy; and, thirdly, Levin and Rappaport
(1988) and Rappaport Hovav and Levin (1992), who analyze -er nomináis
from the perspective of argument structure configurations. In the following
sections, I will have a closer look at the descriptive analyses which they
propose.
1.1. Agentivity and salience
The analysis which Ryder (1991, 1999a, 1999b) proposes to account for the
variety of profiles in deverbal -er nominalizations is "semantic/pragmatic"
in focus (Ryder 1999b: 291). It analyzes -er nominalization "in terms of
semantic case, event structure and prototype reanalysis" (Ryder 1991: 299)
and builds on the assumption that semantic roles "are better viewed as pro-
totypes rather than as absolute categories" (Ryder 1991: 300). While some
participants will thus "be considered agents by everyone, other, less proto-
typical ones will be considered agents to the extent that they can be con-
strued as matching or approximating the prototype" (Ryder 1991: 300).
Ryder argues that the non-agentive cases of -er can all be regarded as ex-
tensions or reanalyses of the agentive -er prototype, reanalysis being possi-
ble when a non-agentive entity is conceived as being salient or particularly
noticeable within the designated event (Ryder 1991: 309; 1999b: 288-289).
Salience is what relates deverbal -er nominalizations to denominal ones
(e.g., prisoner, weekender) in which "the referent of the -er noun ... should
be higher in salience than that of the nominal base" (Ryder 1999b: 290).
The schema which is instantiated by all -er nominalizations - deverbal and
non-deverbal - can thus be summed up as "a person, animal or thing sali-
ently connected with X" (Ryder 1999a: 307).
How did in Ryder's opinion the distinct extensions of the agentive pro-
totype in deverbal -er nominalizations come about? She (1999b: 287) ar-
gues that two factors are important here, viz. an entity's salience and identi-
fiability in the event, and, secondly, its location on the causal chain of the
event in which it figures (Ryder 1999b: 287). The first and most obvious
extensions of Agent -er nomináis are those nomináis that designate Instru-
ments. Instruments can easily be foregrounded as salient, agent-like enti-
ties: "because they are independent of the Agent in many ways, the event
can be construed as having them as the head of the part of the causal chain
that is in focus, and so, like Agents, they are highly salient" (Ryder 1999b:
288). Especially the great proliferation of implements and machines from
the 16th century onwards explains the extension of the almost exclusively
agentive -er suffix to instrumentais in the early Modern English period (as
attested in Dalton-Puffer 1994).
Once -er nomináis came to designate Instruments, "clothing intended to
be worn while performing central actions in an episode" was construed as
similar to an Instrument used in that episode (Ryder 1991: 305). It is the
'identifiability' of clothing items, or the fact that they are "specifically de-
signed for certain activities, and so can be readily identified by those activi-
ties" (Ryder 1999b: 290) which explains why various -er nomináis denot-
ing items of clothing have since been coined: e.g., loafers, sneakers, wad-
ers, loungers, jumper. Patientive -er nomináis referring to food, then, pro-
file structures in which "the agent is outside the boundaries of the episode,
leaving the original patient as the most agent-like element remaining" (Ry-
der 1991: 309): the nominal steamer 'edible clam', for instance, profiles a
clam that steams. Nomináis like broiler, baker and frier are analyzed simi-
larly. Ryder (1991: 309) says they all designate "agent-like active patients".
Comparable to the instrumental and clothing types of -er nomináis, "the
food is sufficiently salient and ... sufficiently identifiable by the verb" to
become the referent of an -er nominalization (Ryder 1999b: 289). The sali-
ence of these items of food is attributed to two features in particular: firstly,
the processes which they are related to refer to a "specialized type of prepa-
ration" (Ryder 1999b: 289), which the animals, vegetables and fruit are
bred or selected for; and, secondly, "the food that is cooking, at least in
modern kitchens, is often the most prominent participant actually present in
the event" (Ryder 1999b: 289). Ryder refers here to the fact that the "per-
son in charge of the meal performs few if any overt actions during the
cooking process, and neither does the device in which or by which the food
is cooked" (Ryder 1999b: 289).
Although originally the referents of the food-related nomináis may have
been interpreted as active Patients, Ryder (1991: 309) claims, they were
reanalyzed as truly patientive (e.g., steamer as a clam that steams was re-
analyzed as a clam that is steamed) and thus paved the way for -er forma-
tions "that have few if any agent-like qualities", e.g., scratcher 'a lottery
ticket that is scratched to reveal the potentially winning patterns' (Ryder
1991: 310) and dipper 'something that is dipped before being eaten' (Ryder
1991: 309). That the latter, in spite of their lack of agent-like properties,
can be profiled by an -er nominalization, Ryder (1999b: 289) argues, is
only due to their being sufficiently "identifiable in their event".
Ryder's analysis of -er nominalization has the merit of trying to offer an
integrated analysis of agentive and non-agentive -er nominalization on the
basis of a wealth of data. It includes interesting observations, for instance
on the relation between instrumental -er nomináis and the agentive proto-
type of -er nominalization. Ryder's approach, however, in my opinion, also
has its weak points and in the next section, I will point to some of them. In
Section 1.1.2 I will discuss Panther and Thornburg's (2002) alternative
cognitive proposal for the analysis of -er nominalization and show that it
significantly improves on the analysis proposed by Ryder and contains
several valuable descriptive observations concerning the semantics of -er
formation. In Section 1.1.3, then, I will argue that, as shown in Lemmens
(1998), a detailed analysis of the nature of the base verbs of deverbal -er
nominalizations is necessary to arrive at an accurate description of their
profile.
1.1.1. -er nominalizations as natural symbolic units
Ryder argues that non-agentive -er nominalizations are extensions of the

agentive prototype which have in common with agentives that they desig-
nate entities which are conceived of as being salient and identifiable within
the designated event. I consider Ryder's analysis in terms of salience and
identifiability problematic in two respects. First, Ryder accounts for non-
agentive -er nominalizations in terms of a semantic property that is typical
of agentives and she thus suggests that the semantics of non-agentives can
be reduced to the vague concept of 'agent-like salience'. Secondly, Ryder
tries to shed light on the semantics of -er nominalizations but she ignores
their constructional properties.
-er nominalizations: Agent names or Subject names? 105
First, Ryder creates the - in my opinion false - impression that agentive

and non-agentive deverbal -er nomináis have blended into one schematic
concept of 'agent-like salience'. She repeatedly stresses that salience is
very much an agentive property: because the most salient participant in the
causal chain is typically the Agent of the process, non-agentive entities that
become "the most salient part of their event's causal chain" are said to be
agent-like or "behave like Agents" (Ryder 1999b: 288). Ryder argues that
this is true of "many Instruments and a few Patients" (Ryder 1999b: 288),
such as the cooking-related nomináis. She admits, however, that nomináis
such as scratcher, dipper and sipper designate "fairly prototypical patients
that have few if any agent-like qualities" (Ryder 1991: 309).
The claim that non-agentive and agentive -er nominalizations realize
one and the same meaning of '(agent-like) salience' conflicts with the lan-
guage user's interpretation of non-agentive -er nominalizations: nomináis
such as kneeler 'a chair which you have to kneel on' and squeezer 'a con-
tainer that you have to squeeze to extract something from it' are not inter-
preted as having an agent-like meaning of salience. Rather, they designate
entities that are done to or affected by the process. That non-agentives have
to be interpreted as non-agentive rather than as salient, agent-like entities is
also revealed by -er nominalizations that constitute clear cases of lexical
ambiguity (Taylor 1995). Nominalizations such as reader and baker, for
instance, have an agentive and a patientive reading {reader either profiles
'someone who reads' or 'a compilation of literature'; baker designates
'someone who bakes' or 'a food - meat, fruit or vegetable - that is suitable
for baking'). Rather than realizing one 'merged' meaning of salience, the
agentive and non-agentive readings of these -er nominalizations constitute
irreducible senses in their own right. The linguistic issue that should be
addressed, therefore, is why one and the same suffix can designate salient
entities as divergent as Agents and Patients, while leaving intact their re-
spective agentive and non-agentive readings.
The notions of salience and identifiability cannot shed light on this is-
sue, because they fail to address the constructional properties of deverbal
-er suffixation. Ryder's analysis is purely semantic in design and does not
consider the unique constructional link which -er establishes between an
entity and a process. I view -er nomináis as natural symbolic units or units
of which the meaning is encoded in the lexicogrammar (see Chapter 3,
Section 2). An account of -er nominalizations, in my opinion, therefore has
to begin with a careful description of their constructional properties and it
will have to distinguish deverbal -er nomináis from -er nomináis with non-
verbal bases. Nominalizations with a verbal root and nominalizations with a
non-verbal root have a compositional value and a meaning that are neces-
sarily different. Denominal -er nominalizations, moreover, do not integrate

verbal and nominal categories into one structure, since they do not involve
reclassification (as defined in Chapter 3, Section 3). In terms of Lan-
gacker's network model of language, a highly abstract schema which gen-
eralizes across all instances of -er nominalization (deverbal as well as oth-
ers) may well be possible; but deverbal -er nominalizations instantiate a
lower-level, less abstract constructional schema, which, among other
things, has to do justice to the variety of process types that can be nominal-
ized by -er and the range of participants in these processes that can be pro-
filed.
1.1.2. -er suffixation as a complex conceptual category
Like Ryder, Panther and Thornburg (2002) adhere to a purely semantic

approach to -er nominalization and reject the possibility of a construction-
ally-based description of it. Unlike Ryder, however, they do not posit a
general abstract meaning for all -er nomináis. Instead they argue that the
system of -er suffixation has a central sense to which most other senses of
-er nomináis are "more or less directly linked" (Panther and Thornburg
2002: 193). The central sense is that of a "human Agent who performs an
action or engages in an activity to the degree that doing so defines a pri-
mary occupation" (Panther and Thornburg 2002: 193). Nomináis that des-
ignate non-human entities which are perceived to be human in some respect
(e.g., retriever, warbler, grasshopper, skyscraper) are metaphorically re-
lated to this central sense through personification. Instrumental -er nomi-
náis, Panther and Thornburg suggest, can either be viewed as strong substi-
tutional metonyms for human Agents or (more plausibly) as historical ex-
tensions of the -er prototype which are "metonymically motivated by con-
ceptual contiguity" (Panther and Thornburg 2002: 177). -Er nomináis with
event referents (e.g., thriller, eye-opener, pageturner), finally, are described
as instances of reiflcation.
Apart from the fact that it presents what is probably the most elaborate
account of -er nominalization thus far, Panther and Thornburg's analysis in
my opinion significantly adds to the existing descriptions of -er nominali-
zation in two ways: firstly, it opens up interesting new perspectives for the
analysis of non-deverbal -er formation and its relation to deverbal -er\ sec-
ondly, it contains some valuable observations with respect to the semantics
of instrumental and truly non-agentive -er formations.
First, in their extensive overview of deverbal and non-deverbal -er
nomináis Panther and Thornburg manage to reveal systematic conceptual
resemblances between denominai and deverbal -er nomináis. They argue

that both verb-based and non-verb-based -er nominalizations evoke action
scenarios which include activities, participants and a specific setting. Verb-
based -er formations evoke such action scenarios directly, i.e. through their
verbal base; non-verb-based nomináis evoke them metonymically, by nam-
ing a component (e.g., Patient, Instrument, Location, Time) of the action
scenario, which comes to function as reference point (Panther and Thorn-
burg 2002: 158). The action must then be metonymically accessed from
that reference point. It is especially among prototypically agentive and in-
strumental -er nomináis that interesting parallels turn out to exist between
the nomináis with a verbal base and those with a non-verbal base: nomináis
such as tinner, slater, hatter, philosopher, for instance, can be said to re-
semble deverbal -er formations like baker, programmer and teacher in that
they also designate a human who occupationally performs an action, be it
an action that is metonymically linked to the nouns on which they are based
(i.e. tin, slate, hat, philosophy). Similarly, nomináis such as threewheeler
and doubledecker are said to metonymically realize Instruments because
their base "names a 'design feature' that evokes a scenario in which the
entity with such a design feature is used' (Panther and Thornburg 2002:
170). In short, Panther and Thornburg (2002) show that denominal -er
structures can realize metonymically many of the meanings which deverbal
-er nominalizations evoke directly.
Interestingly, Panther and Thornburg (2002: 173) also point to a number
of semantic similarities between instrumental and truly non-agentive
deverbal -er nomináis such as broiler, cooker, reader, poster. The latter,
they suggest, are "conceptually fairly close to Instruments" because like
Instruments they are "purpose-designed entities" (Panther and Thornburg
2002: 173). They are either "designed for a special purpose" or they have
"inherent properties that make them suitable for certain purposes" (Panther
and Thornburg 2002: 173): a broiler is a chicken that is suitable for broiling
and a squeezer is a bottle designed to be squeezed.
Panther and Thornburg's observations are, I think, basically correct (the
'purpose-designed' status of most patientive -er nomináis has also been
hinted at in Levin and Rappaport 1988 and Lemmens 1998). However, they
cannot explain why a prototypically agentive suffix such as -er has come to
designate instrumental as well as truly non-agentive entities that are de-
signed or suitable for a specific purpose. The so-called "conceptual close-
ness" of instrumental and non-agentive -er nomináis does not explain why
they make use of one and the same, prototypically agentive suffix to desig-
nate two fundamentally different participants, i.e. agent-like and patientive
ones. The basic extension from agentive to non-agentive nomináis that
seems to have taken place within the system of deverbal -er suffixation is,
in other words, left unaccounted for. hi Chapter 7 I will show that it is only
when the compositional value of instrumental and patientive -er nomináis
is analyzed more closely that the reason behind their semantic resem-
blances becomes clear.
Summarizing, Panther and Thornburg show that denominal -er nomináis
can be profitably examined from the perspective of metonymy and that
their semantics resembles that of deverbal -er nomináis to a certain degree.
They also make interesting observations concerning the semantics of in-
strumental and patientive -er nomináis. In my own analysis of -er nomi-
nalization, I will restrict myself to deverbal -er nomináis because their se-
mantics is encoded in the process-participant configuration which they
designate through the constructional link which they establish between a
verb and the -er suffix. I will, more particularly, argue that it is only the
analysis of the constructional properties of the various types of -er suffixa-
tion that can shed a new light on their semantics, and, ultimately, on the
high-level schema which all deverbal -er nomináis instantiate.
1.1.3. The verbal base of deverbal -er nomináis
Among the lexicogrammatical features that must be looked at in more de-

tail to uncover the meaning of deverbal -er nominalization and shed light
on the occurrence of agentive and non-agentive -er nomináis are those of
the base verb. In Lemmens (1998), it is shown that the distinction between
transitive and ergative processes, as described in Davidse (1991, 1992b,
1998b), is also operative on the morphological level of deverbal -er deriva-
tion. What distinguishes a transitive from an ergative process? Characteris-
tic of transitive processes (e.g., chase, hit, kill) is that they are Actor-
centred: their "most central participant" is the Actor, and the "Actor-
Process complex is grammatically more nuclear and relatively more inde-
pendent" (Davidse 1992b: 110). The basic Actor-Process complex can be
extended only to include a Goal, as in The lion is chasing the tourist. Erga-
tive processes such as break, open and roll, in contrast, are "Medium-
centred", with the Medium as "most nuclear participant" (Davidse 1992b:
110) (e.g., The glass broke). The basic Medium-Process constellation can
only be opened up to include an Instigator, as in The cat broke the glass.
While the transitive Goal is a "totally 'inert' Affected", the ergative Me-
dium "co-participates in the process" (Davidse 1992b: 118). In ergative
one-participant constructions such as The glass broke, this active co-
participation of the Medium in the process is foregrounded and the Medium

is presented as "semi-" or "quasi-autonomous" (Davidse 1998b).32
The differences that characterize the transitive and ergative paradigms
in the clause, Lemmens (1998) argues, are also at work in deverbal -er for-
mation. Lemmens (1998: 127) observes, firstly, that corpus material shows
-er nomináis to be primarily based on transitive processes, which is intui-
tively plausible given the prototypically agentive profile of -er nominaliza-
tion and the Agent-centredness of the transitive paradigm. Nomináis that
profile the Goal of a transitive process are far less common, but nonetheless
exist (e.g., dipper, sipper, scratcher). Importantly, they cannot be attributed
agent-like properties, but designate inert, affected entities (Lemmens 1998:
137).33
Lemmens further holds that the majority of -er nomináis derived from
ergative processes designates the Instigator of the process (e.g., opener,
freezer). The claim that -er nomináis are typically Agent-profiling then
schematizes over nomináis that designate the Actor of a transitive process
and nomináis that profile the Instigator of an ergative one (Lemmens 1998:
131). Against Keyser and Roeper (1984: 395), who regard Medium-
profiling -er nomináis as "idiosyncratic", Lemmens moreover argues that
-er formations designating the Medium of an ergative process can be ac-
counted for as extensions of the agentive prototype. Medium-profiling -er
nomináis are motivated by the Medium's potential for independence, which
is tied up with its co-participating role in the process. Lemmens (1998: 136)
states that the "higher the independence, the more likely it becomes that an
-er formation may profile this participant rather than the Instigator". A
nominal like cracker, for instance, designates the Medium of the ergative
process crack and foregrounds the fact that the particular type of wafer
which it refers to makes a cracking noise when you eat it (Lemmens 1998:
135).
Lemmens manages to refine Ryder's account of -er nominalization in
terms of extension of the agentive prototype by applying the transi-
tive/ergative distinction to it. Especially the non-agentive type of -er nomi-
nalization can now be described more accurately, viz. as involving a Goal-
profiling and a Medium-profiling type, with the former constituting the
most radical departure from the agentive prototype. Medium-profiling -er
nomináis can be said to profile a co-participating participant in the process:
while an Instigator may still be part of the semantic base, they present the
Medium as a semi-autonomous or independent entity. Illustrative of Me-
dium-profiling -er nomináis are the food-related -er nomináis (e.g.,
steamer, broiler, baker) which Ryder identifies as "agent-like active pa-
tients" (Ryder 1991: 309). The fact that they are more agent-like than, for
instance, dipper, sipper and scratcher can now be argued to be due to the
ergative nature of the process which they nominalize and, related to it, to
the Medium which they profile.34 More in general, Lemmens's account of
how the transitive and ergative paradigms operate in -er derivation points to
the importance of considering the contribution of the verb from which -er
nomináis are derived: the compositional value of a deverbal -er nominaliza-
tion largely depends on its verbal base.
To conclude, then, both Ryder's analysis and Lemmens's refinements of
it are indicative of the rich representational or ideational semantics of
deverbal -er nominalization. The analysis of the representational functions
which are realized in a nominalization system forms an essential part of a
multifunctional approach to nominalization (as discussed in Chapter 3,
Section 4). Representational categories are important for the description of
deverbal -er nominalizations in that they reveal the variety that exists
among them and allow us to identify the category's prototype, i.e. the
Agent-profiling type of -er nominal. However, representational categories
seem unable to generalize across all instances of deverbal -er suffixation.
The category of Agent, for instance, fails to account for the non-agentive
instances of -er nominalization. In representational terms, moreover, the
non-agentive type of -er nominalization is necessarily viewed as a periph-
eral, non-prototypical instance of -er nominalization. A schematic view on
-er nominalization which, I argued in Chapter 3 (Section 1.1), necessarily
complements an analysis in terms of prototype cannot therefore be based on
the representational categories of agency and process type. In the following
section, I will examine an attempt at schematization suggested in Levin and
Rappaport (1988) and Rappaport Hovav and Levin (1992).
1.2. Non-agenti ves as external arguments
Levin and Rappaport (1988) and Rappaport Hovav and Levin (1992) claim
that the derivation of -er nomináis "refers not to particular semantic-role
labels but rather to argument-structure configurations" (Levin and Rappa-
port 1988: 1067). -Er nomináis may bear a wide range of semantic relations
to the verbs from which they are derived, but "they most systematically
correspond to the external argument of the base verbs" (Rappaport Hovav
and Levin 1992: 145) and therefore only derive from verbs which have
external arguments (Levin and Rappaport 1988: 1070). What does the term
'external argument' stand for?
The notion of 'external argument' was introduced in Williams (1980,
1981) to refer to a particular, "distinguished argument, not a syntactic posi-
-er nominalizations: Agent names or Subject names ? I l l
tion, a case, or something else" (Williams 1981: 83). The external argument
is the argument "that corresponds to the NP in a sentence of which a [verb]
phrase with that item as its head is predicated" (Williams 1981: 84). The
example which Williams gives to illustrate the notion is that of the verb hit.
Hit has two arguments, viz. one which serves the semantic role of Actor
and another one which is Theme. The Actor is the external argument be-
cause in a clause it "must be specified in a position external to the verb
phrase of which hit is the head" (Williams 1981: 84). The Theme, on the
other hand, "must be specified within the verb phrase" (Williams 1981: 84).
This seems to correspond to what in Levin and Rappaport (1988: 1074) is
understood by 'external argument': the external argument of a verb is de-
fined there as its "underlying (D-structure) subject", while an internal ar-
gument is an "underlying (D-structure) direct object".
Passives, middles and ergative one-participant structures illustrate that
external arguments do not necessarily correspond to the so-called "surface
structure" Subject. In passive clauses, for instance, the arguments of the
verb are reorganized and "the internal argument of the verb occupies the
canonical subject position" (Haegeman and Guéron 1999: 123). In the pas-
sive clause Louise was invited (by Thelma), the internal argument Louise
thus occupies the Subject position, while the external argument Thelma
"may appear optionally in a PP adjunct" (Haegeman and Guéron 1999:
123). Ergative, one-participant structures such as The ball rolls (called 'un-
accusative' in Perlmutter 1978) have only one argument, which functions
as Subject and is not an external, but an internal argument (Levin and Rap-
paport 1988: 1074). In the corresponding two-participant structures, the
internal argument functions as Object (e.g., I roll the ball). Middle clauses
such as This book reads well, finally, constitute yet another type of clause
in which the Subject is the internal rather than the external argument of the
verb (Williams 1981; Keyser and Roeper 1984).
In their analysis of -er nominalizations, Levin and Rappaport (1992:
145) discuss four groups of nomináis which, they claim, corroborate their
"external argument generalization", viz. instrumental -er nomináis, nomi-
náis derived from the spray/load-type of verbs, nomináis based on intransi-
tive verbs, and, finally, nomináis which are related to the middle form of
the verb.
As far as instrumental -er nomináis are concerned, Levin and Rappaport
claim that they must designate "intermediary" instruments or instruments
which are able "to perform the action in some sense autonomously" (Levin
and Rappaport 1988: 1071). The verb open, for instance, can take -er suf-
fixation to refer to an intermediary instrument involved in the opening of
something, as in
(7) a. Doug opened the can with the new gadget,

b. The new gadget opened the can. (opener)
Verbs that take what have been referred to as "facilitating" or "enabling"

instruments (Marantz 1984 and Wojcik 1976, as cited in Levin and Rappa-
port 1988: 1071), in contrast, do not allow instrumental -er nomináis:
(8) a. Bill ate the meat with a fork.

b. *The fork ate the meat. (*eater)
Interestingly, it is also intermediary instruments which in the clause can

function as Subject, as illustrated in The new gadget opened the can and
The key unlocked the door (Levin and Rappaport 1988: 1071). Instrumental
Subjects, Levin and Rappaport claim, "are only found with verbs that have
external arguments independently" (Levin and Rappaport 1988: 1071), i.e.
with verbs that always take an external argument: the instrumentais that
figure as Subjects of these verbs can be said to "satisfy the lexical require-
ments of certain predicators for external arguments" (Levin and Rappaport
1988: 1071). Because instrumental -er nominalizations only profile inter-
mediary instruments or instruments which in the clause can appear as Sub-
jects and designate external arguments, Levin and Rappaport conclude that
they can be analyzed as designating external arguments (Levin and Rappa-
port 1988: 1072).
A second group of nomináis which is claimed to provide support for the
external argument generalization consists of nomináis derived from
spray/load verbs (as illustrated in (9) and (10), based on Levin and Rappa-
port 1988: 1072):
(9) a. Bill loaded cartons onto the truck,

b. Bill loaded the truck with cartons.
(10) a. Jack squirted water on the plants,
b. Jack squirted the plants with water.
Especially the behaviour of the Locatum-argument of spray/load verbs (i.e.

the argument which is either realized as a bare nominal such as cartons and
water, or as a vw/A-phrase, as in with cartons and with water), is interesting
according to Levin and Rappaport: being an internal argument, the Loca-
tum can never appear as Subject of the clause or be profiled by an -er
nominalization (Levin and Rappaport 1988: 1073):
(11) ""Cartons loaded the truck, ("loader)

Nomináis such as loader, sprayer, spreader and sprinkler do not designate

the Locatum-argument of load, spray, spread or sprinkle, but rather profile
an (intermediary) instrument, which in the clause is realized either as Sub-
ject or in the form of a vw'fA-phrase (Levin and Rappaport 1988: 1073):
(12) a. The crane loaded the truck.

b. Bill loaded the truck with cartons with a crane.
With respect to -er nomináis derived from intransitive verbs, Levin and
Rappaport claim that they "should be based on a unergative rather than an
unaccusative verb, since only the single argument of a unergative verb is an
external argument" (Levin and Rappaport 1988: 1074). The distinction
between unergative and unaccusative intransitives goes back to Perlmutter
(1978) and Burzio (1986, as cited in Levin and Rappaport 1988: 1074).
Unergatives largely coincide with what in Davidse (1991, 1992b) are called
'intransitives' (e.g., run, climb, work and sin). The unaccusative category
includes all the ergative verbs of Davidse's model in their 'non-caused'
sense, as in the door opened, the ball rolled, the water froze, as well as
processes such as disappear, happen and die. The single argument of unac-
cusatives "is not an external argument but rather a direct internal argument,
that is, an underlying (D-structure) direct object" (Levin and Rappaport
1988: 1074). It cannot therefore undergo -er nominalization:
(13) *disappearer, *happener, *dier, *exister
-Er nominalizations derived from verbs with both an unaccusative and a

transitive use can only profile the external argument of the transitive use of
the verb (Levin and Rappaport 1988: 1075), i.e. they can profile the Instiga-
tor but not the Medium of the ergative process:
(14) opener, dryer, freezer, heater
Yet another group of -er nominalizations which is claimed to corrobo-

rate the external argument generalization mainly consists of nomináis de-
rived from cooking verbs. They refer to the "theme argument of the base
verbs (the entity that changes state)" (Levin and Rappaport 1988: 1077).
Examples are baker, broiler, fryer, roaster and steamer (Levin and Rappa-
port 1988: 1077). -Er nomináis which are not based on verbs of cooking
but seem to behave comparably are bestseller, loaner and sticker (Rappa-
port Hovav and Levin 1992: 149). They "present a problem for an account
in terms of the notion of external argument" (Levin and Rappaport 1988:
1077), because they are generally assumed to profile an internal rather than
an external argument. Levin and Rappaport (1988: 1078) point out, how-
ever, that these nomináis "do not receive the interpretation that would be
expected if they were derived directly from the transitive or unaccusative
uses of the related verb". Rather, their meaning seems closer to the inter-
pretation that the related verb receives in the middle structure (Levin and
Rappaport 1988: 1078). A nominal like broiler, for instance, designates a
chicken which is bred or intended to be broiled, which comes closer to the
interpretation of a middle construction such as This chicken broils well than
to This chicken broils. Apart from semantic correspondences, middles and
nominalizations in -er also show the same constraints with respect to the
verbs which they can be based on: both of them, it is argued, can only be
derived from verbs with 'affected objects'. A structure like *this planet
sees well and the corresponding -er nominal *seer are, for that reason, im-
possible (Levin and Rappaport 1988: 1079).
Does the relationship with middle formation imply that cooking -er
nomináis, like all -er nominalizations, profile the external and not the inter-
nal argument of the base verb? Levin and Rappaport (1988: 1078) admit
that most accounts of the middle construction consider the single argument
of middles as internal (e.g., Keyser and Roeper 1984; Roberts 1987), but
they argue that the occurrence of -er nomináis like broiler, baker and
steamer might be taken as support for an "alternative analysis of the middle
construction" (Levin and Rappaport 1988: 1078). The analysis which they
propose involves "the externalization of the direct internal argument of the
verb in the lexicon" (Rappaport Hovav and Levin 1992: 149). An -er nomi-
nalization which is related to middle formation therefore profiles an exter-
nal rather than an internal argument.
Levin and Rappaport's analysis is meant to show that -er nominaliza-
tions cannot be generalized across in terms of semantic roles but only in
terms of the notion of 'external argument'. The latter, however, proves to
be an ill-defined notion: what it seems to stand for is 'being the most likely
candidate for serving as Subject', and this, by and large, corresponds to
being the Agent of the process or being an entity that can be conceived as
agent-like. They thus fail to distinguish the representational function of
Agent from the 'interpersonal' or instantiating/grounding function of Sub-
ject (see Chapter 4, Sections 2.2 and 2.3). Levin and Rappaport also bring
in the Subject of middle constructions, but without investigating the true
extent of this category in English. This leaves a number of -er types unac-
counted for. In what follows, I will briefly run through those types of -er
nominalization which Levin and Rappaport fail to elucidate. The main ones
are: -er nomináis designating what I will call 'oblique participants' (Laffut
2000), -er nomináis designating the single participant of an ergative unac-

cusative verb, and, finally, the middle-related type of -er nominal (as a
systematic category). (Note that many of these are not accounted for by
Ryder 1991, 1999b, Lemmens 1998 and Panther and Thornburg 2002 ei-
ther.)
1.2.1. -er nomináis profiling oblique participants
Following Laffut (2000: 139), I will use the term "oblique participant" to
refer to participants which are clausally realized in the form of a preposi-
tional phrase (as in 15). Laffut (2000: 137-156) lists a number of semantic
and formal arguments which show that oblique participants differ from
circumstantials as in (16) in that they are basically "complements" and
elaborate a salient e-site in the verb with which they are used (Langacker
1987: 305) (examples based on Laffut 2000):
(15) a. The pope was killed by a sniper.

b. Spray Static Guard on your comb or brush.
(16) a. The pope was killed on a Sunday.
b. Floyd cut the salami in the kitchen.
The existence of -er nomináis that profile oblique participants counters part
of Levin and Rappaport's analysis of instrumental -er nomináis, of nomi-
náis based on spray/load, verbs and of nomináis derived from unaccusative
verbs. Nomináis like stroller, walker and viewer fail to profile the agent-
like, external argument of the processes of strolling, walking and viewing
(this is also pointed out by Ryder 1999b: 272-273): instrumentais such as
stroller and viewer cannot be analyzed as doing the strolling and viewing
themselves, i.e. a stroller does not stroll and a viewer does not itself view.
Instead, they seem to profile oblique participants.
Apart from oblique participants which indicate the means whereby a
process can be carried out, -er nomináis can also designate the place of a
process, as in:
(17) I work on the stepper, I do squats, bench presses, abdominals, concen-

trated curls - it's all short, sharp shocks and it's very effective. (CB)
(18) Various crashing noises erupted from our pew as our new friend
searched for his lost gold earring beneath the wooden kneelers. (CB)
(19) Our welcoming ensemble features leather wherever your body

touches. Includes a sleeper with innerspring mattress and TV-tilt
headrest. (CB)
These nomináis illustrate that -er nomináis derived from intransitive verbs
(in Levin and Rappaport's terminology, unergative verbs) do not necessar-
ily designate the external argument of the base verb: rather than profiling
the entity which does the stepping/kneeling/sleeping, they designate the
surface that you step on, the chair that you kneel on and the bed that you
sleep in.
Finally, Levin and Rappaport only focus on the Locatum argument of
spray/load verbs (e.g., Bill loaded the truck with cartons, Bill loaded car-
tons onto the truck) and they ignore the Location of the process, as if that
category is not relevant to -er suffixation (e.g., Bill loaded the truck with
cartons, Bill loaded cartons onto the truck). Yet, I have found several -er
nomináis which designate the Location of the verb load:
(20) [about a boat:] She came up the Thames to Chelsea Harbour, and was
craned out on to a low-loader. (CB)
(21) The most popular type of washing machine is the front-loader but
there are still alternatives. (CB)
(22) Top-loaders are useful for a kitchen short on space as they can be
placed at the end of a run of units. (CB)
The Location cannot be analyzed as the external argument of load. The fact
that it can be profiled by an -er nominalization thus forms yet another
counterargument to the external argument generalization which Levin and
Rappaport propose for -er nominalization.35
1.2.2. -er nomináis based on unaccusatives
Levin and Rappaport are right in pointing out that (unaccusative) verbs of
existence and disappearance do not normally take -er suffixation (e.g.,
*exister, *happener, *disappearer). Yet, in some cases they do, witness the
following (attested) examples of -er nomináis with a one-participant unac-
cusative as basis:
(23) When I get busy, I tend to forget to water my plants, so it's a good
thing I've got a plant that's a wilter. The minute it gets a little low on
water, it wilts, so it's a reminder it's time to water everything again.

(Ryder 1999b: 274)
(24) A few thought they had noticed someone resembling the man in the
picture. I wasted two days tracking one of the supposed resemblers
and found no resemblance at all. (Kastovsky 1986b: 410)
An ad hoc formation with disappear and die likewise cannot be excluded:

e.g., He is a good disappearer, You have peaceful diers and troubled diers.
What's more, -er nomináis which nominalize the one-participant or 'unac-
cusative' use of ergative verbs are also found: there is not only the example
of cracker (Lemmens 1998: 135); also the food-related nomináis broiler,
baker, steamer and cooker designate the Medium of the ergative verb
which they are derived from. They cannot therefore be analyzed as external
arguments.
1.2.3. Middle-based -er nomináis
Finally, Levin and Rappaport's attempt to analyze the so-called 'middle-

based' type of -er nominal as profiling the external argument of the base
verb is circular. As pointed out by Ryder (1999b: 275), Levin and Rappa-
port "hypothesize that -er forms must always appear as external arguments
of their root verbs and then ... decide that the subjects of middle verbs are
external arguments primarily based on the possibility of turning the verbs
into bases for -er nomináis". The relation which, they suggest, exists be-
tween certain non-agentive -er nomináis and middle constructions in itself,
however, is intriguing. The middle construction captures the semantics of
certain non-agentive -er nomináis better than any other clausal structure. A
nominal like broiler, for instance, does seem to be more closely related to
the middle construction This chicken broils well than to the non-middle
one-participant structure This chicken broils (Levin and Rappaport 1988:
1078). The relation between the nominal bestseller and the middle con-
struction This item sells best is perhaps even more obvious. Yet another
nominal with a meaning that comes close to that of the corresponding mid-
dle construction is spreader 'butter which is particularly spreadable' (Ryder
1999b: 273): spreader designates a type of butter that spreads well. Even
though, when looked at in terms of its relation to the middle construction,
spreader seems to connect with other (non-agentive) -er nomináis, it can-
not, however, be accounted for by Levin and Rappaport's external argu-
ment analysis: it profiles the Locatum-argument of the verb spread (you
spread butter on your sandwich with a knife), which is an internal rather
than an external argument (Ryder 1999b: 273). That it nonetheless exists

forms yet another indication that the external argument generalization is
untenable and that Levin and Rappaport's analysis of -er nominalization in
terms of the argument structure of the base verb cannot generalize across
the agentive and non-agentive cases of -er nominalization. As with Ryder's
analysis in terms of semantic roles, it is especially with regard to the cate-
gory of non-agentive -er nomináis that many questions remain unanswered.
The external argument generalization may not hold, but the observation
that some non-agentive -er nomináis are close in meaning to middle forma-
tion opens up a highly interesting, new perspective on non-agentive -er
nominalization. As I pointed out in Chapter 3 (Section 5), systematic rela-
tions or relationships of agnation between nominalizations and clauses shed
light on the clausal categories which are realized in nominalizations. With-
out doubt, the most characteristic property of middle constructions is that
they are centred on the relation between a non-agentive Subject and an
active finite verb (e.g., This shirt irons easily, That floor cleans well).
Agentive participants are, in any case, prototypical Subjects. If it can be
shown, then, that, together with agentive -er nominalizations, all the non-
agentive cases of -er suffixation can be related to clausal Subjects, this
would lead to a generalization that covers the distributional properties of
-er nominalization much better than the generalization in terms of semantic
roles which Ryder proposes, the generalization in terms of conceptual con-
tiguity discussed in Panther and Thornburg (2002) or that in terms of argu-
ment structure configurations, suggested by Levin and Rappaport. In the
following section, I will look at how the function of Subject has figured in
descriptive accounts of -er nominalization up till now.
1.3. -er nominalizations as Subject names
The label of 'Subject' has thus far been assigned to -er nominalizations
only sporadically, and then particularly as an empty label representing a
purely syntactic category. The term 'Subject' is in fact often used in one
and the same breath with the semantic role that is typically mapped onto it,
viz. that of Agent. In the same way, patientive -er nomináis are then classi-
fied as belonging to the 'Object' type of -er nomináis.
Marchand's (1969) account o f - e r nominalization nicely illustrates this
specific use of the concept of Subject in the description of -er suffixation.
Marchand (1969) distinguishes between a Subject type of deverbal -er
nominalization, an Adverbial Complement type and an Object type. In the
Subject group, he situates all those nomináis that are agentive, ranging from
human Agents (e.g., preacher, hunter) over non-human, animate Agents

(e.g., retriever, warbler) and material Agents or Instruments (e.g., blotter,
lighter, toaster) to immaterial Agents (e.g., eye-opener, reminder). Nomi-
náis profiling what he calls an 'Adverbial Complement' are slipper 'you
slip into the shoe' and waders 'waterproof boots'. The Object type of -er
nominal, finally, is made up by the patientive cases of -er nominalization,
exemplified by a nominal such as tier 'pinafore' from you tie it (Marchand
1969: 275).
A comparable analysis, but then of Dutch non-agentive -er nomináis is
found in Van Langendonck (1988). Even though he recognizes the general-
izing power of assuming a Subject profile for both agentive and non-
agentive -er nomináis (Van Langendonck 1988: 326), Van Langendonck
prefers to distinguish between what he calls "Subject", "Object" and
"prepositional Object" nomináis. The Object type consists of those -er
nomináis which are patientive in profile; the prepositional Object group
contains circumstantials which are realized by means of a prepositional
phrase (e.g., loper [runner], i.e. 'something that you walk on', based on
Boon 1972: 94). The Subject group, finally, consists of the prototypical,
agentive instances of -er nominalization, as well as of instrumental nomi-
náis and other, inanimate but agent-like cases of -er suffixation. Interest-
ingly, Van Langendonck (1988: 325-326) gives two reasons for not adher-
ing to a generalization in terms of the Subject category: first, he does not
see any evidence in favour of passive structures (with a patientive Subject)
being able to underly nominalization processes. Secondly, the occurrence
of prepositional Object profiling nomináis is considered problematic for a
generalization in terms of Subject, since prepositional phrases cannot be
construed as clausal Subjects. To include the latter type of -er nominaliza-
tion in the analysis of the overall system, Van Langendonck thus proposes
an analysis which distinguishes between Subjects/Agents, Objects/Patients
and prepositional Objects.
In short, descriptive accounts of deverbal -er suffixation in English and
Dutch typically use the Subject-label as if it were identical with that of
Agent (see Chapter 2, Section 4). The notion of 'Subject' is not considered
to generalize across the agentive and the non-agentive instances of -er
nominalization. In fact, arguments are provided against using it to cover
agentive as well as non-agentive -er nominalizations. It is, for instance,
pointed out that (Dutch) -er nomináis that designate oblique participants do
not seem relatable to a clause in which they function as Subject. Patients,
then again, can be realized as clausal Subjects, viz. of middle and of pas-
sive clauses, but not all patientive -er nomináis appear to nominalize proc-
esses that can figure in a clausal middle construction (as pointed out by
Booij 1986), and evidence that morphological processes can be passive in

nature is argued to be sparse.36
So, if the concept of Subject is going to be used to generalize across all
deverbal -er nominalizations, and if the middle construction is going to be
linked up with non-agentive -er nominalization, it will have to be shown
that these three arguments are unjustified: -er nomináis that profile oblique
participants will have to be provided for; non-agentive -er nominalizations
in general have to be shown to be more closely related to middle formation
than to passive clauses, and, finally, non-agentive nomináis derived from
verbs that do not normally occur in clausal middle constructions have to be
integrated in the analysis. The Subject function, moreover, has to be attrib-
uted a semantic function in its own right, which can, but is not always,
mapped onto the representational function of Agent.
2. Lexicalization versus ad hoc nominalization
I have thus far concentrated on the occurrence of agentive as well as non-

agentive -er nomináis. Cutting across the network of deverbal -er nominali-
zation is, however, yet another distinction that has to be accounted for, viz.
that between lexicalized and ad hoc -er nominalizations:
(25) a. saver of lives, as in he is the saver of many lives

b. a lifesaver 'lifeguard' (Levin and Rappaport 1988: 1070)
These -er nominalizations are characterized by different degrees of en-

trenchment and serve distinct functions in discourse. Both these differences
have been described in various ways. A lexicalized nominalization such as
lifesaver has been argued to serve a primarily labelling or naming function
(Kastovsky 1986a, 1986b). Ad hoc nominalizations have been termed "syn-
tactic recategorizations" (Mötsch 1979, cited in Kastovsky 1986b: 411),
"entirely predictable nominalizations" (Quirk et al. 1985: 1547) or "deicti-
cally used" structures (Ryder 1999b: 282). They are said to serve the tex-
tual function of condensing information and of "referring back to some-
thing contained in the previously occurring text" (Kastovsky 1986b: 411;
see also Mackenzie 1990: 141-143). I prefer to use the term 'ad hoc nomi-
nalization' because it captures the fact that these nomináis are not suffi-
ciently entrenched to be lexicalized and therefore have to be derived 'on the
spot', on the basis of the constructional schema of -er derivation. Their
being used for anaphoric reference or reference to something in the previ-
ous text does not distinguish them from lexicalized-er nominalizations,
Lexicalization versus ad hoc nominalization 121
which can likewise be used anaphorically once they come to function in the
structure of the nominal. Still, it cannot be denied that ad hoc formations
are frequently used to rephrase a previous clause by nominalizing it, as
happens in the following examples:
(26) One guy jumped right into the fight, but his friend immediately van-
ished. Hie police came and hauled off the fighter, after which the van-
isher promptly reappeared, laughing, (cited in Ryder 1999b: 283)
(27) "Not him," says Mrs. Cook. "This one, though. A bit of a poseur, but
he moves very nicely." I will not describe the nice mover, to spare his
blushes, but let us say that he had dressed for the part. (CB)
Interestingly, lexicalized -er nomináis and their ad hoc counterparts are

characterized by significant lexicogrammatical differences. As Mackenzie
(1990: 133) points out, lexicalizations by and large function as ordinary
nouns "in that their syntactic behaviour is indistinguishable from compara-
ble words that cannot be related to a verb (author, butcher, etc.)". Ad hoc
-er nominalizations, in contrast, are only "partially nominalized"
(Mackenzie 1990: 136-137). Strang (1968, 1969) calls them "minimal"
nominalizations. Unlike lexicalized -er nomináis, they cannot shift the
characteristic accent position of the verb (as illustrated in 28a). They take
adjectives that resemble adverbiale in that they "do not ascribe a quality to
the referent but rather indicate the manner, frequency or extent of the activ-
ity associated with the referent" (Mackenzie 1990: 137), as in (29):
(28) a. He is a PHOtographer of snow-scenes.

b. He is a professional phoTOgrapher. (Mackenzie 1990: 136)
(29) a. He is a hard worker.
b. He works hard. (Mackenzie 1990: 137)
Ad hoc nominalizations are said to "inherit argument structure" (Levin

and Rappaport 1988: 1069): they can be postmodified by an o/-phrase real-
izing what would clausally have been a participant of the process (e.g.,
saver of lives; destroyer of Del Harris's hopes). And, finally, they seem to
be necessarily agentive in profile (Rappaport Hovav and Levin 1992: 132).
Non-agentive ad hoc -er nominalizations do not appear to exist. Ad hoc
nominalizations do not even allow for an instrumental profile: a nominal
like, wiper, for instance, may receive an instrumental or a human Agent
reading, but "a wiper of windshields may receive only an agentive reading"
(Levin and Rappaport 1988: 1069).
The semantic differences that are linked to these lexicogrammatical

properties have been hinted at in Levin and Rappaport (1988) and Rappa-
port Hovav and Levin (1992). Ad hoc nominalizations are described as
event nomináis because, unlike their lexicalized counterparts, they seem to
designate processes that have been carried out. As Levin and Rappaport put
it, "Someone may be called a lifesaver even if he or she has never saved
anyone, but someone may not be called a saver of lives unless he or she has
actually been involved in saving lives" (Levin and Rappaport 1988: 1070).
While ad hoc nominalizations designate events that have occurred, lexicali-
zations are non-event nomináis or nomináis that do not presuppose the ac-
tual occurrence of the process.
However, as the nominalizations in (30) and (31) (i.e. breather and
sleeper) make clear, Levin and Rappaport's characterization of ad hoc
nominalizations as designating actual events is not accurate: not all of them
designate events that have actually occurred:
(30) Recent research has established that merely breathing smoke-laden air
- passive or 'second-hand' smoking - puts the breather at risk. (CB)
(31) Trouble sleeping is the most common complaint, including an inabil-
ity to get to sleep, terrible, frightening dreams that wake the sleeper in
a state of anxiety, and waking early feeling unrefreshed but unable to
get back to sleep. (CB)
The nomináis breather and sleeper refer to the one who breathes and the
one who is sleeping respectively, but they do not designate specific, actual
occurrences of the processes of breathing or sleeping. As to the "ability to
inherit complement structure" (Rappaport Hovav and Levin 1992: 132),
then, this does not seem to be possible in ad hoc nominalizations only: the
nominalizations in (32) and (33) are both lexicalizations and they are fol-
lowed by o^phrases which periphrastically realize an Object which speci-
fies the range or scope of the process (in 32) or the Beneficiary of the
teaching (in 33):
(32) Mrs. Vogt was a teacher of German and French in a secondary

school.... (CB)
(33) Mrs. Emerson, the second-grade teacher of Tony, the youngest boy
....(CB)
Unlike what is suggested in Levin and Rappaport (1988) and Mackenzie

(1990), the inclusion of an o/-phrase is thus not a distinctive feature of ad
hoc nominalizations.
Towards a coherent account of deverbal -er nominalization 123
Levin and Rappaport's distinction between event and nonevent nomi-

nalizations may be somewhat inaccurately formulated, but it does reveal
that -er nominalizations not only differ in terms of the entity which they
designate, but, that, depending on whether they are lexicalized or not, they
differ in lexicogrammatical properties and realize different meanings. An
account of the system of deverbal -er suffixation has to offer a satisfactory
explanation for these differences and integrate them in the overall sche-
matic network of -er nominalization.
3. Towards a coherent account of deverbal -er nominalizations
The system of deverbal -er nominalization may have started out as an agen-
tive system, it can no longer be characterized as such: non-agentive -er
nominalizations are numerous and too productive to be described as non-
systematic. In the next two chapters, I will defend the claim that deverbal
-er suffixation has evolved from profiling Agents to profiling Subjects. I
will argue that an analysis of -er nominalization in terms of the Subject
function provides a more accurate generalization for the present distribu-
tional properties of -er nominalization than one in terms of agency. In par-
ticular, I will show that the various types of non-agentive -er nominaliza-
tions are systematically equivalent to the Subject of middle constructions
and of so-called "Setting-Subject" constructions (Langacker 1991: 346;
e.g., This tank does not squirt water.). I will moreover argue that deverbal
-er suffixation construes equivalents to the basic grounding or deictic func-
tion of the clausal finite verb. The correspondences between the clausal
Subject-Finite unit and -er nominalization elucidate the semantics of the -er
suffix significantly.
The evolution in the system of deverbal -er nominalization may run par-
allel to certain changes of predominance in the various functional layers of
organization in the clause. Within the Prague School (which has always
integrated the comparative and the diachronic in its functional approach), it
has been hypothesized that the English clause has developed from an
Agent-oriented to a predominantly Subject-oriented form of organization.
Or, as Vachek (1966: 92) puts it, "the main function of the grammatical
subject in an English sentence is no longer to express an agent". Unproto-
typical or non-agentive Subjects as in middle constructions are on the rise.
Significantly, it is this clause type, which combines a non-agentive Subject
with an active finite verb, which turns out to be particularly important for
the analysis of non-agentive -er nominalization. It will therefore constitute
the focus of the next chapter. In Chapter 7, then, I will present my own
analysis of -er nominalization, along the lines of the theoretical-descriptive

framework set out in Part I.
Chapter 6
The middle construction
This chapter37 focuses on the clausal middle construction, as exemplified in

the following sentences:38
(1) More than 50 years later, the book still sells well. (CB)
(2) The new Holden Berlina handles like a junior sports sedan.... (CB)
(3) And it's latex paint, so it cleans up quickly and easily with soap and
water. (CB)
(4) If they wanted to do Eliot, why didn't they do The Magi? It's narrati-
ve, it reads well.... (CB)
Middles typically combine an active verb form with a Subject that is pa-
tientive in nature and thus reminiscent of passive constructions.39 Semanti-
cally speaking, the middle is often associated with a facilitative meaning
(Fawcett 1980; Kemmer 1993): it is said to express how easy or difficult it
is to realize a particular process with respect to the entity construed as Sub-
ject. Prototypical middle constructions are construed with the adverb easily,
as in
(5) When no longer required, the discs remove easily. (CB)

(6) Compact lipstick size design, slips easily into your handbag or pocket.
(CB)
Rather than account for the unique constructional link which middles
realize between a non-agentive Subject and an active verb form, most
analyses of middle formation describe middle constructions by reference to
other constructional paradigms and, as a result, leave many questions unan-
swered. In this chapter, I will start with a brief sketch of the two main exist-
ing approaches to middle formation (Section 1). I will show that they tend
to reduce the representational potential of middle formation and focus on
only part of the middle system. Among the middle constructions that have
been systematically ignored in the literature are, for instance, those based
on intransitive processes, with an oblique participant as Subject. They will
be the main topic of Section 2.1 will then zoom in on two aspects of middle
formation which, in my opinion, are crucial to identify the high-level
schema which middle constructions instantiate: I will first discuss the vari-
126 The middle construction
ous semantic profiles of middle formation, and argue that some middles,
rather than profile the facility with which a process can be carried out, fo-
cus on the destiny of the Subject or on the result of carrying out a particular
process on it (Section 3). In a final section, then, I will elaborate on the
essentially modal nature of the relationship which middles establish be-
tween a Subject and a Finite (Section 4).
1. Middle formation: A state of the art
In the literature on middle formation, some have stressed the active, one-
participant nature of the construction and have claimed agentive status for
its Subject, while others have ascribed passive value to the construction,
emphasizing that its Subject is affected and that an Agent is implied. I will
refer to the former as the ergative and the latter as the passive approach to
middles. Those who situate middle constructions close to ergative one-
participant structures are, among others, Hale and Keyser (1987), Sinclair
et al. (1990), Francis et al. (1996) and Sohn (1998). They argue that mid-
dles resemble ergative one-participant structures in that, firstly, they allow
for the addition of an "event of causation" (Hale and Keyser 1987: 6),
which is generally assumed to be characteristic of ergative one-participant
structures such as the one in (7):
(7) The boat sank. o A torpedo sank the boat.

(8) This pan cleans easily. <=> You clean this pan.
Secondly, like ergative one-participant construals, middles are said to focus

on the participant that is central to the event depicted by the clause and in
which a change of some sort manifests itself (Hale and Keyser 1987: 6). By
being realized in a one-participant structure, this change is presented as
being effected "autonomously" or "without the participation of an external
agent" (Hale and Keyser 1987: 7).
The proponents of the ergative approach attribute varying degrees of
agency to the middle Subject, which is described as "agent-like" in that it
"corresponds to a participant which is, in an intuitively clear sense, respon-
sible for the process depicted in the predicate" (Hale and Keyser 1987: 9;
see also Lakoff 1977). The notion oí responsibility is discussed at length in
Van Oosten (1977, 1986), who even characterizes the middle Subject as
"energy source of the action" (Van Oosten 1986: 85) and as being able to
"bring about the action of the predicate independently" (Van Oosten 1986:
93). The Subject's responsibility is said to follow from its properties,
Middle formation: A state of the art 127
which, Van Oosten claims, represent the linguistic raison d'être of middle
constructions: "the purpose of the patient-subject construction is precisely
to enable the speaker to assert that ... properties of the patient bear the re-
sponsibility for the occurrence of the action of the verb" (Van Oosten 1986:
93). In Lakoffs view as well, "the point of using the patient-subject con-
struction is to say that properties of the patient are more responsible for
what happens than the agent is" (Lakoff 1977: 248).
An alternative analysis of middle formation is found in, among others,
Sweet (1891), Jespersen (1914-1929, 3), Halliday (1967), Smith (1978),
Keyser and Roeper (1984), Fellbaum (1986) and Fellbaum and Zribi-Hertz
(1989). All of them start from the assumption that the middle realizes one
participant overtly, but contains two. The two-participant approach to mid-
dle formation identifies the Subject of the middle as patientive. The Agent
of the process is argued to be necessarily implied. Evidence in favour of the
patientive nature of the middle Subject is said to come from the fact that
middles cannot be used in the imperative (Fellbaum and Zribi-Hertz 1989:
19) (illustrated in 9), and from the fact that the middle Subject, unlike the
Subject of an ergative one-participant construction, cannot be presented as
doing things by itself (see 11) (Keyser and Roeper 1984: 405):40
(9) *Read easily, book!

(10) a. The door closed slowly.
b. The door closed slowly all by itself.
(11) a. The book read slowly.
b. *The book read slowly all by itself.
As Keyser and Roeper (1984: 405) point out, a contradiction exists be-
tween the notion all by itself and the Agent which is implied in middles
(see also Smith 1978): the middle Subject cannot be perceived as agent-like
because the Agent of the process is implied.41 The Subject has moreover
been claimed to be necessarily affected by the process (Levin 1993: 26):
(12) *This poem understands easily.

(13) *The Eiffel Tower sees easily from my window.
Finally, following Sweet (1891), the middle has frequently been compared
to and sometimes even been located within the passive paradigm (Halliday
1967; Keyser and Roeper 1984; Declerck 1991b). At the same time, how-
ever, it is pointed out that middles also differ from ordinary passives. Most
importantly, middle constructions do not allow the implied Agent to be
realized overtly, while passives do (Keyser and Roeper 1984: 406):
(14) John was hit. <=> John was hit by Bill.

(15) Bureaucrats bribe easily. « *Bureaucrats bribe easily by managers.
2. The representational versatility of middle formation42
The deadlock which characterizes the literature on middle formation can

only be broken by a detailed analysis of the different verb types that can
function in the middle and the participant roles that can be mapped onto the
Subject.
First, the construction type of middle formation cannot be reduced to
one or other verb type: it is widely acknowledged - even by those who
situate middles close to the ergative one-participant paradigm - that erga-
tive as well as transitive verbs can function in the middle. Examples (16)
and (17) illustrate the middle use of the transitive verbs remove and wipe;
the middles in (18) and (19) are based on ergatives, i.e. flatten and disperse:
(16) Broiler rack removes easily. (CB)

(17) In Britain, however, the traditional image of schools drama as being
all chalk and blackboards does not wipe easily from people's minds.
(CB)
(18) Nylon carpets are prone to static ... and the pile flattens easily. (CB)
(19) It's heavier, like an emulsion with the texture of chocolate mousse
and it doesn't disperse easily. (CB)
Apart from transitive and ergative verbs, however, also intransitive verbs
can figure in middle constructions, a possibility which, while recognized in
descriptions of the Dutch middle construction, has thus far been almost
unanimously denied in the literature about English.43 The Dutch middles in
(20), (21) and (22) qualify as intransitive. The example in (20) is based on
Haeseryn et al. (1997: 52); that in (21) was extracted from the DDL
corpus44; the middle in (22) is taken from De Vries (1910: 132):
(20) Asfalt fietst prettiger dan grind.

asphalt cycles more pleasantly than gravel
'it is more pleasant to cycle on asphalt than it is on gravel'
(21) Sommige bomen zitten nu eenmaal lekkerder....
some trees just sit more comfortably
'some trees are just more comfortable to sit in'
The representational versatility of middle formation 129
(22) Nieuwe schoenen lopen lastig.

new shoes walk with difficulty
'new shoes are difficult to walk with'
The entities functioning as Subject are normally felt to be outside the

nucleus of the clausal process and the participants that are directly involved
in the process: they are oblique participants of the main verb, providing
either spatial information (e.g., on asphalt, in trees) or information on the
means whereby a process takes place (e.g., with/in new shoes).45
Only Van Oosten (1986), Smith (1978) and Rosta (1995) have hinted at
the existence of English middle constructions of which the Subject is not
patientive. They do not, however, offer a systematic analysis of their
properties, let alone integrate them in the analysis of middle formation in
general. After mentioning the structure This music dances better than the
other one ('to this music you can dance better than to the other one', 1986:
84), Van Oosten, for instance, decides that although "the subject in this
construction does not, then, need to be a patient but can be an instrument or
a locative or hold some other relation to the verb, I will for simplicity
continue to call this the patient-subject construction" (Van Oosten 1986:
85).
The fact that intransitive middles are rare in English has doubtless
contributed to the absence of a coherent description of their properties:
even large-scale corpora such as COBUILD hardly contain intransitive
middles. The following list therefore contains not only middles drawn from
the COBUILD corpus, but also middles that were suggested by native
speakers of English (labelled p.c.), and middles that were found in the
literature on middle formation:
(23) This music dances better than the other one. (Van Oosten 1986: 8)
(24) The Strathayr track is racing consistently nowadays. (Barry Blake
p.c.)
(25) The pitch is playing truly/well. (Barry Blake p.c.)
(26) [about a tennis court:] It is slightly coarser, so it plays a bit slower.
(CB)
(27) "They rolled the green just before the match and it ran three seconds
faster," said Curtis.... (CB)
(28) You wouldn't give my tip a chance of landing Aintree's Martell Red
Rum Chase ... but with the ground riding slower, he should improve
dramatically. (CB)
The Subjects of these middle constructions designate either the means that
is used to carry out the process (music) or they specify its location
(examples 24 to 28). They are normally realized as prepositional phrases:
you dance to music, run on a track, bowl on a pitch, play on a court, run on
a green and ride on a ground. Yet another type of intransitive middle has
an Instrument for Subject (Smith 1978: 103):
(29) The brush paints well.

(30) The knife cuts well.
These middles can be paraphrased as 'with this brush/knife you can

paint/cut well'; they are ambiguous between a middle reading and an
ordinary active reading that profiles the agent-like involvement of the
Instrument in the process.
Middle formation thus turns out to be remarkably versatile as to the verb
classes which it allows for - transitive, ergative and intransitive - and the
roles which can be mapped onto the Subject. The distinct representational
relations in middle constructions form an essential part of the semantics of
middle constructions: the relationship between a Patient and a transitive
process (This meat cuts easily), for instance, is fundamentally different
from that between a Complement and an intransitive process (This court
plays slower), or that between a Medium and an ergative process (This
emulsion doesn't disperse easily). An account of the representational func-
tional strand in the middle thus forms an essential part of the multifunc-
tional approach which in general is vital to come to an accurate description
of structural patterns (see Chapter 2, Section 4).
It is, moreover, only when considering the full range of representational
options which middle formation allows for that it becomes clear that many
of the properties which have in the literature been ascribed to middle for-
mation are not applicable to all middle constructions. For one thing, the
middle cannot be said to systematically alternate with a two-participant
structure in which the Subject is turned into Object (a claim made by pro-
ponents of the ergative approach), because intransitive middles contain
only one participant (e.g., The green ran faster but *You ran the green
faster). Neither can it be analyzed as realizing a kind of passive two-
participant structure (e.g., The green ran faster but *The green was run
faster). Intransitive middle constructions also form evidence against an
agentive interpretation of the Subject in middles: like Patients, the Subject
of intransitive middles cannot be interpreted as energy source of the proc-
ess. In general, it has to be observed that the Subject of middle construc-
tions is not necessarily affected (as argued by the proponents of the 'pas-
The representational versatility of middle formation 131
sive' view on middle formation) and it does not always undergo a change
(as claimed by the adherents of the ergative view on middle formation),
witness the following examples:
(31) Girls don't make it just because they are skinny. It's about how they
photograph and move. (CB)
(32) His earlier short stories don't read so well. (CB)
The girls in (31) are not affected or changed by being photographed. Like-
wise, the short stories in (32) are not affected or changed by being read.
The middle Subject may not be agentive, but is the Agent of the process
then obligatorily implied? As pointed out by Langacker, a middle like This
ice cream scoops out quite easily cannot but be interpreted as involving an
Agent which does the scooping out: we do not picture the ice cream scoop-
ing itself out (Langacker 1991: 334). It is, in fact, the semantics of the tran-
sitive verb scooping out which invokes the Agent: if for This ice cream
scoops out quite easily we do not imagine the ice cream scooping itself out,
it is because we follow the semantic instructions given by the active transi-
tive verb scoop out, which necessarily implies an Agent. What is presented
as most prominent by the middle construction, however, is the relationship
between the process and the non-agentive entity construed as its 'focal par-
ticipant' or Subject (Langacker 1991: 335). Intransitive middles can be
given an analogous semantic analysis: they most prominently present an
intransitive process in its relation to a Location or Means, but they imply an
Agent which actually executes that intransitive process. In, for instance,
The court plays a bit slower, an Agent is invoked which does the playing.
Likewise, in This music dances better than the other one, an Agent is im-
plied which dances to the music.
Van Oosten and Lakoff s emphasis on the agent-like responsibility of
middle Subjects should therefore be toned down: it is widely agreed that
the middle attributes the fact that a particular process can be carried out to
the inherent properties of the entity construed as Subject (apart from Van
Oosten 1977, 1986 and Lakoff 1977, also Fellbaum 1985, 1986; Langacker
1991; Massam 1992; Rosta 1995; Lemmens 1998), but the fact that the
Subject has properties that make a certain process possible should not be
confused with agentivity. Or, as Rosta (1995: 128) puts it:
In The book sold well... the properties of the book are a necessary condi-
tion for the book to sell well, but need not be (and typically are not) a suffi-
cient condition; if it were not for the properties of the book, it would not
sell well, but even though the book has the properties it has it will not sell
well unless some other participant (an agent) gets involved in getting the
book to sell well.
Rather than assign agent-like value to its Subject, the middle, in other
words, foregrounds the fact that the Subject-entity has properties which
influence the occurrence of a particular process. It has been pointed out that
when non-human entities function as middle Subjects, these inherent prop-
erties are often part of their design (Levin and Rappaport 1988; Lemmens
1998). As Lemmens (1998: 80) puts it, "the properties emphasized in the
middle construction are often also those for which the entity has been de-
signed in the first place". In the case of the latex paint in (33) and the lip-
stick in (34), for instance, it is obvious that cleaning up easily with soap
and water and slipping easily into your handbag or pocket is part of what
they are designed for:
(33) And it's latex paint, so it cleans up quickly and easily with soap and
water. (CB)
(CB)
Animals, then again, are often bred for a particular process: a chicken that
broils well has probably been bred to be broiled (Levin and Rappaport
1988: 1078).
3. Towards a semantic typology of middle constructions
Importantly, middle constructions do not only vary according to the process

type which they are centred on and the non-agentive entity which they pro-
file: middles also differ depending on which facet of the interaction be-
tween the non-agentive Subject and the process they highlight (note that
this makes them particularly suitable for being used in advertising, which,
as pointed out by Yoshimura (forthcoming), frequently resorts to middle
constructions). In the literature on middle formation, one has tended to
focus on facility-oriented middles, or middles that specify how easy or
difficult it is to carry out a particular process on the Subject-entity. Facility-
oriented middles are typically construed with the adverb easily. However,
as pointed out in Heyvaert (1997), corpus data reveal the existence of mid-
dle constructions that do not so much specify how the process can be car-
ried out, as indicate that it can be carried out (i.e. its feasibility) or com-
ment on the result of carrying it out. Still other middle structures highlight
Towards a semantic typology of middle constructions 133
the destiny of the Subject-entity by pointing out where it can be placed.

Interestingly, as with the facility-oriented type of middle, the semantic
properties of the result- and destiny-oriented types of middle formation can
be systematically related to the adverbials or the oblique participants which
they are used with. Feasibility-oriented middles are characterized by the
absence of a specific type of adverb.
The facility-oriented reading highlights the fact that the properties of the
Subject entity are such that the process can be carried out either easily or
with difficulty:
(35) The window opened only with great difficulty. (Langacker 1991: 334)
(36) [about conditioning milk:] ... rinses easily away and really works!
(CB)
Facility-oriented middles thus zoom in on the carrying out of the process

and specify whether that is hindered or facilitated by the properties of the
Subject. Structurally speaking, facility-oriented middles stand out in that
they typically take the adverb easily. Other adverbials, such as with great
difficulty in (35), are also found, though much less frequently.
A type of middle which is closely related to facility-oriented middles in
that it also focuses on the process being carried out is that of middles which
provide a more general indication of the way in which it can be carried out.
This type includes middles which incorporate an adverbial expressing a
quality judgement as in (37) and (38), and middles which contain a com-
parison of quality as in (39) and (40) (Kemmer 1993: 147):
(37) That is easily done because the car handles superbly. (CB)
(38) If they wanted to do Eliot, why didn't they do The Magi? It's narrati-
ve, it reads well.... (CB)
(39) The new Holden Berlina handles like a junior sports sedan.... (CB)
(40) Her life story reads like the Hollywood movies she's starred in. (CB)
Still other middle constructions comment on the properties of a particular

entity by pointing out how much time it takes to carry out a certain process
on it:
faster," said Curtis .... (CB)
(42) [about a cosy car seat protector:] Quickly attaches/removes with ela-
stic straps and velcro tabs. (CB)
(43) This title usually ships within 2-3 days, (www.amazon.com)
The middle constructions discussed so far all zero in on the effect of the
properties of the Subject-entity on how the process can be carried out.
However, the nucleus of non-agentive Subject and active verb phrase in the
middle can also focus on various other stages. I distinguish between three
other possible foci of middle constructions: middles can simply highlight
the feasibility of the process (i.e. whether or not the process can be carried
out); they can zero in on what the process is destined for; or they can focus
on what the result is like when the process is carried out.
First, middles of the feasibility type (Davidse 1991: 44) merely focus on
whether the properties of the entity construed as Subject make a process
possible. They do not specify how this can be done or whether or not it can
be done easily. Examples are:
(44) [about a cook book holder:] Folds up and packs away for convenient
storage. (CB)
(45) This umbrella folds up. (Fellbaum 1986: 9)
(46) I thought we were out of gas, but the car DRIVES! (Fellbaum 1986:
9)
(47) This dress buttons. (Fagan 1992: 57)
Examples like these illustrate that the most prototypical cases of middle
formation may be those with an adverb (such as easily, welt), but that the
presence of an adverb is not an absolute prerequisite: middles which focus
on whether it is feasible to carry out a certain process on a particular entity
do not seem to be characteristically accompanied by specific adverbs.
Some middle constructions contain a locative oblique participant (Fagan
1992: 80). Interestingly, the occurrence of such Locatives appears to ac-
count for the destiny-oriented focus which characterizes these middles
(Davidse p.c.):
(48) [about a children's coat:] Outer flap wraps around little hands and
secures with Velcro.... (CB)
(49) The purpose of the device is to alert deer of your approach, not the
other way around. It is a little whistle that attaches to your car with
self-adhesive tape. (CB)
(50) The ultimate travel pillow. Resteaz fixes to the headrest providing
comfortable head and neck support. (CB)
(51) Playset folds up into a storage case with handle for easy carrying.
(CB)
Towards a semantic typology of middle constructions 135
Destiny-oriented or 'telic' (Dirven p.c.) middle constructions either focus

on where an entity has to be placed to make it function or they merely spec-
ify where it can be stored when it is not being used. In either case, the "de-
liberately designed properties" (Lemmens 1998: 80) that characterize the
Subject of many middle constructions are prominently present. Examples
of destiny-oriented middles which specify where an entity has to be placed
in order to function are those in (48), (49) and (50): the middle in (48) tells
the parent who buys the children's coat to wrap the outer flap around the
child's hands·, in (49), it is pointed out to the person who buys the whistle
that, in order to use it, it has to be attached to the car. Likewise, in (50), it
is specified that the travel pillow has to be fixed to the headrest. The middle
in (51), then, does not tell you to use the playset by folding it up into a
storage case; instead, it indicates that the playset has been designed in such
a way that when you do not use it, you can fold it up into a storage case.
Notice that destiny-oriented middle constructions often indicate not only
where an entity can or should be placed (Location), but also mention how
this can be realized (Means), as in (48) and (49) {with Velcro and with self-
adhesive tape).
Finally, there exist middle constructions which comment on the result of
carrying out a certain process on the entity construed as Subject. In accor-
dance with Heyvaert (1997) I will call them result-oriented middles. An
example of a result-oriented middle is (52), in which the woman function-
ing as Subject is characterized as not photographing well. This does not
mean that it is difficult to photograph her (facility-oriented meaning), but
that, if you photograph her, the pictures resulting from it are hardly ever
beautiful. As also pointed out by Fellbaum (1986: 10), the middle in (53)
emphasizes that the result of washing the item of clothing was good: there
was little shrinkage and no puckering.
(52) She does not photograph well.... (CB)

(53) [about a piece of clothing:] It washed well with little shrinkage and no
puckering. (CB)
Notice the difference in meaning between the adverb well as it is used in

(52) and (53) and the well of middle constructions such as the following
(Fellbaum 1986: 10):
(54) This book sells well.

(55) This car drives well.
These middles do not mean that the result of selling the book or driving the
car is good. Rather, the meaning of well in (54) comes close to fast or in
large quantities, whereas in (55), well can be paraphrased as easily,
smoothly. The adverb well, in other words, has various different contextual
entailments, which in the middle either trigger off a facility- or quality-
oriented reading (as in 54 and 55) or a result-oriented one (as in 52 and 53).
Other result-oriented middles are:
(56) [about a bag:] It measures 37 χ 60 cm ... and folds up neatly. (CB)

(57) She does not photograph well, and the portraits of her pinched fea-
tures snatched outside the High Court only added to the picture of a
shrew. (CB)
The middle in (56) emphasizes that the result offolding up the bag is likely
to be neat. As in middle constructions with an inanimate Subject in general,
it is implied that the Subject-entity has been designed in such a way that
this result is possible. The middle in (57) emphasizes that photographing
the Subject-entity will probably not lead to nice pictures.
The facility-, quality-, feasibility-, destiny- and result-oriented middle
structures which I have thus far discussed all 'target' (Langacker p.c.) one
particular facet of the complex event that is profiled by the middle con-
struction: feasibility-oriented middles highlight the beginning of the event,
facility- and quality-oriented ones its middle stage, while destiny- and re-
sult-oriented ones zoom in on the end stage of the event. Corpus data reveal
that middle constructions often combine two or more of these foci, and this
by integrating adverbs and oblique participants which are characteristic of
different semantic types. Consider the following middle constructions:
(58) Cards and shapes store neatly in the desk. (CB)

(59) Wood valance attaches in seconds, magnetically, to a standard cur-
tain rod. (CB)
(60) A full hood, which neatly stows behind the after sets, would make the
cockpit a part of the accommodation and keep off the rain or sun.
(CB)
(61) [about a kitchen timer clock:] Magnetically attaches to any metal
surface .... (CB)
(62) [about an emergency ladder:] Compact so that it can be stored under
the bed or in a wardrobe, the ladder fixes quickly and easily under the
window ledge. (CB)
(CB)
Middle constructions and modality 137
The middle in (58), for instance, highlights the destiny of the cards and
shapes (i.e. in the desk) and it specifies that the result of storing them in it
is likely to be neat. The middle construction in (59), then, focuses on the
time span in which the process can be carried out (in seconds), the way in
which it can be carried out (magnetically) and the destiny of the wood val-
ance (to a standard curtain rod).
To conclude, middle constructions may be prototypically patientive and
facility-oriented, but they allow for other processes and Subject-entities as
well and they can highlight a wide range of facets of the interaction be-
tween the non-agentive entity construed as Subject and the process. What
seems to characterize all middle constructions, irrespective of the process
which they are centred on, the entity which they construe as Subject and the
facet(s) which they select for highlighting, is that the properties of the Sub-
ject-entity are presented as making (or not making) it conducive to a par-
ticular process. This conduciveness, I will argue in the following section, is
basically a modal meaning which is realized through the unique construc-
tional link which middles establish between a non-agentive Subject and an
active finite verb: it is the specific way in which middle constructions code
the relation between a process and a non-agentive entity - be it Patient,
Location, Means or Instrument - which typifies the construction and pro-
vides the right generalization.
4. Middle constructions and modality
It has been argued before that middles imply a modal meaning (Fellbaum
1985; Fagan 1992; Massam 1992; Iwata 1999). In Massam (1992: 122), it
is posited that all middle constructions contain what is called a null element
"which is in essence equivalent to the modal can". Like the auxiliary can,
middles are argued to realize the modality of possibility/ ability (Massam
1992: 124; Fagan 1992: 54). Fagan (1992) thus suggests the following
paraphrases for the middle constructions in (64a):
(64) a. [about a kind of siding:] It nails easily. It cuts easily.

b. It can be nailed easily. It can be cut easily. (Fagan 1992: 54)
However, no evidence has thus far been adduced in favour of a modal in-
terpretation of the middle: the claim that middles realize the modality of
possibility/ability is only based on the observation that middles seem to be
paraphrasable by clauses with can followed by a passive verb. In this sec-
tion, I will try to provide arguments for the modal nature of middle con-
structions by analyzing their semantic and lexicogrammatical properties in

a more systematic way.
First, middle constructions do not imply epistemic possibility. This be-
comes clear when we add modal forms like possibly or it is possible that to
them:
(65) Possibly, this pan cleans easily.
The meaning of this middle deviates from that of This pan cleans easily, in
which no adverbial modal is present: rather than create "double modality"
(Halliday 1970a: 330fnl2) and reinforce the meaning of the middle This
pan cleans easily, the epistemic modality of possibility that is expressed by
the adverbial possibly introduces a new scope and forms a middle which
comes closer to the structure
(66) This pan may clean easily.
in which the epistemic modality of possibility is explicitly realized by

means of the modal auxiliary may. Unlike in (65), adding the adverbial
modal possibly to (66) results in a double modal structure in which the mo-
dal may is reinforced:
(67) Possibly, this pan may clean easily.
The epistemic modality of possibility, in other words, only appears to be

present in those middles which contain an overt indication of it, in the form
of a modal auxiliary (e.g., may), a modal adverbial (e.g., possibly), or both.
Other attested examples of explicit epistemic modality in middle construc-
tions are the following:
(68) Plunge under cold water: the thin elastic membrane and any pieces of
fat should peel off easily. (CB)
(69) This book could sell. (Roberts 1987: 233)
The modal should in (68) realizes the modality of virtual certainty, and, as
Roberts points out, the modal could in (69) "can only be epistemic here",
i.e. it can only mean it is possible for the book to sell and not the book was
able to sell (Roberts 1987: 233). Including the speaker's comment on the
possibility or probability of the process is, in short, an option in the system
of middle formation that requires the explicit realization of a modal auxil-
iary or adverbial. It does not shed light on the relation between the Subject
and the finite verb in middle formation in general.
Do middle constructions then implicitly realize the deontic or root mo-
dality of ability? Research on middle formation has in general never got
beyond the observation that middles can be paraphrased by passive clauses
with the modal auxiliary can. As also hinted at in Roberts (1987), the mid-
dle itself is unable to contain the modal can in its ability meaning:
(70) *This book can sell.

(71) *This pan can clean easily.
In my view, passive clauses with the modal auxiliary of ability can do not
agnate with middle constructions because they cannot be systematically
related to them: as I argued earlier, not all middle constructions can be re-
lated to a passive construction:
(72) *The green was run faster.

(73) *That music was danced better.
Passive caw-clauses also fail to shed light on the constructional link

which middles establish between a non-agentive Subject and an active Fi-
nite. The modality of ability, moreover, especially foregrounds the Agent of
the process: as a rule, the modality of ability "relates to and is intrinsic to
the actor" (Halliday 1970a: 339). In clauses with active voice, the Subject
is thus "actor with respect to the modality as well as with respect to the
process" (Halliday 1970a: 339): in the clause Jones can drive, Jones is
identified as the one who has the ability to drive. In passive clauses the
relation of ability is between the implied or fry-Agent and the process,
rather than between Subject and process: in, for instance, any problem can
be solved, it is not the problem that is characterized as having the ability to
solve, but the implied Agent.46 Middle clauses, however, focus not so much
on the implied Agent, as on the non-agentive Subject and its properties.
In the next section, I will formulate an alternative modal analysis of
middle formation in which the middle is viewed as realizing a force-
dynamically interpreted type of objective modality. In a first part (Section
4.1), I will briefly elaborate on the specific view of modality that lies be-
hind the analysis which I propose, viz. Langacker's (1991) and Talmy's
(2000) force-dynamic approach to modality as well as the view of modality
which posits that all clauses realize a schematic form of modality, as de-
scribed in Chapter 4 (Section 2.3.1; based on Verstraete 2002). In a second
part, then, I will give arguments in favour of my interpretation of the mid-
die, referring to the general semantics of the construction type and to a

number of constructions that seem to be systematically related to the mid-
dle (Section 4.2).
4.1. The modality of 'letting'
The modality that is characteristic of middle constructions can be described

as objective, with the Subject as the locus of potency directed towards the
realization of the process, not by bringing it about itself, but by letting it be
carried out. Where exactly can we situate this form of modality in the over-
all system of modality?
First, Langacker (1991: 269) argues that one of the properties of the his-
torical antecedents of the modern English modals is that their Subject is
"the locus of some kind of potency directed at the landmark process, i.e. a
physical or mental force that, when unleashed, tends to bring about an oc-
currence ofthat process" (Langacker 1991: 270). Which are these historical
antecedents? Diachronically speaking, the English modals started out as
main verbs: the modal adverbial will, for instance, derives from a verb with
the meaning of want. Similarly, the modal can originally meant that its
Subject had the knowledge or mental ability to do something (Langacker
1991: 269). The original expressions are comparable to such modern Eng-
lish locutions as 'want/know how' to swim. In the case of ' want/know how'
to swim, then, it is the Subject which has the potency to unleash a force
towards the realization of the process of swimming.
The historical evolution from main verb, through root or deontic modal-
ity, to epistemic modality can be described "as a matter of the locus of this
potency becoming progressively less salient and well-defined" (Langacker
1991: 272). While at the main verb-stage, the locus of potency is the Sub-
ject-entity, i.e. "an objectively construed participant accorded maximal
salience" (Langacker 1991: 272), the notion of potency has in later stages
undergone subjectiflcation, i.e. it has become increasingly implicit. It seems
to me, however, that in the 'dynamic' types of modality (i.e. volition or
willingness and ability), the conception of potency is still located in the
entity construed as Subject. The modality of willingness, for example, as
realized by the modal auxiliary will, is always Subject-oriented and ex-
presses the will or willingness of the entity designated by the Subject. In
the clause He will do anything to please her, the //¿-person is the one that
is willing to please, i.e. that is directed towards the realization of the proc-
ess of pleasing. Likewise, in the passive clause He won't be helped, it is the
Subject-entity that is not willing to be helped (Declerck 1991b: 363). In
clauses which express the modality of ability, it is clear that the potency is
located in the Subject as well (e.g., He can play the piano).
I propose that the modality of middle constructions comes close to the
kind of modality realized in the Modern English dynamic modals. In the
middle construction as well, it is the Subject which forms the locus of po-
tency directed towards the realization of the process. However, the potency
of the middle Subject is not the one which is typically associated with cau-
sation, i.e. it does not itself bring about the process. Rather, the Subject has
properties that make it conducive to a particular process: by being con-
strued as Subject of a middle construction, a non-agentive entity is pre-
sented as having the necessary properties or potential to let a process be
carried out. Interestingly, 'letting' has in Talmy's force-dynamic frame-
work (Talmy 2000: 443) been characterized as being part of the "greater
modal system" of language. The letting-foTce which is unleashed by the
middle Subject is, unlike the locus of potency that unleashes it, typically
realized maximally subjective, i.e. it is not explicitly realized in the clause.
I will show that one of the arguments in favour of the presence of an objec-
tive, dynamic type of modality in middle constructions comes from the
occurrence of a subtype of middle construction in which it is explicitly
realized: structures such as The door would not open express "negation of
willingness" (Quirk et al. 1985: 229) or the absence of inclination on the
part of the Subject. They thus explicitly deny the letting modality that is
implied in the middle construction.
Like the dynamic modals of willingness and ability, middle construc-
tions realize a kind of modality that is objective·, the modality of letting is
internal to the proposition and specifies the representational meaning of the
relationship that is established in it. Like dynamic modals, middle construc-
tions therefore also have schematic epistemic modality (as described in
Chapter 4, Section 2.3.1, based on Verstraete 2002). This may either be
realized as indicative mood, or it takes the form of an explicit epistemic
modal, expressing the modalities of possibility, probability or certainty and
situating "the process at varying distances from the speaker's position at
immediate known reality" (Langacker 1991: 246), as illustrated in (74) and
(75):
(74) Plunge under cold water: the thin elastic membrane and any pieces of
fat should peel off easily. (CB)
(75) This book may sell, (personal example)
Because it can take explicit epistemic modals, the modality that is realized
in the middle comes closest to the periphrastic type of dynamic modals
such as be able to and be willing to, which can also be used with explicit
epistemic modals (e.g., He may be able to come, He may not be willing to
do it).
When middles are used in the indicative (which expresses the speaker's
commitment to the truth of the proposition, see Verstraete 2002; Davies,
2001), they can be part of immediate or 'proximal' reality, which typically
coincides with the present tense; or they can be situated in non-immediate
reality, i.e. in the past (Langacker 1991: 245).47 Examples of past middles
are:
(76) The lap portion of the car seat belt slotted into place easily, but the
belt clip at the back was rather difficult to thread. (CB)
Prototypically, however, middle constructions have the present tense and

designate a characteristic or state of the Subject-entity, which "is in princi-
ple instanced by a number of dynamic situations, but the simple present
does not actually refer to any of these" (Declerck 1991b: 182). Because
they do not receive a "single event reading" (Declerck 1986: 172), middles
like these can be described as generic. Either they designate a characteristic
of an individual, as in (78), or they characterize a type or class of entities,
as in (79):
(78) She does not photograph well.... (CB)

(79) The new Holden Berlina handles like a junior sports sedan .... (CB)
In short, the modality of 'letting' which is realized in the middle con-

struction can be situated close to the dynamic modalities of willingness and
ability: like them, it situates the locus of potency that is directed towards
the realization of the process in the Subject; it forms part of the representa-
tional meaning of the proposition; it also has schematic epistemic modality
and it can be situated in the past. Like the periphrastic modals be willing to
and be able to, the modality of letting can, moreover, take explicit epis-
temic modality. I therefore suggest describing the letting modality of mid-
dle formation as representing a dynamic type of modality which designates
being conducive to: when an entity is construed as the Subject of a middle
construction, it is said to be conducive to a particular process. In Section
4.2, I will go more deeply into some of the characteristics of middle con-
structions which support their analysis as 'letting' constructions.
4.2. The middle as 'letting' construction
Which arguments corroborate the analysis of middle constructions as real-

izing the meaning of 'letting', viewed as an instance of dynamic modality?
First, the letting analysis appears to offer an explanation for the centrality
of the Subject-function in middle formation, i.e. the fact that the non-
agentive Subject-entity is felt to be responsible for the process and the em-
phasis that is placed on the role of the properties of the Subject in the carry-
ing out of the process. The middle Subject forms a clear example of what I
described in Chapter 4 (Section 2.4) as being the speech functional role of
the Subject. The Subject in middle constructions is held responsible for the
epistemic modal position that is taken by the speaker: it is pointed out as
being responsible for the epistemic position of truth that characterizes mid-
dles without explicit modal auxiliaries or adverbials and for the epistemic
possibility/probability that is expressed by middles which contain an ex-
plicit modal. The (objective) properties of the Subject-entity, however,
clearly motivate its being construed as Subject by the speaker. Importantly,
this speech functional type of responsibility of the middle Subject should
not be confused with causal or agentive responsibility, as has been done in
Hale and Keyser (1987) and Van Oosten (1986).
The letting analysis entails that the Subject in middle constructions is
identified as the locus of the potency or force that is directed towards the
process. The middle Subject has that force due to its properties, which are
typically physical in nature: they seem to prototypically involve one's abil-
ity to manipulate things (e.g., That rack removes easily, That lipstick slips
easily into your handbag, This conditioner rinses easily away), but they can
also be visual (as in That girl photographs well), auditory (e.g., That music
dances better) or they can involve one's sense of taste (e.g., That wine
drinks like water, This chicken broils well). An example in which the mid-
dle Subject's potency can, in one reading, be interpreted as primarily men-
tal in nature is That novel reads well. All middle constructions, including
the less prototypical ones, are thus centred on the potency of the Subject.
The Subject's potency and the fact that that potency is directed towards
the realization of the process is what was characterized earlier as the con-
duciveness of the middle Subject to the process. Importantly, the relation-
ship of conduciveness that characterizes middle constructions is not to be
interpreted as one of agentivity, but as one of 'letting'. The middle Subject
is not itself the energy source of the process (it is not agentive), but it is -
or is not - conducive to the implied Agent carrying out the process. This
approach remedies the uneasy vacillation between the straight agentive and
patientive interpretations that are found in the literature on middle forma-
tion (discussed in Section 1 of this chapter as the ergative and the passive
interpretations of middle formation). The non-agentive entity construed as
Subject of a middle structure is a letting-entity or a conducive entity, i.e.
one whose properties either let through the energy flow instigated by the
Agent, or blocks it.
Conduciveness, and consequently also the notion of 'letting', can be ar-
gued to constitute general properties of middle constructions in that they
schematize over the various process types and Subject-entities which the
middle can take, as well as over the wide range of facets of the interaction
between them which middles can select for highlighting. The schematic
characterization in terms of letting and conduciveness counterbalances the
traditional focus on the prototypical middle construction with a facility-
oriented profile and a patientive Subject. It constitutes a schema which is
instantiated by all middle constructions, prototypical as well as non-
prototypical. In, for instance, the green ran faster in (80), the properties of
the green are claimed to have let one run faster on it; the middle in (81)
conveys that the properties of the ladder are such that it lets one fix it
quickly and easily under the window ledge. Likewise, the middle in (82)
expresses that the properties of the Subject-entity are such that it lets the
process of washing be carried out with a good result. The umbrella in (83),
finally, is characterized as letting one fold it up.
(81) ... the ladder fixes quickly and easily under the window ledge. (CB)
(82) It washed well and didn't shrink much .... (CB)
(83) This umbrella folds up. (Fellbaum 1986: 9)
Additional evidence in favour of the letting analysis of middle formation

comes from two related constructions which pick up on the meaning of
letting. A first related construction consists of what is basically a middle (it
has a non-agentive Subject and takes an active verb form), but then with the
negative modals won't and wouldn't. It is illustrated in the following exam-
ples:
(84) ... Mr. Fabius chose to exit by a false door. ... the scenery began to
shake in response to the ex-prime minister's desperate efforts, but still
the door would not open. (CB)
(85) ... Mrs. Dambar was having some difficulty with the new dress she
had bought .... For some reason, it wouldn't zip all the way up in the
back. (CB)
(86) I tried to open the door, but the key wouldn 't turn. (Quirk et al. 1985:
229)
(87) I am at α sentence that will not write. (Jespersen 1914-1929, 3: 349)
It has been pointed out in the literature that the negation of will/would in
these clauses seems to "have something of the personificatory force of 're-
fusal'" (Quirk et al. 1985: 229; see also Ehrman 1966; Huddleston 1969):
mrs. Dambar's new dress in (87), for instance, as it were refuses to be
zipped. The modal auxiliary won 't/wouldn't in this use denies the typical,
letting properties of the Subject and shifts the focus to the fact that the Sub-
ject instead hinders the implied Agent in carrying out the process. When
the modal will in middle constructions is followed by internal negation, its
meaning seems to move more towards the epistemic, even though one can
also argue that there is a notion of preventing or resisting something that is
not desired, as in (88) and (89):48
(88) Just press and personalize clothes in minutes with clear, legible im-
prints that will not wash out. (CB)
(89) Place gifts in cardboard boxes that won't crush easily. (CB)
Interestingly, the notion of 'hindrance' which is, in some cases, implied

in the use of the modals won't and wouldn't can also be realized by the
lexical verb refuse:
(90) ... although the key turns, the door will refuse to lock (or to unlock).
(CB)
(91) ... when the matches refused to strike. (Jespersen 1914-1929, 3: 348)
These constructions may be considered as variants of middle clauses with

won't and wouldn't (as in 92 and 93), but then with the notion of hindering
which the latter imply realized lexically:
(92) The door won't lock.

(93) The matches wouldn't strike.
Another construction which helps to shed light on the middle construc-

tion is the laat zieh 'let itself - construction in Dutch. Dutch has a construc-
tion type that makes use of the auxiliary laten 'let', followed by the reflex-
ive pronoun zieh 'itself and an active infinitive (the examples were ex-
tracted from the IDL corpus):
(94) GSM-antwoordapparaat laatzich gemakkelijk kraken.

mobile phone-answering machine lets itself easily crack
'mobile phone-answering machine cracks easily'
(95) Haar veelbewogen Hollywood-leven laat zieh lezen als een span-
nend meisjesboek.
her turbulent Hollywood-life lets itself read like an
exciting girl's book
'her turbulent Hollywood-life reads like an exciting girl's book'
Significantly, this construction type has many features in common with

middle formation: the participant functioning as Subject is non-agentive;
the verb form is active and it is usually accompanied by some modification
of how the process can be carried out with respect to the Subject-entity, but
it may just as well express mere feasibility, as in the following examples
(from the LDL corpus):
(96) De uitkomsten daarvan laten zieh als volgi samenvatten ....

the results of that let themselves as follows summarize
'the results can be summarized as follows ....'
(97) Een beetje bitter, maar het laat zieh drinken.
somewhat bitter, but it lets itself drink
'even though it is somewhat bitter, it drinks'
In Fagan (1992: 211), the German variant of this type of 'let itself -
construction is elaborately discussed and classified as a type of middle con-
struction:
(98) Der Wagen läßt sich angenehm fahren,

that car lets itself pleasantly drive
'that car is pleasant to drive'
The Dutch 'let itself - construction is mentioned and likewise classified as a

type of middle structure. I prefer to treat it as a close agnate of middle for-
mation. Interesting about it is that it lexicalizes the notion of 'letting' which
is claimed here to be implied in all middle constructions. It appears as if in
Dutch, the 'letting' aspect of middle formation can either be realized highly
subjectively (implicitly), or objectively, i.e. in the form of a lexical verb.
Conclusion 147
5. Conclusion
I have suggested that the existing analyses of middle formation can be re-
fined in three ways. First, the construction's representational potential can
be mapped out more accurately and turns out to include also intransitive
processes. Secondly, middle constructions vary according to which facet of
the designated event that they highlight: in addition to facility-oriented
middles, also feasibility-, destiny- and result-oriented middles can be dis-
tinguished and linked up with the occurrence of specific adverbs or oblique
participants. And, finally, the link between a non-agentive Subject and an
active Finite, rather than restriction to a specific verb class, typifies the
middle construction. The essentially modal value of the relationship which
middles establish, and, more particularly, the letting force-dynamic status
of it, allows us to siring all the characteristics of middle formation together.
The middle's emphasis on the properties of the Subject-entity ties in with
the Subject being the locus of the potency of the letting modality; the fact
that the middle profiles a letting relationship explains the meaning of con-
duciveness which all middles have in common, as well as the non-agentive
nature of the Subject and the need for an implied Agent to actually carry
out the process. By being construed as Subject of a middle construction, a
non-agentive entity thus acquires speech functional responsibility and its
properties and the fact that they let one carry out certain processes (in a
particular way) are highlighted. In Chapter 7,1 will show how these proper-
ties of the clausal middle construction shed light on the category of non-
agentive -er nominalization and, ultimately, on the category of deverbal -er
derivation itself.
Chapter 7
A multifunctional approach to -er nominalization
Within a schematic network ..., certain nodes and relationships are far
more prominent and important than others, both cognitively and linguisti-
cally. In particular, special significance attaches to the nodes that function
as the category prototype and as the highest-level schema. The prototype is
significant because of its developmental priority and notable cognitive sali-
ence. As the primary basis for extension, it defines the center of gravity for
the category. The highest-order schema is significant because it embodies
the maximal generalization that can be extracted as a characterization of
the category membership. (Langacker 1987a: 380-381)
0. Introduction
The analysis of -er nominalization which is presented here is aimed at iden-

tifying the high-level schema which deverbal -er nominalizations instanti-
ate. It essentially posits that deverbal -er suffixation implies a relationship
between an entity and a process which is strikingly similar to the Subject-
Finite relation in the clause. Like the clausal Subject, the entity which is
profiled through -er nominalization is either agentive or non-agentive; like
the Finite in clauses, deverbal -er suffixation can ground the process type
which it nominalizes in a variety of ways. As to the actual profile of -er
derivations, I will argue that it is equivalent to that of the clausal Subject. It
is the function of Subject that will be identified as the high-level schema
which all deverbal -er nominalizations realize.
Nominalizations which at first sight seem to resist a Subject interpreta-
tion are those designating non-agentive entities. Among the arguments that
have been raised against viewing them as Subject-profiling are, firstly, that
there are no grounds for assuming that a morphological process such as -er
nominalization can have a patientive entity as Subject; that oblique partici-
pants such as Locatives and certain Instruments are not easily construed as
Subjects; and, finally, that not all non-agentive -er nomináis can be related
to a middle construction.49 The hypothesis which I put forward is nonethe-
less that deverbal -er nominalization originated as an agentive system, but
that the productivity of non-agentive nomináis suggests that it has now
turned into a Subject-oriented system. The rearrangement of options which
seems to have taken place within the network of -er formations may well
150 A multifunctional approach to -er nominalization
have run parallel with the increased productivity of clausal middle forma-
tion, and even with the Subject-centred nature of contemporary English as
such. Possibly, the rise in -er nomináis based on nouns and other, more
complex expressions (such as hardliner, down-and-outer, lefthander) will
constitute another factor of instability within the system of -er nominaliza-
tion in general, which will again refashion the system and may have reper-
cussions for deverbal -er nominalization as well. But for now, the Subject
generalization seems to be the most accurate generalization for deverbal -er
nominalizations.
I claim that non-agentive -er nominalizations can be accounted for in
terms of subjecthood, provided that the link between non-agentive -er
nominalization and middle formation is recognized for what it is, viz. as a
relationship of agnation which is based on a fundamental correspondence
between the speech event-related choices made within the Subject-Finite
unit of the middle and those made in non-agentive -er nominalization.
Middle constructions are thus much more fundamental to the analysis of
non-agentive -er derivation than has been suggested thus far (in Levin and
Rappaport 1988; Rappaport Hovav and Levin 1992; Lemmens 1998). To
elucidate their role, however, it is essential that their own Subject-Finite
structure is interpreted correctly, i.e. as being centred on an active finite
verb and allowing for a wide range of verb types, including intransitives. If
non-agentive -er nomináis are related to middle constructions, then, this
would solve two things. Firstly, non-agentive -er nomináis construing a
Patient are not linked to a verb in the passive voice. Secondly, oblique par-
ticipants such as Locatives also fall within the generalization and can be
said to relate to the Subject function of the middle. The relationship of ag-
nation between middle formation and non-agentive -er nominalization
should, moreover, be seen as a basically schematic relation between sys-
tems. Non-agentive -er nomináis instantiate schemata which are partly
identical to those instantiated by middle formation. Each system, however,
also has its own schemata, which account for the unique character of its
instantiations. In short, if analyzed precisely, the relationship of agnation
between the middle construction and non-agentive -er nominalization can
refute all arguments raised against a Subject generalization for deverbal -er
nominalization.
To map out the overall schematic network of deverbal -er suffixation
and unveil the schematic choices which it has in common with the Subject
and the Finite in clausal constructions, a multifunctional analysis is re-
quired: the representational semantics of -er nominalization has to be inte-
grated with the constructional (or, in Halliday's terms, the interpersonal or
speech-event related) status of the relationship which -er nomináis imply
Representational semantics 151
and with their textual functioning (see Chapter 4, Section 3). It is only
through a multifunctional analysis that we can string together the form and
meaning aspects of -er nominalization and come to an account of its prop-
erties which is truly natural.
The outline of my analysis is as follows: first (in Section 1), I will give a
brief overview of the representational options which -er nominalization
offers and list some of the subschemata which can be said to instantiate the
schematic categories of agentive and non-agentive deverbal -er suffixation.
In a second part, I propose a semantic typology of non-agentive -er nomi-
nalization, modelled after that of middle clauses (Section 2). In the third -
and most substantial - section, I present an elaborate account of the con-
structional relationship which -er nomináis establish between an agentive
or non-agentive entity and a process type. I go more deeply into the distinc-
tion between lexicalizations and ad hoc nominalizations and point to the
different kind of interpersonal relation which they imply between an entity
and a process and the differences in textual functioning that seem to corre-
late with it. In a fourth and final section, then, I elaborate on the Subject-
generalization which I propose for the system of deverbal -er nominaliza-
tion.
1. Representational semantics
Depending on the process type that is nominalized, -er nomináis profile

different participant roles. At the extreme, "every verb defines a distinct set
of participant roles that reflect its own unique semantic properties" (Lan-
gacker 1991: 284). If we distinguish between agentive and non-agentive -er
nominalizations, then, this is necessarily very much a schematic distinction,
which covers a wide range of profiles. In Systemic-Functional Grammar, it
has, for instance, always been emphasized that not all two-participant
clauses can be analyzed in terms of a 'causation' model, and that, strictly
speaking, only clauses which realize an action model and take 'material'
verbs can be said to construe an Actor-Goal type of semantics. Material
processes or 'processes of the external world' prototypically designate
'outer experience', i.e. actions and events. They are to be distinguished
from mental processes or processes of consciousness, related to "inner ex-
perience" (Halliday 1994: 107), and relational processes, which relate two
items of experience to one another, either by classifying ('this is a kind of
the other') or by identifying ('this is the same as that') (Halliday 1994:
107). The participants that are involved in relational and mental process
clauses are 'Agents' and 'Patients' only in a very general, schematic sense
and more refined characterizations can be provided for them: in a mental

clause such as I admire you, for instance, the /-participant can be thought of
as the Senser and the Fou-participant the Phenomenon (Halliday 1994:
117). In a relational process clause such as John is the leader, John can be
identified as the Token and the leader as the Value which is supplied for it.
In short, if in my analysis of -er nominalizations, I use the notions of 'agen-
tive' and 'non-agentive', it is implied that they cover a whole range of enti-
ties which differ depending on the type of verb which the nominalization
derives from. In the next section, I will sketch the main options within each
schematic category. My analysis elaborates on Lemmens' (1998) account
of the transitive and the ergative paradigms and their impact on -er nomi-
nalization. The criteria for discriminating between the various subtypes
which I discuss are based on Hopper and Thompson (1980), Davidse
(1991) and Halliday (1994).
1.1. Agentive -er nominalization
An -er nominalization can be high in transitivity and designate the voli-

tional, human Agent of a transitive process with an affected Object (e.g.,
carver, cutter). -Er suffixation can also foreground the Agent of a process
that is lower in transitivity, but still two-participant (e.g., writer, dancer). In
the latter case, the processes underlying the nominalization are 'pseudo-
effective' (Davidse 1991), i.e. they can contain an Object, but only one
which specifies the "range or scope of the process" (Halliday 1994: 146), as
in write a book and dance the Charleston. The book and the Charleston are
not 'done to' (*What he did to the book was write it, *What they did to the
Charleston was dance it). When an agentive -er nominalization derives
from an ergative process, it profiles the Instigator of the process and an-
other entity is understood to be co-participating in the process (e.g., burner,
opener, freezer). Finally, an -er nominalization can also profile the Agent
of a truly one-participant or intransitive process (e.g., sinner, laborer).
Interestingly, some -er nomináis designate an Agent which does not it-
self carry out the process, but causes someone else to carry it out, such as
yawner in Normally, certification of the votes by the Secretary of State is a
yawner (CNN-news, 27 November 2000) and pageturner 'a well-written,
exciting book' (for more examples of this type, see Ryder 1999b: 276).
They designate what is actually a second-order Agent (Halliday 1994:
172): they focus not so much on the Agent which carries out the process,
but on the entity which causes it to act. A pageturner, for instance, is a
book that is so good that it almost forces the reader (i.e. the first-order
Representational semantics 153
Agent) to turn the pages. Similarly, a weeper is a movie that makes you
weep and may, in fact, have been designed to make you weep (Panther and
Thomburg 2002: 181).
These constitute the main subtypes of agentive -er nominalization.
Cross-cutting this classification and further refining it are various other
distinctions. One can, for instance, distinguish between human Agents (e.g.,
baker, preacher), animals (e.g., wood-pecker, retriever), inanimate agent-
like entities or Instruments which designate material processes (e.g.,
toaster, stapler, scraper) and inanimate Agents based on immaterial proc-
esses (e.g., pointer 'a piece of advice or information which helps you to
understand something or find a way of making progress', reminder,
thriller) (based on Marchand 1969; see also Panther and Thornburg 2002).
Somewhat more peripheral in terms of agentivity, but nonetheless closer to
the agentive than to the non-agentive group of -er nomináis are those
nomináis that derive from mental processes or processes of sensing (e.g.,
admirer, lover, believer) (Halliday 1994). They designate the Senser of the
mental process which they are based on. Nomináis like container and
holder are derived from relational processes expressing possession and
part-whole relations and belong to the agentive rather than to the non-
agentive schematic category because they do not designate the things that
are in the container/holder, but the entity which contains/holds them. The
schematic category of agentive -er nominalization, in short, spans a wide
range of participant roles.
1.2. Non-agentive -er nominalization
Non-agentive -er nomináis fall apart into two main categories: there are
those that designate a patientive participant, and there are -er nomináis that
profile oblique participants such as Locatives and Instruments (for the no-
tion of oblique participant, see Chapter 5, Section 1.2.1). Within the first
group, a further distinction can be made between nomináis that profile pro-
totypical Patients or inert, affected entities (e.g., scratcher, squeezer),
nomináis that designate entities which merely specify the Range of the
process (Halliday 1994) (e.g., reader, leaner), and nomináis which profile
the Medium or central participant of an ergative process, which is not
purely patientive but coparticipates in the process (e.g., broiler, cooker,
steamer).50
The other group of -er nomináis can be further divided into, firstly,
nomináis profiling oblique locative participants (e.g., kneeler 'a kind of
chair which you have to kneel on', stepper 'a raised surface on which you
have to step as a way of working out' Jotter 'something which you can jot
things on or in'); and, secondly, nomináis designating instruments which
lack agent-like properties, such as walker, stroller and viewer.
Without doubt, more and finer-grained representational categories can
be distinguished among agentive and non-agentive -er nominalizations.
This is, however, as far as I will take the representational description: the
main aim of my analysis is to integrate the overall representational seman-
tics of deverbal -er nominalization with the system's constructional proper-
ties. In the following section, I will elaborate on the semantics which all
non-agentive -er nominalizations have in common, and I will propose a
semantic typology of non-agentive -er derivation much along the lines of
that which I set out for middle constructions in Chapter 6. The semantic
correspondences between non-agentive -er nominalization and middle for-
mation will then in Section 3 be shown to derive from a more fundamental,
lexicogrammatical equivalence between the Subject-Finite unit in middle
constructions and the relationship which deverbal -er suffixation estab-
lishes in non-agentive nomináis.
2. A semantic typology of non-agentive -er nominalizations
By turning a non-agentive entity into the Subject of an active Finite, mid-

dles were shown to foreground the conduciveness of the Subject-entity
towards the Agent's carrying out of the action. I claim that the link which
non-agentive -er nominalizations establish is based on a largely equivalent
mechanism: the constructional link which the -er suffix imposes on the
representational relation between entity and process also leads to a general
meaning of conduciveness which characterizes non-agentive -er nomináis.
In this section, I will discuss the various semantic profiles which non-
agentive -er nominalizations can have. In Section 3, I will show that the
semantic correspondences between middle formation and non-agentive -er
nominalization (as hinted at in Levin and Rappaport 1988 and Lemmens
1998) are more systematic than assumed till now and shed a new light on
various non-agentive -er nominalizations that were thus far considered
problematic.
In general, non-agentive -er nominalizations designate entities of which
the properties are conducive to a specific process. As pointed out by Lem-
mens (1998: 139), in the case of non-agentive -er nomináis these character-
istics often "almost predetermine" the profiled entities for a particular proc-
ess: as with the middle construction, the properties of the profiled entity
have typically been designed or intended for a particular process, but much
A semantic typology of non-agentive -er nominalizations 155
more than middle clauses, -er nomináis focus on this process, which forms
the ultimate "destination" of the profiled entity (Lemmens 1998: 138).
Think, for instance, of Levin and Rappaport's (1988: 1078) analysis of
broiler as 'a chicken bred to be broiled'. Another example is killers, desig-
nating 'cattle ready for slaughter' (Oxford English Dictionary on CD-rom,
as cited in Lemmens 1998: 138):
(1) As killing cattle or killers (cattle ready for killing) they are inferior to
corn-fed stock, (attested in 1937, Supplement to the Oxford English
Dictionary 1972)
As pointed out by Lemmens (1998: 138), killers "is not to be interpreted as

indicating that these animals possess qualities that facilitate the actual kill-
ing". Rather, killers refers to animals that have been fattened in a particular
way, so that they are suitable for being killed for their meat.
The orientation towards the process which they have been designed/bred
for seems characteristic of the large majority of non-agentive -er nominali-
zations and goes hand in hand with the nomináis' overall emphasis on the
properties of the profiled entity (see also Panther and Thornburg 2002). I
claim that the semantics of non-agentive -er nominalizations can be further
differentiated according to the facet(s) of the interaction between the non-
agentive entity and the process which they profile. This means that middle
formation and non-agentive -er nominalization show interesting parallels in
terms of the lower-level schemata which they instantiate (for the distinction
between high-level and low-level schemata and its relevance to nominaliza-
tions, see Chapter 3, Section 1.1). Unlike middle constructions, in which
adverbials and prepositional phrases can provide indications as to which
facet of the designated event is highlighted, however, non-agentive -er
nominalizations are typically characterized by a certain indeterminacy with
respect to the facet that they focus on. Most nominalizations do not contain
any overt indication as to whether they highlight the beginning (feasibility),
the middle (facility or quality) or the end (destiny or result) of the event. As
a result, they are open to more than one interpretation or they combine
various facets of the designated event in their semantics.
Only two groups of non-agentive -er nominalizations are more explicit
as to which facet they zero in on: one of them consists of nomináis such as
bestseller, easy-rider and easy-walker which specify how (well) the proper-
ties of the profiled entity let the process be carried out. The other group
includes the nomináis front-loader, top-loader and low-loader and zooms
in on where the properties of the profiled entity allow the process of load-
ing to be carried out. First, when a book or other item is called a bestseller,
what is being referred to is that it sells well/ easily: the -er nominalization
picks up on the facility- or quality-oriented meaning which is prototypically
associated with middle formation and specifies how easy it is to carry out
the process on the entity construed as Subject or how fast it is sold. As
pointed out by Lemmens (1998: 139), even a nominal like seller, in which
the modal adverbial has been left out, is oriented towards the facility/speed
with which the action of selling is carried out: if you describe a particular
thing as a seller, it is implied that it sells well. Two other examples of non-
agentive -er formations that highlight the way in which a process can be
carried out are easy-rider and easy-walker. An easy-walker is the latest
type of stroller, which has properties that facilitate walking with it because
it has only three wheels which are highly manoeuvrable. An easy-rider is a
kind of motorbike which rides easily or smoothly, as in
(2) The 1.41 rides smoothly while the 1.6 is much firmer. (CB)
Nomináis such as top-loader, front-loader and low-loader, then, do not

focus on the ease or difficulty with which the process of loading is carried
out, but rather specify where a particular entity has to be loaded to make it
serve the function it has been designed for: front-loader and top-loader
refer to a type of washing machine or clothes drier which has to be loaded
by inserting the clothing at the front or from the top. A low-loader desig-
nates a kind of vehicle of which the carrying platform is low, so that you
can load things onto it from a low position. Of course, the fact that what is
loaded onto the low-loader does not have to be lifted to a high position
greatly facilitates the loading. That is not, however, what the -er nominal
highlights: like front-loader and top-loader, low-loader primarily indicates
how the profiled entity can be put to use, by specifying where the loading
has to take place. As such, these nomináis resemble the destiny-oriented
type of middle construction, which was argued to be typically accompanied
by a Locative, as in (3):
(3) It is a little whistle that attaches to your car with self-adhesive tape.
(CB)
Most non-agentive -er nominalizations do not specify which facet they

foreground. Some of them seem to combine feasibility and facility/quality·.
a viewer, for instance, enables you to view photographic transparencies
(more easily); a walker is a frame which helps babies or disabled or ill peo-
ple to walk (more easily); a stroller allows you to take your baby out and
stroll (comfortably) with it; a sleeper in a train is a bed or sofa in which you
A semantic typology of non-agentive -er nominalizations 157
can sleep (well); a train that is referred to as a sleeper is a train which has
all the facilities to let travellers sleep.51 The various non-agentive -er nomi-
náis that designate items of clothing also seem to profile especially if {fea-
sibility) and how (facility) they allow a certain process to be carried out:
waders are boots with which you can wade; slippers are heelless shoes
which one can slip into easily.52 Other non-agentive -er nominalizations
also express feasibility, but especially foreground the fact that undergoing
the process on which the nominalization is based is what the profiled entity
has been designed for (Lemmens 1998). It seems to me that nomináis such
as kneeler, squeezer, jotter and stepper especially foreground the processes
which the entities that they profile have been designed for. A kneeler thus
designates a type of chair which is designed in such a way that you can
kneel on it; a squeezer is a kind of bottle that one has to squeeze to extract
something from it; a jotter is meant to jot things down in; a stepper is a
raised surface that one has to step on as a way of working out.
A final group of non-agentives, then, can be argued to resemble the re-
sult-oriented type of middle formation. I am thinking here of the -er nomi-
nalizations that profile items of food, such as cooker, steamer, broiler and
fryer. They designate animals and vegetables that have, in many cases,
been bred or raised to be prepared in a certain way.53 But they also imply
that, if a particular process is carried out on the item of food in question, the
result will be good. Middle structures such as those in (4), which Levin and
Rappaport (1988: 1078) associate with non-agentive -er nomináis referring
to food, primarily hint at how to prepare the food to achieve the most deli-
cious or the best possible result: the adverbial well is used here in its result-
oriented meaning. A boiler, for instance, designates a type of chicken that
is best boiled and used for soup; a cooker is an apple which is best cooked
and used for, for instance, applesauce.
(4) this type of chicken broils well: broiler

this type of apple cooks well: cooker
this type of clam steams well: steamer
this type of chicken/ rabbit fries well: fryer
In short, the semantic distinctions that were introduced for the analysis
of middle formation seem pertinent to the description of non-agentive -er
nominalization as well. Importantly, they elucidate what all non-agentive
-er nominalizations have in common, viz. the conduciveness of the profiled
entity towards a specific process and the fact that the properties of the pro-
filed entity are responsible for it. It is the properties of the easy-walker that
make it easy to walk with it; it is because it has an opening at the front, that
a front-loader can be loaded via the front; a viewer has properties that allow
you to view photographic transparencies with it; a squeezer is squeezable,
i.e. it has properties which enable you to squeeze it when you have to ex-
tract something from it; a broiler has certain properties which turn broiling
into a good way of preparing it. Like middle clauses, non-agentive -er
nominalizations may even foreground specific facets of the interaction be-
tween the process and the non-agentive entity which they profile, ranging
from the feasibility of the process, the facility or speed with which the
process can be carried out, and the place it is oriented towards, to the result
of carrying it out. However, since few non-agentive -er nominalizations
explicitly indicate the facet which they want to highlight, vagueness with
respect to the distinction between facility-, destiny- and result-orientation
seems to be part of the system of non-agentive -er nominalizations.
3. The constructional properties of -er nominalization
In this section, I will show that, if the lexicogrammatical properties of

agentive and non-agentive -er nominalizations are considered more closely,
evidence can be provided for the claim that it is the Subject profile which
constitutes the basic commonality between agentive and non-agentive -er
nominalization. Importantly, cutting across the constructional properties of
deverbal -er nominalizations is the distinction between lexicalizations and
ad hoc nominalizations: lexicalizations can be agentive or non-agentive in
profile, while ad hoc nominalizations are always agentive (Levin and Rap-
paport 1988). Before we try to offer a comprehensive account of the system
of deverbal -er nominalization in English, therefore, the lexicogrammatical
commonalities between non-agentive -er nominalizations and agentive
lexicalizations have to be analyzed and the properties of ad hoc nominaliza-
tions have to be accounted for. I will first look into non-agentive -er nomi-
nalizations and consider their relation with agentive lexicalizations. In Sec-
tion 3.2, the category of ad hoc nominalization will be described. It is not
until Section 4, then, that I consider the system of deverbal -er suffixation
in its entirety and elaborate on the generalization which I propose for it.
The constructional properties of -er nominalization 159
3.1. Lexicalized -er nominalizations
3.1.1. Non-agentive -er nominalization
In this section, I want to put forward the claim that the semantic correspon-
dences between non-agentive -er nominalization and middle formation
which I described in Section 2 reflect a more fundamental level of corre-
spondence between both construction types. I argue that non-agentive -er
nominalizations establish a relation between a non-agentive entity and a
process which is equivalent to that of the Subject-Finite unit in middle
clauses: like middle clauses, non-agentive -er nomináis designate a rela-
tionship that is characterized in terms of objective, dynamic modality, and,
like middle constructions, non-agentive -er nominalizations have a sche-
matic, epistemic modal claim mapped onto this relationship:
(5) broiler 'a chicken that broils well'

(6) bestseller 'a book that sells well'
(7) easy-rider 'a motorbike that rides easily'
First, I propose that the semantics of conduciveness which characterizes

non-agentive -er nominalizations derives from the same type of modality as
that expressed by the middle. I argued in Chapter 6 (Section 4) that the
relationship which middle constructions establish between a non-agentive
entity and a process is one of objective modality (Halliday 1970a): rather
than construing the speaker's position with regard to obligation or likeli-
hood, the modality realized in middle constructions was argued to form
"part of the content of the clause, expressing conditions on the process re-
ferred to" (Halliday 1970a: 338). It was moreover classified as an instance
of dynamic objective modality, because, like the Subject of clauses with the
dynamic modalities of volition and ability, the middle Subject forms the
(objectively realized) locus of potency directed towards the realization of
the process. Unique about the middle construction, then, was that the mid-
dle Subject does not bring about the process itself, but lets an implied
Agent bring it about. Thus, the meaning of conduciveness that typifies
middle constructions was argued to derive from the objective, dynamic
letting modality expressed by their Subject-Finite unit.
Non-agentive -er nominalization can in my opinion be argued to instan-
tiate a type of objective, dynamic modality which is similar to that of mid-
dle formation. As in middle constructions, this modality is objective or part
of the ideational content structure of the non-agentive -er nominalization,
where it significantly adds to the semantics of the process-entity constella-
tions that are designated and serves to derive a more elaborate meaning: a
cooker is therefore more than an apple that cooks and a broiler is more
than a chicken that broils. In particular, the objective type of modality that
is realized in non-agentive -er nomináis comes close to the dynamic mo-
dalities of volition and ability, because it relates a process to an entity
which is conceived as the locus of potency of the modality: the profiled
entity in non-agentive -er nominalizations may be non-agentive, but it is
the locus of potency of the process because it lets a process be carried out,
i.e. it is conducive to a specific process. A kneeler is thus conducive to
kneeling, i.e. it enables you to kneel on it; a sleeper is a train that has beds
that let you sleep; a walker is a frame that makes it possible for you to walk,
it lets you walk. Importantly, these correspondences with the middle's Sub-
ject-Finite complex suggest that non-agentive -er nominalizations can be
analyzed as profiling what is clausally construed as the locus of potency of
a dynamic modal relationship, i.e. they profile a subject-like entity.
Like middle constructions, non-agentive -er nomináis moreover imply a
schematic, epistemic modal claim or position in addition to the objective
modality of their propositional content (see Chapter 4, Section 2.3.1). As in
middle structures, the epistemic modal position of the speaker in non-
agentive -er nominalizations relates to the 'truth' of the proposition: the
relationship which is designated by non-agentive -er nomináis is presented
as 'known to be true' and it is the profiled entity that is held responsible for
the truth of it (Chapter 4, Section 2.4). In particular, non-agentives imply
that the entity which they focus on has properties that make it 'modally' or
'speech functionally' responsible for the truth of the proposition: the fact
that the profiled entity is conducive to a particular process and lets it be
carried out is, as in middle constructions, attributed to the properties of the
profiled (Subject) entity. Kneelers, walkers and sleepers thus have proper-
ties that make them conducive to the processes of kneeling, walking and
sleeping. The fact that non-agentives imply that the modal responsibility of
the relationship which they designate is assigned to the entity which they
profile is reminiscent of the modal responsibility of the Subject in the
clause. It therefore provides additional support for the claim that non-
agentives relate to the clausal Subject profile.
It is the modal value of the relationship which non-agentive -er nomi-
nalizations imply which explains why they are so-called 'non-event' nomi-
nalizations, i.e. why they do not presuppose that the process which they
nominalize has actually taken place (Levin and Rappaport 1988; Rappaport
Hovav and Levin 1992). Something can be called a squeezer before anyone
has actually squeezed in it precisely because the -er nominal makes a mo-
dal claim about the entity which it profiles: the entity is presented as having
properties that make it conducive to a particular process, but, crucially, this

does not entail that this process has already been carried out.
In section 3.1.2, I zoom in on agentive lexicalizations. Agentive -er
nominalizations do not present a problem for the Subject-generalization
which I propose: Agents are prototypically mapped onto the clausal Subject
(see, for instance, Langacker 1991; Halliday 1994). However, it is interest-
ing to see that the properties of agentive lexicalizations support the analysis
which I have suggested for non-agentive -er nominalizations: like non-
agentive -er nominalizations, agentive lexicalizations imply a relationship
between a (Subject-) entity and a process which is basically modal in na-
ture.
3.1.2. Agentive -er nominalizations
Agentive -er nominalization is typically associated with occupational terms

such as teacher, baker, preacher, life-saver, fire-fighter, trainer. Levin and
Rappaport (1988) classify these occupational terms as 'non-event' nomi-
nalizations. They argue that many of them can be used to refer to people
"before they have engaged in the activity denoted by the related verb"
(Levin and Rappaport 1988: 1070): one can, for instance, be a life-saver
and not have saved any lives yet. What seems to lie behind Levin and Rap-
paport's characterization of agentive lexicalizations as non-event nominali-
zations is, once again, the modal nature of the relationship which agentive
-er nominalizations realize.
It appears to me that, like non-agentive -er nominalizations, agentive
lexicalizations tend to establish a relationship between an entity and a proc-
ess which is basically modal in nature. The human agentive prototype of -er
nominalization (which diachronically speaking is the oldest type of -er
nominal) seems to instantiate types of modality which come close in mean-
ing to what Langacker (1991) identifies as the historical antecedents of the
modern English modals. Especially the antecedent of can, which indicated
"that its subject had the knowledge or mental ability to do something" and
that of may, which specified that the Subject "had the necessary strength or
physical ability" (Langacker 1991: 269) are of interest to the analysis of
human agentive lexicalizations such as baker, fisher, engraver, wood-cutter
and painter, all of which are found in Old English (Kastovsky 1971).
Importantly, many Modern English agentive -er nominalizations still as-
sign some degree of ability to the Agent which they profile, irrespective of
whether this is a human being, an animal or an inanimate entity. Like
clauses that realize dynamic modality, agentive -er nomináis moreover add
a schematic or epistemic claim to the objective modality of the relationship

which they designate, presenting it as 'known to be true to the speaker'. As
in clausal structures, the profiled entity is held responsible for the truth of
the proposition: it is the properties of the profiled entity which are respon-
sible for the fact that the relationship of ability is 'true'. A life-saver thus
refers to a person who knows how to save lives·, a fire-fighter designates
someone who knows how to fight a fire. A retriever is able to retrieve birds
and other animals which hunters have shot; a warbler is a kind of bird
which can warble or sing pleasantly. Likewise, a transmitter can transmit
television and radio programmes and a toaster can toast bread. It is the
modal nature of the relationship which agentive nomináis like these estab-
lish which accounts for their so-called 'non-event' reading: a life-saver is
essentially a type of person who has the ability or knows how to save lives.
He or she need not have saved any life yet.
Notice that it is implied in many agentive lexicalizations that the pro-
filed entity brings about the process regularly, a baker is someone who
knows how to bake bread, but will not be called a baker unless he or she
bakes bread regularly (professionally). When referring to animate entities,
agentive -er nominalizations in general tend to designate characteristic or
habitual behaviour: apart from the dynamic modality of ability, these nomi-
náis seem to make the epistemic claim of 'permanent habit or characteris-
tic' which can be realized by the clausal indicative as well (e.g., He bakes
bread can be understood as expressing a characteristic of the Subject, De-
clerck 1991b: 183). Nominalizations such as drinker, gambler and wood-
pecker, then again, characterize the profiled Agent in terms of a persistent
habit: a drinker is someone who will drink; a gambler is someone who will
gamble. They thus realize the dynamic modality of "strong volition" (De-
clerck 1991b: 361).
Importantly, the type of modality that seems to be characteristic of agen-
tive -er nominalizations in general is dynamic and situates the locus of po-
tency which it implies in the profiled (Subject-) entity: in agentive -er
nominalizations, the profiled entity is presented as the locus of potency
directed towards the realization of the process. Unlike in non-agentive -er
nominalizations, however, the potency associated with the profiled entity of
agentive nominalizations does not let someone else carry out the process,
but it brings the process about itself. In Talmy's terms, rather than implying
the force of letting, agentive -er nominalizations thus tend to realize that of
causing which is linked to the modalities of volition and, especially, ability.
As I pointed out earlier, the causality of agentive -er nominalizations
can be either first-order or second-order, as in laugher 'a movie or book
that makes people laugh' and pageturner 'a book that makes you turn the
pages'. Rather than being a direct causer, the profile of an agentive -er
nominalization may also be an 'instigative setting' (Vandenberghe 2001):
nominalizations such as gusher 'a flowing oilwell' and squirter 'a container
that ejects liquid in a jet from a narrow opening' relate to what have been
termed "Setting-Subject" clauses by Langacker (1991: 346), as in
(8) A supertanker continues to gush oil off the coast of Spain. (CB)
(9) Tank is non-functional and does not squirt water. (CB)
It is argued in Vandenberghe (2001) that Setting-Subject constructions like

those in (8) and (9) are based on ergative verbs (Davidse 1992b) and allow
for an apparent setting to be used in Subject-position. By being construed
as Subject, however, the (agent-like) instigative force that characterizes the
setting in its function of Emitter of the flow of the substance, is fore-
grounded. As an Emitter, the setting then "either instigates the motion as
part of its functioning, or due to its malfunctioning as a container, for in-
stance a crack or hole ..." (Vandenberghe 2001: 10).
In conclusion, agentive lexicalizations have in common with non-
agentive -er nominalizations that the relationship between Subject and verb
which they imply tends to be a modal one, with the profiled entity or Sub-
ject as the locus of potency. Characteristic of agentive -er lexicalizations,
then, is that rather than profiling a subject-like entity which lets an Agent
carry out the process, they profile an agentive Subject or a subject-like en-
tity which brings about the process itself.
3.1.3. A note on instrumental -er nominalization
Interestingly, the differences and correspondences between agentive and

non-agentive -er lexicalizations also shed a new light on the unique posi-
tion of the large group of instrumental -er nominalizations within the sys-
tem of deverbal -er nominalization. Instrument-denoting nomináis have
often been analyzed as inanimate extensions of the human agentive proto-
type of -er nominalization (e.g., Marchand 1969; Ryder 1991, 1999b).
Many of them can indeed be related to clausal agent-like participants, as
they are mapped onto the Subject function:
(10) a computer computes data

a printer prints data
a transmitter transmits data
a carrier carries people and things
However, I pointed out before that a significant number of instrumental

-er nomináis cannot be attributed agent-like properties. Nomináis like
knocker and poker, for instance, seem to be most acceptable in an instru-
mental with meaning (Ithe knocker knocks on the door; Ithe poker pokes
the wood in the fire, but: you knock on the door with the knocker, you poke
the wood with the poker). In the case of stroller, viewer and walker, an
agentive reading is totally excluded: a stroller does not stroll·, a walker does
not walk and a viewer does not itself view.
Now that the agentive and the non-agentive type of lexicalized -er
nominalizations have been described and their respective properties have
been accounted for, it becomes clear that instrumental -er nomináis find
themselves in a unique position in the system of deverbal -er suffixation:
instruments by definition hover between being able to carry out a process
themselves (as agent-like participants) and letting others carry it out (as
non-agentive entities). Those instrumental -er nomináis that are non-
agentive thus resemble middle constructions in that they profile an entity
that has properties that let an implied Agent perform a particular action
(e.g., stroller, viewer, walker). Agent-like instrumentais, on the other hand,
foreground the agent-like ability of the tool which they refer to (as in
transmitter, toaster).
A large group of instrumental -er nomináis lies in between the agentive
and the non-agentive type of -er nomináis: depending on which perspective
is chosen, they can be conceived as agentive or as non-agentive. Even In-
struments which strongly depend on the intervention of a human Agent in
order to fulfill their function can, in certain contexts, be presented as agent-
like, witness the following attestations ofpeeler and grater:
(11) All Purpose Peeler - Peels fruit and vegetables quickly and safely.
Simply the best. (CB)
(12) ... a clever stainless steel double grater that grates ingredients finely
or coarsely.... (CB)
Instrumental -er nominalizations, in short, embody one of the funda-

mental choices that is offered by -er suffixation, i.e. that between an agen-
tive and a non-agentive, letting profile. The fact that their position in the
system of deverbal -er nominalization is significantly elucidated by the
relationship of agnation which I propose for non-agentives and middle con-
structions and by the modal nature of agentive lexicalizations, once again
provides support in favour of a middle-related analysis of non-agentives
and a Subject generalization for deverbal -er nominalizations in general.
In the next section, I will show that the analysis of ad hoc nominaliza-
tions helps to complete the picture of the Subject-Finite-like relationship
which deverbal -er nominalizations encode. While lexicalizations were
shown to prefer modalized Subject-Finite relationships, it will be argued
that ad hoc nominalizations exploit the other option which the Subject-
Finite unit in clauses offers and designate primarily temporally grounded
instances of a process type.
3.2. Ad hoc nominalizations
I pointed out earlier that the distinction between lexicalized -er nominaliza-
tions and ad hoc formations is centred on differences in the degree to which
-er nominalizations are 'entrenched': while lexicalizations are sufficiently
automatized to be themselves included as units or fixed expressions in the
language system, ad hoc formations are derived 'ad hoc', on the basis of a
schematic unit or structural pattern (Chapter 3, Section 1.2). Especially
interesting for my analysis of deverbal -er suffixation is that lexicalizations
and ad hoc formations also differ in their semantic and formal properties.
As to the meaning of ad hoc nominalizations, I referred earlier to Levin and
Rappaport's (1988) 'event' interpretation, according to which ad hoc nomi-
nalizations designate events or processes that have actually occurred, in
contrast with the so-called 'non-event' value of lexicalizations. Regarding
the formal properties of ad hoc nominalizations, I have in Chapter 5 (Sec-
tion 2) pointed to the frequent occurrence of adverbial-like adjectives (e.g.,
He is a hard worker, Mackenzie 1990: 137), the impossibility to change the
accent position of the verb (e.g., He is a PHOtographer of snow-scenes,
Mackenzie 1990: 137), and the frequent use of postmodifying structures
(e.g., the Welsh destroyer of Del Harris's hopes, CB).
I will argue that, in spite of their semantic and structural differences,
lexicalized and ad hoc -er nominalizations instantiate the same high-level
constructional schema of deverbal -er nominalization. Like lexicalized -er
nominalizations, ad hoc ones can be shown to establish a relationship
which is similar to that which in the clause is realized between the Subject
and the Finite. Of the options that are available within the Subject-Finite
constellation, however, ad hoc nominalizations tend to choose options
which are partly different from those realized in lexicalizations. Ad hoc
nominalizations, in other words, realize a low-level schema which is differ-
ent from that of lexicalizations. More precisely, unlike lexicalizations, ad
hoc nominalizations tend to establish a relationship between an entity and a
process which is primarily temporal (see also Strang 1968, 1969).
To map out the properties of ad hoc nominalizations and establish their

precise position within the system of deverbal -er nominalization, it is im-
portant to distinguish between two types of ad hoc nominalizations. Ad hoc
nominalizations seem to follow either one of two strategies:
ished. The police came and hauled off the fighter, after which the van-
isher promptly reappeared, laughing. (Ryder 1999b: 283)
(14) I'm not a frequent resigner. (CB)
In (13), the -er nominalizations vanisher and fighter have definite, specific
reference and refer back to participants which were introduced in the pre-
ceding discourse context. I will call them 'phoric ad hoc nominalizations'.
Phorie ad hoc nominalizations differ from ad hoc nominalizations such as
resigner in (14), which is indefinite and part of a non-referential nominal:
rather than establish the presence of a referent in the universe of the dis-
course, it ascribes a quality to the entity construed as Subject of the copular
clause (Kuno 1970). In what follows, I will consider the properties of both
types of ad hoc nominalizations more closely.
3.2.1. Phoric ad hoc nominalizations
What characterizes phoric ad hoc nominalizations like the ones in (13) is

that they realize specific reference to a referent that has been established in
the discourse or in the situational context. The identity of the referent of ad
hoc instances can typically be retrieved exophorically (i.e. in the context of
situation) or anaphorically (in the preceding text) (Halliday and Hasan
1976). An example of a nominalization which refers to an instance that can
only be identified in the context of situation is when someone who is listen-
ing to an opera asks Who is the singer? (Jespersen 1914-1929, 6: 231). The
nominalization singer then refers to a specific instance in the situational
context, i.e. the one who is singing now.54 The nomináis in (15) illustrate
anaphoric reference or reference to an entity that was introduced in the
preceding discourse (see also 13):
(15) One of them was faking. ... Could the faker keep up free association
...? The faker, whichever he was, had practised or had natural talents.
(Kastovsky 1986b: 410)
Phorie ad hoc nominalizations are characterized by the fact that they

designate an actual relationship which is situated within the temporal
framework of the discourse. Rather than being entrenched common nouns
with a conventional, modal meaning, they establish an ad hoc relationship
which depends on the context for its interpretation and precise temporal
grounding. Unlike non-phoric ad hoc nominalizations (which I will discuss
in Section 3.2.2), they realize a relationship which is always temporally
grounded. The nominal singer in Who's the singer?, for instance, refers to
the one who is singing (now). The nomináis fighter and vanisher in (16)
designate the one who was fighting and the one who had vanished respec-
tively.
ished. The police came and hauled off the fighter, after which the van-
isher promptly reappeared, laughing. (Ryder 1999b: 283)
It is the temporally grounded status of phoric ad hoc formations like these

which lies, I think, behind Levin and Rappaport's (1988) notion of 'event'
nomináis.
The nominal strategy which phoric ad hoc nominalizations use to estab-
lish specific reference is that of definitely grounded common nouns with
relative quantification. Phoric ad hoc nominalizations cannot opt for the
nominal strategy of indefiniteness. The context always makes clear what
the relevant reference mass is. In the case of the nomináis in (15), for in-
stance, the ad hoc formation faker quantifies universally over the reference
mass (which consists of only one instance, as indicated by one of them) by
profiling a quantity that is identical to it. Other examples which illustrate
the definite status of phoric ad hoc nominalizations are:
(17) "Not him," says Mrs. Cook. "This one, though. A bit of a poseur, but
he moves very nicely." I will not describe the nice mover, to spare his
blushes, but let us say that he had dressed for the part. (CB)
(18) Recent research has established that merely breathing smoke-laden air
- passive or 'second-hand' smoking - puts the breather at risk.
(CB)
Lexicalized -er nominalizations, in contrast, can be used with the vari-

ous grounding strategies that are available for ordinary common nouns as
well and they are either definite or indefinite. They can be definite and have
relative quantification, as illustrated by the second occurrence of teacher in
(19):
(19) I was a teacher at an approved school and I saw a staff member beat
up a boy. I was asked to turn a blind eye but I couldn't and because of
that I was shunned by colleagues who expected me to support the
teacher. (CB)
The nominalization the teacher has definite reference and refers back to the
staff member mentioned earlier. The contextually defined reference mass of
teacher thus consists of only one instance, over which the teacher quanti-
fies universally. Lexicalizations can also have indefinite reference. As
pointed out in Chapter 4, when it is used indefinitely, the nominal type
specification does not define a reference mass, but serves a purely "classifi-
cational" function (based on Davidse 2000b: 1113). Indefinite lexicalized
nomináis may present the designated entity as indefinite, but specific (and
therefore paraprasable as 'a certain X', Dirven p.c.), as in:
(20) But it was when a teacher realized the lead singer of the group,
Freddie Mercury, was a bisexual who had died of AIDS last fall, that
the controversy began. (CB)
Often, however, no specific instance is referred to, as in (21):
(21) At my children's school, they're looking for a teacher.
Phorie ad hoc formations, which designate specific, contextually identifi-

able instances, differ most radically from lexicalized -er nominalizations.
The difference between lexicalizations and ad hoc formations that are used
non-referentially is more subtle: the non-phoric ad hoc nominalizations
which I will discuss in Section 3.2.2 in some respects come closer to lexi-
calizations than to phoric ad hoc nominalizations. They are not only indefi-
nite, but the type of relationship which they designate is not always tempo-
ral and can imply modal grounding as well. Indefinite ad hoc formations
thus seem to be non-entrenched derivations which are modelled closely
after lexicalizations.
3.2.2. Non-phoric ad hoc nominalizations
In this section, I will show that nomináis such as resigner in I'm not a fre-
quent resigner are situated in between phoric ad hoc -er nominalizations
and true lexicalizations. Unlike the former, they have indefinite grounding,
but in contrast with the latter, they tend to make their type specification
more informative, by elaborating it with pre- and postmodifiers.
As argued in Davidse (2000b, 2001), the type specification in indefinite
constructions has a purely classificational function and it "has to enable the
hearer to conceptualize instances corresponding to the categorization used
by the speaker" (Davidse 2001: 13). For this reason, the speaker will take
care to use a type specification which is sufficiently identifiable or clear. In
the case of ad hoc formations, I would suggest, this tends to lead to more
elaborate type specifications: because these -er formations are not them-
selves entrenched or lexicalized units, they tend to be turned into more
elaborate, higher-order type specifications. The higher-order type specifica-
tion in ad hoc formations is typically realized by means of explicit markers:
as in ordinary nomináis, a higher-order type specification is derived when
pre- and/or postmodifying structures are added to the head noun, i.e. the -er
derivation as such:
(22) ... he could swing the ball and was a precocious taker of wickets.
(CB)
(23) She was, in that relatively brief time, a girlfriend, a pregnant woman,
... the carrier of an unwanted second child,.... (CB)
(24) I was too self-absorbed to be a giver of pleasure. (CB)
(25) To get herself back out, she befriends a myriad of oddballs including
... a professional loser of boxing matches. (CB)
A more elaborate, higher-order type specification may also be derived con-

textually, without the inclusion of pre- or postmodifiers. In that case, the
borderline between ad hoc formations and lexicalizations is reached.
Let's start with the use ofpremodifìers in ad hoc formations. Apart from
ordinary epithets, indicating some quality of the type of thing that is being
profiled (as in 26 and 27), indefinite ad hoc -er nominalizations frequently
take classifying adjectives, which indicate the subclass of the entity being
designated (Halliday 1994: 184). The fact that these -er derivations are
deverbal noun type specifications may explain the frequent use of what
Mackenzie (1990: 137) calls "adverbial adjectives" or classifying adjec-
tives that "do not ascribe a quality ... but rather indicate the manner, fre-
quency or extent of the activity", as in (28), (29) and (30):
(26) Minnie is no sentimental praiser of the past. (Jespersen 1914-1929, 6:

229)
(CB)
(28) Legal secretary about her boss: I swear, the moment I need to talk to
Max, he's suddenly gone. I'm beginning to think he's a professional
vanisher. (Ryder 1999b: 282)
(29) Ask a frequent flyer like Margie Crase .... (CB)
(30) I'm not & frequent resigner. (CB)
The use of adjectives such as frequent, regular, professional is notewor-

thy in that they can be argued to realize lexically what many 'entrenched'
-er nominalizations incorporate in the grounding that is implied in them,
viz. the fact that a specific process is carried out habitually, regularly or
professionally. In ad hoc -er nominalizations, the relationship that is estab-
lished is temporal, but notions of habituality and professionality can be
expressed by adjectives. As noted above, these adjectives serve to derive
higher-order type specifications: in the nominal a frequent flyer like Margie
Crase, for instance, frequent derives the higher-order type specification
frequent flyer and Margie Crase is identified as an instance of the type of
people that can be classified as frequent flyers. Likewise, in (28), the he-
person is identified as an instance of the class of professional vanishers.
Type specifications such as frequent flyer and professional vanisher may
not be entrenched or conventional, but semantically they often function in a
way which is very similar to lexicalized -er nominalizations.
What is special about the postmodifying elements in ad hoc formations
is that they tend to be of the restrictive type, i.e. they restrict "the head
noun's type specification" (Langacker 1991: 432): they provide informa-
tion which is essential for the -er derivation to develop into the higher-
order or more elaborate type specification which is necessary for it to per-
form its classifying function, associated often not just with indefinite
grounding in the nominal, but also with the clausal function of predicative
Attribute as in (31) to (33):
(31) You don't have to be a passionate lover of opera to adore Verona's

opera season. (CB)
(32) Miss Garden was no giver of confidences. (CB)
(33) John Kamm is head of Occidental Chemical in Hong Kong and a
frequent traveller to China. (CB)
Unlike what has been suggested in, for instance, Levin and Rappaport
(1988) and Mackenzie (1990), postmodification is also possible in nomi-
náis with lexicalized -er nouns. However, because the head of such nomi-
náis is a lexicalized -er noun, the postmodifier may in that case serve vari-
ous functions: it may be restrictive and derive a higher-order type specifica-
tion which is still more elaborate than the type specification of the lexicali-
zation (as in 34). Or it may be non-restrictive, as in the nominalization in
(35):
(34) ... Allen has told his lover of 12 years that when the custody hearing
begins next week there would be nothing of her left standing. (CB)
(35) The teacher, of West London, was led away to the cells .... (CB)
Notice that some of the more elaborate type specifications, consisting of an

ad hoc -er derivation followed by a postmodifier, seem to have become
entrenched or 'lexicalized' themselves, and are therefore explicitly included
in the dictionary (that it is not necessarily words that can become en-
trenched or lexicalized was already pointed out in Chapter 3, Section 1.2).
The pattern 'be a lover o f , as in As a young man, he was a lover of high-
speed sports (CB), is thus included in the Collins Cobuild Dictionary, in
addition to the lexicalization lover 'someone with whom one has a sexual
relationship'. Another example is reminder 'letter sent to tell you that you
have not done something, such as pay a bill' and 'be a reminder of as in
The British are about to be given a sharp reminder of what fighting abroad
really means (Collins Cobuild Dictionary, 1995).
As argued above, the frequent inclusion of postmodifying structures and
of premodifiers in ad hoc nominalizations seems to be motivated by the fact
that for non-entrenched, ad hoc -er derivations to function as noun type
specifications, it is necessary to derive a higher-order, more elaborate type
specification first. When ad hoc nominalizations do not take any (explicit)
pre- or postmodifiers, they are difficult to distinguish from true lexicaliza-
tions and only differ from them in that they are not entrenched. Typically,
they exploit the habitual meaning which is implied in some agentive lexi-
calizations. Consider, for instance, the clause Sinatra told the judge his old
friend was 'a giver, not a taker'. The nominalizations giver and taker do
not belong to the entrenched units of the language system and, as illustrated
in (36) and (37), they typically need elaboration if they are to function as
noun type specifications:
(36) Miss Garden was no giver of confidences. (CB)

(CB)
In his oldfriend was 'a giver, not a taker ', however, they are used as if they
are lexicalizations and designate the type specification 'someone who has
the habit of giving/taking'.
The fuzziness of the categories which non-phoric ad hoc formations re-

alize suggests that they constitute the transition area between phoric ad hoc
formations and entrenched nominalizations: they are modelled after lexical-
izations, but they are not sufficiently automatized yet to function as lexical-
ized units. They thus illustrate that lexicalizations and ad hoc nominaliza-
tions "mark the end points of a functional scale and are completely dissoci-
ated only in certain extreme cases" (Kastovsky 1986b: 412). Like phoric ad
hoc derivations, the relationship which they imply is often temporal in na-
ture, as is shown by Levin and Rappaport's (1988: 1070) examples of life-
saver and saver of lives : Levin and Rappaport in my opinion correctly point
out that a life-saver need not have saved any lives yet, while a saver of lives
as in He is a saver of many lives implies that he has saved many lives.
Some non-phoric ad hoc nominalizations, however, come closer to being
lexicalizations in that they establish a relationship which has modal charac-
teristics. Being agentive in profile, ad hoc nominalizations then mostly
imply the modality of ability, as illustrated by the nominalizations in (38)
and (39):
(CB)
3.2.3. Ad hoc -er nominalization: Conclusion

To conclude, ad hoc -er nominalizations represent derivations which are
not entrenched, fixed expressions of the language system (and hence not
evoked unconsciously, Dirven p.c.), but which are consciously derived ad
hoc, to serve a specific discourse need. They turn out to imply temporal
grounding and they follow either one of two nominal grounding strategies.
Ad hoc nominalizations are either definitely grounded, phoric nomináis and
imply a relationship between a particular, context-identifiable entity and a
process which receives specific temporal grounding in the discourse con-
text. Or they are used indefinitely (typically non-phorically) and may also
imply modal grounding. Indefinite ad hoc nominalizations designate in-
stances of a type which, in order to discharge its 'classifying' function,
tends to be elaborated by pre- and/or postmodifiers.
3.3. The constructional properties of -er nomináis: Summary
To summarize, I have successively discussed the lexicogrammatical proper-

ties of non-agentive -er nominalizations, agentive lexicalizations, and
phoric versus non-phoric ad hoc nominalizations. Two of their properties
stand out: first, all of them imply a Subject-Finite relationship, but the exact
nature of it differs for each of them; and, secondly, the types of nominal
grounding which the nomináis containing the -er derivation receive also
add to our understanding of these different types.
As for non-agentive -er nominalizations, the various semantic corre-
spondences between the system of middle formation and non-agentive -er
nominalization were shown to be based on a fundamental lexicogrammati-
cal equivalence: like middle constructions, non-agentive -er nouns were
argued to establish the objective, dynamic modality of letting. The relation-
ship of agnation between middle constructions and non-agentive -er nomi-
nalizations thus turned out to elucidate the properties of the latter signifi-
cantly and it helped to identify the clausal categories which non-agentives
imply (see Chapter 3, Section 5.3). Agentive lexicalizations were also
found to have objective, dynamic modality, viz. that of ability or volition.
They were moreover shown to typically realize the force-dynamic model of
causality. Ad hoc nominalizations all seem to imply temporal grounding. If
they are used phorically, they are necessarily temporal. When used in a
non-phoric way, they were shown to come close to being lexicalizations
and, in some cases, imply modality.
Lexicalized -er nominalizations can be used with specific, definite ref-
erence, as in (40a) and (40b) and with specific (41a) or non-specific (41b)
indefinite reference; or they are non-referential, as in (42a) and (42b):
(40) a. Mrs. Emerson, the second-grade teacher of Tony .... (CB)

b. Various crashing noises erupted from our pew as our new friend
searched for his lost gold earring beneath the wooden kneelers.
(CB)
(41) a. But it was when a teacher realized the lead singer of the group,
Freddie Mercury, was a bisexual who had died of AIDS last fall,
that the controversy began. (CB)
b. In that town, there's a diner where the portions are gigantic.
(42) a. I was a teacher at an approved school .... (CB)
b. Is it an eater or a cooker and how long does it keep? (CB)
Phoric ad hoc nominalizations choose the specific, definite type of the

common noun strategy:
(43) One of them was faking. ... Could the faker keep up free association
...? The faker, whichever he was, had practised or had natural talents.
(Kastovsky 1986b:410)
Finally, non-phoric ad hoc nominalizations seem to be associated especially

with indefinite, non-referential nomináis:
Thus, the nominal strategies of grounding of-er derivations seem to corre-

late to a certain extent with the types distinguished on the basis of the kind
of Subject-Finite relationship which the -er noun implies: in lexicalized -er
nouns the clausal layer is modal and the nominal layer is typically indefi-
nite; in phoric ad hoc formations, on the other hand, the clausal layer is
temporal and the nominal layer turns out to be more typically definite.
In the following and last section of this chapter, I will attempt to put
more semantic flesh on the Subject-generalization which I have proposed
for deverbal -er nominalizations.
4. Deverbal -er suffixation as Subject-profiling
This section is aimed at identifying what all deverbal -er nominalizations

have in common. In previous sections, I have foregrounded the various
possible Subject-Finite relationships which the noun type specification
formed through deverbal -er suffixation can imply and the range of repre-
sentational roles and processes that can be mapped onto these relationships.
Different subtypes of -er nominalization were identified on the basis of
correspondences with distinct Subject-Finite relations such as modal and
temporal grounding in agnate clausal constructions. Here I want to describe
the impact of -er suffixation in general by zooming in on what all -er
nominalizations have in common. To come to a schematic characterization
of -er nominalization, it is essential that a distinction be made between the
various Subject-Finite relations which -er nouns have in their base, on the
one hand, and the actual profile of -er formations, on the other: different
types of Subject-Finite relationships may form the base of an -er nominali-
zation, but -er suffixation always foregrounds the Subject-entity of these
relationships. Unlike clauses, -er nominalizations do not profile the proces-
sual relationship between the Subject, Finite and clausal head, but they
zoom in on the entity that is involved in it (which itself comes to function
as a noun type specification that can be instantiated and grounded in the
Deverbal -er suffixation as Subject-profiling 175
speech event). It is therefore in the Subject profile that the semantic com-
monality of all -er nominalizations lies.
I pointed out in Chapter 5 that the notion of Subject has figured in the
analysis of -er nominalization before, but then as a more or less empty no-
tion, the meaning of which is primarily linked to the representational role
that is prototypically mapped onto it, viz. that of Agent (e.g., Marchand
1969; Bauer 1983; Keyser and Roeper 1984; Levin and Rappaport 1988;
Mackenzie 1990; Rappaport Hovav and Levin 1992). Halliday's multifunc-
tional analysis of the clause, however, makes clear that the functions of
Subject and Agent form part of two distinct layers of organization (see
Chapter 2, Section 4). The Subject function clusters together with the vari-
ous options which are available to the language user to ground the proposi-
tion and assume a modal position with regard to it. As such, the Subject
serves an 'interpersonal' function. The Agent, on the other hand, expresses
how the speaker experiences the world which he describes and forms part
of the 'representational' metafunction of language. If we are to come to an
accurate account of the Subject function, and, related to it, of the profile of
-er nominalizations, therefore, the Subject function has to be described in
terms of the interpersonal functions which it serves.
I argued in Chapter 4 that for the description of the function of Subject,
a syntagmatic or configurational analysis is required which considers the
different steps in the organization of the clause as a construction related to
the speech event, viz. the functions of process type specification, quantifi-
cation and grounding. I thus identified three distinct functions of the clausal
Subject: it was argued to function as instantiator of the clausal process type
specification; it was shown to co-operate with the finite element of
grounded clauses by establishing objective or explicit person deixis; and,
finally, it was identified as the modally responsible element of the clause
(Halliday 1994).
I suggest that what deverbal -er nominalizations profile is comparable to
the function of the Subject in clauses. Firstly, -er nominalizations charac-
terize people and things in terms of a particular process type which they
instantiate. -Er suffixation applies to a process type specification and es-
sentially represents a mechanism whereby a process type (given by the base
verb) is instantiated by tying it to an entity (marked by -er). As in the
clause the instantation is not necessarily causal in nature but may also in-
volve other types of force, such as as that of 'letting'. Secondly, -er nomi-
nalizations imply categories of grounding and thus also profile the 'gram-
matical person' to which the process is tied. More specifically, the person
deixis which they establish is that of the 'third person' (Davidse p.c.). Fi-
nally, like the clausal Subject, the entity which comes to be profiled
through -er suffixation is held 'responsible' for the proposition that is im-
plied in -er nominalization: in it is vested the truth of the proposition. The
entity either has a certain potential for carrying out a process or for letting
others carry it out (in lexicalizations); or it is an entity that has done/ is
doing/ will do something (in phoric ad hoc nominalizations). Lexicalized
-er nominalizations in addition typically profile entities in which, as in the
Subject of the dynamic modals of ability, volition and letting, a certain
potency is located: the profiled entity has the necessary skills/ knowledge/
physical appearance/ design/ inherent properties to carry out or let others
carry out a process. In phoric ad hoc nominalizations that establish a rela-
tionship of which the precise temporal status has to be retrieved from the
context no such potency is implied.
5. Conclusion
In conclusion, by analyzing the correspondences between the clausal Sub-

ject-Finite unit and -er nominalization more closely, a new perspective on
-er nominalization has been offered. While so far, its semantics had been
approached as representational semantics only, I have shown that the most
fundamental generalization behind -er nominalization is constructional: -er
nomináis profile an entity that is related to a process much like a clausal
Subject is related to the finite verb. In agentive -er nomináis, the profiled
entity is typically also the 'doer' of the process. In non-agentive -er nomi-
náis, by contrast, we can observe the added meaning that is associated with
the Subject of the clausal middle construction, viz. that of conduciveness.
By positing a relationship of equivalence between non-agentive -er
nominalizations and the Subject-Finite unit in middle formation, a new light
was shed on many longstanding controversies: non-agentive cases of -er
nominalization (patientives, as well as oblique participants) were accounted
for and identified as full members of the system of deverbal -er nominaliza-
tion, rather than as idiosyncratic or peripheral formations. The large group
of instrumental -er nomináis was described more accurately as hovering
between being able to carry out a process itself (like agentive -er nomináis)
and enabling others to carry it out (like non-agentives). And, finally, the
analysis of the Subject-Finite complex of middle constructions also re-
vealed the commonalities between agentive and non-agentive lexicaliza-
tions, viz. their dynamic modal status and the identification of the Subject
as the locus of potency of the modality. The force-dynamic characteristics
of the modal relationships which agentive and non-agentive -er nominaliza-
tions establish, then again, revealed their basic difference (i.e. letting vs.
Conclusion 177
causing). Finally, the modal value of lexicalized -er nomináis was shown to
contrast with the predominantly temporal focus of ad hoc nominalizations.
To conclude, a multifunctional approach to deverbal -er nominalization,
or an approach which is not restricted to the representational properties of
the system, but includes the constructional categories which it realizes and
the textual functioning of the nouns that result from it, manages to offer a
new perspective on the mechanisms of -er derivation. It unveils the interac-
tion between the nominal and clausal interpersonal categories of type speci-
fication, instantiation and grounding and it shows that the high-level se-
mantics of -er nominalization is identifiable only in terms of these catego-
ries.
Part III
Factîve nominalization
Chapter 8
Factive nominalization: Towards a descriptive
position
0. Introduction
The gerundive, that- and the fact i/zaZ-constructions that I want to focus on
in Chapter 9 are of the following types:
(1) To this day he regrets that one of the unanticipated side effects of the
Welfare State was the disappearance of granny and grandpa from the
family hearth into the old people's homes. (CB)
(2) I don't like his having to do so much .... (CB)
(3) One characteristic regarding farm holidays is the fact that the different
offers are so varied. (CB)
Structures like the ones in (1) and (2) have traditionally been described as
complement clauses, or as clauses functioning as Subject or Object in an-
other clause (e.g., Kiparsky and Kiparsky 1971; Menzel 1975; Noonan
1985; Wierzbicka 1988).55 All three construction types are typically charac-
terized as being 'factive', a feature generally thought of as the "presupposi-
tion by the speaker that the complement of the sentence expresses a true
proposition" (Kiparsky and Kiparsky 1971: 345).
Strikingly, the system of factive nominalization tends to be primarily as-
sociated with ¿Aaí-structures, which are conceived as prototypical instances
of factive nominalization. As to the position of gerundive constructions and
the fact /Aai-structures, opinions diverge. Lees (1968: 59-60), for instance,
includes only ίΑαί-structures in his category of 'factive nomináis':
(4) That he came was obvious.
Yet, in his analysis of 'gerundive nomináis' he acknowledges that some of

them are factive in meaning (Lees 1968: 71-72):
(5) His eating vegetables is surprising.
The Kiparskys (1971), on the other hand, seem to view gerundives (as in
6b) primarily as 'tests' for identifying factive predicates and their that-
182 F active nominalizatioti: Towards a descriptive position
complements: "Only factive predicates allow the full range of gerundial

constructions ... to stand in place of the that-clause" (Kiparsky and Kipar-
sky 1971: 346). The fact ^/-constructions (as in 6c) are likewise primarily
viewed as periphrastic structures testing for the factivity of /Aa/-structures
(Lees 1968: 65; Kiparsky and Kiparsky 1971: 346-347):
(6) a. that the dog barked during the night

b. the dog's barking during the night
c. the fact that the dog barked during the night
In the transformational literature on factives, it has even been suggested

that the head noun fact is present in the so-called 'deep structure' of all fac-
tive nominalizations, also in those that lack a factive head noun in their
'surface structure' (Kiparsky and Kiparsky 1971: 356; Stockwell et al.
1973: 584; Menzel 1975: 89).
Halliday (1994: 266, 269) characterizes the fact /^/-constructions
straightforwardly as constituting a subtype of factive nominalization. Ge-
rundives he analyzes as borderline cases between factive and non-factive
complementation: he says that there exists only a minimal distinction and
perhaps even blending between structures like
(7) She liked the snow falling, [factive]

(8) She watched the snow falling, [non-factive]
Halliday consequently focuses exclusively on the fact that- and that-

structures in his discussion of factive constructions. Davidse (1991, 1994),
finally, takes Halliday's analysis one step further and lists a number of ar-
guments in favour of excluding gerundives from the category of factive
complementation all together, basically arguing that factive nominaliza-
tions are necessarily derived from grounded clauses (Davidse 1991: 355).
Why is it that /Aa/-nominalizations are generally regarded as factive pro-
totypes, while the linguistic status of other nominalized constructions with
a factive meaning is contentious? The existing analyses of factive nomi-
nalization have all tended to concentrate on one construction type. I will
consider factive gerundives, /¿art-structures and the fact /Aa/-constructions
together and show that it is only by taking into account the structural prop-
erties of each of these agnate constructions that one can chart the options
that exist within the system of factive nominalization and gain more insight
into the 'factive presupposition'. In this chapter, I will sketch the most im-
portant analyses that have thus far been given of both the external and the
internal functioning of the various types of factive nominalizations. In
Factivity as truth presupposition 183
Chapter 9 I will then propose a new, radically functional description of the

internal properties of factive nominalizations.
1. Factivity as truth presupposition
Without doubt the best-known attempt to describe factivity is that by the

Kiparskys (1971). They argue that factivity arises when the "speaker pre-
supposes that the embedded clause expresses a true proposition, and makes
some assertion about that proposition" (Kiparsky and Kiparsky 1971: 348).
They show that the notion of factivity distinguishes verbs such as regret,
resent, and ignore, which are factive and combine with factive comple-
ments, from non-factive predicates such as claim, assert, suppose (Kipar-
sky and Kiparsky 1971: 345-347). They observe a similar distinction
among copular clauses: while clauses with be significant, odd, exciting are
factive in meaning, adjectives such as likely, possible, true and false give
rise to a non-factive reading:
(9) a. I regret that it is raining, [factive]

b. I suppose that it is raining, [non-factive]
(10) a. It is significant that he has been found guilty, [factive]
b. It is likely that he has been found guilty, [non-factive]
Two aspects of the analysis of factivity presented in Kiparsky and Ki-

parsky (1971) are especially noteworthy: first, the Kiparskys' analysis re-
veals a correlation between the factive category and specific formal fea-
tures and it points to a number of formal differences between factive and
non-factive complementation. Secondly, it suggests that factivity results
from a combination of features which relate to the factive nominalization
itself, the predicate of the clause in which it is embedded and the role of the
speaker in the clause.
First, Kiparsky and Kiparsky list a number of agnates of factive com-
plements which help to distinguish factive complements from non-factive
ones. As I pointed out earlier, //¿«/-constructions are claimed to be system-
atically replaceable by the fact íAírt-structures and gerundive nominaliza-
tions (see 6) (Kiparsky and Kiparsky 1971: 346). The Kiparskys also argue
that, unlike factive nominalizations, most non-factive nominalizations can
have their Subject turned into the Subject of the main clause, as illustrated
in (1 lb) (Kiparsky and Kiparsky 1971: 346):
(11) a. It is likely that he will accomplish even more, [non-factive]

184 Factive nominalization: Towards a descriptive position
b. He is likely to accomplish even more.

But c. It is relevant that he has accomplished even more, [factive]
d. *He is relevant to have accomplished even more.
Factive nominalizations cannot be turned into an 'accusative and infinitive

construction' either, as is made clear in (12d) (Kiparsky and Kiparsky
1971: 358):
(12) a. I understand that Bacon is the real author, [non-factive]

b. I understand Bacon to be the real author.
But c. I regret that Bacon is the real author, [factive]
d. *I regret Bacon to be the real author.
Moreover, extraposition is shown to be optional in clauses with factive

complements, but obligatory in those with non-factive complements (Ki-
parsky and Kiparsky 1971: 346):
(13) a. That there are porcupines in our basement makes sense to me.
[factive]
b. It makes sense to me that there are porcupines in our basement.
But c. *That there are porcupines in our basement seems to me. [non-
factive]
d. It seems to me that there are porcupines in our basement.
Finally, Kiparsky and Kiparsky (1971: 362) show that both factive and
non-factive clauses can be replaced by it, which is the "pro-form of noun
phrases" (see 14a and 14b), while only non-factive clauses can take so, or
the "pro-form of sentences", as illustrated in (15):
(14) a. John supposed that Bill had done it, and Mary supposed it, too.
[non-factive]
b. John regretted that Bill had done it, and Mary regretted it, too.
[factive]
(15) a. John supposed that Bill had done it, and Mary supposed so, too.
[non-factive]
b. *John regretted that Bill had done it, and Mary regretted so, too.
[factive]
What is interesting about the Kiparskys' approach is not only that they
situate factivity within a specific formal paradigm. Important is also that
they do not restrict the notion of factivity to one structural element only: in
Factivity as embedded projection 185
their discussion of the factive constructions in (16) and (17), for instance,
the Kiparskys first state that the "factive sentence ... carries with it the pre-
supposition 'it is raining'" (Kiparsky and Kiparsky 1971: 348). Then they
continue by saying that the "speaker presupposes that the embedded clause
expresses a true proposition" (Kiparsky and Kiparsky 1971: 348). Their
conclusion, finally, is that "predicates which behave syntactically as fac-
tives have this semantic property" (Kiparsky and Kiparsky 1971: 348).
(16) It is odd that it is raining.

(17) I regret that it is raining.
As Delacruz (1976) points out, the Kiparskys thus use the notion of the fac-
tive presupposition in three different senses: to refer to the sentence as a
whole, to the main verb that functions in it, and, thirdly, to the speaker ut-
tering the clause (Delacruz 1976: 180). And, of course, the complement or
nominalization itself is also called 'factive' (Kiparsky and Kiparsky 1971:
356). There is no reason why the four interpretations of factivity should be
incompatible.56 In fact, they illustrate that, rather than being traceable to
one structural unit only, factivity results from afusion of several factors.
The Kiparskys may point to a number of interesting lexicogrammatical
differences between clauses which are factive in meaning and clauses
which are not, but the precise relationship between the formal characteris-
tics of the factive category and the notion of factivity is never made clear.
As Davidse (1994: 260) puts it, "tracing systematically different paradig-
matic affiliations to establish distinct constructions is α heuristic technique,
not α grammatical explanation". The notion of factivity should be linked
up with the category's grammatical behaviour, and because they refrain
from doing so, the Kiparskys ultimately fail to shed light on it. The Kipar-
skys' characterization of facts as propositions that are 'presupposed true'
thus remains vague and requires further elaboration.
2. Factivity as embedded projection
The analysis of factive constructions given in Halliday (1968, 1994) and

Halliday and Hasan (1976) is centred on two basic constructs, viz. that of
embedding (vs. hypotaxis), on the one hand, and that of projection (vs. ex-
pansion), on the other. The Kiparskys' factive paradigm is in Halliday's
approach linked to the system of embedded projection. What do the notions
of embedding and hypotaxis, projection and expansion stand for?
186 F active nominalization: Towards a descriptive position
2.1. Embedding vs. taxis
First, the term 'embedding' is not new: it is also found in mainstream lin-
guistics, where it is used in a meaning which is closely associated with that
of 'complementation': embedded clauses are defined as complements or
subordinated clauses that function as arguments of another clause (Noonan
1985: 42). Traditionally, the /^-clauses in (18) and (19), for instance, are
both analyzed as embedded clauses or complements (examples taken from
Noonan 1985: 90):
( 18) I think that Floyd didn 't hit Roscoe.

(19) I regret that Floyd didn't hit Roscoe.
Halliday, however, posits a fundamental difference between clauses such as

those in (18) and (19) and he uses the notion of embedding for the that-
clause in (19) only. He claims that the undifferentiated concept of 'subordi-
nation' which is used in mainstream linguistics actually covers two distinct
types of clausal relationships, viz. embedding and hypotaxis.
In Halliday's view, embedded clauses are clauses that have been rank-
shifted in that they are made to function at a rank which is lower than that
of the clause (see Chapter 3, Section 3.2.1). Halliday argues that it is only
embedded or rankshifted clauses which can function as constituents within
another clause and serve a function in it: the clause [that Floyd didn't hit
Roscoe] in (19) thus forms an embedded constituent within the clause I re-
gret that Floyd didn't hit Roscoe and serves the function of Object in it.
Embedded or rankshifted clauses should be distinguished from hypotac-
tic clauses (as in 18). Instead of being embedded or structurally integrated
as a constituent in another clause, hypotactic clauses depend on the clause
with which they are used (Halliday and Hasan 1976: 136, 221). While a
sentence such as I regret that Floyd didn't hit Roscoe thus consists of only
one clause which has another one embedded in it, a clause such as I
think/said that Floyd didn't hit Roscoe consists of two clauses, of which the
second is dependent on or hypotactically related to the first. Hypotaxis is,
in other words, a relation between clauses that form a 'clause complex',
whereas embedding involves a simple clause in which another clause is
downranked. Notice that, in addition to being hypotactic, the relation be-
tween clauses in a clause complex can also be paratactic (traditionally re-
ferred to as 'coordination'). Unlike hypotactic clauses, paratactic clauses
are equal in status: there is no dependent and dominant clause, but only an
'initiating' and a 'continuing' one (Halliday 1994: 218). An example of
paratactic clause combining, Halliday (1994: 250) argues, is formed by

quotes:
(20) Brutus said: 'Caesar was ambitious'.
The claim that the traditional notion of subordination actually covers

two distinct types of clausal relations has been made by Halliday as early as
1968. Various other studies have since suggested that a distinction should
indeed be made between, on the one hand, part-whole relationships or em-
bedding, and, on the other hand, part-part or tactic relationships (i.e. para-
taxis and hypotaxis) (McGregor 1997). With regard to quotes, for instance,
it has been argued that "the quoted sentence is not syntactically or semanti-
cally a part of the sentence that contains it" (Partee 1973: 418; see also
Munro 1982). As to the construct of subordination, Haiman and Thompson
(1984: 510) argue that it is "at best a negative term which lumps together
all deviations from some 'main clause' norm, and thus treats as unified a
set of facts which we think is not a single phenomenon". They point out
that the systemic-functional framework is 'exceptional' in that it has been
careful to distinguish between hypotaxis and embedding (Haiman and
Thompson 1984: 521; see also Matthiessen and Thompson 1988; Hopper
and Traugott 1993). Before I consider the arguments that can be given in
favour of Halliday's distinction between embedding and taxis and before I
turn to the analysis of factive nominalizations as embedded constituents,
another distinction has to be introduced, viz. that between projection and
expansion.
2.2. Projection vs. expansion
Cross-cutting the categories of embedding and taxis are those of projection

and expansion. Halliday (1994: 219) argues that the relationship between
clauses within a clause complex or, in the case of embedding, within a sim-
ple clause cannot only be described in terms of dependency and embed-
ding, but also in terms of more general semantic concepts. He posits the
existence of two fundamental types of semantic relationships between
clauses, viz. those of expansion and projection.
When clauses are combined and each clause represents "the phenomena
of experience themselves" (Halliday and Hasan 1976: 131), they are said to
be related through expansion. Expansion thus implies that each of the
clauses involved in the relationship forms a "direct representation of (non-
linguistic) experience" (Halliday 1994: 250). In the clause complex When-
ever the horse stopped, he fell o f f , for instance, both clauses are direct rep-
resentations of experience. When, on the other hand, one of the clauses in
the relationship represents experience not directly but instead designates
"linguistically processed phenomena" (Halliday 1968: 195), Halliday pre-
fers to speak of projection. Projections are metaphenomena (Halliday 1994:
249): they are representations "of a (linguistic) representation" (Halliday
1994: 250). They represent "phenomena already encoded in language"
which then participate in other linguistic structures (Halliday and Hasan
1976: 131). Among the most important subtypes of projection which Halli-
day distinguishes are the following (Halliday 1994):
(21) 'Was Caesar ambitious?' asked Mark Antony, [quote]

(22) Mary said/thought that she wanted to go home, [report]
These clause complexes differ in terms of the process type that functions in
their main clause and in terms of the relationship that exists between the
two clauses which they consist of: quotes are paratactically related to
clauses which typically contain a verbal process (i.e. asked in 21). Reports
are hypotactically related to the 'dominant' clause: in (22), that she wanted
to go home thus depends on the main clause Mary said/thought. Reports
can follow mental processes (e.g., thought), as well as verbal processes
(e.g., said). When they are construed with a mental process, they are tradi-
tionally referred to as 'indirect speech' representations.
2.3. Facts as embedded projections
Halliday (1968, 1994) analyzes facts as projections which are not tactically
related to the clause with which they are used, but are embedded in it as a
constituent. Facts are said to typically combine with clauses that contain
mental and relational processes, and, less frequently, verbal processes. An
example of a factive nominalization that functions in a relational clause or a
clause that sets up a relation between two entities (Halliday 1994: 119) is
given in (23):
(23) Mr. Archer said it was significant that Queensland, a growing sector
in the Australian property industry, had been granted this important
recognition. (CB)
Mental processes taking factives typically express a feeling of affection

(Halliday 1994: 118):
(24) I regret the role of mother isn't generally more highly regarded by
everyone. (CB)
Finally, in some cases, also verbal processes can be followed by a fact:
(25) Its owners cheerfully admitted that the bird was mad, and seemed
genuinely offended when I suggested it needed to be put down. (CB)
Characteristic of factive nominalizations is that they are not projected by

the process of the clause in which they function, but come "as it were ready
packaged in projected form" (Halliday 1994: 264). A clause with a factive
nominalization in it thus represents some form of interaction with a pre-
packaged proposition: a fact "is an entity interacted with, or mentally ma-
nipulated", it is "the source of a specific affect ... or something which is
cognitively or linguistically reacted to", as in Don't forget [that you are a
Christian] and He didn 't want to admit [that he was a Christian] (Davidse
1994: 276). Non-factive projections, in contrast, are projected by the proc-
ess of the clause which they combine with: rather than being factive, the
projected clause gives the 'content' of the locution or idea that is implied in
it. In sentences such as those in (26) and (27), the ¿Ααί-structure is thus hy-
potactically projected by the verbs think and point out and it specifies the
thoughts and locutions of the main process's Subject; in the factive con-
structions in (28) and (29), on the other hand, the ¿Ααί-constructions are not
projected by but responded to or interacted with by the Subject of the main
verb:
(26) David thinks that all championships would be better run on Grand
Prix lines. (CB)
(27) King pointed out that Meredith's admission was merely a crumb
thrown black Americans in lieu of real equality for all. (CB)
(28) Probably you resent that peace demonstrations include kids who can
wait out the war in college. (CB)
(29) I regret that it all happened. (CB)
Where do embedded projections come from, then? Whose ideas do they

represent? In Kiparsky and Kiparsky (1971) it was argued that facts are
presupposed true by the speaker, i.e. that it is always the speaker who is
committed to the truth of the factive proposition. As pointed out by Davidse
(2003), however, 'speaker factivity' is only one possible interpretation of
factivity and two other types of factivity need to be recognized, i.e. 'proc-
esser factivity' and 'speaker-processer factivity'. Where do these notions
come from and what do they stand for? In his (1976) article, Delacruz ob-
serves that the sentence Bill regrets that John resigned can receive a read-
ing in which both the speaker and Bill regard the proposition John resigned
as a fact, as well as a reading in which John resigned is a fact to Bill (i.e.
the processer of the mental process regret) only. Building on Delacruz's
observation and on the basis of a corpus study of factive constructions,
Davidse (2003) suggests that there are three basic interpretations of the no-
tion of factivity and that these interpretations can be linked to a number of
general semantic schemata. Firstly, facts may be facts for the speaker only
when they function in discourse contexts that imply no other consciousness
than that of the speaker (e.g., in relations of 'addition' or 'causality',
Davidse 2003; e.g., Add to that the fact that over half the menus the NRA
analyzed include sandwiches, and you 've got a trend worth noting, CB). In
environments which either represent or imply a 'second consciousness'
distinct from the speaker's, by contrast, facts may be facts either to both the
speaker and the processer that is interacting mentally with the proposition
or to the processer only. In The council have woke [sic] up to the fact that
Glasgow and a lot of areas are slums (CB), for instance, the use of the verb
wake up conveys that the speaker was subscribing to the factivity of the
proposition that Glasgow and a lot of areas are slums before the council
came to accept it (Davidse 2003). A pure 'processer fact', on the other
hand, can be found in What the 37 year-old finds most soul destroying
about Haiti is the fact that things have not changed since his childhood
(CB): the processer the 37year-old is depicted here in his or her emotional
reaction to a proposition which s/he perceives as a fact but which is not
necessarily a fact to the speaker (Davidse 2003). Note that even facts that
are facts to the processer only represent propositions which are not pro-
jected by the clause with which they combine - as are reports and quotes -
but which are 'pre-existent to' (Davidse 2003) and 'manipulated by' the
relation in which they participate.
By characterizing facts as projections, Halliday manages to elucidate
their commonality with clause complexes that have typically figured in the
discussion of factives, i.e. those involving direct and indirect speech and
thought. It is the status of factive nominalizations as embedded constituents
which turns out to be crucial to distinguish them from the other types of
projection. In what follows, I will go more deeply into the arguments that
have been given in the literature to support the analysis of factives as em-
bedded in rather than hypotactically related to the clause with which they
combine. The discussion is based on Halliday (1968, 1994), Halliday and
Hasan (1976) and Davidse (1991, 1994).
F activity as embedded projection 191
Halliday's claim that factive clauses are embedded entails that they are
viewed as downranked constituents which serve a function in another
clause. Which arguments can be given in favour of the downranked status
of factive structures? Evidence that ί/ζαί-factives fill a nominal slot in the
clause comes from the observation that, unlike their hypotactic counterparts
in (31), the nominalized ίΑαί-constructions in (30) can be coded as Subject
of the passive (as in 30a), they can be given thematic prominence by being
fronted (e.g., 30b), and they can be extraposed as Objects, while the Object
slot is filled by it (see 30c):
(30) a. That they had to wait in line was resented by people.

b. That they had to wait in line people resented.
c. Most people resented it that they had to wait in line.
(31) a. *That they had to wait in line was insisted by Drew.
b. *That they had to wait in line Drew insisted.
c. *Drew insisted it that they had to wait in line.
Factive that-structures also seem to be semantically entitized or reified:

as Davidse (1994: 271) points out, "there is a strong general tendency for
facts to be replaceable by nominal groups": while in (32) and (33), for in-
stance, the /Aai-factives can be said to systematically relate to ordinary
nominal groups, the hypotactic ί/ζαί-structures in (34) and (35) do not allow
for such an alternation (examples based on Davidse 1994: 271):
(32) a. Most people resented that they had to wait in line.

b. Most people resented the long waiting period.
(33) a. Tom regrets that they 're getting divorced.
b. Tom regrets their divorce.
(34) a. Drew insisted that they had to wait in line.
b. *Drew insisted the long waiting period.
(35) a. Sue thinks that they 're getting divorced.
b. *Sue thinks their divorce.
The conceptual reification of the clause in ίΑαί-factives is, moreover, ac-

companied by internal structural behaviour which is different from that of
ordinary or non-rankshifted clauses. First, downranked ίΑαί-structures are
limited in terms of the thematic freedom which they allow their constituent
parts. Unlike hypotactically projected that-clauses (as in 36), for instance,
that-factives (as in 37) cannot have their main verb thematized (Hooper
1975: 99):
(36) a. Drew insists that wait they shall/will.

b. Sue thinks that divorce they will.
(37) a. *People resent that wait they will.
b. *Tom regrets that divorce they will.
Another structural characteristic which distinguishes embedded from hypo-

tactic /^/-structures is that the former readily accommodate non-sequent
tenses or deictically absolute tenses relating directly to the speaker-now
rather than being coded relative to the projecting clause (Kiparsky and Ki-
parsky 1971: 359). An example is Ijust remembered you are a socialist, in
which the factive /Aa/-clause is not grounded relative to the main process
remembered, but has direct deictic ties to the speaker-now.
It should be observed that the fact that //zai-factives can be fitted in with
the nominal paradigm is not to say that they do not also differ from non-
nominalized or ordinary nomináis. 77aa/-factives are nominalized to a large
degree, but being embedded clauses, they will always be non-prototypical
nomináis. Unlike with ordinary nomináis, for instance, the use of factive
iAa/-structures as Subject-nominal in interrogatives is restricted (e.g., Kuno
1973: 363; Noonan 1985: 82; Huddleston 1988: 145fn; Davidse 1994:
270):
(38) ? Is that she did it now known? (Davidse 1994: 270)

(39) ? Is that Floyd left town significant? (Noonan 1985: 82)
As far as extraposition is concerned, by contrast, that-factives are more

flexible than ordinary nomináis in that they can be extraposed:
(40) It's significant that Floyd is drunk. (Noonan 1985: 83)

(41) It makes sense to me that there are porcupines in our basement. (Ki-
parsky and Kiparsky 1971: 346)
Moreover, as is pointed out by Davidse (1994: 270), a factive /Aa/-structure

like that in (42) does not allow for the passive:
(42) That you work so hard pleases us.

*We are pleased by that you work so hard.
In spite of the fact that they do not take over all of the functions which an
ordinary nominal can fulfill in the clause, however, factive /Λα/-structures
clearly function as nomináis and their embedded or nominal constituent
status shows up in many of their lexicogrammatical properties.
F activity as embedded projection 193
Because it fails to distinguish embedding from hypotaxis, some of the

early literature on factive complementation is riddled with examples of
what are actually non-factive clauses or clauses that are hypotactically re-
lated to the main clause. Examples of hypotactic structures that have thus
wrongly been situated in the factive paradigm are, for instance, the follow-
ing constructions cited in Lees (1968: 59-60):
(43) I know that he came.

(44) I told her that he saw us.
It also becomes clear now that the 'tests' which the Kiparskys use to
distinguish factive from non-factive clauses primarily pick up on the dis-
tinction between hypotaxis (non-factive) and embedding (factive). That
factive clauses cannot be substituted by so, for instance, has to do with their
embedded, nominalized status: as shown by Halliday and Hasan (1976:
131-134), substitution can only be used with the same function and class as
the wording which it substitutes for. In the case of hypotactic clauses, so is
possible because the clause which it replaces serves the function of clause.
By contrast, embedded or rankshifted clauses have lost their clausal status
and function first and foremost as nomináis in another clause (Davidse
1994: 274). Reference, in contrast, does not depend on whether the refer-
ence item has the same function or class as the clause which it refers to
(Halliday and Hasan 1976; Davidse 1994). It is therefore possible to have
substitution in (45), but not in (46). In (46) only reference is acceptable:
(45) Do we really have to wait in line? It seems so. Drew says so.
(46) I regret that you had to wait in line. I regret it/*so too.
Further, it is the nominalized status of factive clauses that makes front-

ing possible, either as Subject of the passive, or as a marked Theme of the
clause (Davidse 1991: 337). Hypotactic structures do not normally have the
same flexibility, because they do not represent constituents of the clause,
but depend on it (see 48):
(47) That he had been wrong was admitted by him.

(48) IThat he had been wrong was said by him.57
Finally, the fact that factive nominalizations cannot have their Subject
turned into the Subject of the main clause (as shown in 49a) or figure in
'accusative and infinitive constructions' (as illustrated in 49b) is a direct
result of their embedded status as well (Davidse 1991: 354). Because fac-
tive clauses are rankshifted into nominal slots as constituents of the clause,
they cannot move beyond the boundaries of the nominal element of struc-
ture.
(49) a. *He is relevant to have accomplished even more, [factive]

b. *I regret them to have accepted, [factive]
2.4. Conclusion
Halliday's interpretation of facts as embedded projections is innovative and

elucidates the relationship between the meaning of factivity and the lexico-
grammatical properties of factive constructions. By identifying facts as pro-
jections, Halliday manages to shed light on the resemblances which factive
constructions show with clause complexes in which speech and though act
verbs figure (e.g., factive He resents that she did it vs. non-factive He
says/thinks that she did it). Halliday's approach thus situates fact clauses
within their natural habitat, i.e. that of metaphenomena, ideas or linguisti-
cally processed phenomena, as opposed to direct representations of the way
in which the language user experiences reality. Importantly, by breaking up
the undifferentiated mainstream concept of 'subordination' into embedding
and hypotaxis, Halliday also succeeds in singling out the main difference
between facts and projecting clause complexes with reports: evidence is
provided in favour of the embedded, constituent status of facts and the hy-
potactic, dependent status of reports. Factive fAai-clauses turn out to be
downranked nominalized constituents which display much of the flexibility
of ordinary nomináis with regard to assuming Subject or Ob-
ject/Complement position in the clause and they are to a certain degree also
semantically entitized. Rather than being 'presupposed true by the speaker',
it was argued, facts are propositions which are 'pre-existent' to the relation
in which they participate (Davidse 2003). Their being pre-existent rather
than projected by the main clause is reflected in their embedded status. The
concept of embedded projection thus significantly elaborates and refines
the Kiparskys' somewhat vague characterization of facts as being 'proposi-
tions that are presupposed true by the speaker'.
However, Halliday's account of factive nominalization also leaves many
questions unanswered. I see two main lacunae, both of which are basically
concerned with the delineation of the category of factive nominalizations:
first, the distinction between hypotaxis and embedding is not always clear-
cut and it is sometimes hard to draw the line between truly factive (embed-
ded) projections and non-factive (hypotactic) projections. Secondly, Halli-
Delineation of the fact category 195
day focuses exclusively on finite factives (i.e. on ίΑαί-constructions and the

fact /Aa/-nominals) and leaves the category of gerundive nominalizations
undiscussed. Yet, the analysis of gerundive nominalizations is highly perti-
nent to the description of factivity and the borderline between factive ge-
rundives or gerundives which are embedded projections, on the one hand,
and non-factive gerundives, on the other, is at times hard to draw. In what
follows, I will briefly elaborate on some of the remaining problems with
respect to the delineation of the category of factive nominalization defined
as embedded projection.
3. Delineation of the fact category
The delineation of the category of factive nominalization is problematic in

several respects. First, there exist constructions which combine properties
of both the factive and the non-factive category of predicates. Constructions
belonging to this type have been called 'semi-factives' (e.g., Karttunen
1971; Hooper 1975). There also exist a large number of ambiguous predi-
cates or predicates that can be interpreted either factively or non-factively.
Finally, the analysis of gerundive nominalizations poses problems of de-
lineation as well.
3.1. Ambiguity and vagueness
Kiparsky and Kiparsky (1971: 360-361) point out that there exist numerous
predicates that are ambiguous and have both a factive and a non-factive
use. A sentence like that in (50a), for instance, is said to have a factive and
a non-factive reading, evidence for which is provided by the minimal pairs
in (50b) and (50c) (Kiparsky and Kiparsky 1971: 360):
(50) a. They reported that the enemy had suffered a decisive defeat.
b. They reported the enemy to have suffered a decisive defeat, [non-
factive]
c. They reported the enemy's having suffered a decisive defeat,
[factive]
The sentence in (50c) implies "that the report was true in the speaker's
opinion" (Kiparsky and Kiparsky 1971: 360). It is factive and consists of a
systematic agnate of the factive type of ¿/¿«/-structure, viz. a gerundive
nominalization. The to-infinitive in (50b), on the other hand, "leaves open
the possibility that the report was false" (Kiparsky and Kiparsky 1971:
360). As I argued before, the accusative and infinitive construction is only
possible with non-factive clause complexes, because it requires a direct
relationship between two clauses. The Kiparskys' claim that the projected
clause is not presupposed true and may be 'false' picks up on the hypotactic
nature of the clause complex: the clause the enemy to have suffered a deci-
sive defeat is hypotactically projected by the process of the main clause and
represents the projection of the main clause Subject, rather than that of the
speaker. The hypotactic clause in (50b) thus implies that the fact that the
enemy had suffered a decisive defeat is only what they reported; it does not
necessarily coincide with the speaker's opinion.
An example of a verb which can be used in different verb senses and
which, depending on the meaning in which it is used, is either factive or
non-factive is explain (Kiparsky and Kiparsky 1971: 361):
(51) I explained Adam's refusing to come to the phone, [factive]

(52) I explained that he was watching his favorite TV show, [non-factive]
In (51), explain is used in the meaning of 'give reasons for', while in (52) it
means 'say that S to explain X' (Kiparsky and Kiparsky 1971: 361). In Hal-
liday's model, (51) is considered factive because the structure Adam's re-
fusing to come to the phone functions as embedded projection: rather than
being projected or directly asserted by the clause I explained, the I-person
expresses some assertion about the (pre-existent) proposition. The that-
clause in (52), on the other hand, is a non-factive, hypotactic projection of
the main predicate: it is directly projected by the Subject of the main proc-
ess and represents the locution that is implied in explain.
The Kiparskys rightly observe that the list of verbs with both a factive
and a non-factive reading is quite extensive: many constructions are am-
biguous, even those that contain what seem to be prototypically factive
verbs like regret·, the sentences in (53) and (54), for instance, differ in
meaning from the one in (55): regret is factive in (55) and expresses a kind
of affect, with the that-clause an embedded projection. In (53) and (54), on
the other hand, the ίΑαί-clause is hypotactically projected by regret, which
is used in the meaning of 'regret to say'. The /Aa/-clauses in it consequently
realize a locution rather than a fact.
(53) Irma regrets that she cannot answer your letters personally. (CB)
(54) The Times also regrets that enclosures accompanying letters cannot
be returned. (CB)
(55) He trained during the Gulf War and even regrets that he was never
called up to take part in any action. (CB)
Apart from ambiguous predicates, there exist also predicates which are
characterized by vagueness and which have features in common with both
the factive and the non-factive category. Among the processes that have in
the literature been argued to belong to this type are verbs describing "proc-
esses of knowing or coming to know" (Hooper 1975: 117), such as know,
discover, find out, learn, and predicates such as deny, doubt, it is possible,
it is true, it is false (see, among others, Kiparsky and Kiparsky 1971; Kart-
tunen 1971; Hooper 1975; Wierzbicka 1988).
(56) Anyone who has seen Internet knows that it uses amazing technology.
(CB)
(57) ... she also found out that her father has been in touch with Derek all
this time. (CB)
(58) ... he discovered that Sharon's parents, both known drug takers, had
been repeatedly attacking their daughter. (CB)
(59) The agent of a sacked Coronation Street star denies that she is on the
verge of a nervous breakdown. (CB)
(60) It is true that in the Seventies there was no shortage of 'stupid gym-
nasties', as the Bhagwan once called the act of lovemaking. (CB)
(61) ...it is possible that this year we may suffer again. (CB)
It has repeatedly been pointed out in the literature that predicates like the
ones in (56) to (61) are located somewhere in between 'true' factives and
non-factives. The process know has thus been described as "semantically
factive", but "syntactically non-factive" (Kiparsky and Kiparsky 1971:
348).58 It is, for instance, impossible to say *I know the fact that John is
there or *I know John's being here, which seems to suggest that know is
non-factive. Also, know takes the accusative and infinitive construction,
which is normally only possible with hypotactic or non-factive structures
(e.g., I know him to be there, based on Kiparsky and Kiparsky 1971: 348).
Similar observations have been made with regard to it is possible and it is
true·, like factive predicates, they cannot have their Subject raised to the
Subject-position of the main clause (e.g., It is possible that he will accom-
plish even more cannot be turned into *He is possible to have accomplished
even more). Like factives, moreover, they may or may not take extraposi-
tion (e.g., It is possible that we may suffer again or That we may suffer
again is possible). Unlike factives, however, clauses construed with predi-
cates such as it is possible and it is true do not alternate with the fact that-
constructions and gerundives (e.g., *The fact that we may suffer again is
possible/true·, *John's being ill is possible/true) (Kiparsky and Kiparsky
1971).
For one thing, structures like these show that the distinction between
hypotactic projection and embedded projection is not always clear-cut and
that analyses of particular configurations are likely to confront us with the
existence of categories that lie in between hypotactic and embedded projec-
tion. Constructions like the ones in (56) to (61) thus reveal the need for fur-
ther clarification of the way in which factive /^-structures are related to
the clause with which they combine and ultimately, their semantic and
structural properties will have to be integrated in a description of factive
nominalization that is aimed at being exhaustive. My own analysis of fac-
tive nominalization, however, will focus on the internal functional proper-
ties of the factive category, rather than on the delineation of the external
contexts in which it occurs.
3.2. Gerundive nominalizations: Acts or facts?
The problems related to the delineation of the factive category within the
group of gerundive nominalizations were already pointed out by Lees
(1960) in his groundbreaking study on nominalization in English. In his
analysis of what he calls 'gerundive nomináis', Lees (1968: 58) shows that
a distinction should be made between the following two subtypes:
(62) a. Eating vegetables is healthful.

b. * Eating vegetables was a great surprise.
(63) a. His having eaten vegetables was a great surprise,
b. *His having eaten vegetables is healthful.
The nominalization in (62) is said to illustrate a type of gerundive nominal

that is 'action-referrent'. It is said to be characterized by the fact that it
"cannot have any expressed subject, nor any auxiliary" (Lees 1968: 72).
The structure his having eaten vegetables in (63), on the other hand, is clas-
sified with the 'fact-referrent' type of gerundive nominalizations. It is said
to occur "with or without the genitive subject ... or the auxiliaries" (Lees
1968: 72). Lees repeatedly stresses that the contrast between both types of
gerundive nomináis is "as yet imperfectly understood" (Lees 1968: 58) and
that "little beyond these bare facts is as yet known about this further dis-
tinction among these nomináis" (Lees 1968: 72).
Lees's description of gerundive nomináis makes clear that problems of

delineation can be expected in the case of nomináis that lack an explicitly
realized Subject: such nomináis can be either action-referrent or factive.
When subjectless gerundive nomináis also lack auxiliaries, it becomes even
more difficult to identify their precise import, as is shown in the examples
in (64) to (67):
(64) I think my only regret, as I say, was not being able to train for the job
I wanted to do. (CS)
(65) Threatening people will get you nowhere. (Halliday 1994: 248)
(66) Tearing up my new paper dolls was mean. (Thompson 1973: 378)
(67) Gathering quietly around the campfire each evening was nice.
(Thompson 1973: 377)
The situation becomes even more complicated when it turns out that, unlike
what Lees claims, the action type of gerundive nomináis (i.e. the type of
gerundive nominal that is not factive but 'action-referrent') can in some
cases take a possessive or periphrastic Subject, as in:
(68) Techniques for doing this include the caregiver's exposing the infant
to various contingent and discrepant experiences. (CB)
(69) My husband speaks very well, but his job involves my answering the
phone on his behalf quite a bit of the time. (CB)
Lees's analysis of gerundive nominalizations is interesting in many re-

spects. First, it reveals that within the category of gerundive nominaliza-
tions, a further distinction can be made between different subtypes, and that
the identification of these subtypes is problematic. The existence of an ac-
tion-referrent and a factive type of gerundive nominalization and the prob-
lems that are associated with their delineation show that gerundive nomi-
nalizations are highly interesting to the analysis of factive nominalization
and explore the borderline between factivity and non-factivity.
Secondly, Lees's criteria for distinguishing between factive and non-
factive gerundives are intriguing but clearly require further elaboration.
Importantly, the notion of 'embedded projection' as defined by Halliday
(1994) does not suffice to further clarify the distinction either. Especially
the notion of 'projection' turns out to be insufficiently understood to make
a precise demarcation possible between embedded projections and embed-
ded expansions: while the difference between projection and expansion in
hypotactic clause complexes is fairly straightforward, in the case of embed-
ded constructions, it is not always obvious whether a structure represents a
direct representation of experience or a metaphenomenon. Do the embed-

ded structures threatening people and gathering quietly around the camp-
fire each evening in (65) and (67), for instance, represent direct representa-
tions of experience or are they facts, i.e. ideas or linguistically processed
phenomena? And which lexicogrammatical properties motivate their inter-
pretation? Like the occurrence of ambiguous and vague predicates, the
problems of delineation which characterize the category of gerundive
nominalizations thus reveal the need to further refine the analysis of factive
nominalization as embedded projections.
Finally, the analysis which Lees gives of action and factive nominaliza-
tions may not suffice to draw the line between factivity and non-factivity,
but it does make clear that an analysis of the internal structural properties
of factive nominalizations might be able to contribute to their analysis as
embedded projections or facts. By pointing to the tendency for action ge-
rundives to occur without Subject and auxiliaries, while factive gerundives
can take auxiliaries and frequently include a Subject, Lees suggests that
there exists a correlation between a nominalization's meaning and its inter-
nal structural behaviour. My approach to factive nominalizations builds on
Lees's insights and will largely consist in analyzing the internal, nominal
properties of the various types of factive nominalizations. In the following
section, I will therefore consider the proposals that have thus far been made
with respect to the internal properties of factive constructions.
4. Towards an internal, nominal analysis
The interest in the external structural behaviour of factive nominalizations

and the attempts to map out the various contexts in which factive clauses
can be embedded, pioneered by Kiparsky and Kiparsky (1971), have eluci-
dated many aspects of the system of factive nominalization. One of the
drawbacks of the focus on the external functioning of factive clauses has
been the lack of attention devoted to their internal structural properties. The
internal features of factive constructions have mostly been left unanalyzed,
and have not been integrated with the way in which the nominalization
functions externally as a nominal constituent. They have, moreover, tended
to be treated as irrelevant to the semantic interpretation of factive nominali-
zations.
If the internal structural properties of factive nominalizations are con-
sidered, it is the presence of clausal features which is typically highlighted:
factives are conceived of as "imperfect nomináis" or as nomináis "in which
the verb is still alive as a verb" (Vendler 1967: 131). 7%a/-structures in par-
Towards an internal, nominal analysis 201
ticular are analyzed as being fully clausal internally. Their nominalized

status is attributed solely to the fact that they function as nomináis exter-
nally and that they can serve as Subject and Object in the clause: as Kruis-
inga and Erades (1953: 108) put it, embedded /Aa/-structures show "no con-
nection between function and form". The various labels that have been used
in the literature for embedded iAa/-structures are all meant to capture this
apparent contradiction between their internal, clausal properties and their
external, nominal-like behaviour: /Aaf-structures have been referred to as
'noun clauses' (Sweet 1891: 170), 'Subject/Object clauses' (Poutsma 1929;
Kruisinga 1932; Kruisinga and Erades 1953), 'clausal nominalizations'
(Comrie 1976; Cortine and Thompson 1985) and 'nominal clauses' (Quirk
et al. 1985: 1047).
Unlike ίΑαί-structures, gerundive nomináis have typically been argued
to have an internal structure that is partly nominal.59 The discussion of the
internal structure of factive gerundives has, however, generally been re-
stricted to listing its clausal and nominal aspects (an elaborate overview of
the nominal and verbal features of gerundive nomináis can be found in,
among others, Lees 1960; Chomsky 1970; Pullum 1991). With the excep-
tion of Lees (1960), linguists generally seem to have tacitly accepted that
the internal structure of factive gerundives, like that of ί/ζαί-factives, does
not contribute to the factive meaning with which it is associated. Analyses
that consider the internal properties of the various types of factive nomi-
nalizations and try to link them up with the factive meaning of the category
are sparse. In the next section, I will successively discuss the analyses of-
fered by Halliday (1994) and by Langacker (1991) and show that both of
them stand out in that they try to reconcile the nominal-like external behav-
iour of factive nominalizations with the properties of their internal organi-
zation.
4.1. Halliday's dependency analysis
As I have shown earlier, Halliday's analysis of factive nominalizations as

embedded constituents is primarily elaborated in terms of the relation that
exists between an embedded structure, on the one hand, and the larger con-
figuration which it forms part of, on the other. Also included in his descrip-
tive analysis of factives, be it somewhat less prominent, is an analysis of
the internal structure of embedded ¿Λαί-structures. Behind it is the idea that
an embedded clause functions in the structure of a nominal, i.e. as a "con-
stituent WITHIN" it (Halliday 1994: 242). More specifically, Halliday pos-
202 Factìve nominalization: Towards a descriptive position
its that a factive /¿¿¡/-structure either functions as head of the nominal, or as

postmodifier to a head noun (Halliday 1994: 242, 266):
(70) That you 're sorry isn't enough.

(71) The fact that you 're sorry isn't enough.
Huddleston (1988: 145fn) criticizes Halliday's internal analysis of that-

factives and argues that, rather than functioning as head of a nominal, that-
structures are immediately Subjects or Objects in the clause. The main ar-
gument which Huddleston gives against Halliday's analysis is that embed-
ded finite structures "cannot enter into construction with any of the Pre- or
Postmodifiers that are found in genuine nominal groups" (Huddleston
1988: 145fh). McGregor (1997) voices a comparable criticism. As I pointed
out in Chapter 3 (Section 3.2.1), McGregor argues that downward rankshift
necessarily involves internal reclassification, whereby the downranked unit
"takes on the grammatical and semantic properties inherent to the lower
ranking unit" (McGregor 1997: 127). Unlike non-finite clauses, McGregor
argues, finite clauses do not show any sign of reclassification: semantically
speaking, a finite clause designates not a thing, but an ongoing phenome-
non located at some point in space and time; structurally speaking, it is not
characterized by any reduction in the number of grammatical options that
are available for anchoring the situation to a specific occurrence. McGregor
concludes that /¿«/-structures are not internally reclassified and cannot
therefore be analyzed as rankshifted.
Even though Huddleston has a point when he argues that /¿«/-structures
cannot take pre- or postmodifying structures and should not therefore be
analyzed as heads, the alternative analysis which he proposes basically goes
back to the early definitions of embedded /¿¿//-structures as Subject or Ob-
ject clauses that show no correlation between function and form (e.g., Pout-
sma 1929; Kruisinga 1932; Kruisinga and Erades 1953). McGregor's
analysis likewise results in inexplicable discrepancies between the external
functioning of finite embedded clauses (e.g., as Subject in [That he did it]
came as no surprise) and their internal structural features, which are
claimed to be those of a non-rankshifted clause. By analyzing finite clauses
as non-rankshifted and non-finite clauses as rankshifted or reclassified,
McGregor moreover fails to clarify the semantic correspondences between
them. In the next section, we will see how Langacker's functional analysis
of the nominal group and his 'type-instance' approach to nominalizations
does allow one to grasp both the external and internal reclassification of
factive /¿«/-clauses as nomináis.
4.2. Type vs. instance nominalization
Langacker (1991: 23) suggests that nominalizations of the types illustrated

in (72) can be profitably examined from the standpoint of "the internal or-
ganization of the relational predication, in particular whether it represents
the conception of a process type or a specific instance of that type" (exam-
ples based on Langacker 1991: 422):
(72) a. Zelda's reluctant signing of the contract surprised the entire crew.
b. Zelda's reluctantly signing the contract surprised the entire crew.
c. That Zelda reluctantly signed the contract surprised the entire crew.
The nominalization in (72a) is called an 'action nominal' in Lees (1968:

64) and is to be distinguished from the 'action' type of gerundive nominal
which I discussed earlier (e.g., his job involves [my answering the phone on
his behalf quite a bit of the time] and [Eating vegetables] is healthful). The
structure in (72b) represents a factive gerundive, and that in (72c) is a fac-
tive /Aa/-structure.
Langacker (1991) zooms in on the internal functional organization of
these nominalizations and argues that the structural differences that exist
between them follow from the fact that they derive from three different
"levels of organization in the assembly of a finite clause" (Langacker 1991:
33), i.e. a verb stem, an 'intermediate processual structure' and a finite
clause. These structural levels, he says, correspond to a type, an un-
grounded and a grounded instance respectively. The action nominalization
Zelda's reluctant signing of the contract is analyzed as a 'type' nominaliza-
tion because the nominalization process involves the reification of a verb
stem (sign) into a noun (signing), of which the "semantic function ... is
limited to specifying a type" (Langacker 1991: 33). Of course, if that noun
is used as head of a nominal, it becomes an instance and the nominal means
of instantiation and grounding apply. The gerundive nominalization in
(72b), on the other hand, is claimed to derive not from a verb stem realizing
a process type, but from an intermediate processual expression or un-
grounded instance (sign the contract), which Langacker (1991: 34) defines
as "a structure that is like a finite clause except for the absence of an ex-
plicit subject and a predication of tense or modality". The structure's in-
stantiated status is said to follow from the fact that it "incorporates a fully
specified direct object" (Langacker 1991: 34). The result is a 'complex
noun' which functions as head noun, the Subject of which is expressed by
means of possessive periphrasis (Zelda's signing the contract). Finally,
unlike gerundive nomináis, /AaMiominals are said to derive from a
204 Factive nominalizatioti: Towards a descriptive position
grounded instance or a finite clause, whose internal structure is retained in

the nominalization.
Interestingly, Langacker also attempts to link up the structural properties
of the action and gerundive types of nominalizations with the meanings
which they express. In the case of action nominalizations, which start from
a simple process type specification and are turned into simple nouns, he
points out that they behave like nouns also semantically and can therefore
profile various facets of the reified event, such as the manner in which it is
carried out (73a), its duration (73b), its propriety (73c) and even the factu-
ality of the event (73d) (Langacker 1991: 32):
(73) a. Harvey's taunting of the bear was merciless.

b. Harvey's taunting of the bear lasted three hours.
c. Harvey's taunting of the bear was ill-advised.
d. Harvey's taunting of the bear came as a big surprise.
Compare this with the meanings in which the simple noun party can be
used (Langacker 1991: 33):
(74) a. The party was boisterous.

b. The party lasted three hours.
c. The party was ill-advised.
d. The party came as a big surprise.
As to the factive meaning of certain gerundive nominalizations, Lan-

gacker remarks that the "construal of a reified process as representing a
particular instance renders it compatible with contexts that presuppose its
factuality" (Langacker 1991: 34). He argues that a construction such as
Harvey's taunting the bear can be interpreted factively because it is derived
from the intermediate processual expression taunting the bear, which
"represents an instance and not just a type" (Langacker 1991: 34). Because
it is not finite and therefore not grounded, however, a gerundive nominal
can also occur in non-factive contexts. Compare the factive use of Harvey's
cruelly taunting the bear in (75) with the non-factive use in (76) (Lan-
gacker 1991: 34):
(75) Harvey's cruelly taunting the bear was a severe blow to his campaign
for presidency.
(76) Harvey's cruelly taunting the bear is something that could simply
never happen.
In short, Langacker's analysis of action nominalizations, gerundive

nominalizations and that-structures suggests that an analysis of the internal
organization of nominalized constructions can benefit from taking into ac-
count the functional level of assembly of their processual starting-point. It
also shows that the functional constructs of type specification, instantiation
and grounding are relevant to the description of nominalizations and shed
light not only on many of their structural properties, but also on the mean-
ings that are linked to, for instance, action and factive nominalizations.
Some of the details of Langacker's type/instance analysis of nominaliza-
tions, however, are in need of further clarification: especially the analysis
of gerundive nominalizations can be refined. Langacker's discussion of
gerundive nomináis moreover only deals with gerundives that have a peri-
phrastic Subject. As could be deduced from Lees's analysis of gerundive
nominalization, however, it is especially in the group of gerundives that
lack an explicit Subject that problems of delineation between factive and
non-factive gerundives occur. The divergent internal properties of all the
different types of gerundive nominalizations should be included in the
analysis of factivity.
In the next chapter, I will try to give a precise and systematic descriptive
analysis of the internal functional properties of factive gerundives, that-
factives and the fact fAa/-constructions. As well, I will show that a func-
tional analysis of all three subtypes of factive nominalization is a precondi-
tion to come to a precise analysis of the category of factive nominalization
and to link up the notion of factivity with the lexicogrammatical properties
of the nominalized structures that realize it.
Chapter 9
A functional analysis of factive nominalizations as
nominal constructions
This chapter will present an analysis of the internal organization of factive

nominalizations as nominal constructions. It aims at counterbalancing the
prevailing focus on the external nominal contexts in which factive nomi-
nalizations figure as clause constituents. Assuming those as one aspect of
the rank shift involved in factive nominalizations, it will complete the pic-
ture by tracing the internal reclassification involved in the various types of
factive nominalizations, by virtue of which they fulfil nominal functions
such as type specification and grounding. The factive nominalizations that I
will consider are either based on full finite clauses (i.e. íAaí-factives and the
fact /Aa/-constructions) or on an 'intermediate processual expression' (in
the case of gerundive factives). I will try to account for their internal func-
tional organization in terms of the functional model of the nominal and the
clause set out in Chapter 4.
In a first part (Sections 1, 2 and 3), I will successively zero in on that-
factives, the fact /Aai-constructions and factive gerundive nominalizations
and describe how each of them makes the transition from processual start-
ing-point to nominal construction. In Section 4, the main results of these
descriptions and the implications for the analysis of the system of factive
nominalization in general will be pointed out.
1. ίΑαί-factives: From finite clause to proper name strategy
The clausal level of assembly which ίΑαί-nominalizations start from is that

of a full finite clause or a clause which profiles a grounded instance of a
process type (Langacker 1991: 34). It was shown in Chapter 4 that a finite
clause grounds its process type specification in two ways: it establishes a
temporal/modal link with the speech event, and it relates the process type to
a grammatical person. Person deixis, it was argued, is expressed subjec-
tively on the Finite (by means of Subject-Finite concord) and is objectively
realized by the Subject (Davidse 1997). While the point that /Aa/-structures
start from finite clauses is obvious and not controversial, the claim that the
resulting nominalized //^/-constructions show parallels with specific nomi-
nal paradigms is more contentious.
208 Factive nominalizations as nominal constructions
I pointed out before that it is widely agreed that factive /Aai-structures

serve essentially nominal functions in the clause, viz. functions such as
those of Subject and Object (e.g., among others, Sweet 1891; Poutsma
1929; Kruisinga 1932; Lees 1960; Kiparsky and Kiparsky 1971; Comrie
1976; Quirk et al. 1985). That ίΑαί-structures also fit into the nominal para-
digm internally seems less obvious. The following two points are crucial to
come to terms with the internal properties of íAaí-structures. First, it is im-
portant to recognize the downranked character of the reclassification which
nominalized ί/ζαί-structures realize (Halliday 1994). And, secondly, it has
to be acknowledged that the nominal paradigm allows for structures which
do not realize the functions of type specification, quantification and
grounding iconically.
1.1. Downranking as functional reclassification
Let us start with the downranked status of iAcz/-constructions. As argued by

Halliday (1994), factive ί/ζαί-structures are clauses that have been made to
function at a rank which is lower than that of the clause, i.e. that of the
nominal. It is because /Aa/-factives are downranked clauses that they have
retained so many of their clausal properties. Deverbal -er nomináis and ac-
tion nominalizations of the type the opening of the door simply contain re-
classifications of a verbal unit into a noun, i.e. a unit which belongs to the
same rank but realizes a different class. By contrast, fAai-structures are re-
classified through rank shift and the finite clause which they start from is
nominalized as a whole: a clause that is downranked is downranked with its
internal clausal properties untouched. A downranked finite clause thus re-
tains its finite status. The only effect which downranking has on the internal
clausal properties of the downranked clause is that they are as it were
viewed through an additional, nominalized viewing frame which construes
the process in it "holistically" rather than scanning it sequentially or step-
by-step (Langacker 1991: 440, 449) (for Langacker's notions of 'sequen-
tial' vs. 'holistic' scanning, see also Chapter 4, Section 2.1.1).
Even though a downranked clause retains most of its clausal properties,
its being downranked implies that it has to discharge the functions of the
accommodating class. Downranking implies functional reclassification
'from below'. For factive ί/ζαί-clauses this means that they, in one way or
other, discharge the nominal functions of type specification, instantiation,
quantification and grounding. In the next section, I will show how, follow-
ing Langacker (1991), embedded /Aa/-structures can be fitted into the
nominal paradigm and linked up with nominal strategies.
that-/actives: From finite clause to proper name strategy 209
1.2. The nominal paradigm
In Chapter 8,1 described how Halliday analyzes //^/-structures as basically

realizing the common noun strategy. In his view, /Aa/-structures realize the
head of the nominal (Halliday 1994: 266). He also claims that /Aa/-factives
alternate with nomináis in which they do not function as heads themselves,
but postmodify a nominal head, as, for instance, in the fact that he left her.
As correctly pointed out by Huddleston (1988: 145fn), Halliday's analysis
of /Aa/-structures as head of the nominal is questionable since, unlike ordi-
nary nominal heads, /Aa/-structures cannot take any pre- or postmodifica-
tion. While I agree with Huddleston's criticism and reject Halliday's analy-
sis of /Aa/-structures as nominal heads, I do not follow Huddleston in the
conclusion which he draws, viz. that /Aa/-structures cannot be analyzed as
nomináis. I believe that they can be fitted into the nominal paradigm, if
only one does not restrict the nominal strategies which a nominalized
clause can adopt to that of common nouns only.
I thus come to the second precondition for identifying the internal
nominal functioning of //¡«/-structures: it has to be acknowledged that there
exist nomináis which do not realize the functional schema of 'grounded
instance of a type' iconically or by means of a nominal head and determin-
ers. In fact, it is by no means exceptional to have nomináis which do not
need determiners to instantiate and ground the type specification (Lan-
gacker 1991: 148, 150). Pronouns and proper names, for instance, are in-
herently definite. As to postmodifiers, they characterize a type of Thing in
terms of some process within which it is directly or indirectly a participant
(Halliday 1994: 188). Inherently definite units such as pronouns and proper
names do not take this kind of postmodification.
Huddleston's rejection of the nominal status of //ia/-structures on the ba-
sis of the absence of pre- and postmodifiers can now be argued not to hold:
even among ordinary nomináis, examples are found of nomináis without
any pre- or postmodifiers. Nouns such as pronouns and proper names are
uniquely identifiable and fully characterized. They function as fully
grounded nomináis in their own right (Langacker 1991). As pointed out by
Langacker (1991: 149), /Aa/-factives are part of this nominal paradigm:
embedded /AaZ-structures do not function as heads of the nominal, but they
are fully grounded nomináis themselves. Factive that-structures have in
common with pronouns and proper names that they are fully characterized
and uniquely identified in relation to the speech context. Langacker (1991:
149) in my opinion rightly points out that in the case of nominalized that-
structures, "the unique circumstances of the grounding relationship are in-
corporated", as a result of which the "specified type ... has only a single
210 F active nominalizations as nominal constructions
instance, with the consequence that the derived noun is inherently definite".
Like pronouns and proper names, therefore, /Aai-clauses do not require
"separate grounding" to function as nomináis (Langacker 1991: 148). I will
refer to the nominal strategy which embedded /Aa/-structures employ as the
'proper name strategy'.
The unique combination of clausal and nominal properties which that-
factives realize can now be mapped out in more detail. With ¿Aaí-structures,
the nominalization process applies to a finite clause. By being downranked,
this finite clause is reified and construed as an abstract entity, with its proc-
ess scanned holistically rather than sequentially. Rather than effecting fun-
damental changes to the clausal internal properties of the finite clause, the
downranking mechanism reclassifies the clause into a nominal, capable of
discharging nominal functions such as grounding according to the proper
name strategy. Like proper names, factive ίΑαί-clauses establish an entity
which is definite in itself and can therefore function as a fully grounded
nominal.
Thus, by following Langacker's analysis of factive iAa/-constructions,
one can analyze them as nominal-like not only externally, but also inter-
nally. This counters both Huddleston's and McGregor's analysis of embed-
ded ίΑαί-structures. McGregor (1997: 127) argues that finite clauses cannot
be rankshifted because rankshift or reclassification always implies that the
downranked unit "takes on the grammatical and semantic properties inher-
ent to the lower ranking unit". Finite clauses "show none of the formal
properties ... which are characteristically associated with nominalizations"
(McGregor 1997: 131). It is especially the finite or fully grounded status of
/Aai-clauses which, according to McGregor, prevents them from being
downranked as nomináis: he argues that finite clauses "always represent
situations as situations, not as entities" (McGregor 1997: 131).
Against McGregor (1997), it can now be argued that that-
nominalizations do function as nomináis internally. First, as pointed out by
Langacker (1991: 448), it is essential to acknowledge that with embedded
/Aa/-structures "there are multiple levels of organization to consider, and
multiple paths through which the contents of a /Aa/-clause are accessible
from the ground". Reclassification need not be marked by means of a re-
duction of the grounding options that are available within the downranked
clausal unit. Nor does it always require that the reclassified unit adopts par-
ticular morphological and syntactic properties that are characteristic of the
accommodating class. Instead, reclassification is junctional in nature.
In the following section, I will zoom in on the fact ¿/¡^/-constructions
and show how they provide additional evidence for the 'proper name'
analysis of í/zaí-structures.
the fact ihat-constructions: A case of apposition 211
2. the fact /Art-constructions: A case of apposition
As pointed out in Chapter 8 (Section 4.1), Halliday analyzes the fact that-
constructions like those in (1) and (2) as that-factives functioning as post-
modifiers of the head noun:
(1) He was, I think, troubled by the fact that the Democratic Party
couldn't put together a majority on the national level.... (CB)
(2) And I don't like the fact that it's been carnivalized. (CB)
Quirk et al. (1972: 647, 1985: 1260) likewise treat the that-clauses in the
fact ¿Aírf-constructions as postmodifiers of the fact-noun. To distinguish
them from the restrictive type of relative postmodifying clauses (as in 3),
they describe /Aai-structures in the fact iAaf-constructions as 'appositive'.
(3) The news that appeared in the papers this morning was well received.
(Quirk et al. 1985: 1244)
As the most obvious difference between relative and appositive postmodifi-

cation they point out that appositive that functions as a conjunction, rather
than as a relative pronoun. Halliday also discusses the different status of the
embedded that-clause in restrictive relative clauses and in the fact that-
clauses. In restrictive relative clauses, the clause is said to form a direct
representation of reality and of how the language user experiences it (Halli-
day 1994: 242). The relationship between the restrictive relative that-clause
and the antecedent is consequently one of expansion. The that-c\mse in the
fact f/^/-constructions, in contrast, itself forms a projection or metaphe-
nomenon (see Chapter 8, Section 2.2). Rather than representing reality in a
direct way, it represents a phenomenon that has already been linguistically
processed (Halliday 1994: 264). As indicated in Chapter 8, in Halliday's
view, facts come as 'ready-packaged' projections, i.e. they are not pro-
jected by the clause in which they function.
Yet another type of that-clause which is considered to be postmodifying
is exemplified in (4) and (5):
(4) the thought that she might one day be a queen (Halliday 1994: 264)
(5) the news that the team had won (Quirk et al. 1985: 1244)
Quirk et al. (1985) group these //¡ai-structures with the appositive type of
postmodifying clauses. Halliday (1994: 263) calls them "embedded locu-
tions and ideas", functioning as postmodifiers within a nominal group. He
argues that, like ί/ζαί-clauses postmodifying the noun fact, they represent
projections or metaphenomena, but, unlike in the fact iAaf-structures, "the
projecting element is the noun that is functioning as Thing" (Halliday 1994:
263). In the nominal in (4), for instance, it is the noun thought which pro-
jects the that- clause that follows it.
The analyses which Quirk et al. (1972, 1985) and Halliday (1994) sug-
gest for the relationships between embedded /Aa/-clauses and the noun
classes that precede them can now be summarized as follows:
Table 4. The analysis of 'postmodifying' ίΑαί-clauses as presented in Quirk et al.

(1972, 1985) and Halliday (1994)
restrictive relative
the fact that the thought/claim that
clauses
apposition relative apposition
head-postmodifier head-postmodifier head-postmodifier
/ expansion projection
the ίΑαί-clause is an em- the that-clause is an em- the /Aa/-clause is an em-
bedded projection or bedded direct represen- bedded projection or
metaphenomenon tation of experience metaphenomenon
Notice that all embedded fÄa/-clauses are analyzed as postmodifiers, in

spite of the fact that embedded projections realize an appositive relation
and restrictive relative clauses a relative relation. Secondly, it can be ob-
served that Halliday analyzes the semantic relationship of the that-clause
and the noun that precedes it only for restrictive relative clauses and em-
bedded locutions/ ideas: the relation between the fact and the that- clause in
the fact ¿/zai-constructions is not further specified.
In the next section, I will focus on the relationship of apposition which
has been argued to exist between the the fact ίΑαί-nominal and the that-
clause. In a first part (Section 2.1), I will discuss the structural characteris-
tics of restrictive apposition and compare them with those realized by re-
strictive relative clauses and in the thought/claim /Aa/-constructions. I will
suggest that Quirk et al.'s (1985) and Halliday's (1994) analysis of the fact
^¿-constructions in terms of modification is incompatible with their re-
strictive appositive status. In Section 2.2, then, I will zoom in on the func-
tional characteristics of the appositive units and compare the features of the
fact iAa/-constructions with those of non-nominalized cases of apposition.
the facttiaat-constructions:A case of apposition 213
2.1. The structural characteristics of the fact /Aaf-constructions
Quirk et al. (1972: 647, 1985: 1260-1261) analyze the fact ίΑαΖ-clauses as
consisting of two units which are related to each other in a restrictive, ap-
positive way (see also Langacker 1991: 149). But what characterizes re-
strictive apposition? Apposition is said to be a relationship between "con-
stituents of the same level" which are "identical in reference" (Quirk et al.
1972: 620). In, for instance, Paul Jones, the distinguished art critic, both
Paul Jones and the distinguished art critic refer to the same person: the
relation between them can therefore be described as appositive. Appositive
units are restrictively linked when they form a single information unit, i.e.
when they are uttered on a single tone unit and are not separated by com-
mas in writing: the apposition between Paul Jones and the distinguished art
critic in Paul Jones, the distinguished art critic can thus be analyzed as
non-restrictive. Examples of restrictive apposition are the famous critic
Paul Jones, the singer Robeson and the River Thames.
The characterization of the fact ίΑαί-structures as appositive nominal
groups thus implies that they establish a relation between two units which
are of equal status. At the same time, however, Quirk et al. (1985: 1261)
classify the /¿¿¡/-clauses in the fact /Aû/-constructions as postmodifiers. I
argue that the appositive nature of the fact /Aaf-constructions is not recon-
cilable with their being described in terms of modification (see also Biber et
al. 1999: 645) and that, consequently, the use of the notion of postmodifica-
tion in the description of the fact /Aai-constructions is unjustified.
Let us have another look at Halliday's account of the structural relations
that may exist between members of a complex. Halliday (1994: 221) argues
that the notions of hypotaxis/parataxis and expansion/projection (described
in Chapter 8, Sections 2.1 and 2.2) "define complexes at any rank" and can
also be used to describe complex relations at the level of the nominal. In
general, parataxis is the "linking of elements of equal status", while hypo-
taxis is the "binding of elements of unequal status" (Halliday 1994: 221).
The initiating and continuing element in a paratactic relationship are 'free'
"in the sense that each could stand as functioning whole", while in a hypo-
tactic relationship only the dominant element is free (Halliday 1994: 221).
Parataxis thus forms an alternative to a relationship of modification, which
always involves one element modifying another (Halliday 1994: 218).
As far as I can see, the the fact ίΑαί-nominal and the ίΛαί-clause are
paratactically and not hypotactically linked to each other: they are not re-
lated through modification, but form elements on the same level, with each
of them having the potential to stand on its own. The construction the fact
that Caesar was dead can, for instance, be replaced by the fact or by that
Caesar was dead without affecting the acceptability of the clause in which
it is embedded:60
(6) a. The fact was obvious to all.

b. That Caesar was dead was obvious to all.
Rather than functioning as a postmodifier, therefore, the /Aa/-clause in the

fact fAai-constructions has a status which is equal to that of the the fact
that-nominal. The that-clause cannot be analyzed as postmodifier, and nei-
ther can it be classified as Qualifier to the fact-noun in terms of its repre-
sentational semantics: the /AaZ-clause does not characterize the type of
Thing that precedes it in terms of a process in which it is a participant (as
argued in Halliday 1994: 188). Instead, it represents a second Thing.
To fully grasp the nature of the paratactic relationship that exists be-
tween the the fact that-nomm&X and the /Aa/-clause, its semantic status
should be analyzed as well (Halliday 1994: 243). The main distinction to be
made here is that of expansion versus projection, expansion being a rela-
tionship between direct representations of experience and projection in-
volving metaphenomena or linguistically processed ideas (Halliday 1968;
1994). Halliday distinguishes three types of expansion relationships. Elabo-
ration implies that one element restates another element in other words,
specifies it in greater detail or exemplifies it. A clausal example of elabora-
tion is given in (7a). When two elements are related by means of extension
the second element adds some new element to the first one, gives an excep-
tion to it or offers an alternative (e.g., 7b). Enhancement, finally, involves a
second clause which qualifies the first one with some circumstantial feature
of time, place, cause or condition (e.g., 7c) (Halliday 1994: 220).
(7) a. John didn't wait; he ran away, [elaboration]

b. John ran away, and Fred stayed behind, [extension]
c. John was scared, so he ran away, [enhancement]
Each of these expansion types can be combined either with parataxis (as in
all the examples in 7) or with hypotaxis. In fact, in considering the various
possible mappings, Halliday defines 'apposition' as a combination of
elaboration and parataxis (Halliday 1994: 225): when two items are in ap-
position, the second item elaborates the first and specifies it in more detail.
The type of semantic relationship which the fact ίΑαί-constructions es-
tablish can now be identified. Halliday (1994) only gives a semantic char-
acterization of embedded /AaZ-clauses: they are projections without explicit
projecting clause. However, when a that-factive comes to function as the
the fact that-constructions: A case of apposition 215
continuing element in a paratactic, appositive relationship with a nominal

containing the fact noun, it also enters into a semantic relationship with that
NP. It seems to me that the paratactic relationship between the that-
structure and the fact NP is one of elaboration : the ίΑαί-clause restates and
specifies the 'fact' referred to by the fact NP; it clarifies what exactly the
fact is about.
We can now determine the precise position of the fact /¿ai-structures
with respect to the other construction types described as 'postmodifying' by
Quirk et al. (1985) and Halliday (1994). First, restrictive relative clauses
can be analyzed as postmodifiers. They have the function of qualifying the
type of Thing designated by the head: the general type news in the news
that appeared in the papers this morning is thus further subcategorized as
(news) having appeared in the papers. Importantly, as pointed out by Lan-
gacker (1991: 432), a restrictive relative clause "restricts the head noun's
type specification", i.e. it constitutes an element of the type specification.
Nominal grounding therefore applies to the combination of head noun and
restrictive relative clause (see also Davidse 2000b). Because they function
as postmodifiers, restrictive relative clauses cannot function on their own:
(8) a. The news was well received.

b.*That appeared in the papers this morning was well received.
Finally, restrictive relative clauses establish a semantic relationship which

can be characterized in terms of elaboration (with the relative wA-element
in the embedded clause restating the nominal antecedent, Halliday 1994:
243) or as one of extension (whereby the relative clause is viewed as adding
a new element to the nominal type, thus turning it into the higher-order type
specification that is grounded by the nominal):
(9) the man who came to dinner
As to embedded locutions and ideas or íAa¿-clauses following nouns

such as assertion, claim, rumour, thought, belief, view, news, Halliday
(1994: 263) suggests that they be analyzed as containing "projecting
nouns": the relation between the two appositive units which they establish
is in that case one of projection. Embedded locutions and ideas are said to
differ from the fact /^-constructions in that they can be systematically
related to clause complexes in which the that-structures are ranking rather
than embedded projections:
(10) the assertion that/ They asserted that Caesar was ambitious.
(11) the thought that/ They thought that Caesar was ambitious.
(12) the claim that/ They claimed that Caesar was ambitious.
However, I feel that there is an important distinction in status between

the /^¿-structures in the projecting clause complexes in (10) to (12) and the
/Aa/-clauses that are embedded in the corresponding nomináis. While in the
projecting clause complexes, the relationship between the two clauses is
hypotactic, embedded locutions and ideas are appositive constructions and
paratactically combine two units which are of equal status. This can be de-
duced from the fact that the two components can function on their own, as
illustrated in (13) and (14) (the existence of /Àaf-clauses which represent
embedded locutions and ideas was first pointed out in Davidse 1994):
(13) a. The rumour that he killed her is not true,

b. That he killed her is not true.
(14) a. Just the thought that he may have killed her is unbearable,
b. That he may have killed her is unbearable.
In view of the fact that there is no reason to assume that a noun can serve a
projecting function, I propose to analyze constructions of this type also as
establishing a semantic relationship of elaboration. Like the fact that-
structures, the /Aa/-clauses in the claim/thought /Aai-constructions restate
and specify the nominal that precedes it. Like the fact /^/-constructions,
the claim/ thought /^/-constructions thus fit into the paradigm of apposi-
tion and realize a combination of parataxis and elaboration.
It follows that the appositive relationship established in a nominal such
as the thought that Caesar was ambitious does not differ from that in the
fact //^/-constructions: both are based on embedded projections, both para-
tactically relate the two units which they consist of, and in both the that-
clause restates and specifies the nominal that precedes it. In general, the
relationships that are established in the fact //¡^/-constructions, restrictive
relative clauses and constructions of the type the claim that can now be
summarized as follows, with the shaded areas indicating that the proposed
analysis deviates from the analysis suggested in Quirk et al. (1972, 1985)
and Halliday (1994) (see Table 4):
the fact Ubat-constructìons: A case of apposition 217
Table 5: An alternative analysis of embedded ίΑαί-clauses
restrictive relative
the fact that the thought/claim that
clauses
apposition relative apposition
head-postmodifier, i.e.
parataxis parataxis
hypotaxis
expansion: extension/
expansion: elaboration expansion: elaboration
elaboration
the ίΑαί-clause itself is the that-ciause is an the f/zaf-clause itself is an
an embedded projection embedded direct repre- embedded projection or
or metaphenomenon sentation of experience metaphenomenon
The overview of the various structural and semantic relationships that

can be established between nominal and that-clause makes it possible to
demarcate the properties of the category of the fact iAaf-constructions more
accurately. What characterizes the fact /Aa/-structures is that they establish
a structural relationship between two elements which are considered to be
of equal status, viz. the fact and the iAa/-clause. The semantic relation be-
tween them is one of elaboration: the ίΑαί-clause restates and further speci-
fies the the fact-nommàì. It is especially their paratactic status which turns
out to distinguish appositive structures such as the fact /Aa/-constructions
from restrictive relative clauses.61 Interesting to observe is also that, unlike
what is claimed in Halliday (1994), there is no fundamental structural dif-
ference between the fact /Aa/-structures and the thought/claim that-
constructions. As I will show in the next section, the precise difference be-
tween these constructions only becomes clear when their functional proper-
ties are considered more closely. To come to a comprehensive account of
the properties of the fact ¿ΑαΖ-structures, therefore, the functional organiza-
tion of the fact íAaí-nominalizations has to be looked into as well.
2.2. The functional properties of the fact í/zaí-constructions
As far as their functional organization is concerned, appositive structures

stand out for several reasons. Most strikingly, appositive nomináis contain
two units "each of which is itself a nominal". As Langacker (1991: 149)
puts it, a composite structure such as the singer Robeson is "a doubly-
characterized, doubly-grounded thing". For nomináis to be able to function

in the most prototypical type of restrictive apposition (e.g., the famous
critic Paul Jones, the singer Robeson, the number three) they should,
moreover, "display minimal definite determination" (Van Langendonck
1999: 116): in other words, the nomináis that figure in restrictive apposition
are necessarily definite. Either they have a definite article (e.g., in the fa-
mous critic, the singer) or they do not take a determiner because they are
fully grounded themselves (as in Paul Jones, Robeson).
Because they "designate the same conceived entity but describe it in dif-
ferent ways" (Langacker 1991: 432), there are also interesting differences
between nomináis in apposition. Restrictive apposition of the type the
singer Robeson is characterized by the fact that the first unit "is preceded
by a definite determiner ... and is more general than the second appositive"
(Quirk et al. 1972: 638). Unlike the second nominal, the first appositive
unit can take modifiers freely, e.g., famous in the famous critic Paul Jones
(Lee 1952; Quirk et al. 1972, 1985). The second appositive seems to neces-
sarily lack an explicit determiner and it cannot take modifiers (Haugen
1953; Van Langendonck 1999). It is, for instance, impossible to have *the
poet (the) great Burns instead of the great poet Burns (Van Langendonck
1999: 117).
These structural differences have been argued to reflect important dif-
ferences in function: while the first element in a construction like the poet
Burns is classificational, the second is specific (see, for instance, Lee 1952;
Haugen 1953; Quirk et al. 1972, 1985; Langacker 1991; Schmid 2000).
Importantly, Van Langendonck (1999: 113) claims that restrictive apposi-
tion, or 'close apposition', represents "a crucial criterion for proper names".
In close apposition, he argues, "the unit with the potentially most specific
reference is a proper name" (Van Langendonck 1999: 116), while the other
one functions as a common noun and helps to "specify the subclass to
which the proper name belongs" (Van Langendonck 1999: 117). Behind
this claim lies the view that proper names are essentially identified by the
way in which they function in a construction. Van Langendonck (1999)
argues that a distinction should be made between proprial lemmata, on the
one hand, and the grammatical category of proper names, on the other. Pro-
prial lemmata represent 'dictionary entries' and may be used in various
ways. In the case of the name John, for instance, it is possible to use it with
unique reference, i.e. to refer to a uniquely identifiable person with that
name (as in 15), or it can be used as a common noun, as illustrated in (16)
(Van Langendonck 1999: 100):
(15) John attended a meeting today.

the fact Oast-constructions: A case of apposition 219
(16) You are talking about a different John.
What constitutes a proper name, then, is derived from the way in which an
item functions in a construction and units which are not proprial lemmas
may also be used as proper names (Van Langendonck 1999: 96). The close
apposition test thus identifies not only proprial lemmas (e.g., Robeson in
the singer Robeson), but also structures of which the primary function is
not that of proper names. Van Langendonck (1999: 118) gives the follow-
ing examples:
(17) a. The song 'The Seven Drunken Knights'

b. The question 'What does it mean to live in modern society?'
As Van Langendonck (1999: 118) points out, "semantically and syntacti-

cally" structures like these "behave as proper names in these appositional
structures in the same way as the units in the standard appositional syn-
tagms": they show individual reference and they can function on their own
syntactically.
The restrictive type of apposition realized by the fact /Actf-structures sig-
nificantly resembles that found in constructions like the singer Robeson and
the poet Burns. As in ordinary nominal appositive structures, the first nomi-
nal (the fact) and the second nominal (the ίΑαί-clause) differ in terms of
how they describe the fact which they designate, but they are identical in
reference. Notice that the that-clause in the fact ίΛαί-constructions thus
does not specify the nominal type fact in greater detail (as do restrictive
relative ίΑαί-clauses), but relates to the nominal the fact as a whole. Like
restrictive appositive structures such as the poet Burns, moreover, a the fact
ί/ιαί-structure starts with a common noun which is preceded by a definite
determiner and which can be premodified, as in:
(18) The ugly fact that he was holding a gun indicated his guilt. (Quirk et
al. 1985: 1262)
The primary function of the first nominal in the fact ί/ζαί-structures is

classificational: it indicates which 'class' or type of things the second
nominal belongs to. As Schmid (2000: 14) points out, the first nominal
(which he calls the "shell noun") serves the "semantic function of charac-
terizing and perspectivizing complex chunks of information which are ex-
pressed in clauses or even longer stretches of text". Van Langendonck
(1999: 120) describes the function of the common noun in restrictive appo-
sition as designating the 'basic level' meaning or the category of the second
nominal. Like the second nominal in the appositive structure the poet
Burns, then, the /AaZ-clause in the fact //¡aZ-structures is more specific and
specifies which fact is being referred to. Because the finite //ia/-clause is
uniquely grounded, it functions as a full nominal (see Section 1). In terms
of Van Langendonck's analysis of restrictive appositive constructions, the
//¡¿¡/-clause can be said to serve the function of proper name in appositive
the fact fAa/-constructions.
The position of the fact that-structures in the overall system of factive
nominalization can now be clarified: they classify the (factive) /Aa/-clauses
which they contain explicitly in terms of the nominal class of fact; they
identify the class which the embedded /Aa/-constituent belongs to as being
that of facts. It is here that the main difference with embedded locutions
and ideas lies: in the thought/claim /Aa/-structures, the //¡«/-clause is identi-
fied as realizing the category of locutions and ideas, rather than that of
facts. In contexts that allow for a factive as well as a non-factive reading, a
/Aa/-clause which stands on its own as nominal may therefore be inter-
preted either as an embedded locution or as a fact:
(19) a. That she refused the job makes sense to me.

b. The fact that she refused the job makes sense to me.
c. The news/rumour that she refused the job makes sense to me.
The inclusion of the fact-noun, moreover, helps to "activate one single idea,
because nouns have a higher potential for reification than nominal that-
clauses" (Schmid 2000: 367). Because they include an explicit and nominal
classification of the //¡«/-clause, it can be argued that the fact //¡«/-structures
come closer to non-nominalized or ordinary nomináis, which might explain
why they can be used in clausal contexts that do not allow ordinary that-
factives (see, for instance, Davidse 2003), as in (20) and (21):
(20) a. We are pleased by the fact that you work so hard. (Davidse 1994:
270)
b. *We are pleased by that you work so hard.
(21) a. ... it's going to take some adjusting to the fact that Anchorage is a
one-paper town. (CB)
b. *... it's going to take some adjusting to that Anchorage is a one-
paper town.
Finally, the fact //¡«/-constructions also confirm the analysis of that-

factives as full nomináis that use the proper name-strategy. In the fact that-
structures, the embedded /Aa/-clauses function as second nominal in the ap-
Gerundive facts 221
positive relation to the fact. As such, they enter into a relation with the
nominal paradigm that normally functions in that particular position. Be-
cause in non-nominalized cases of appositives (like the poet Burns), the
type of nominal that figures as second element in a restrictive appositive
relation has the properties of a proper name (e.g., Burns), the /Aa/-clause in
the fact ί/ιαί-constructions can be said to connect with proper names. The
analysis of the nature of the relationship which the fact /¿«/-constructions
realize thus confirms what I suggested in my analysis of ίΑαί-factives, viz.
that /Aa/-factives are paradigmatically related to nomináis realizing the
proper name strategy.
2.3. the fact /Aaí-struc tures: Conclusion
The transition from clausal starting point to nominal structure realized by

the fact ¿Ααί-structures can be summarized as follows: like /Aai-factives, the
fact ίΛαί-constructions start from a finite clause which is downranked and
functionally reclassified as a nominal, more specifically, one following the
proper name strategy. Because it functions as a proper name, it can also
enter into a relationship of apposition with other nomináis. The appositive
construction which the that-clause comes to function in then categorizes or
classifies the that-clause as being a fact.
3. Gerundive facts: From atemporal clausal head to common noun or

proper name strategy
To reveal the differences and the similarities between that-factives (includ-

ing those that function in appositive the fact ίΑαί-constructions) and gerun-
dive factives such as those in (22) and (23), I propose that it is necessary to
consider three things: firstly, the clausal level of assembly which gerundive
nomináis start from; secondly, the nominal strategies which the embedded
clausal unit in gerundive nominalizations adopts and, thirdly, the way in
which the nominal and clausal properties interact in the resulting nominal-
ized structure. In this section, I will elaborate each of these aspects in more
detail.
(22) Since old Nanapush had saved her from death ..., her visiting the man
stood to reason. (CB)
(23) She appreciated the clerk's taking the time to be so friendly ... (CB)
3.1. From atemporal clausal head...
It is fairly obvious that the main difference between iAai-factives and ge-
rundive factives is that the former are based on a finite clause, while the
latter are not. Which clausal level of assembly it is that gerundive nomi-
nalizations derive from, however, is unclear. In his analysis of nominaliza-
tions, Langacker (1991: 33) argues that gerundives start from an "interme-
diate level which profiles an instance that is left ungrounded". A gerundive
nominalization such as Zelda's signing the contract is thus claimed to de-
rive from the processual expression sign the contract. I will propose an al-
ternative analysis (see also Heyvaert 2000) which views gerundive nomi-
nalizations as being based on an 'atemporal clausal head' (as discussed in
Chapter 4, Section 2.1). A nominalization such as Zelda's signing the con-
tract will be analyzed as deriving from the atemporal clausal head signing
the contract. Rather than being an ungrounded instance, the processual
starting point of gerundive nominalizations is thus an atemporalized proc-
ess type specification. This is relevant for the analysis of gerundives as
nominal constructions: I will argue in Section 3.2. that it is the iype-status
of the clausal head which determines the nominal strategies that gerundive
nominalizations can adopt.
We will start from a number of well-known observations concerning the
internal status of gerundive nominalizations. As Lees (1968: 65-66) points
out, gerundive nominalizations can take auxiliaries (as in 24a), they can be
used with an adverbial (illustrated in 24b) and they are construed with par-
ticipants which are not periphrastically realized but as clausal constituents
(see 24c) (see also, among others, Fraser 1970; Chomsky 1970; Pullum
1991; Hye Suk Yoon 1996; Malouf2000):
(24) a. his having brought up the box

b. his drawing the picture rapidly
c. his drawing the picture
Gerundive nominalizations differ in these respects from so-called 'action

nomináis' like the signing of the contract (Lees 1960). Action nominaliza-
tions are generally considered to be more 'nominal' in nature: they cannot
take any auxiliaries (as shown in 25a), they are used with adjectives rather
than with adverbials (see 25b), and they realize their participants periphras-
tically (illustrated in 25c):
(25) a. *his having brought up of the box

b. his rapid drawing of the picture
Gerundive facts 223
c. his drawing of the picture
It has been deduced from these lexicogrammatical properties that in action

nomináis, the suffix -ing itself has nominalizing power and creates a noun.
Declerck (1991b: 496) refers to this type of nominalization as 'nominalized
gerunds'. The -ing structure found in gerundive nominalizations, in con-
trast, seems to be processual in nature (e.g., among others, Quirk et al.
1985; Declerck 1991b; Pullum 1991; Hye Suk Yoon 1996; Taylor 1996).
If we compare the processual properties of gerundive nominalizations
with those of atemporal clausal heads (see Chapter 4, Section 2.1), some
striking similarities can be observed. First, as in atemporal clausal heads,
the -ing suffix in gerundive nominalizations can attach to both perfective
and imperfective verbs. Lees (1968: 66) already pointed out that gerundive
nominalizations can be formed on all types of verbs, including 'stative'
ones, while action nomináis such as the signing of the contract require what
he calls 'action' verbs:
(26) a. his having a hat, but *his having of a hat

b. his resembling his mother, but *his resembling of his mother
c. his believing her, but *his believing of her
Moreover, the -ing structure in gerundive nomináis can be systematically

related to a fo-infinitive, which constitutes the atemporal alternative to -ing
in atemporal clausal heads used in clauses as well:
(27) a. his having a hat -» for him to have a hat

b. his resembling his mother -» for him to resemble his mother
These to-infinitive constructions can be used in many factive contexts in

which gerundive nomináis can also be used:
(28) For him to have a hat/His having a hat surprised me.

(29) For him to resemble his mother/ His resembling his mother came as a
surprise.
Significantly, the processual constructions found in gerundive nominali-

zations include clause-like realizations of Objects: it was argued in Chapter
4 that in the clause, Objects are part of the processual type specification
and, thus, of the atemporal clausal head (Davidse 1997). As we will see in
Section 3.3, the processual expression in gerundive nominalizations can
also take a (periphrastically realized) Subject, as illustrated in John's sign-
ing the contract. In non-finite clauses as well, atemporal clausal heads

combine with Subjects, as in:
(30) Those who ate the apples ran off wild and mad into the forest - the
apples having been poisoned by a woman who had loved Merlin ....
(CB)
Finally, like atemporal clausal heads, the clausal structure in gerundive

nominalizations can include a range of auxiliaries:
(31) Their having to attend it sometimes puts upon them. (CB)

(32) ... her having been able to follow it into full, ecstatic awareness of
God was sheer grace. (CB)
(33) ... its originality and sheer brilliance vindicate completely her having
been declared Doctor of the Church. (CB)
Importantly, the modal auxiliaries have to in (31) and be able to in (32) and
the perfect auxiliary in having been declared in (33) do not themselves
ground the clausal head: they are atemporalized by the -ing suffix and
therefore essentially belong to an atemporalized process type specification
in which they further modify the simple verb stem.
In short, the lexicogrammatical properties of gerundive nominalizations
suggest that the processual level of assembly which they start from is that
of an atemporal clausal head. This analysis ties in with well-known obser-
vations about the clause-like status of gerundive nominalizations. As to the
precise effect of the -ing suffix on the verb it is added to, then, I would ar-
gue that, as in atemporal clausal heads, -ing imposes an atemporal profile
on the process and forms an alternative to grounding it. Rather than relating
the process to the speech event by means of modality/tense, the atemporal
profile of the -ing suffix confirms the ungrounded status of the process type
specification and serves to derive a higher-order type specification. -Ing
only changes the way in which the process is looked at: rather than scan-
ning the component states of the process in the step-by-step or sequential
way that is characteristic of finite clauses, -ing scans the component states
of the process type holistically. As also argued in Chapter 4 (Section 2.1.1),
it activates the component states of the verb cumulatively, through sum-
mary scanning.
Because they realize summary scanning rather than sequential scanning,
atemporal clausal heads are likely candidates for nominalization: as pointed
out before in the analysis of that-factives (Section 1.1), the nominalizing of
clause-like units (such as finite clauses) turns them into unitary entities of
Gerundive facts 225
which the component states are scanned simultaneously or holistically and

which can be manipulated and used in other clauses (Langacker 1991: 440).
By using atemporal clausal heads in nominal units, an additional viewing
frame is, in other words, imposed on them. Because internally, atemporal
clausal heads already offer a holistic perspective on the component states of
the process type, they come closer to the holistic end-profile of the nomi-
nalized construction in which they figure than do embedded //^/-structures
(where the holistic, nominalized perspective applies to finite clauses or
clauses which have sequential scanning internally). It is probably their in-
ternal atemporal status which explains why gerundive factives, like the fact
//^/-constructions, can figure in nominal contexts which are not open to
//^/-structures. The example in (34a), for instance, alternates with a the fact
/Aa/-construction, but not with a /Aa/-structure:
(34) a. Last year, Mgr. Jacques Gaillot ... was ousted by the Vatican
for openly promoting the use of condoms to prevent the spread of
Aids. (CB)
b. ... was ousted for the fact that he openly promotes the use of
condoms to prevent the spread of Aids.
c. *... was ousted for that he openly promotes the use of condoms to
prevent the spread of Aids.
Importantly, describing gerundive nominalizations as containing a spe-

cific functional level of assembly of the clause provides an alternative to
the traditional analyses of gerundive nominalizations which describe them
vaguely as 'clause-like' or 'verbal' (e.g., among others, Lees 1960; Quirk et
al. 1985; Pullum 1991; Hye Suk Yoon 1996; Malouf 2000). The analysis
proposed here makes clear that gerundive nominalizations are 'clause-like'
because they derive from the functional level of the atemporal clausal head,
which includes Objects, adverbials and non-grounding auxiliaries of the
clause. The processual levels of assembly which deverbal nominalizations
can start from (Langacker 1991) can now be identified as being either a
verb stem or simple process type specification, a higher-order, atemporal-
ized process type specification, or a full finite clause. In my analysis of ge-
rundive nominalizations, I will show that there is a fourth level of assembly
which nominalizations can be derived from, viz. that of ungrounded in-
stances or atemporal clausal heads with a Subject in the objective case
(Section 3.3).
It has to be noted that the analysis of gerundive nominalizations as being
based on a clausal head that realizes a clausal type specification deviates
from the analysis which Langacker suggests for them. As I pointed out be-
fore, Langacker (1991: 33) analyzes the gerundive nominal Zelda's signing
the contract as being derived from the 'ungrounded instance' sign the con-
tract. It is the inclusion of the Object which is held responsible for the in-
stantiated nature of the structure sign the contract·. "The fact that the nomi-
nalized structure incorporates a fully specified direct object supports the
claim that it represents an instance and not just a type" (Langacker 1991:
34). Langacker adds that even though "-ing attaches to the verb from which
this structure inherits its processual profile, it is the structure as a whole -
with its clause-like internal organization already established - that is nomi-
nalized" (Langacker 1991: 32).
I see two main differences between Langacker's analysis and the one
which I propose. First, my analysis considers the Object as part of the
clausal head or type specification and assigns an instantiating role to the
Subject only. Arguments in favour of these claims were given in Chapter 4,
Section 2.2 (see also Davidse 1997). Secondly, in my description of gerun-
dive nominalizations, the functional level of assembly at which the nomi-
nalization process takes place is that of signing the contract rather than sign
the contract. Of course, the atemporal clausal head signing the contract
itself is based on the simple process type specification sign the contract.
Crucially, however, sign the contract is not the structure which the nomi-
nalization applies to. Langacker's analysis remains vague about the precise
nature of the -ing of gerundive nominalizations and the difference with the
nominalizing -ing in action nomináis such as the signing of the contract. By
describing gerundive nominalizations as being based on an atemporal struc-
ture which includes the atemporal marker -ing, I explicitly identify the na-
ture of the -ing suffix in them as being verbal rather than nominalizing.
It can be pointed out here that the verbal status of the -ing suffix in ge-
rundive nominalizations such as Zelda's signing the contract is not recog-
nized by everyone. While in many of the descriptions of gerundive nomi-
nalizations it is posited that the suffix -ing is verbal (e.g., among others,
Horn 1975; Schachter 1976; Quirk et al. 1985; Pullum 1991; Hye Suk
Yoon 1996), some have claimed that it has nominalizing value (Baker
1985; Milsark 1988). As Milsark (1988: 622), for instance, puts it: "The
structures that have classically been called gerundives - that is, those that
occur in subject position and in object position of prepositions and of verbs
such as regret and enjoy - are ... those whose verbs have assumed nominal
categorization as a result of the affixation of -ing". At the same time, Mil-
sark considers the -ing form in gerundive nominalizations to be functioning
as head of a gerundive clause, and he argues that "the nominal features of
the derived verb form percolate up its projection path, ultimately relabeling
the entire gerundive clause as NP" (Milsark 1988: 613). In my opinion,
Gerundive facts 227
Milsark's analysis confuses the atemporal effect of the -ing suffix with
nominalization. As I pointed out before, the holistic perspective which -ing
offers on the component states of the process comes close to the holistic
viewpoint which nominalization in general has on clause-like structures
(including finite structures) (see Langacker 1991). The two should not,
however, be confused: the atemporal status of -ing does not itself effect
nominalization, witness the occurrence of atemporal -ing in subordinate (or
non-downranked) non-finite clauses. The atemporal status of -ing does,
however, seem to be conducive to nominalization and an atemporal struc-
ture can, when nominalized, function in contexts that are more typically
'nominal' in nature than can nominalized expressions which are finite.
Milsark's analysis may confuse atemporalization with nominalization,
but it does make clear that it remains to be explained how an atemporal
clausal head can be nominalized and fulfil nominal functions in the clause.
What is it that turns an atemporal clausal head into a gerundive nominaliza-
tion with nominal-like behaviour? In Hye Suk Yoon (1996), it is suggested
that the atemporal clausal component which gerundive nominalizations de-
rive from is nominalized by a process of 'zero derivation', which leaves
intact the internal processual status of the component but enables it to func-
tion as a nominal in larger configurations. I do not see any reason for ana-
lyzing the nominal behaviour of atemporal clausal heads in gerundive
nominalizations as resulting from what is basically a morphological proc-
ess, i.e. that of zero-derivation. Rather, I propose that, like the finite clause
in rAa/-factives and the fact ί/ιαί-constructions, the atemporal clausal head
in gerundives is downranked, with its internal processual status left intact.
More specifically, it is only when Halliday's construct of downranking
is interpreted as functional reclassification that the lexicogrammatical
properties of factive gerundives can be accounted for and the differences
and similarities with 'action' nominalizations can be explained. Both turn
out to involve the reclassification of a processual structure, but they realize
two different types of reclassification. The -ing of an action nominalization
such as the signing of the contract realizes the reclassification of a verb
stem (sign) into a noun (signing): it nominalizes the verb and changes the
class it belongs to, but it does not alter the rank which the verb functions
on: both the base verb and the derived noun belong to the rank of the word.
Action nominalization thus realizes a type of reclassification which does
not involve downranking. In the gerundive nominalization signing the con-
tract, in contrast, the nominalization process is not realized by -ing: the -ing
suffix is merely an atemporal verbal marker and does not shift the class of
the verb it attaches to from verb to noun. Rather, the entire atemporal
clausal head is downranked within the structure of the nominal.
Crucially, the downranked status of gerundive nominalizations is, like

that of ^«/-structures, primarily functional in nature and the downranked
clausal head adopts the functional properties of a nominal head. In the fol-
lowing section, I will go more deeply into the precise nature of the func-
tional reclassification of atemporal clausal heads in gerundive nominaliza-
tions. As in my analysis of that-factives and the fact //¿ai-constructions, I
will zoom in on the paradigmatic relations which the nominalized proces-
sual structure enters into and try to identify the precise nature of the nomi-
nal strategy which it connects with. To fully grasp the properties that char-
acterize factive gerundives, I will situate factive gerundives in the broader
system of gerundive nominalization and consider non-factive types of
downranked gerundive constructions as well.
3.2. ... to common noun or proper name strategy
In Pullum (1991), the rather provocative claim is defended that gerundive

nominalizations are nomináis with a verb phrase head: the gerund in ge-
rundive nominalizations is claimed to be "a verb buried inside the nonfinite
verb-phrase head of a noun phrase" (Pullum 1991: 786). Pullum bases his
claim on the observation that the head of gerundive nominalizations does
not behave like ordinary nominal heads in a number of respects (Pullum
1991: 768-770): the head of gerundive nominalizations can, for instance,
take clausal negation not, which cannot be used with ordinary nominal
heads (illustrated in 35); gerundive nominalizations cannot take attributive
adjectival modification (see 36), nor do they permit restrictive relative
clause formation (as illustrated in 37). Finally, unlike in ordinary nomináis,
when the head of a gerundive nominalization follows a genitive, it cannot
be ellipsed and interpreted anaphorically (see 38):
(35) a. my not having left, but *my not books

b. John's not destroying the city, but ""John's not destruction of the
city
(36) his kind offer, but *his kind walking me home
(37) *his leaving her that you predicted
(38) a. I was amazed at Stacy's eagerness and at Morgan's too.
b. *I was amazed at Stacy's being so eager and at Morgan's too.
Against Pullum (1991), I claim that gerundive nominalizations do not

result from some kind of 'mismatch' of a verbal head and a nominal con-
stituent. I will argue that the downranked atemporal clausal heads that fig-
Gerundive facts 229
ure in gerundive nominalizations are functionally reclassified and come to

function either as nominal heads or as full nomináis. In fact, as in embed-
ded ίΑαί-clauses, there exists a correlation between the internal level of
grounding of the embedded processual unit and the nominal (grounding)
strategy which the downranked unit adopts. In my analysis of that-factives,
I have argued that the proper name strategy which a downranked finite
clause adopts corresponds to its profile of uniquely identified, grounded
clause (see also Langacker 1991). In the previous section I have shown that
gerundive nominalizations derive from a level of assembly which profiles
what is basically a type specification. The nominal behaviour of gerundive
nominalizations will now be revealed to be in accordance with the specific
status of its head.
I will start with a brief recapitulation of Lees's (1960) description of ge-
rundive nominalizations and the distinction which he makes between 'ac-
tion' and 'factive' gerundive nomináis. I will then go more deeply into the
'non-transformational' analysis which Schachter (1976) offers of the inter-
nal nominal functioning of gerundive nominalizations (Section 3.2.1). In
section 3.2.2, then, I will set out my own view of gerundive nominalization.
3.2.1. Gerundive nominalizations: from transformation to nominal

constituent
I explained in Chapter 8 that Lees (1960) identifies two semantic subtypes

of gerundive nominalizations, i.e. factives and action-referrent gerundives
(the latter category is to be distinguished from Lees's category of 'action
nomináis' such as the signing of the contract). Some of the examples which
he gives of the factive type are reproduced in (39). In (40), gerundive nomi-
nalizations of the action type are given:
(39) a. His eating vegetables is surprising,

b. His dressing himself is funny.
(40) a. Eating vegetables is healthy,
b. Dressing oneself is fun.
Gerundive nominalizations which are action-referrent, Lees (1968: 72) ar-

gues, "cannot have any expressed subject, nor any auxiliary". Factive ge-
rundives, in contrast, can occur "with or without the genitive subject" and
they can take auxiliaries (Lees 1968: 72). An example of a factive nomi-
nalization without Subject is the following:
230 Factìve nominalizations as nominal constructions
(41) Having gone pleased us.
Lees specifies that, if in factive gerundives the Subject is not realized

overtly, the Subject of the gerundive nominalization will be interpreted as
being similar to that of the matrix clause, or to another participant realized
in it: in (41), for instance, the Subject of having gone is interpreted as being
the we that is implied in the us of the matrix clause.
Schachter (1976) tries to formulate an alternative to the transformational
approach given by Lees and argues in favour of "an analysis in which ge-
rundive nomináis are assigned an underlying structure more like that of
ordinary noun-headed noun phrases" (Schachter 1976: 206). Two aspects
of his account of gerundive nominalizations are particularly interesting:
first, he suggests that the interpretation of subjectless gerunds "has to do
with the more general matter of the interpretation of nomináis with and
without overt determiners" (Schachter 1976: 210). Secondly, he compares
the nominal behaviour of gerundives with that of uncount nouns and plural
count nouns. He argues that subjectless gerunds "are actually determiner-
less gerundive nomináis" (Schachter 1976: 211) which are either related to
the generic use of uncount nouns and plural count nouns, as in (42) or des-
ignate specific instances, as in (43) (Schachter 1976: 214-215):
(42) a. Going to the beach is enjoyable.

b. Milk does something for every body.
(43) a. I enjoyed going to the beach yesterday.
b. There's milk all over the kitchen floor.
The gerundive nominal in (42a) is said to refer to a class of activities, rather

than to a specific instance. It is comparable to the generic use of uncount
nouns. The gerundive nominal in (43a), in contrast, designates a specific
instance and is comparable to the non-generic use of the uncount noun in
(43b). Schachter (1976: 215) concludes that the "determinerless NP itself is
thus merely a class name, and this name may be used to refer either to the
whole of the class or to some specific instance of it, according to the con-
text".
Schachter's analysis of gerundive nominalizations breaks new ground in
that it is the first to point out systematic resemblances between gerundive
nominalizations and ordinary, non-nominalized nominal structures·, it com-
pares the downranked clausal head in gerundive nominalizations with heads
of ordinary nomináis and points out correspondences that exist between the
determination of gerundives and that of uncount nouns and plural count
nouns. By bringing into the picture the nominal properties of gerundive
Gerundive facts 231
nominalizations, Schachter's analysis breaks quite radically with the lin-

guistic tradition of approaching nominalizations as 'transformed clauses'
and as consisting of clausal categories only. His description of the nominal
status of gerundives, however, also leaves room for improvement:
Schachter's analysis can be broadened out and other types of gerundive
nominalizations should be included in the description. Certain aspects of
the analysis which he proposes for determinerless gerundives can moreover
be refined.
In the following section, I will show how a new light can be shed on ge-
rundive nominalizations by tracing how the downranked clausal heads
function in them as reclassified full nomináis or nominal heads. I will argue
that a downranked clausal head can either adopt the common noun strategy
(using determiners or quantifiers) or the proper name strategy (with the re-
classified clausal head functioning as full nominal).
3.2.2. Towards a systematic account of the nominal properties of gerundive

nominalizations
The various subtypes of gerundive nominalizations that will be analyzed

are illustrated in the following examples:
(44) a. Only that could account for their having a suitcase with them.
(CB) [possessive pronoun, factive]
b. Techniques for doing this include the caregiver's exposing the
infant to various contingent and discrepant experiences. (CB)
[genitive, non-factive]
(45) a. ... showing one's feelings too clearly (or being highly sensiti-
ve) is regarded as a 'feminine' trait. (CB) [no determiner, non-
factive]
b. Having attended Harvard can be helpful. (Schachter 1976: 215)
[no determiner, factive or non-factive]
(46) a. Of course I regret getting pregnant. (CB) [no determiner, factive]
b. Going there was fun. (Postal 1970: 478) [no determiner, non-
factive]
(47) a. This burning the midnight oil of yours has got to stop. (Schachter
1976: 218) [definite determiner, non-factive]
b. Any talking loudly on your part will be punished. (Schachter
1976: 218) [indefinite determiner, non-factive]
By considering the nominal functioning of the downranked atemporal

clausal heads in these nominalizations, I will show that each of them consti-
tutes a different subtype of gerundive nominalization.
I will start with what are considered to be the prototypical instances of
factive gerundive nominalizations, i.e. gerundive nomináis containing a
possessive determiner or a genitive preceding the downranked atemporal
clausal head, as illustrated in (44a), (48) and (49):
(48) I appreciate your taking the time to talk to us. (CB)

(49) I think the country's making a statement is a lot bigger than partisan
politics (CB).
It is this type that is most often referred to when the co-occurrence of nomi-
nal and clausal properties in gerundive nominalizations is pointed out. It is
also this type of gerundive nominalization which most clearly shows that
the atemporal clausal head has been reclassified, in this case as a nominal
head. As such, it adopts the common noun strategy and can function like
any other nominal head: it can be preceded by either a possessive pronoun
(as in 48) or a genitive (see 49), which serve to identify a definite and spe-
cific subset of it. The instance which gerundive nominalizations like these
refer to is part of the concrete, physical world of instances and is identified
in terms of the system of person, defined from the standpoint of the speaker
(Halliday 1994: 181). As regards the use of the genitive in factive gerun-
dive nominalizations, it should be pointed out that factive gerundive nomi-
nalizations do not seem to occur with the indefinite type of genitive illus-
trated in (50):
(50) Is there a/some diligent student's notebook here?
As argued by Woisetschlaeger (1983; cited in Davidse 1999: 224), the

identity of a/some diligent student's notebook is not retrievable and non-
specific. The fact that I have not come across factive gerundives with this
type of genitive suggests that the genitive that functions in gerundive nomi-
nalizations is necessarily referential.
Importantly, the definite, specific type of reference which is realized by
the factive nominalizations in (48) and (49) is not unique to factive nomi-
nalizations: there exist non-factive or action-referrent gerundive nominali-
zations with a possessive or a genitive preceding the atemporal clausal head
as well. This 'action' type of gerundive nominalizations - even though oc-
casionally mentioned in the literature (e.g., Quirk et al. 1985: 1064; De-
Gerundive facts 233
clerck 1991b: 497) - has always been overshadowed by the factive gerun-
dive prototype. Consider the following examples:
(51) Techniques for doing this include the caregiver's exposing the infant
to various contingent and discrepant experiences. (CB)
(52) My husband speaks very well, but his job involves my answering the
phone on his behalf quite a bit of the time. (CB)
(53) And then all of the sudden, it fractures again and a third time, and the
third time, there's an enormous ripping sound, and that is her having
the stroke. (CB)
(54) Some researchers have felt that having a mental illness would lead to
the individual's occupying a position in a lower social class .... (CB)
None of the nominalizations illustrated here can be replaced by a factive

the fact that or the fact «^construction. Rather than being facts, they repre-
sent a type of gerundive nominalization which differs from the category of
action gerundives that is described in Lees (1960) and Schachter (1976) in
that it does not profile a class or type, but links the nominal head to a spe-
cific instance the identity of which is defined by the possessive or genitive
which precedes it.
Let us turn to another type of gerundive nominalization. Which nominal
strategy can action gerundive nominalizations of the type illustrated in
(45a), (55), (56) and (57) be said to follow?
(55) The bedrock of character is self-discipline. A related keystone is be-

ing able to motivate and guide oneself, whether in doing home-work,
finishing a job or getting up in the morning. (CB)
(56) Self-awareness also involves recognising one's strengths and weak-
nesses and seeing oneself in a positive but realistic light.... (CB)
(57) But others, like this junior, admitted that knowing about AIDS and
protecting oneself from it are two very different things. (CB)
Schachter (1976: 214) compares the nominal behaviour of gerundive nomi-

nalizations like these to the generic use of uncount nouns (take, for in-
stance, the use of milk in Milk does something for every body)·, like the re-
lated generic type of mass nouns, these nomináis are said to name a kind or
class of activity or condition, rather than a specific instance of it. In
Davidse (1991: 364), nominalizations such as those in (55) to (57) are char-
acterized as "merely specifying a 'type' of a process". I will call them the
generic category of action gerundives.
Which kind of nominal grounding strategy do generic action gerundives

follow? I refer here to the analysis of generically used mass nouns and plu-
ral count nouns in Davidse (1999). Following Carlson (1978), Davidse pos-
its that mass nouns and plural count nouns which are generic in profile be-
have like proper names with respect to quantification and grounding: their
generic reading excludes in principle explicit quantification, because they
designate just one instance, viz. the kind or type which they refer to as
such. Compare, for instance, the generic mass noun in (58) with its non-
generic counterpart in (59):
(58) a. Milk helps to make your bones stronger.

b. *Some milk helps to make your bones stronger.
(59) a. There's milk on the table.
b. There's some/much milk on the table.
Similar to proper names, moreover, the use of the bare nominal is sufficient
to identify which kind is being referred to: definite grounding is thus part of
the internal structure of a generically used mass noun (Davidse 1999: 211).
Rather than referring to an indefinite instance in physical space (as does
milk in 59) or to a uniquely identified instance in physical space (as do or-
dinary proper names), however, generically used mass nouns designate
specific instances in type space. They are basically names for kinds that are
presented as instances. Notice that the definite status of generic action ge-
rundives is supported by the fact that they can occur as second nomináis in
close apposition:
(60) Politics is the act of reallocating power. (CB)

(61) Thirdly, the act of getting married is expensive. (CB)
It can be concluded that the category of action nominalizations that refers

to kinds is not only, as Schachter correctly argues, generic in profile, it is
also essentially definite in nature and functions in the clause as a definitely
grounded nominal.
Interestingly, Schachter (1976) also points to the existence of generi-
cally used gerundives which contain the auxiliary of perfect aspect have.
Remember that in Lees's analysis, the use of such an auxiliary is explicitly
excluded from the action-referrent type of gerundive nomináis. Schachter
(1976: 215) gives the following example:
(62) Having attended Harvard can be helpful.

Gerundive facts 235
Schachter (1976: 215) argues that this nominal "seems to name a type of
condition resulting from some completed activity". I believe Schachter's
claim to be essentially correct: some action gerundives contain the auxiliary
have. In my opinion, however, the clause in (62) can receive not only a ge-
neric reading, but also a reading in which a specific instance is being re-
ferred to. Rather than being necessarily generic, in other words, the nomi-
nal is ambiguous. On the generic interpretation, no particular instance is
envisaged. The atemporal clausal head having attended Harvard - which
designates a type specification - then comes to function as a definitely
grounded nominal with generic reference. Notice that, as I argued in Chap-
ter 4 (Section 2.2), the presence of the auxiliary of perfect aspect have does
not itself instantiate or ground the process type specification. Of course, the
secondary tense which have realizes is indirectly situated with respect to
the ground. Yet, it does not itself instantiate or ground the process type.
The nominalization having attended Harvard in (62) can, in my opinion,
also be analyzed as designating a specific instance, the identity of which is
retrievable from the context of speech. If, for instance, I am telling some-
one that I have attended Harvard, and that person answers that Having at-
tended Harvard can be helpful, the gerundive nominalization is factive in
meaning and designates a specific instance. As such, it agnates with the
nominal the fact of your having attended Harvard. Factive gerundives of
the type having attended Harvard will be discussed in more detail later on
in this section.
An example of a gerundive nominalization which resembles that in (62)
in that it is ambiguous between a generic, action reading and a factive in-
terpretation is that in (63):
(63) Do you consider not having been to art school a disadvantage?
Firstly the Subject of this nominal may be coreferential with the Subject of
the matrix clause; the speaker then asks the yow-person how he or she feels
about the fact that he/she has not been to art school. While this reading
might be the most obvious one here because there is a nominal in the ma-
trix clause, viz. you, which can readily be interpreted as identifying the
Subject of the gerundive nominalization (which is not the case in 62), a ge-
neric reading cannot be excluded. One can, for instance, imagine the ques-
tion being directed to the head of an art school, who has been to art school
himself: the nominalization in that case does not refer to a specific instance,
but it designates a kind of condition. The nominal not having been to art
school thus comes to be interpreted as generic, on a par with having at-
tended Harvard in its non-specific meaning. A generic reading, finally,
seems to be the most likely reading for the gerundive nominalization in

(64):
(64) Gandhi ... also reminded Americans of color not to be ashamed of

their heritage of slavery; there was dishonor only in having been
slaveholders. (CB)
Unlike in (63), the Subject of the gerundive nominalization cannot be re-

trieved from the matrix clause: it is not the 'Americans of color' which
should be ashamed of having been slaveholders, but rather the white in-
habitants of America. The nominalization having been slaveholders may
therefore be analyzed as profiling a kind of condition resulting from some
completed activity, rather than a particular instance. As such, it has definite
generic reference.
Let us now turn to the type of gerundive nominalization that is exempli-
fied in (65) and (66):
(65) I resent being tagged a problem boy. (CB)

(66) Being dropped by Scotland before the Five Nations was a disap-
pointment .... (CB)
In Schachter's (1976) view, these nominalizations designate specific in-

stances and are comparable to the non-generic use of the mass noun milk in
There's milk on the kitchen table. It was pointed out in Chapter 4 (Section
1) that mass nouns such as milk in There's milk on the kitchen table are
used indefinitely and take cardinal quantification (Davidse 1999: 222), as
can be inferred from the fact that they alternate systematically with absolute
or cardinal quantifiers such as some and much:
(67) a. There's milk on the table.

b. There's some/ much milk on the table.
Schachter's analysis of these nomináis thus implies that they are indefinite
and take cardinal quantification. However, it is impossible to add quantifi-
ers such as some and much to the gerundives in (65) and (66). (As I will
show later on in this section, however, action-referrent gerundives can in
some very marked cases take indefinite quantifiers.) Conceiving them as
indefinite moreover seems to conflict with the fact that they designate spe-
cific instances of a process type. Against Schachter, I therefore argue that
nominalizations like the ones in (65) and (66) function as proper names,
much like the factive ίΑαί-structures that were discussed in Section 1 of this
Gerundive facts 237
chapter: they designate an individual instance of the reclassified process

which is instantiated in physical space.
The proper name analysis which I propose for the factive gerundive
nomináis in (65) and (66) is supported by their semantic and formal behav-
iour. First, semantically speaking, the downranked atemporal clausal head
is conceived as being instantiated and the instantiator of the process is re-
trievable from the context: it is, for instance, clear that being tagged a prob-
lem boy in (65) refers to the /-Subject of resent·, similarly, in (66), it is im-
plied that it is the speaker who was dropped by Scotland before the Five
Nations. The nominal into which the instantiated process type is reclassi-
fied can therefore be argued to function as a nominal with definite refer-
ence: as is the case with gerundives taking a genitive or a possessive, it is
clear which instance is being referred to. Unlike in gerundive nominaliza-
tions with a genitive/possessive, however, there is no explicit determiner
indicating the instantiator of the process and signalling the definite status of
the nominalization. Rather than following the common noun strategy,
nominalizations like those in (65) and (66) follow the proper name strat-
egy: they designate an inherently definite instance of a process type. Be-
cause they function as definite proper names, they can occur as second
nominal in a restrictive appositive structure:
(68) ... had decided to boycott me for the mere fact of having written a
book on Franco that was not a denunciation. (CB)
(69) But, as he was now recognizing, the fact of having registered as a
conscientious objector effectively closed the door .... (CB)
It should be pointed out that, like gerundive nomináis with a posses-

sive/genitive, this type of gerundive is not necessarily factive: there also
exists an action-referrent type of gerundive that designates a specific in-
stance of a process type but lacks a possessive or genitive, as in:
(70) Getting together quickly for coffee would be fun. (Thompson 1973:
377)
(71) Adjourning immediately at 4:00 was impossible. (Thompson 1973:
377)
As with the factive gerundives in (65) and (66), the specific instance of the
process type that is being referred to can be retrieved from the context. Ac-
tion gerundives of this type are also definite, and they can therefore func-
tion in appositive constructions:
(72) Prue drew a momentary blank, still flushed with guilt over being
caught in the act of searching Luke's shack. (CB)
(73) Ayatollah Khomeini had said the act of severing ties with the United
States was a correct move. (CB)
The final category of gerundive nominalizations which I will discuss as-

sumes a rather marginal position within the overall system of gerundive
nominalizations. It can be illustrated by the following example given by
Schachter (1976: 218):
(74) Any talläng loudly on your part will be punished.
We can also think of examples such as
(75) More/some/0 reporting from you would be welcome, (personal ex-

ample)
These nominalizations follow the common noun strategy in which an in-

definite quantity of the nominal head is designated by the use of determin-
ers such as any, some, 0 and more. As Schachter (1976: 218) points out,
this constructional schema can also be used with definite determination, as
in (76):
(76) That burning the midnight oil of yours has got to stop.
Remarkably, the head can take postmodification (from you, on your part
and of yours), which specifies the instantiator of the downranked clausal
process type. The gerundive nominalizations in (74) to (76) therefore des-
ignate, as observed by Schachter, instances of a process type, rather than
being generic in focus. However, gerundives of this type which do not
specify the instantiator of the process periphrastically seem to viate more
towards generic reference, as in
(77) More/some/0adjourning at 4.00 p.m. would be welcome, (personal

example)
It should also be noted that this - rather marginal - type of gerundive seems
possible only with action gerundives, not with factives.
Gerundive facts 239
3.2.3. The nominalfunctioning of gerundive nominalization: Conclusion
Gerundive nominalizations are interesting in that, on the one hand, they

apply the proper name strategy (found also with factive that-
nominalizations) which reclassifies the clausal structure as a fully grounded
nominal. On the other hand, they also display the common noun strategy,
by downranking the atemporal clausal head into nominal head position and
integrating it with nominal determiners.
Used as proper names, gerundives can refer either to classes or individ-
ual instances of process types (depending on whether the process itself is
contextually instantiated or not). As for the gerundives that follow the de-
terminer plus common noun constructional schema, the common type with
possessive/genitive always refers to individual instances of processes, pre-
cisely because the possessive/genitive designates the instantiator of the
process. The marked type with indefinite (e.g., some, any, 0 ) or definite
determiner allows periphrastic expression of the instantiator of the process
by postmodification (from you, of yours, on your part). If this is not the
case, this type seems inclined to generic reference. All the options listed
above are represented by the horizontal rows in Table 6.
Factive gerundives turn out to display more restrictions with regard to
these options than action ones. Action gerundives have all possibilities, but
factive gerundive nominalizations never designate classes of a process -
they always refer to definite instances of a process, either by means of the
proper name or the common noun strategy. The options available for fac-
tive and action gerundives are charted in the columns of Table 6.
Table 6. An overview of the nominal strategies adopted by gerundive nominaliza-

tions
ACTION gerundives FACTIVE gerundives
generic reference; Eating vegetables is

/
proper name strategy healthy.
reference to an
individual situation;
Adjourning at 4:00 was I resent being tagged a
impossible. problem boy.
proper name strategy
definite, ... his job involves my ... the country's making

reference specific answering the phone on a statement is a lot big-
to an his behalf. ger than partisan politics.
individual
situation;
common
noun This burning the mid-
strategy definite or night oil of yours has got
indefinite, to stop. /
specific
Some reporting from you
would be welcome.
The analysis of the internal nominal behaviour of gerundive nominaliza-

tions thus helps to distinguish the various subtypes of gerundive nominali-
zation more accurately, and it provides valuable information about the na-
ture of factive gerundives, as opposed to gerundives which designate ac-
tions. In this respect, some headway has been made with the aim of com-
plementing the functional analysis of gerundives 'from above', e.g., as
complements of the clause, with one 'from below', looking at the functions
Gerundive facts 241
of grounding, quantification, instantiation and type specification of the

nominal constructions which gerundives form.
In the next section, I will concentrate on how precisely the functional
properties of the clausal level of assembly which factive gerundive nomi-
náis derive from are integrated with the functional organization of the
nominal.
3.3. Factive gerundive nominalization: Integrating the clausal and nominal

paradigms
I have thus far argued that factive gerundive nominalizations are based on
atemporal clausal heads which are functionally reclassified and down-
ranked either as the common noun head of a nominal (with a possessive or
a genitive preceding it), or as proper names (with no determiner). Impor-
tantly, even when gerundives come to function as common nouns, they do
not behave as prototypical members of the common noun category. Pullum
(1991) rightly observes that gerundives fail to display certain properties
that characterize non-nominalized common noun heads (e.g., adjectival
modification and restrictive relative clauses are impossible; see Section
3.2). I claim that it is only by analyzing how the processual functional or-
ganization of the downranked clausal head is integrated with the nominal
structure of factive gerundives that we can understand such restrictions and
the phenomenon of factive gerundives as such.
Let us recapitulate the nominal forms of determination which factive ge-
rundives can establish. We have seen in the previous section that factive
gerundives always function as definitely grounded nomináis: they function
either as common nouns that are grounded by a possessive or a genitive in
front of the atemporal clausal head (e.g., Only that could account for [their
having a suitcase with them], CB); or they function as proper names and
the instantiator of the specific instance which they designate is retrievable
from the matrix clause or the context (e.g., [Being dropped by Scotland
before the Five Nations] was a disappointment...., CB). Strikingly, factive
gerundive nominalizations cannot take any of the other ordinary nominal
determiners: structures like [Caroline's/her opening the door] surprised me
danài regret [opening the door] do not relate to
(78) a. *The opening the door surprised me.

b. *I regret that opening the door.
It should also be noted that the genitive of gerundive nominalizations alter-

nates with nouns which are in the objective or common case, as shown in
examples (79) and (80) (for a detailed overview of the contexts in which
the common case is used rather than the genitive, see Poutsma 1929 and
Kruisinga and Erades 1953).62 This type of gerundive structure constitutes a
fourth clausal level of assembly from which deverbal nominalizations can
be derived, i.e. that of the non-finite clause.
(79) As matters stand, Tory MPs who resent John Major outmanoeuvring
them in last year's leadership coup will always find journalists more
than willing to indulge their conspirational ambitions. (CB)
(80) I don't approve of the crew's only remaining man going there. (Lees
1968:72)
How can the non-occurrence of articles and demonstrative pronouns and

the possibility of using the objective case besides the genitive be explained?
In the literature on gerundive facts, the possessive and the genitive have
mostly been referred to as the Subjects of the nominalized process (e.g.,
among others, Kruisinga and Erades 1953; Jespersen 1914—1929, 3; Lees
1960; Wasow and Roeper 1972; Thompson 1973; Schachter 1976). Like
Subjects, the possessive and genitive can indeed designate various experi-
ential roles: they can realize an agent-like participant (e.g., the clerk's tak-
ing the time to be friendly·, her visiting the man) or a Patient (e.g., his being
dropped from the ticket). Depending on which participant is profiled, the
atemporal clausal head which they are used with can be active as in (81) or
passive as in (82):
(81) I think the country's making a statement is a lot bigger than partisan
politics. (CB)
(82) Its volume alone would be impressive, but its originality and sheer
brilliance vindicate completely her having been declared Doctor of
the Church. (CB)
Like the clausal Subject, moreover, the Subject in gerundive nominaliza-

tions can be in the objective case, as was shown above. Significantly, those
gerundive nomináis that are factive in meaning but lack an explicit deter-
miner all turn out to necessarily imply a Subject. So, while there is no ex-
plicit Subject in the nominalization having introduced us in he regrets hav-
ing introduced us, the Subject is understood to be the Ae-person of the ma-
trix clause.
Gerundive facts 243
The importance of the Subject to factive gerundive nominalizations is

particularly clear in ambiguous nominalizations such as Do you consider
[not having been to art school] a disadvantage? and [Having attended
Harvard] can be helpful. I pointed out earlier that these nominalizations are
either generic action gerundives or they are interpreted as factive nominali-
zations. Crucially, their interpretation as factives depends on the identifica-
tion of a Subject: if, for instance, not having been to art school is addressed
to someone who has been to art school himself and the you of the matrix
clause cannot therefore be interpreted as instantiating the process type, and
no Subject can be identified in the context, then the nominalization is ge-
neric rather than factive. If having attended Harvard is used after someone
has told the speaker that he/she attended Harvard, the Subject of the nomi-
nalization is known and the nominalization is factive in meaning.
Factive nominalizations, in other words, seem to necessarily have a Sub-
ject, which is either explicitly identified or it is implied but retrievable. It is
obvious that the central role which the Subject plays in factive nominaliza-
tions primarily relates to the clausal level of organization in the nominaliza-
tion: the Subject function is an essentially clausal function. In my discus-
sion of atemporal clausal heads in Chapter 4 1 have identified the role of the
Subject as being that of instantiator: by being tied to a Subject, an atempo-
ral clausal head was argued to convert into the conception of an instance.
The obligatory explicit or implicit inclusion of a Subject in factive gerun-
dive nominalizations can thus be argued to instantiate the atemporal clausal
head that forms their starting point. Factive gerundives, then, can choose
between the nominal common noun or proper name strategy, but on the
clausal level, they necessarily establish instantiation of the atemporalized
process type which they contain.
The integration of the clausal instantiation which factive gerundives re-
alize and the nominal strategies which they use to ground the reclassified
clausal head can be schematically represented as follows:
Table 7. An overview of the integration of clausal instantiation and nominal

grounding in factive gerundive nominalizations
NOMINAL CLAUSAL
instantiation, quantification, grounding instantiation
definite common noun possessive/genitive Subject realized

proper name / Subject implied
3.4. Factive gerundive nominalization: Conclusion
To conclude, in my analysis of factive gerundives, I have first identified the

clausal level of assembly which they start from as being that of an atempo-
ral clausal head. The fact that gerundive nomináis include clausally realized
Objects, auxiliaries and adverbials rather than adjectives was thus given a
natural explanation. It was then argued that, like the finite clause in that-
factives, the atemporal clausal head of gerundives is downranked and func-
tionally reclassified into a unit of the nominal class. I focused on the nomi-
nal paradigms which the downranked atemporal clausal head in gerundive
nominalizations comes to function in and described in detail which nominal
grounding strategies the various types of gerundive nominalizations follow.
I have also set up a typology of gerundive nominalizations which clarifies
the position of factive gerundives in the overall system of gerundive nomi-
nalization.
In a final part, then, I have shown that the restrictions that exist on the
internal nominal behaviour of factive gerundives follow from the influence
of their clausal layer of organization. I have pointed out that gerundive fac-
tives require a Subject that instantiates the reclassified clausal type specifi-
cation, and this Subject is either explicitly realized in the form of a posses-
sive or genitive, or it is implied and can be retrieved from the context. Fac-
tive gerundives, it can be concluded, successfully manage to integrate their
clausal layer of instantiation (of an atemporalized process type) with the
nominal grounding that is required for a noun to function in the speech
event.
4. The internal organization of factive nominalizations: Conclusion
The analysis of the internal functional properties of /Aai-factives, the fact

¿Ααί-constructions and gerundive factives provides clear evidence in favour
of my claim that nominalization essentially involves the functional reclassi-
fication of a non-nominal into a nominal unit, whereby the nominalized
unit not only takes over the external functions of nominal constituents in
the clause but also adopts the latter's internal nominal behaviour: all three
types of factive nominalizations start from a clausal level of assembly but
realize nominal strategies of grounding. The clausal structure which factive
nominalizations reclassify was shown to come to function either as nominal
head or as a full nominal itself. Crucially, the precise nominal strategies
followed by the various types of factive nominalizations were shown to be
in accordance with the status of the clausal unit which they reclassify. That-
The internal organization offactive nomináis: Conclusion 245
factives incorporate uniquely grounded processual instances. As nomináis,

they refer to these uniquely grounded instances of process types, by using
the proper name strategy. 7%a/-factives are thus fully grounded nomináis.
The appositive status of the fact //^/-constructions was argued to support
the analysis of /Aa/-factives as proper names, while the the /àc/-nominal
was shown to indicate the category or 'basic level' meaning of the that-
structure. Finally, gerundive factives are all derived from an atemporalized
clausal head. As a clausal structure this clausal head is not grounded, but in
gerundive factives it is always instantiated by a Subject. The Subject of
gerundive factives may be either expressed explicitly by a possessive de-
terminer, or by the genitive or objective case, or it may be implied and re-
trievable from the matrix clause or context. In the case of factive gerun-
dives with possessive determiner or genitive, clausal and nominal instantia-
tion were shown to be integrated seamlessly with each other.
The description of the three types of factive nominalizations also illus-
trated the importance of taking a functional perspective to the internal or-
ganization of nominalized constructions. The structural analysis of that-
factives given in Halliday (1994), according to which embedded that-
structures are the heads of the nominal turned out to be inconsistent with
the radically functional analysis of ^/-structures as proper names (see also
Langacker 1991). 77ta/-factives thus also illustrate the need for a functional
understanding of the concept of rank shift, rather than one based on struc-
tural constituency (see Chapter 2, Section 3.2). In the analysis of the fact
/Aa/-structures as well, Halliday's head-postmodifier analysis was shown to
be unsatisfactory and to overlook the peculiarities of the appositive rela-
tionship realized in them. Functionally, the paratactic relation which the
fact /Aitf-structures establish is one of double grounding, with both the fact
and the /Aa/-structure realizing a fully grounded and definite nominal (Lan-
gacker 1991). In terms of their representational semantics, the that-
structures in the fact //¡^-constructions were shown to function not as
Qualifiers elaborating the type designated by the head noun, but as Things
themselves. In my analysis of gerundive nominalizations, then, I pointed
out that they follow either the common noun strategy, with the atemporal
clausal head functioning as head of the nominal; or they realize the proper
name strategy, with the atemporal clausal head functioning as fully
grounded nominal. Factive gerundives were shown to be always grounded
definitely. It was also shown that for the nominalization to have the poten-
tial of being factive in meaning, the atemporal clausal head has to be linked
to the category of instantiation, i.e. to the Subject. In terms of their repre-
sentational semantics, gerundive nominalizations turned out to be derivable
246 Factive nominalìzatìons as nominal constructions
from both perfective and imperfective verbs and, if realized, their Subject
could be mapped onto both agentive and non-agentive participant roles.
Importantly, the analyses that were presented of the internal functional
organization of factive nominalizations point to the need to consider not
only prototypical or canonically coded nomináis but also more peripheral
realizations of the nominal paradigm. I mean two things by this. First, when
analyzing the nominal behaviour of nominalizations, it is essential that one
does not only look for canonical types of nominal coding. The discussion of
the internal nominal behaviour of nominalized constructions has often
foundered because nominalizations were expected to behave as it were
more 'typically nominally' than do ordinary, non-nominalized nomináis
themselves. Gerundive factives preceded by a possessive or a genitive, for
instance, have thus generally been considered to display some degree of
nominal internal behaviour, and this because they realize the canonical type
of nominal coding whereby the functions of type specification and ground-
ing are realized iconically, by separate components. Bare gerundive fac-
tives and í/zaí-factives, in contrast, do not realize grounding iconically and
have therefore hardly been recognized as being nominal-like internally as
well. It is only by taking into account the existence of non-canonical nomi-
nal codings such as non-separate symbolization of a function that one can
move beyond the existing descriptions of nominalizations and come to a
coherent analysis of their external and internal functional properties.
However, it is not only vital to consider non-canonical nominal codings
in the analysis of nominalizations; it should also be acknowledged that
nominalizations which realize the reclassification of a clausal structure will
always remain peripheral members of the nominal paradigm, i.e. members
which do not realize all of the properties that characterize prototypical
nomináis. Nominalized constructions cannot be expected to behave exactly
like non-nominalized nominal structures: their processual starting point
necessarily influences much of their internal organization. To conclude
from this that nominalizations like the ones which I discussed in this chap-
ter are not nominal-like internally, however, is too drastic. Nominalizations
can be said to realize the same schematic construction or configuration of
functions as do ordinary nomináis. The functions invoked throughout the
analysis of that-factives and factive gerundives are those of type specifica-
tion, instantiation, quantification and grounding.
What about the notion of factivity, then? What do the various types of
factive nominalizations have in common semantically, and how can this be
related to their lexicogrammatical properties? I believe that the obligatory
presence - explicit or implicit - of a Subject in gerundive factives sheds a
new light on the notion of factivity. Langacker (1991: 34) already noted
The internal organization of factive nomináis: Conclusion 247
that the "construal of the reified process as representing a particular in-

stance renders it compatible with contexts that presuppose its factuality".
While in Langacker's viewpoint, however, instantiation is not exclusively
linked to the Subject function, the analysis of the internal, functional or-
ganization of gerundive nominalizations shows that clausal instantiation
depends on the presence of a Subject. Factivity does not require finiteness,
as is suggested by the tendency to focus on factive fAa/-structures only
(e.g., Lees 1960; Kiparsky and Kiparsky 1971; Halliday 1994; Davidse
1994). Rather, it is about being conceived as an instance. My analysis of
the internal functional properties of ίΑαί-factives, the fact íAútf-constructions
and gerundive factives reveals that what they all have in common and thus
seems to constitute the minimal condition for factive nominalization is the
instantiation of the process type which they reclassify. While gerundive
factives profile the instantiated nature of the nominalized process type (and
consequently foreground the role of the Subject), /Aai-structures realize
process types which are also grounded in the speech event.
Yet, the internal organization of factive nominalizations cannot itself
exhaustively explain how the meaning of factivity arises. Think, for in-
stance, of the gerundive nominalizations which I described as 'instantiated
actions'. Table 6 shows them to realize exactly the same options as do fac-
tive gerundives: they can follow the 'instantiation-of-a-type' model that is
characteristic of common nouns (e.g., ... and that is [her having a stroke]),
or they can follow the proper name strategy and function without determin-
ers (e.g., Going there was fun). In spite of the fact that their internal organi-
zation does not seem to differ from that of factive gerundives, these nomi-
nalizations are not factive: rather than representing metaphenomena, they
designate direct representations of experience. That these two subtypes of
gerundive nominalization realize different systems of nominalization only
becomes clear when we consider the constructions which they can be said
to agnate with: factive gerundives agnate systematically with the fact that-
constructions and /Aai-factives; instantiated action nomináis belong to a
paradigm which also contains generic action nominalizations designating a
class. The lack of internal functional differences between factive gerun-
dives and instantiated action gerundives illustrates that the analysis of fac-
tive nominalization cannot suffice with descriptions of the internal func-
tional properties of its various subtypes. As already suggested in Kiparsky
and Kiparsky (1971), factivity results from a combination of features and
does not depend on the nominalized construction only: the context in which
it is embedded is crucial as well. I hope to have shown, however, that the
almost exclusive focus on the external functioning of factive constructions
in the literature should be counterbalanced by careful descriptive analyses
of the nominalized constructions' internal functional organization. The de-

scription of the internal functioning of factive nominalizations does not
only establish the unique position of each subtype of factive nominalization
within the overall system, it also contributes to the identification of the
schematic category of factivity and of the various lexicogrammatical reali-
zations of it.
Chapter 10
Conclusion
In this study I have tried to show that it is in the dialectic between theory
and description that lies the key to an analysis of nominalization which
moves beyond existing descriptions and is able to shed a new light on spe-
cific nominalization types and on the process of nominalization in general.
My study has been based on the assumption that an accurate description of
specific nominalization patterns presupposes a coherent theoretical view on
the language system and an insightful description of nominal and clausal
categories. A significant part of it has therefore been devoted to the discus-
sion of a number of basic theoretical notions and to the introduction of
nominal and clausal functional categories. However, this study also makes
clear that it is descriptive analyses of nominalized constructions which
identify the theoretical constructs that are pertinent to the analysis of nomi-
nalization. It is also only through the description of nominalization patterns
that we can come to a better understanding of the constructional mecha-
nisms that lie behind the process of nominalization in general. The largest
part of this book has therefore been devoted not to theory, but to the de-
scription of two specific systems of nominalization.
The analysis of deverbal -er nominalization and of the various types of
factive nominalizations has provided new insight into the general process
of deverbal nominalization. Deverbal nominalization can now be defined as
the functional reclassification of a specific functional level of assembly in
the clause into a construction which does not only serve nominal functions
in larger configurations but internally makes use of nominal means of in-
stantiation, quantification and grounding. The analysis of nominalization
requires the following major steps: firstly, the level of assembly which the
nominalization derives from has to be identified; secondly, it has to be de-
termined whether the reclassification that applies to it involves rank shift or
not; thirdly, we have to identify the exact nominal strategy which the nomi-
nalized unit adopts; and, finally, the precise lexicogrammatical integration
of the clausal starting point with the nominal end-status of the nominalized
construction has to be considered: by bringing together their internal lexi-
cogrammatical features and their external functional behaviour, we can de-
scribe the unique semantic import of specific nominalization systems.
First, the various clausal levels of assembly which nominalization can
apply to turn out to coincide with the main functional levels in the organi-
250 Conclusion
zation of the clause (as identified in Chapter 4): nominalization can apply
to a simple process type specification or verb stem (e.g., bake -» baker,
sign -» the signing of the contract)·, it can start from a complex process
type specification or atemporal clausal head (e.g., playing the piano -»
[Her playing the piano] surprised us); nominalization can be based on a
non-finite clause (e.g., John playing the piano —» [John playing the piano]
surprised us), or it can apply to a finite clause (e.g., John plays the piano —>
[That John plays the piano] surprises us). The clausal categories which
deverbal nominalizations, unlike non-nominalized nomináis, imply (think,
for instance, of the categories of modal and temporal grounding in deverbal
-er nominalizations) are closely related to this clausal starting point of the
nominalization process.
Through nominalization, then, the various clausal levels of assembly are
turned into nominal constructions, i.e. they are reclassified into the nominal
class. While the nominal status of nominalizations derived from verb stems
is unproblematic, the internal nominal character of nominalizations based
on more elaborate units of clausal patterning is contentious. As my analysis
of gerundive nominalizations and /Aa/-factives reveals, the internally re-
classified nature of clause-like nominalizations shows up only when, firstly,
their downranked status is acknowledged and, secondly, when non-iconic
types of nominal coding are considered. The notion of downranking and,
behind it, the idea that language consists of levels of functional organiza-
tion which together make up a hierarchy or rank scale, is vital to the analy-
sis of nominalizations which contain units of clausal patterning: such
nominalizations establish reclassification by shifting a clausal configuration
of functions - with its internal clause-like structure intact - from clausal to
nominal rank. Through rank shift, a set of functions which normally figures
in the clause thus comes to function at nominal rank and is fitted into the
functional organization of the nominal. Generalizing, we can say that
deverbal nominalization always involves reclassification: in some cases
reclassification relates two units of the same rank (i.e. reclassification from
verb to noun). In other cases, reclassification is realized through rank shift.
To recognize the precise nominal paradigm in which rankshifted clausal
units come to function, a view of nominal organization is needed which is
radically functional and describes nominal composition in terms of compo-
nent functions which are not necessarily realized by discernible component
structures (see Chapter 2, Section 3.1). As the description of gerundive and
iAaZ-nominalizations makes clear, nominalizations, just like 'ordinary'
nomináis, do not necessarily realize the prototypical, iconic form of coding
that is characteristic of common nouns. An analysis of nominalizations
which tries to force them into what is felt to be the nominal prototype (i.e.
Conclusion 251
the common noun strategy) is therefore bound to fail and will not be able to
reveal the internal nominal status of the majority of nominalizations that
involve downranking.
The descriptive breakthrough needed to account for both iconically and
non-iconically coded nomináis and, consequently, to identify the internal
nominal properties of downranked clausal structures, was shown to be pro-
vided by Langacker's functional analysis of the nominal in terms of type
specification, instantiation, quantification and grounding. I have then
shown that, by systematically analyzing the internal organization of nomi-
nalizations from this functional perspective, their classification as nominal
can be based not only on their external functioning, but also on their inter-
nal functional behaviour.
The precise nominal paradigm which reclassified units fit into varies ac-
cording to the clausal level of assembly which is being nominalized. If a
verb stem or simple process type is nominalized, it is necessarily turned
into a common noun or nominal type specification (see Langacker 1991),
and this irrespective of the categories of instantiation and grounding that
may be implied in the relationship that is established in the resulting nomi-
nalization (think, for instance, of the analogy between the semantics of -er
nouns and the Subject-Finite unit). An atemporal clausal head is either
turned into a common noun (which is then instantiated and grounded by
means of determination) or, when instantiation is implied, it functions as a
full nominal or proper name. If the atemporalized process type specification
is construed as a unique instance, but then in type space, it follows the
proper name strategy of generic mass nouns. Likewise, a downranked finite
clause is reclassified and comes to function as a proper name, as also
pointed out by Langacker (1991). The same can, in my opinion, be argued
for nominalizations which apply to the fourth level of assembly which I
have identified, viz. that of ungrounded instances or non-finite clauses with
their Subject in the objective case (e.g., John playing the piano —» [John
playing the piano] surprised us).
Importantly, the functional characteristics of the clausal level of assem-
bly are thus integrated with those of the nominal construction in which the
reclassified unit comes to function. Downranked clausal units adopt in
some cases the common noun strategy (i.e. in gerundive nomináis of the
type his opening the door and that burning the midnight oil of yours), but
far more often, they function as full nomináis or proper names. When
nominalization involves rank shift, in other words, the default nominal
strategy seems to be that of proper names, which provide a unique name for
an instantiated process or for a unique process type (in generic gerundives).
In gerundive nomináis, it is only when the Subject-entity instantiating the
252 Conclusion
downranked process is foregrounded, that the common noun strategy ap-

plies and the Subject is explicitly realized as determiner (as in his opening
the door) or as postmodifier (as in any talking loudly of yours).
In short, rather than being inexplicable combinations of nominal-like ex-
ternal behaviour and clause-like internal categories, nominalizations are
basically nominal constructions, externally as well as internally. A full de-
scription of a nominalized construction, therefore, requires an analysis of
its external, nominal functioning, of the clause-like categories which it still
contains, as well as of the way in which the nominalized construction fits
these clause-like categories into the nominal paradigm. It is only when the
latter, third level of analysis is included in the description of nominaliza-
tions that their internal nominal status can be revealed and the lexicogram-
matical integration of clausal and nominal categories can be identified.
It is moreover in the integration of clausal and nominal categories which
nominalizations realize that important clues to their semantics lie. In my
analysis of -er nominalization, I have, for instance, argued that the seman-
tics of the noun type specification resulting from -er derivation is basically
analogous to that of the clausal Subject. The integration which -er nomi-
nalizations establish between a process type and an entity was shown to be
strikingly similar to the relationship which in the clause is realized by the
combination of Subject and Finite (and the categories of temporal and mo-
dal grounding realized in it). In my analysis of gerundive nominalizations,
then again, I showed that for gerundives to be able to receive a /active read-
ing in the clause, they have to integrate the clausal conception of an instan-
tiated process with a definitely grounded nominal structure. I argued that
gerundives can realize this integration in two ways: either they follow the
common noun strategy and use a determiner which does not only ground
the nominal but also instantiates the downranked process type specification;
or they construe the atemporal process type as a full (definite) nominal
whose instantiation of the downranked process type by a Subject is contex-
tually realized. Like factive ^/-clauses, factive gerundives were shown to
necessarily designate instances rather than types. Like them, they were
shown to nominalize these instances as definitely grounded nomináis. The
factive reading, I concluded, is not necessarily linked to finite structures,
but to instantiated processes which function as definite nomináis in factive
clausal contexts.
To summarize, the approach to nominalization which has been pre-
sented in this study breaks away from tradition in several respects. It shows
that reclassification is both external and internal: nominalized structures
that function as nomináis in larger configurations are reclassified also in-
ternally and take on the nominal behaviour of particular nominal strategies.
Conclusion 253
My approach thus complements the prevailing focus on the clausal catego-

ries that figure in nominalizations and considers nominalizations first and
foremost as nominal constructions which integrate the clausal and nominal
categories which they contain into one nominalized structure. Importantly,
my analysis of nominalization also moves away from the descriptive frag-
mentation that characterizes the literature on nominalization and offers a
unified account of 'morphological' types of nominalization (such as -er
derivation) and nominalizations which involve more elaborate, clause-like
structures. Finally, the analysis of deverbal -er nominalizations and of fac-
tive nominalization shows it is possible to link up the lexicogrammatical
properties of nominalizations with the meanings which they realize.
The study of nominalization patterns that has been presented here and
the overall framework which I have proposed for the analysis of nominal-
ized constructions is based on nominalization systems that exemplify the
various clausal levels of assembly which nominalization can start from.
Many nominalization types have, however, been left undiscussed. Un-
doubtedly, a detailed analysis of their lexicogrammatical and semantic
properties will modify and refine the present model. Among the nominali-
zation types that might be particularly interesting are nominalizations de-
rived by -ee suffixation and derived nomináis of the type the destruction of
the city. Action nominalizations such as the signing of the contract should
also be considered more closely. Unlike gerundive nominalizations, action
nominalizations and derived nomináis necessarily realize their participants
by means of nominal elements (i.e. nominal pre- and postmodifiers) and
determination in them is not restricted to the possessive type, but it includes
the article as well. As to -ee derivation, the increased productivity of agen-
tive -ee nomináis (e.g., escapee, attendee) has to be accounted for and the
resemblances and differences with -er derivation have to be identified.
Also, by elaborating especially on the internal structure of nominalizations,
I have chosen to focus on only part of the story: the analysis which I pro-
pose for the internal organization of nominalizations has to be supple-
mented with a systematic overview of the external contexts in which nomi-
nalizations can occur and of the broader, textual motives that inform their
occurrence.
Appendix
Non-agentive -er nominalizations
baker 'a food (meat, fruit or vegetable) that is suitable for baking'
bestseller 'a book or other item that sells well'
boiler 'a chicken suitable for boiling'
broiler 'a young chicken suitable for broiling'
cooker 'an apple for cooking*
diner 'aplace where you can dine'
dipper 'vegetable/fruit/other snack that has to be dipped before
being eaten'
dunker 'a doughnut that is preferably dunked before being eaten'
front-loader 'an appliance, such as a washing machine or a clothes drier
which allows for material to be inserted at the front'
fryer 'a young chicken or rabbit suitable for frying'
gusher 'a flowing oilwell'
jotter 'a small notebook'
keeper 'something that should be kept'
killers 'cattle suitable for killing'
kneeler 'a kind of chair, which you have to kneel on'
knocker 'a piece of metal on the front door of a house with which
you can knock on the door, to attract the attention of the
people inside*
loaner 'something that is loaned to someone'
low-loader 'a vehicle in which the carrying platform is kept low for
ease in loading'
poker 'a metal bar to move coal or wood in a fire'
reader 'a (course)book consisting of a compilation of literature '
roaster 'a kind of meat suitable for roasting'
scratcher 'a lottery ticket that one has to scratch to reveal the win-
ning patterns'
sipper 'a drink that one has to sip'
sleeper 'a train you can sleep in'
sleeper 'a bed or sofa in a train, in which you can sleep'
slipper(s) 'a kind of shoes that can be slipped into'
squeezer 'a kind of container that you have to squeeze in order to
extract something from it'
squirter 'a container that ejects liquid in a jet from a narrow open-
ing'
steamer 'an edible clam'
stepper 'a raised surface on which you have to put your foot, as a
way of working out'
256 Appendix
stroller 'a small chair on wheels, in which a baby or small child

can sit and be wheeled around'
top-loader 'a washing machine or clothes drier which can be loaded
from the top'
viewer 'a device for viewing photographic transparencies'
walker 'a special kind of frame which is designed to help babies or
disabled or ill people to walk'
Notes
1. It can be pointed out here that, as McGregor (1997: 2) remarks, in claiming

that grammar is a semiotic or symbolic system, "there is no implication that
every identifiable pattern in the organization of a language is necessarily se-
miotic". It is, for instance, possible that the fact that a particular morpheme
always occurs in final position does not have any semiotic significance. The
claim that language is semiotic should rather be taken to mean that "a certain
non-trivial subset of the organizing patterns are semiotically significant"
(McGregor 1997: 2).
2. Note that Halliday's interpretation of Hjelmslev (1961) is to a large extent
based on his own theoretical views on grammar: Hjelmslev's Prolegomena to
a Theory of Language itself hardly refers to grammar.
3. For this section, I have benefitted from discussion with Kristin Davidse.
4. Notice that, strictly speaking, even a nominal like those three black cats is not
fully canonical because the expression of quantity which it contains extends
beyond the (well-defined) constituent three to the demonstrative pronoun
(those instead of, for instance, this) and the head noun (cats instead of cat)
5. In Halliday (1979: 58), it is argued that the organization of a linguistic unit
may be considered from three different perspectives: at its own level ("its re-
lation to other elements identified at the same level as itself'), from above, i.e.
in terms of "its relation to elements at the next (or some) higher level" and
from below, or in "its relation to elements at the next (or some) lower level".
Halliday thus considers a third level of analysis of linguistic units to consist of
the syntagmatic relationships which a unit holds to other units at the same
(rank) level. I include syntagmatic relations under the 'external' functional re-
lations of a unit and define the third level differently, i.e. as the level of the
unit itself, where all external and internal relations come together.
6. The metaphor which Halliday uses to describe the rank scale is, in fact, remi-
niscent of the 'building-block' metaphor (Langacker 1987a, 1999): the vari-
ous levels of the rank scale are claimed to appear as it were "one inside the
other, in a sort of one-dimensional Chinese box arrangement" (Halliday 1961:
250).
7. Note that in Langacker's model, grammatical classes are, like constructions,
conceived as symbolic units or as bipolar linguistic symbols: the class of
nouns, for instance, "is defined by a schema that we can represent as
[[THING]/[...]], where [THING] is a schematic semantic unit (...) and [...] a
schematic phonological unit" (Langacker 1987a: 81).
8. Nomináis with a common noun as head are thus considered to be more proto-
typical than proper names. Van Ltingendonek (1999) is of the opinion that it is
not common nouns which are prototypical, but proper names, and this for
258 Notes
several reasons: because of their primary referring or identifying function, be-

cause the meanings which proper names have are of a presuppositional nature
and therefore more easily processed, and, finally, because proper names dis-
play the unmarked nominal feature of definiteness and are prototypically sin-
gular, countable, non-generic and concrete. Van Langendonck argues that it
is proper names rather than common nouns which receive a 'truly iconic moti-
vation', namely as nouns which mostly occur without an article (Van Langen-
donck p.c.). (Note that the characterization of prototypical nouns as serving
the pragmatic function of reference is due to Croft (2000), who argues that
cross-linguistically, certain unmarked or prototypical combinations of prag-
matic function and class can be observed: a noun, for instance, has the pri-
mary function of 'reference to an object', verbs have the function of 'predica-
tion of an action', and adjectives that of 'modification by a property'. It seems
to me, however, that Croft's pragmatic/ communicative functions are not
meant to point to prototypical instances within a class, but rather characterize
classes in general, as opposed to other classes. As such, they form a highly
schematic type of characterization and are valid for all members of the class.
Like Langacker's description of nomináis as 'things or regions in some do-
main', the characterization of nomináis as 'referring to an object' therefore
does not distinguish between nouns and nomináis. The schematic function of
'reference' as used by Croft (2000) should be distinguished from that of
'grounding', which applies to the link which nomináis - but also clauses - es-
tablish with the speech event. It is with respect to this link, or more specifi-
cally, with respect to the structural implementation of it, that common nouns
and proper names can be said to differ: unlike common nouns, proper names
do not need individually symbolized grounding predications tagged onto them
in the nominal to be grounded in the speech event; they are inherently
grounded.)
9. I will in Chapter 4 present a more elaborate analysis of the Subject. Note that
the analysis of the Subject that is given here is comparable to that presented in
Langacker (1991: 305-321), where prototypical Subjects are argued to rank
highly with respect to the so-called "topicality factors" (based on Givon 1976,
1978, as cited in Langacker 1991: 306): they are agentive, human, definite and
they constitute "the figure within the profiled relationship" (Langacker 1991:
308). However, Langacker also argues that a general characterization of the
Subject cannot be based on the notion of topic, but "must be subjective in na-
ture (i.e. a matter of construal rather than conceptual content" (Langacker
1991: 312). The definition which he proposes is that of 'relational figure'.
10. In Dik (1980), a distinction is made between 'semantic', 'syntactic' and
'pragmatic' functions. Semantic functions are functions such as Agent, Goal,
Recipient; the Subject and Object are regarded as syntactic, and, finally, func-
tions such as Theme and Topic are considered to be pragmatic. As Dik him-
self points out, however, the terms are somewhat unfortunate and chosen
"mainly for traditional reasons" (Dik 1980: 13). The so-called 'syntactic'
Notes 259
functions of Subject and Object, for instance, rather than being treated as se-
mantically 'empty', are given a semantic interpretation.
11. Like Halliday, moreover, McGregor (1997) relates the logical metafiinction to
the syntagmatic relation of dependency.
12. Importantly, in Cognitive Grammar, as in Systemic-Functional Grammar, it is
posited that there is "no fundamental distinction between morphological and
syntactic constructions, which are fully parallel in all immediately relevant
aspects" (Langacker 1987a: 82; see also Halliday 1961). Both morphological
and syntactic structures realize meaning in a natural or non-arbitrary way.
13. A similar distinction can be observed in the system of -ing nominalization: the
nominal opening in (1) is lexicalized, whereas opening in (2) forms an ad hoc
formation based on a schematic unit:
(1) He squeezed through a narrow opening in the fence. (CB)
(2) It entails a kind of patient regard, an opening of oneself toward
the subject.... (CB)
14. However elucidating the analysis of a construction's order of assembly may
be, it does not (always) tell the whole story: as pointed out by Langacker
(1987a, 1991, 1998, 1999), composite structures tend to exhibit partial rather
than full compositionality and, secondly, the contribution of their component
parts is not always recognized, i.e. they are not always fully analyzable. First,
as far as compositionality is concerned, the fact that components can be rec-
ognized within an assembly of symbolic units "does not entail that these com-
ponents exhaust its characterization", i.e. the composite whole "may have
properties above and beyond those of its components, which may in turn be
manifested in it only imperfectly" (Langacker 1987a: 87). The composite
structure is then either more specific than what can be computed from its
components, or it forms an extension of the expected compositional value
(Langacker 1999: 153). Nominalizations - particularly nominalizations at the
level of the word - are constant reminders of the partial nature of the compo-
sitionality of constructions. A structure like ruler, for instance, is more fre-
quently understood as a device for measuring things than as a device used for
ruling lines and its meaning thus significantly extends the expected composi-
tional value; similarly, a computer is much more than just 'something that
computes' (Langacker 1999: 128). In short, any assembly of symbolic struc-
tures must be treated as a separate entity in its own right and "with many pos-
sible sources of extracompositionality" (Langacker 1998: 4), such as dis-
course and context, particular domains of knowledge, mental spaces, meta-
phor and metonymy. The discrepancy between an expression's expected com-
positional value and the actual meaning that becomes its conventional value
facilitates a decline in the analyzability of the construction (Langacker 1999:
128): the language user becomes less cognizant of the components of the as-
sembly or no longer co-activates the composite structure and its components
(Langacker 1987a: 462). Analyzability is a matter of degree: while in a nomi-
nalization like teacher, the components may still be perceived, this is less true
for, for instance, nomináis like propeller and drawer. Many of the -er nomi-
260 Notes
nalizations that designate items of clothing likewise seem to have lost much of
their analyzability, to the extent even that nomináis like jumper, sneakers and
loafers no longer seem to be recognized as being derived from the processes
ofjumping, sneaking and loafing.
15. The systemic-functional view of nominalizations is closely tied up with the
concept of 'grammatical metaphor': nominalization is presented as a major re-
source for the creation of metaphorical rather than 'typical' or congruent lexi-
cogrammatical realizations of semantic categories. A nominalization such as
the cast's brilliant acting, for instance, is thus viewed as the metaphorical
counterpart of the clause the cast acted brilliantly (Halliday and Matthiessen
1999: 229). To describe the relationship between the congruent and the meta-
phorical realizations, Gleason's concept of agnation is used and it is claimed
that every metaphorical structure has "an agnate form corresponding to its
congruent realization" (Ravelli 1988: 141).
16. Agnates that help to distinguish gerundive nominalizations from action nomi-
nalizations are, for instance, non-finite modifiers (e.g., Any boy [cleaning the
kitchen] will be rewardedfor it) and non-finite clauses (e.g., With [Tom clean-
ing the kitchen], we can be sure that it will be very clean). Both structures ag-
nate systematically with gerundive nominalizations but not with action nomi-
náis: like gerundive nomináis, for instance, they allow for imperfective proc-
esses (conpare, for instance, any boy [being prepared to help] and with [Tom
being prepared to help] with the gerundive nominal [Tom's being prepared to
help] surprised me and with the action nominal * Tom's being of prepared;
*Tom's being of a good husband.
17. Plurals and non-plural mass nouns are grouped together by Langacker (1991)
because they behave alike in numerous respects. They can, for instance, both
occur as full nomináis without quantifier or overt grounding predication (e.g.,
They're looking for diamonds/ gold); they do not tolerate the indefinite article
(e.g., *a diamonds/ gold); and both can be used with a number of quantifiers
not permitted with count nouns (e.g., most diamonds/ gold/ *diamond; a lot of
diamonds/ gold/ *diamond) (see Langacker 1991: 77).
18. Note that for a plural, "the size of the instance will be given by the number of
its component entities", but collectively, these entities "constitute just one in-
stance of the plural-noun type" (Langacker 1991: 81).
19. As pointed out by Langacker (1991: 54), even in those three black cats, the
different structural levels are not realized entirely separate from each other:
number is indicated not only by the quantifier three, but also by the grounding
predication those and by the head noun.
20. Only when the clausal head is progressive in profile, is -ing impossible: the
auxiliary pair be playing can thus not be turned into an -ing form (*being
playing). Langacker (1991: 232) attributes this to the fact that "simpler alter-
natives are available that do essentially the same job, i.e. they provide effec-
tively equivalent information without creating potentially awkward ambigui-
ties". The -ing form playing is a simpler alternative for being playing and
therefore makes the latter redundant.
Notes 261
21. Important evidence in favour of the type status of atemporal clausal heads
comes from gerundive nominalizations and the way in which they function in
the structure of the nominal. I will not, however, go into their properties here
and refer to Chapter 9 for a discussion of the status of gerundive nominaliza-
tions as atemporalized process type specifications.
22. While both Halliday (1994) and Langacker (1991) agree on the grounding
function of the finite element in the clause, they differ as to where exactly
they draw the boundary between the grounding and the non-grounding part of
the verb group: Langacker considers only the actual markers of tense and mo-
dality as grounding; in Halliday's analysis, auxiliaries like did/does/was/
had/have are considered to be part of the finite or grounding element, and this
primarily because they are picked up in tags, such as didn't you?, has he?,
wasn't she?. In Langacker's model, the verb group has responded can thus be
divided into -s (which grounds the verb group) and have responded (which
constitutes the non-grounding, head-part of the verb group). In Halliday
(1994), that same verb group is analyzed as has (grounding) and responded
(non-grounding). When no auxiliaries are present, Halliday's analysis of the
verb group coincides with that in Langacker (1991): a verb group such as re-
sponded is divided into -ed and respond.
23. In Davidse ( 1997) and Heyvaert (2000), the view is defended that the clause is
instantiated (as well as grounded) by the Subject-Finite complex, rather than
by the Subject only. Especially on the basis of my analysis of gerundive
nominalizations (presented in Chapter 9), I have come to an analysis which is
slightly different in that it considers only the Subject to fulfill an instantiating
role. I view the function of the finite or grounding element of the verb group
as being that of grounding the instance in the speech event (which, of course,
presupposes instantiation, but crucially does not foreground it). As argued by
Davidse (1997), the Subject then co-operates in the grounding by establishing
objective or explicit person deixis (see Section 2.3.2). The Subject thus serves
two distinct, though related functions: it instantiates the process type specifi-
cation of the clause and, if the clausal head is finite, it co-operates in the
grounding.
24. The notions of 'subjective' and 'objective' as they are used here in the context
of modality should be distinguished from the use which Langacker (1985,
1990b) makes of them: in Langacker's work they do not refer to whether or
not a linguistic element is related to the speaker, but they relate to how explic-
itly the speaker is present in an utterance. The more explicit the speaker fig-
ures in the utterance, the more 'objective' its status is said to be. As Verstraete
(2001: 1513) points out, the notions of subjectivity and objectivity used by
Langacker are not entirely unrelated to the notions used in the classification of
modality: "they can be regarded as an additional dimension for the subjective
pole, depending on how explicitly the speaker-relatedness figures in the utter-
ance".
25. The subjective deontic modality of obligation/permission is expressed by the
prototypical uses of deontic auxiliaries such as must, should, shall, may.
262 Notes
26. The objective deontic modality of obligation and permission is typically real-
ized by non-Speaker-Hearer-oriented deontic modal auxiliaries, such as have
to, ought to, can.
27. Verstraete (2002: 65) observes that, when the objective modal auxiliary is a
periphrastic modal such as have to or be able to, the schematic, epistemic mo-
dality that accompanies objective modals can surface in the form of an epis-
temic modal auxiliary, as in He may have to leave his house; She might be
able to come over after all.
28. Note that the notions of 'subjective' and 'objective' are used here in a mean-
ing that is related to the way in which they are used by Langacker (1985,
1990b): the Subject is said to establish person deixis objectively because it
puts the person deixis 'on stage' as a maximally salient entity; the person
deixis that is realized in the finite element, on the other hand, is subjective, or
minimally salient: it remains implicit (based on Langacker 1991: 93).
29. In Langacker (1991: 247), a similar claim is made with respect to Subject-
verb agreement: "what is traditionally known as 'subject-verb agreement' is
analyzed as being part of the grounding predication".
30. An extensive list of English non-agentive -er nominalizations has been in-
cluded in the appendix.
31. The distinction between lexicalized and 'schematic' or 'ad hoc' nominaliza-
tions was introduced in Part I, Chapter 3, Section 1.2.
32. This view on English transitivity, in which a fundamental split is posited be-
tween transitive and ergative clause construals is proposed in Davidse (1992b,
1998b) and Davidse and Geyskens (1998). Davidse's view is rooted in the
British structural-functional tradition of Halliday (1967, 1968), Lyons (1969)
and Cruse (1972, 1973, as mentioned in Davidse and Geyskens 1998). It
should be distinguished from Perlmutter's (1978) distinction between 'unerga-
tive' and 'unaccusative' verbs: broadly speaking, one can say that Perlmut-
ter's 'unaccusative' category includes all the ergative verbs of the structural-
functional tradition, as well as intransitive verbs such as fall, die and stumble
(for a detailed comparison of the two traditions, see Davidse and Geyskens
1998).
33. Following Levin and Rappaport (1988), Lemmens claims that Goal-profiling
-er nomináis are related to the middle use of the transitive base verb: "as with
the middle construction the focus is on the properties of the Affected" (Lem-
mens 1998: 137). Lemmens's account of the relation between Goal-profiling
-er nomináis and middle formation will be discussed in Chapter 7, Section 2.
34. Interestingly, a category which is not included in Lemmens's account (it was
not accounted for in Ryder's approach either) is that of -er nomináis that pro-
file so-called 'oblique participants' such as jotter, stepper, kneeler, viewer,
stroller (I will explain the notion of 'oblique participant' (based on Laffut
2000) in Section 1.2). Together with the Goal- and Medium-profiling group of
-er nomináis, they make up the non-agentive type of -er nominalization. I will
go more deeply into this type of -er nominalization in Section 1.2.1.
Notes 263
35. In fact, it has been pointed out by Laffut (2000: 145) that the Locatum is the
Patient or primary affected of the process, but "because of the involvement in
the locative process, the Location can be analyzed as a second affected", both
when it is realized prepositionally and when it functions as a bare noun
phrase.
36. Deverbal nominalization types which have been analyzed in the literature as
offering an active and a passive voice option are those of -ion and -ment suf-
fixation, as in:
(1) a. the critics' hostile reception of the play (active, 'they received
the play')
b. the play's hostile reception by the critics (passive, 'the play was
received') (Quirk et al. 1985: 1289)
(2) a. the committee's appointment of John (active, 'they appointed
him')
b. John's appointment by the committee (passive, 'he was appointed')
(Lees 1968: 68)
37. Many aspects of the description of the middle proposed in this chapter are also
discussed in Heyvaert (1997), Heyvaert (1998) and Davidse and Heyvaert (in
press).
38. Structures like these have been variously referred to as activo-passives (Jes-
persen 1914-1929, 3: 347), medio-passives (e.g., Declerck 1991b: 203), proc-
ess-oriented passives (Halliday 1967), facility-oriented passives (Fawcett
1980: 148), patient-subject constructions (Van Oosten 1977: 459) and pseudo-
intransitives (Lyons 1969: 366; Smith 1978: 103). I will use the term 'middle
construction' (as used by, among others, Keyser and Roeper 1984; Fellbaum
1986; Hale and Keyser 1987; Fagan 1992; Kemmer 1993).
39. Hence labels like activo-passive, medio-passive and patient-subject construc-
tion.
40. Even if The door closed slowly is analyzed as a middle construction, it is
turned into an ergative one-participant construction as soon as the phrase all
by itself is added to it: The door closed slowly all by itself czanot but be inter-
preted as an ergative one-participant construal.
41. Note that the necessarily unexpressed nature of the Agent in middle construc-
tions has inspired some authors to call them pseudo-intransitives (Lyons
1969; Smith 1978).
42. For the notion of 'representational' or 'ideational' fonctions, see Chapter 2,
Section 4.
43. The intransitive type of middle is, in fact, well established in Dutch and has
received a fair amount of attention in the literature on Dutch middle
constructions (e.g., De Vries 1910; Holierhoek 1980; Van den Toorn 1982;
Hoekstra and Roberts 1993; Ackema and Schoorlemmer 1994; Peeters 1999).
44. The IDL corpus is the 38 million word corpus of the Instituut voor Neder-
landse Lexicografie 'Institute for Dutch Lexicography' in Leiden.
45. Intransitive middles in Dutch agnate with impersonal constructions in which
the oblique participant is realized as a prepositional phrase:
264 Notes
(1) Het fietst prettiger op asfalt dan op grind.

it cycles more pleasantly on asphalt than on gravel
'it is more pleasant to cycle on asphalt than gravel'
(2) Het zit lekkerder in sommige bomen.
it sits more comfortably in some trees
'it is more comfortable to sit in certain trees'
(3) Het loopt lastig in nieuwe schoenen.
it walks difficult in new shoes
'it is difficult to walk in new shoes'
46. Ehrman (1966: 74) points out that passive clauses with can often highlight
that "there are certain positive qualities of the subject such that the way is
cleared for the predication", as in These engines can be removed from a boat
with relative ease (Ehrman 1966: 14). Passives like these resemble middles in
that they focus on the properties of the non-agentive Subject entity. The fact
that they make use of a passive verb, however, constitutes an important differ-
ence with middle constructions.
47. Middles have often been claimed to be necessarily generic (e.g., Keyser and
Roeper 1984; Roberts 1986; Fagan 1992; Masam 1992; Sohn 1998). As also
pointed out in Rosta (1995) and Iwata (1999), this is not true: middle con-
structions can just as well designate specific events. An example of a non-
generic middle construction is The lap portion of the car seat belt slotted into
place easily (CB).
48. Observe that will can also be used in middles to make "a prediction or a gen-
eralization about someone's behaviour" (Declerck 1991b: 413-414), as in Her
face is her best asset because she has very good bone structure which will
photograph well (CB).
49. These objections were raised in discussions of Dutch -er nominalization in
Van Langendonck 1988 and Booij 1986 (discussed in Chapter 5). They can,
however, be argued to apply to English -er suffixation as well.
50. Notice that, unlike Ryder (1991, 1999b) and Lemmens (1998), I categorize
nomináis related to processes of cooking (broiler, boiler, baker, cooker) with
the non-agentive group of -er nominalizations. I have two reasons for that.
Firstly, even though these nomináis profile participants which co-participate
in the process, they clearly also have a (human) Agent in their base which
does the broiling and cooking. Secondly, as I will show in Section 2 of this
chapter, the semantics of -er nomináis of cooking comes closer to that of non-
agentive, middle-related -er nomináis than to that of agentive nomináis (as
also pointed out in Levin and Rappaport 1988).
51. A comparable reading is plausible for the nominal diner, which may have
been construed analogously to sleeper, in the Oxford English Dictionary,
diner is defined as 'a railway dining car' and the attested examples which are
given of it are all drawn from the railway context, e.g., A new dining car
which ...is the first diner... built by that company.
52. As I pointed out in Chapter 3, many of the -er nominalizations designating
items of clothing are characterized by a certain loss of analyzability (Lan-
Notes 265
gacker 1987a, 1999) and are no longer recognized as -er derivations: are
nomináis such as jumper, loafers and sneakers still analyzable as deriving
from the processes ofjumping, loafing and sneaking? Or have they lost their
verbal meaning and evolved into simple nouns?
53. The idea that food is raised to be processed in a particular way is clearly be-
hind the following corpus example: ... can you supply me with any informa-
tion about this apple? Is it an eater or cooker and how long does it keep?
(CB).
54. The nominal singer can of course also have its lexicalized meaning here and
designate a professional singer.
55. I follow Noonan's (1985) definition of (clausal) complementation here, which
is in line with the literature on factive complementation in general (e.g.,
Rosenbaum 1967; Kiparsky and Kiparsky 1971; Stockwell, Schachter and
Partee 1973; Menzel 1975): "By complementation we mean the syntactic
situation that arises when a notional sentence or predication is an argument of
a predicate. For our purposes, a predication can be viewed as an argument of a
predicate if it functions as the subject or object of that predicate" (Noonan
1985: 42). Clausal complements differ in this respect from non-clausal or
nominal complements, which are prototypically conceived of as referring to
Objects, rather than to both Objects and Subjects (e.g., Radford 1997: 498,
519; Haegeman and Guéron 1999: 21, 124).
56. As Delacruz (1976: 180) puts it, "Possibly whenever the use of a predicate in
a sentence involves a presupposition one could say that the sentence as a
whole has or carries a presupposition. And if a sentence carries a presupposi-
tion, one might say that whoever asserts it presupposes whatever is presup-
posed by the sentence".
57. As pointed out in Davidse (1994: 280-281), there is a special sense associated
with fronted locutions. The sentence That he had been wrong was definitely
said by him, for instance, has a semantic value which differs from that of He
said that he had been wrong: "On the one hand, ... it specifies the content of
the Sayer's speech act. On the other hand, the embedded status of the locution
entails speaker commitment to the 'fact' that this locution was produced by
the Sayer" (Davidse 1994: 281). Also structurally speaking, fronted locutions
come close to facts: they can function as Subjects of the passive; they cannot
be substituted, but only be referred to (That was definitely said by him). They
can function as second nominal in appositive constructions (The words: 'Iwas
wrong' were definitely said by him, embedded locution: quote). Davidse
therefore proposes to consider the fronted type of locutions as embedded con-
stituents, rather than as tactically related clauses.
58. As correctly pointed out by Wierzbicka (1988: 49), this is "a contradiction in
terms", which "undermines the whole idea that semantic factors (such as fac-
tivity) 'account for the complement system of English, and other languages as
well'".
266 Notes
59. This has inspired Noonan (1985: 60) to coin the term 'nominalized comple-
ments' for them, to be distinguished from fAai-structures and infinitivals
which are termed 'complements'.
60. As I will point out in Section 2.2, not all clausal contexts allow for replace-
ment of a the fact ίΑαί-clause by a that-clause. This has primarily to do with
the higher degree of 'nominality' of the fact /Aai-clauses: they can be fitted
into more clausal contexts than can /AaZ-clauses. In Subject position, however,
the fact that- structures are typically replaceable by í/taí-structures.
61. Structures can be ambiguous between a restrictive relative reading and an ap-
positive, factive one, witness the nominal the fact that she forgot in The fact
that she forgot is unimportant (Langacker 1991: 432). On the relative clause
interpretation, the fact is interpreted as a participant in the relative clause: she
forgot the fact. In its factive reading, in contrast, that she forgot is what con-
stitutes the fact.
62. The possessive pronouns in gerundive nominalizations can also be turned into
objective case (see, among others, Sweet 1898; Poutsma 1929; Lees 1960;
Declerck 1991b): e.g., We have always been a close couple and shared every-
thing, so I never resented [him leaning on me when he was depressed]. (CB);
I do not like [him coming here so ofien], (Sweet 1898: 120).
References
Databases:
COBUILD corpus, Birmingham
INL corpus, Leiden
Ackema, Peter, and Maaike Schoorlemmer

1994 The middle construction and the syntax-semantics interface. Lin-
gua 93: 59-90.
Bauer, Laurie
1983 English Word-Formation. Cambridge: Cambridge University
Press.
Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad, and Edward
Finegan
1999 Longman Grammar of Spoken and Written English. London:
Longman.
Booij, Geert
1986 Form and meaning in morphology: the case of Dutch 'agent
nouns'. Linguistics 24: 503-517.
1992 Morphology, semantics and argument structure. In Thematic
Structure. Its Role in Grammar, I. M. Roca (ed.), 47-67. Berlin:
Foris Publications.
Boon, René
1972 -£r-afleiding in modern Nederlands. Synchronische analyse en
inleidend onderzoek [-Er derivation in modern Dutch. A syn-
chronic analysis and introductory investigation]. M.A. thesis,
Department of Linguistics, University of Leuven.
Carlson, Greg
1978 Reference to Kinds in English. Bloomington: Indiana University
Linguistics Club.
Chomsky, Noam
1970 Remarks on Nominalization. In Readings in English Transforma-
tional Grammar, R. A. Jacobs, and P. S. Rosenbaum (eds.), 184—
221. Waltham, Massachusetts: Ginn.
Comrie, Bernard
1976 The syntax of action nomináis: a cross-language study. Lingua
40: 177-201.
Comrie, Bernard, and Sandra A. Thompson
1985 Lexical nominalization. In Language Typology and Syntactic De-
scription III, T. Shopen (ed.), 349-407. Cambridge: Cambridge
University Press.
268 References
Croft, William
2000 Parts of speech as language universale and as language-particular
categories. In Approaches to the Typology of Word Classes, P.
M. Vogel, and B. Comrie (eds.), 65-102. Berlin/New York:
Mouton de Gruyter.
Dalton-Puffer, Christiane
1994 Are Shakespeare's agent nouns different from Chaucer's? - On
the dynamics of a derivational sub-system. In Studies in Early
Modern English, D. Kastovsky (ed.), 45-58. Berlin/New York:
Mouton de Gruyter.
Davidse, Kristin
1991 Categories of experiential grammar. Ph. D. diss., Department of
Linguistics, University of Leuven.
1992a On the real function of Subject. Paper delivered at the 19th Inter-
national Systemic-Functional Congress, Sydney, 13-18 July
1992.
1992b Transitivity/ergativity: the Janus-headed grammar of actions and
events. In Advances in Systemic Linguistics: Recent Theory and
Practice, M. Davies, and L. Ravelli (eds.), 105-135. London:
Pinter Publishers.
1994 Fact projection. In Perspectives on English. Studies in Honour of
Professor Emma Vorlat, K. P. Carlon, K. Davidse, and B.
Rudzka-Ostyn (eds.), 259-286. Leuven: Peeters.
1996 Turning Grammar on itself: Identifying Clauses in Linguistic
Discourse. In Meaning and Form, Systemic Functional Interpre-
tations, Meaning and Choice in Language: Studies for Michael
Halliday, M. Berry, C. Butler, R. Fawcett, and G. Huang (eds.),
367-393. Norwood, New Jersey: Ablex Publishing Corporation.
1997 The Subject-Object versus the Agent-Patient asymmetry. Leuven
Contributions in Linguistics and Philology. 86 (4): 413-431.
1998a Agnates, verb classes and the meaning of construals. The case of
ditransitivity in English. Leuven Contributions in Linguistics and
Philology. 87 ( 3 ^ ) : 281-313.
1998b On transitivity and ergativity in English, or on the need for dia-
logue between schools. In English as a Human Language: to
Honour Louis Goossens, J. Van der Auwera, F. Durieux, and L.
Lejeune (eds.), 95-108. Muenchen: Lincom.
1999 The semantics of cardinal versus enumerative existential con-
structions. Cognitive Linguistics 10 (3): 203-250.
2000a Semiotic and possessive models in relational clauses: thinking
with grammar about grammar. Revista Canaria de Estudios In-
gleses 40: 13-35.
2000b A constructional approach to clefts. Linguistics 38 (6): 1101—
1131.
References 269
2001 Postdeterminers: their secondary identifying and quantifying

functions. Preprint nr. 177, Department of Linguistics, University
of Leuven.
2003 A corpus check of the factive presupposition. In Configurations
of Culture. Essays in Honour of Michael Windross, A. Remael,
and K. Pelsmaekers (eds.), 115-126. Antwerpen: Garant.
Davidse, Kristin, and Sara Geyskens
1998 Have you walked the dog yet? The ergative causitivization of
intransitives. WORD 49(2): 155-180.
Davidse, Kristin, and Liesbet Heyvaert
in press On the so-called 'middle' construction in English and Dutch. In
Empirical Approaches to Contrastive Linguistics and Transla-
tion Studies, S. Granger, J. Lerot, and S. Petch-Tyson (eds.).
Amsterdam: Rodopi.
Davies, Eirian
1979 On the Semantics of Syntax. Mood and Condition in English.
London: Croom Helm.
2001 Propositional attitudes: a semantic model of epistemic mood and
modality in English. Functions of Language 8 (2): 217-250.
Declerck, Renaat
1986 The manifold interpretations of generic sentences. Lingua 68:
149-188.
1991a Tense in English. Its Structure and Use in Discourse. London:
Routledge.
1991b A Comprehensive Descriptive Grammar of English. Tokyo: Kai-
takusha.
Delacruz, Enrique B.
1976 Factives and proposition level constructions in Montague Gram-
mar. In Montague Grammar, B. H. Partee (ed.), 177-199. New
York: Academic Press.
De Vries, Wobbe
1910 Opmerkingen over Nederlandsche syntaxis. [Comments on
Dutch syntax]. Tijdschrift voor Nederlandsche Taal- en Letter-
kunde [Journal of Dutch Linguistics and Literature] 29: 122-165.
Dik, Simon C.
1967 Some critical remarks on the treatment of morphological struc-
ture in transformational generative grammar. Lingua 18: 352-
383.
1980 Studies in Functional Grammar. London: Academic Press.
1997 The Theory of Functional Grammar. Part II. Complex and De-
rived Constructions. K. Hengeveld (ed.). Berlin/New York:
Mouton de Gruyter.
Ehrman, Madeline E.
1966 The Meanings of the Modals in Present-day American English.
The Hague: Mouton.
270 References
Fagan, Sarah
1992 The Syntax and Semantics of Middle Constructions. Cambridge:
Cambridge University Press.
Fawcett, Robin P.
1980 Cognitive Linguistics and Social Interaction. Towards an Inte-
grated Model of a Systemic Functional Grammar and the Other
Components of a Communicating Mind. Exeter/ Heidelberg:
Julius Groos Verlag and Exeter University Press.
Fellbaum, Christiane
1985 Adverbs in agentless actives and passives. Papers from the
Parasession on Causatives and Agentivity at the Twenty-first Re-
gional Meeting of the Chicago Linguistic Society. 21-31.
1986 On the Middle Construction in English. Indiana: Indiana Univer-
sity Linguistics Club.
Fellbaum, Christiane, and Anne Zribi-Hertz
1989 The Middle Construction in French and English: A Comparative
Study of its Syntax and Semantics. Indiana: Indiana University
Linguistics Club Publications.
Francis, Gill, Susan Hunston, and Elizabeth Manning
1996 Collins COBUILD Grammar Patterns 1: Verbs. London:
HarperCollins.
Fraser, Bruce
1970 Some Remarks on the Action Nominalization in English. In
Readings in English Transformational Grammar. R. A. Jacobs,
and P. S. Rosenbaum (eds.), 83-98. Waltham, Massachusetts.
Gleason, Henry A.
1965 Linguistics and English Grammar. New York: Holt, Rinehart
and Winston.
Haas, William
1954 On defining linguistic units. Transactions of the Philological
Society. 54-84.
Haegeman, Liliane, and Jacqueline Guéron
1999 English Grammar. A Generative Perspective. Oxford: Blackwell
Publishers.
Haeseryn, Walter, K. Romijn, Guido Geerts, J. de Rooij, and M. C. van den Toorn
(eds. )
1997 Algemeen Nederlandse Spraakkunst [General Dutch Grammar].
Groningen: Martinus Nijhoff, Deurne: Wolters Plantyn.
Haiman, John, and Sandra A. Thompson
1984 'Subordination' in Universal Grammar. Proceedings of the Tenth
Annual Meeting of the Berkeley Linguistics Society. 510-524.
Hale, Kenneth, and Jay Keyser
1987 A View from the Middle. Lexicon Project Working Papers 10.
Cambridge: MIT.
References 271
Halliday, Michael A. K.
1961 Categories of the Theory of Grammar. Word 17 (3): 241-292.
1966a Some notes on 'deep' grammar. Journal of Linguistics 2: 57-67.
1966b The concept of rank: A reply. Journal of Linguistics 2: 110-118.
1967 Notes on transitivity and theme in English. Part I. Journal of
Linguistics 3 (1): 37-81.
1968 Notes on transitivity and theme in English. Part III. Journal of
1970a Functional diversity in language as seen from a consideration of
modality and mood in English. Foundations of Language 6 (3):
322-361.
1970b Language structure and language function. In New Horizons in
Linguistics, J. Lyons (ed.), 140-165. Harmondsworth: Penguin
Books.
1976 System and Function in Language. G. R. Kress (ed.). London:
Oxford University Press.
1979 Modes of meaning and modes of expression: types of grammati-
cal structure, and their determination by different semantic func-
tions. In Function and Context in Linguistic Analysis: Essays Of-
fered to William Haas, D. J. Allerton, E. Carney, and D. Hold-
croft (eds.), 57-79. Cambridge: Cambridge University Press.
1985 An Introduction to Functional Grammar. London: Edward Ar-
nold.
1988 On the ineffability of grammatical categories. In Linguistics in a
Systemic Perspective. J. D. Benson, M. J. Cummings, and W. S.
Greaves (eds.), 27-51. Amsterdam/Philadelphia: John Benja-
mins.
1992 How do you mean? In Advances in Systemic Linguistics. Recent
Theory and Practice, M. Davies and L. Ravelli (eds.), 20-35.
London: Pinter.
1994 An Introduction to Functional Grammar (2nd ed.). London: Ed-
ward Arnold.
Halliday, Michael A. K., and Ruqaiya Hasan
1976 Cohesion in English. London: Longman.
Halliday, Michael A. K., and Christian M. I. M. Matthiessen
1999 Construing Experience Through Meaning. A Language-based
Approach to Cognition. London: Cassell.
Haugen, Einar
1953 On resolving the close apposition. American Speech 28: 165-
170.
Hengeveld, Kees
1989 Layers and operators in Functional Grammar. Journal of Linguis-
tics 25: 127-157.
272 References
Heyvaert, Liesbet
1997 Patientive -er nomináis. Leuven Contributions in Linguistics and
Philology 86 (4): 433-456.
1998 Nonagentive deverbal -er nominalization in English and Dutch: a
contrastive analysis. Languages in Contrast 1 (2): 211-243.
2000 Gerundive nominalization: from type specification to grounded
instance. In Constructions in Cognitive Linguistics. Selected Pa-
pers from the Fifth International Cognitive Linguistics Confer-
ence, Amsterdam, 1997, A. Foolen, and F. van der Leek (eds.),
103-121. Amsterdam/Philadelphia: John Benjamins.
2001 Nominalization as an interpersonally-driven system. Functions of
Language 8 (2): 287-329.
in press Nominalization as grammatical metaphor: on the need for a radi-
cally systemic and metafìinctional approach. In Metaphor in Sys-
temic-Functional Perspectives, L. Ravelli, A.-M. Simon-
Vandenbergen, and M. Taverniere (eds.). Amsterdam/ Philadel-
phia: John Benjamins.
Hjelmslev, Louis
1961 Prolegomena to a Theory of Language. Translated by F. J. Whit-
field. Madison: University of Wisconsin Press.
Hoekstra, Teun, and Ian Roberts
1993 Middle constructions in Dutch and English. In Knowledge and
Language. Volume 2. Lexical and Conceptual Structure, E. Reu-
land, and W. Abraham (eds.), 183-220. The Netherlands: Kluwer
Academic Publishers.
Holierhoek, C. M. G.
1980 Werkwoorden van Waarneming. Aspekten van hun Systematiek
en Ontwikkeling [Verbs of perception. Aspects of their function-
ing and development]. Den Haag: Universitaire Pers Leiden.
Hooper, Joan B.
1975 On assertive predicates. In Syntax and Semantics. Volume 4,
J.P. Kimball (ed.), 91-124. New York: Academic Press.
Hopper, Paul J., and Sandra A. Thompson
1980 Transitivity in grammar and discourse. Language 56: 251-299.
Hopper, Paul J., and Elizabeth C. Traugott
1993 Grammaticalization. Cambridge: Cambridge University Press.
Horn, George M.
1975 On the Nonsentential Nature of the POSS-ING Construction.
Linguistic Analysis 1 (4): 333-387.
Huddleston, Rodney
1969 Review article on Madeline Ehrman, 'The meanings of the mo-
dals in present-day American English'. Lingua 23: 165-176.
1988 Constituency, multifunctionality and grammaticalization in Hal-
liday's functional grammar. Journal of Linguistics 2: 137-174.
References 273
Hye Suk Yoon, James

1996 Nominal gerund phrases in English as phrasal zero derivation.
Linguistics 34: 329-356.
Iwata, Seizi
1999 On the status of an implicit arguments in middles. Journal of
Linguistics 35: 527-553.
Jespersen, Otto
1914-1929 A modern English grammar on historical principles. 7 volumes.
London: George Allen and Unwin.
Karttunen, Lauri
1971 Some observations on factivity. Papers in Linguistics 4: 55-69.
Kastovsky, Dieter
1971 The old English suffix -er(e). Anglia. Zeitschrift für Englische
Philologie. 89: 285-325.
1985 Deverbal nouns in Old and Modern English: from stem-
formation to word-formation. In Historical Semantics: Historical
Word-Formation, J. Fisiak (ed.), 221-261. Berlin/New York:
Mouton de Gruyter.
1986a The problem of productivity in word formation. Linguistics 24:
585-600.
1986b Diachronic word-formation in a functional perspective. In Lin-
guistics across Historical and Geographical Boundaries. In
Honour of Jacek Fisiak on the Occasion of His Fiftieth Birthday.
Volume 1: Linguistic Theory and Historical Linguistics, D.
Kastovsky, and A. Szwedek (eds.), 409-421. Berlin/New York:
Mouton de Gruyter.
Keenan, Edward L.
1971 Two Kinds of Presupposition in Natural Language. In Studies in
Linguistic Semantics, C. Fillmore, and D. Langendoen (eds.),
45-52. New York: Holt, Rinehart and Winston.
1976 Towards a universal definition of 'Subject'. In Subject and
Topic, C. N. Li (ed.), 303-333. New York: Academic Press.
Kemmer, Suzanne
1993 The Middle Voice. Amsterdam/Philadelphia: John Benjamins.
Keyser, Samuel J., and Thomas Roeper
1984 On the middle and ergative constructions in English. Linguistic
Inquiry 15: 381—416.
Kiparsky, Paul, and Carol Kiparsky
1971 Fact. In Semantics: An Interdisciplinary Reader in Philosophy,
Linguistics, and Psychology, D. D. Steinberg, and L. A. Jakobo-
vits (eds.), 345-369. Cambridge: Cambridge University Press.
Kruisinga, E.
1932 A handbook of present-day English. 5th edition. Groningen:
Noordhoff.
274 References
Kruisinga, E., and P.A. Erades

1953 An English Grammar. 8th edition. Groningen: Noordhoff.
Kuno, Susumu
1970 Some properties of non-referential noun phrases. In Studies in
General and Oriental Linguistics. Presented to Shirô Hattori on
the Occasion of His Sixtieth Birthday, R. Jakobson, and S. Ka-
wamoto (eds.), 348-373. Tokyo: TEC Company.
1973 Constraints on Internal Clauses and Sentential Subjects. Linguis-
tic Inquiry 4 (3): 363-385.
Laffiit, An
2000 The locative, image impression and material/product construc-
tions. A functional description of the impact of NG-PrepP varia-
tion. Ph. D. diss., Department of Linguistics, University of Leu-
ven.
Lakoff, George
1977 Linguistic Gestalts. Papers from the 13th Annual Regional Meet-
ing of the Chicago Linguistics Society. 236-287.
1987 Women, Fire, and Dangerous Things: What Categories Reveal
about the Mind. Chicago: University of Chicago Press.
Langacker, Ronald W.
1985 Observations and Speculations on Subjectivity. In Iconicity in
Syntax, J. Haiman (ed.), 109-150. Amsterdam/Philadelphia: John
Benjamins.
1987a Foundations of Cognitive Grammar. Theoretical Prerequisites.
Stanford: Stanford University Press.
1987b Nouns and verbs. Language 63: 53-94.
1988 A Usage-Based Model. In Topics in Cognitive Linguistics, B.
Rudzka-Ostyn (ed.), 127-161. Amsterdam/Philadelphia: John
Benjamins.
1990a Concept, Image, and Symbol. The Cognitive Basis of Grammar.
Berlin/New York: Mouton de Gruyter.
1990b Subjectification. Cognitive Linguistics 1: 5-38.
1991 Foundations of Cognitive grammar. Descriptive Application.
Stanford: Stanford University Press.
1993 Reference-point constructions. Cognitive Linguistics A: 1-38.
1998 Conceptualization, Symbolization and Grammar. In The New
Psychology of Language. Cognitive and Functional Approaches
to Language Structure, M. Tomasello (ed.), 1-39. Mahwah, New
Jersey: Lawrence Erlbaum Associates.
1999 Grammar and Conceptualization. Berlin/New York: Mouton de
Gruyter.
Lee, Donald W.
1952 Close apposition: an unresolved pattern. American Speech 27:
268-275.
References 275
Lees, Robert Β.
1960/1968 The Grammar of English Nominalizations. Bloomington: Indiana
University Research Center in Anthropology, Folklore and Lin-
guistics, publication 12.
Lemmens, Maarten
1998 Lexical Perspectives on Transitivity and Ergativity. Causative
Constructions in English. Amsterdam/Philadelphia: John Benja-
mins.
Levin, Beth
1993 English Verb Classes and Alternations. Chicago: University of
Chicago Press.
Levin, Beth, and Malka Rappaport
1988 Nonevent -er nomináis: a probe into argument structure. Linguis-
tics 26: 1067-1083.
Lyons, John
1969 Introduction to Theoretical Linguistics. Cambridge: Cambridge
University Press.
1977 Semantics. Volume 2. Cambridge: Cambridge University Press.
Mackenzie, Lachlan J.
1990 First Argument Nominalization in a Functional Grammar of Eng-
lish. Linguistica Antverpiensia 24: 119-147.
Malouf, Robert
2000 Verbal gerunds as mixed categories in head-driven phrase struc-
ture grammar. In Syntax and Semantics. Volume 32. The Nature
and Function of Syntactic Categories, R. D. Borsley (ed.), 133—
166. San Diego: Academic Press.
Marchand, Hans
1969 The Categories and Types of Present-day English Word-
formation. A Synchronic-Diachronic Approach. 2nd edition.
München: C. H. Beck'she Verlagsbuchhandlung.
Massam, Diane
1992 Null objects and non-thematic subjects. Journal of Linguistics
28: 115-137.
Matthews, Peter H.
1966 The concept of rank in 'Neo-Firthian' grammar. Journal of Lin-
guistics 2: 101-109.
Matthiessen, Christian M. I. M.
1995 Lexicogrammatical Cartography: English Systems. Tokyo: In-
ternational Language Sciences Publishers.
Matthiessen, Christian M. I. M., and Sandra A. Thompson
1988 The structure of discourse and 'subordination'. In Clause Com-
bining in Grammar and Discourse, J. Haiman, and S. A. Thomp-
son (eds.), 275-329. Amsterdam/Philadelphia: John Benjamins.
276 References
McGregor, William B.
1991 The concept of rank in systemic linguistics. In Functional and
Systemic Linguistics: Approaches and Uses, E. Ventola (ed.),
121-138. Berlin/New York: Mouton de Gruyter.
1997 Semiotic Grammar. Oxford: Clarendon.
Menzel, Peter
1975 Semantics and Syntax in Complementation. The Hague: Mouton.
Milsark, Gary L.
1976 Existential Sentences in English. Bloomington: Indiana Univer-
sity Linguistics Club.
1977 Toward an explanation of certain peculiarities of the existential
construction in English. Linguistic Analysis 3: 1-29.
1988 Singl-ing. Linguistic Inquiry 19 (4): 611-634.
Munro, Pamela
1982 On the transitivity of 'Say' verbs. In Syntax and Semantics. Vol-
ume 15. Studies in Transitivity, P. J. Hopper, and S. A. Thomp-
son (eds.), 301-318. New York: Academic Press.
Noonan, Michael
1985 Complementation. In Language Typology and Syntactic Descrip-
tion. Volume 2. Complex Constructions, T. Shopen (ed.), 42-140.
Cambridge: Cambridge University Press.
Partee, Barbara H.
1973 The syntax and semantics of quotation. In A festschrift for Morris
Halle, P. Kiparsky, and S. R. Anderson, 410-418. New York:
Holt, Rinehart and Winston.
Peeters, Roger J.
1999 The Adjunct Middle Construction in Dutch. Leuven Contribu-
tions in Linguistics and Philology 88 (3—4): 355-401.
Perlmutter, David M.
1978 Inpersonal passives and the unaccusative hypothesis. Proceed-
ings of the Fourth Annual Meeting of the Berkeley Linguistics
Society. 157-189.
Postal, Paul M.
1970 On Coreferential Complement Subject Deletion. Linguistic In-
quiry 1 (4): 439-500.
Poutsma, H.
1929 A Grammar of Late Modern English. 2nd edition. Groningen:
Noordhoff.
Pullum, Geoffrey K.
1991 English nominal gerund phrases as noun phrases with verb-
phrase heads. Linguistics 29: 763-799.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik
1972 A Grammar of Contemporary English. London: Longman.
1985 A Comprehensive Grammar of the English Language. London:
Longman.
References 277
Radford, Andrew
1997 Syntactic Theory and the Structure of English. A Minimalist Ap-
proach. Cambridge: Cambridge University Press.
Rappaport Hovav, Malka, and Beth Levin
1992 -ER nomináis: implications for the theory of argument structure.
In Syntax and Semantics. Volume 26: Syntax and the Lexicon, T.
Stowell, and E. Wehrli (eds.), 127-153. New York: Academic
Press.
Ravelli, Louise J.
1988 Grammatical metaphor: an initial analysis. In Pragmatics, Dis-
course and Text. Some Systemically-inspired Approaches, E.
Steiner, and R. Veltman (eds.), 133-147. London: Pinter.
Roberts, Ian G.
1987 The Representation of Implicit and Dethematized Subjects.
Dordrecht: Foris.
Rosenbaum, Peter S.
1967 The Grammar of English Predicate Complement Constructions.
Cambridge, Massachusetts: MIT Press.
Rosta, Andrew
1995 'How does this sentence interpret?' The semantics of English
mediopassives. In The Verb in Contemporary English. Theory
and Description, B. Aarts, and C. F. Meyer (eds.), 123-144.
Cambridge: Cambridge University Press.
Ryder, Mary E.
1991 Mixers, mufflers and mousers: the extending of the -er suffix as
a case of prototype reanalysis. Proceedings of the Berkeley Lin-
guistics Society 17: 299-311.
1999a Complex -er nomináis. Where grammaticalization and lexicaliza-
tion meet? In Between Grammar and Lexicon, E. Contini-
Morava, and Y. Tobin (eds.), 291-331. Amsterdam/Philadelphia:
John Benjamins.
1999b Bankers and blue-chippers: an account of -er formations in pre-
sent-day English. English Language and Linguistics 3 (2): 269-
297.
Saussure de, Ferdinand
1972 Cours de linguistique générale. Edition critique préparée par
Tullio de Mauro. Paris: Payot.
Schachter, Paul
1976 A nontransformational account of gerundive nomináis in Eng-
lish. Linguistic Inquiry 7 (2): 205-241.
Schmid, Hans-Jörg
2000 English Abstract Nouns as Conceptual Shells. From Corpus to
Cognition. Berlin/New York: Mouton de Gruyter.
278 References
Sinclair, John, Gwyneth Fox, Stephen Bullón, Ramesh Krishnamurthy, Elizabeth

Manning, and John Todd
1990 Collins COBUILD English Grammar. London: HarperCollins.
Smith, Carlota
1978 Jespersen's 'move and change' class and causative verbs in Eng-
lish. In Linguistic and Literary Studies in Honor of Archibald A.
Hill, M. Jazayery, E. Polomé, and W. Winter (eds.), 101-109.
The Hague: Mouton.
Sohn, Joong-Sun
1998 The English middle and inchoative. Kansas Working Papers in
Stockwell, Robert P., Paul Schachter, and Barbara H. Partee
1973 The Major Syntactic Structures of English. New York: Holt,
Rinehart and Winston.
Strang, Barbara M. H.
1968 Swift's agent-noun formations in -er. In Wortbildung, Syntax und
Morphologie: Festschrift zum 60. Geburtstag von Hans Mar-
chand, H. Brekle, and L. Lipka (eds.), 217-229. The Hague:
Mouton.
1969 Aspects of the history of the -er formative in English. Transac-
tions of the Philological Society. 1-30.
Strawson, Peter
1959 Individuals: An Essay in Descriptive Metaphysics. London:
Routledge.
Sweet, Henry
1891 A New English Grammar. Logical and Historical. Oxford: Clar-
endon.
Taeldeman, Johan
1990 Afleidingen op -seh semantiek, produktiviteit en integratie in een
globale verantwoording van deverbatieve nomina [Derivations in
-sel: semantics, productivity and integration in an overall account
of deverbal nominalizations]. In Betekenis en Produktiviteit.
Gentse Bijdragen tot de Studie van de Nederlandse Woordvor-
ming [Meaning and productivity. Gent contributions to the study
of Dutch word-formation], J. De Caluwe (ed.), 77-115. Studia
Germanica Gandensia.
Talmy, Leonard
2000 Toward a Cognitive Semantics. Volume 1. Concept Structuring
Systems. Cambridge, Massachusetts: MIT Press.
Taylor, John R.
1995 Linguistic categorization: prototypes in linguistic theory. 2nd
edition. Oxford: Clarendon.
1996 Possessives in English. An Exploration in Cognitive Grammar.
Oxford: Clarendon.
References 279
Thompson, Sandra A.
1973 On subjectless gerunds in English. Foundations of Language 9:
374-383.
Vachek, Josef
1966 The Linguistic School of Prague. Bloomington: Indiana Univer-
sity Press.
Vandenberghe, Wim
2001 Instigative Setting-constructions: Force Dynamic Research on
'New' Types of Agency. Leuven Contributions in Linguistics
and Philology 90: 365-390
Van den Toorn, M.C.
1982 Nederlandse Grammatica [Dutch grammar]. Groningen: Wolters
Noordhoff.
Van Langendonck, Willy
1988. -ER als akkusatief en -LING als ergatief suffix in een Zuidbra-
bants dialekt [-ER as an accusative and -LING as an ergative suf-
fix in a South-Brabant dialect], Spektator 17 (4): 317-327.
1999 Neurolinguistic and syntactic evidence for basic level meaning in
proper names. Functions of Language 6 (1): 95-138.
Van Oosten, Jeanne
1977 Subjects and agenthood in English. Papers from the 13th Annual
Regional Meeting of the Chicago Linguistics Society. 451-471.
1986 The Nature of Subjects, Topics and Agents: A Cognitive Expla-
nation. Bloomington: Indiana University Linguistics Club.
Vendler, Zeno
1967 Linguistics in Philosophy. Ithaca/New York: Cornell University
Press.
Verstraete, Jean-Christophe
2001 Subjective and objective modality: Interpersonal and ideational
functions in the English modal auxiliary system. Journal of
Pragmatics 33: 1505-1528.
2002 Interpersonal grammar and clause combining in English. Ph. D.
diss., Department of Linguistics, University of Leuven.
Wasow, Thomas, and Thomas Roeper
1972 On the subject of gerunds. Foundations of Language. Interna-
tional Journal of Language and Philosophy 8: 44—61.
Wierzbicka, Anna
1988 The Semantics of Grammar. Amsterdam/Philadelphia: John Ben-
jamins.
Williams, Edwin
1980 Predication. Linguistic Inquiry 11 (1): 203-238.
1981 Argument Structure and Morphology. The Linguistic Review 1:
81-114.
280 References
Yoshimura, Kimihiro
forthcoming The Middle Construction in English: A Cognitive Linguistic ana-
lysis. Berlin/New York: Mouton de Gruyter.
Index
action nominalization, 46-47,48, 51, double grounding, 70, 217-220

52, 203-204 vs. postmodification, 211-217
ad hoc nominalization, 45, 57,100- argument
101,117, 120-123,158, 165- external, 110-115
172,259 internal, 111-114
event reading, 122,165, 167, assembly
172 construction, 20
nominal grounding, 167-169, level of, 6,41,49, 94-95, 222,
172,173-174 225-226, 242, 244-245, 249-
non-phoric, 168-172 252
phoric, 120-121, 166-168, 172 order of, 48,226
postmodification, 121-122, atemporal clausal head, 73-76, 81,
170-171 86, 94, 222-228,243
premodification, 121, 169-171 auxiliaries
temporal grounding, grounding vs. non-grounding,
166-167,172, 173-174 71,224
Agent
Actor vs. Instigator, 108-109, categorization
152 elaboration/ instantiation, 12-
diachronic origin of deverbal 15, 27, 34
-er, 99, 123 extension, 5, 12, 27, 34,43-44
modal value of agentive -er, schématisation, 5, 12, 14,27,
161-163 43
profile of deverbal -er, 99-110, class
161-163 external functional properties,
schematic category, 151-153 19, 23-24, 26, 50, 92
systemic-functional view of, functional approach to, 4,19,
29-30 23-28, 50-51, 92
agnation internal functional properties,
construct of, 35-36 19, 24-26, 92
descriptive heuristic, 37-40, notional definition of, 23, 26
55-56 clause
networks of, 54-55 correspondences with nominal,
relevance to nominalization, 6, 6, 70, 71, 81,92-93
7, 52-58,94 functional approach to, 6, 70-
schematic nature of, 56-58, 150 92
Transformational Grammar, grounding, 71, 81-87
53-55 instantiation, 78-79, 81
analogy, 45-46 quantification, 79-81
apposition, 213,217-218 type specification, 71-78
282 Index
clausal head the construct of, 35-40

finite vs. atemporalized, 73-76, partial enation, 36
81, 86, 94 non-enation, 36, 39
inclusion of Objects, 76-77 entrenchment
Cognitive Grammar, 4-5, 11, 13, 15, conventionalization, 12, 13, 172
17, 91, 258 degrees of, 5, 15,44-46, 100-
conventional, 12, 13 101, 165
common noun strategy ergativity, 108-110, 111, 262
nominalizations, 95, 201-202, Instigator and Medium, 108-
209, 232-233, 238,240, 241- 110
244, 246, 257-258 unaccusatives, 113-117
ordinary nouns, 61-62, 81 -er nominalization
comparison, 11, 12 agentive profile, 52, 99-110
complexity agnation with the middle, 7, 14,
conceptual category, 101, 106- 117-118, 150,154-158, 159-
108 161,262
membership of a family of argument structure, 110-118
structures, 14, 15, 26-27,43 conduciveness, 154-158, 159
symbolic complexity, 18, 19, complex conceptual category,
44-45, 47 106-108
composition, 11-14,47-49, 259 denominal, 102-103, 106-107,
analytic definition of, 21 150
building-block view of, 20, 257 destiny-oriented, 156
component functions, 19, 20 facility-oriented, 155-157
as integration, 21, 95, 251-252 feasibility-oriented, 156-157
radically functional approach high-level schema, 100, 110
to, 19-28 low-level schemata, 44
synthetic definition of, 21 modal value of agentives, 161-
constituency, 3, 19, 20, 21-23, 25, 163
28 modal value of non-agentives,
construction 159-161
construct of, 13 result-oriented, 157
integration, 21, 95, 251-252 salience, 102-106
radically functional approach Subject profile, 118-120, 123,
to, 19-28 154,158,159, 174-176
verbal base of, 108-110
downranking expansion, 185, 187-188, 211-217
or rank shift, 5, 22, 49-51, 95, elaboration, 214-217
186, 208,227-228,239,249- extension
250,251 of the prototype, 12-13, 14-15,
26-28,43-44
elaboration
instantiation, 12, 14 type of expansion, 214-217
embedding, 185-187, 188-194
enation factive
embedded projection, 185-195
Index 283
external functioning, 183-185, Subject, 242-244, 244-248

185-187, 191-194,247 typology of, 231-240
instantiation, 204-205, 241- ground, 62, 71, 81
244, 245-248 grounding
internal functioning, 200-205, clausal, 6, 57-58, 81-92, 260-
207-248 261
nominalization, 181-205, 207- definite vs. indefinite, 66-68,
248 167-169,173-174
presupposition, 181, 183-185, Finite, 81-85
189-190 interpersonal status, 31, 93, 95
finite element nominal, 6, 25-26, 66-69, 167-
grounding, 81-85, 93 169,172, 173-174
modality, 81-85 quantification, 64-66
tense, 81, 84-85 Subject, 85-86
fixed expression, 5,44, 101, 165,
172 holistic scanning, 51, 75, 208,210,
224-225
generic head
bare plurals and mass nouns, clausal, 71-78
65, 234 nominal, 62-63, 92
middle, 142, 264 hypotaxis, 185-188, 190-194, 196-
gerundive nomináis, 198, 229- 198,213-217
230,233-236, 238, 240
gerundive nominalization iconic encoding
action, 16,46-47, 199, 232- vs. non-iconic encoding, 4, 20-
233, 237-238, 240, 247 21,24, 25, 28,51,62, 63,69-
atemporal clausal head, 222- 70, 81, 92, 94, 208, 209-210,
228,243 246,250-251
common noun strategy, 232- ideational
233,238,240, 241-244 metafunction, 6, 30-31, 38-39,
factive, 6, 8,46-47,48, 181- 52, 95
182, 183, 195-196, 198-200, modality, 83-84
201,203-205, 221-244,244- -ing, 72, 74-76, 224-227, 260
248 instantiation
generic action, 198, 229-230, clause, 6, 78-81, 86
233-236,238, 240 elaboration, 12, 13, 14, 27, 34,
-ing, 224-227 43
instantiation, 204-205, 241- gerundive nomináis, 204-205,
244, 245-248 241-244,245-248
internal functioning, 200, 203- nominal, 6,25-26, 62-62
205, 222-244 usage event, 10, 27, 33
possessive/genitive, 232, 242, Instrumental -er, 100, 103, 107,
266 111-112, 115, 149, 163-164
proper name strategy, 233-238,
239-240,241-244
284 Index
interpersonal quality-oriented, 133,136

metafunction, 6, 31, 34, 38-39, result-oriented, 135-136
52, 61, 76, 93, 95, 150,175 Subject profile, 131-132, 137,
modality, 82-84 140-147
modality
letting, see modality agentive -er, 161-163
lexicalization, 5,44-45, 100-101, deontic, 82-85, 89-91, 139
120-123,158, 159-164 dynamic, 83-85,140-141, 159,
non-event reading, 122, 160, 161
172 epistemic, 82-85, 89-91, 138,
postmodification, 122 159
lexicogrammar 'letting', 141-147
construct of, 16-17,45 middle constructions, 137-142
relation with semantics, 17-19 non-agentive -er, 159-161
objective vs. subjective, 83,
metafunctions 140-141, 159, 261
as clusters of lexicogrammatical schematic, 82, 89-91, 141,
options, 30, 33, 37-39 159-160
as configurations of functions, modalization, 83
31-33 modulation, 83
interpersonal/constructional, 6, multifunctional approach, 4, 6,29-
31,34, 38-39, 52,61,76, 93, 33, 52, 95, 150, 177
95, 150, 175
representational/ideational, 6, natural
30-31, 38-39, 52, 92-93, 95, description of language, 15-17,
110, 128-132, 150-154 20
textual, 6, 31, 52, 93, 151 non-arbitrary relation between
metaphor, 106-108 lexicogrammar and semantics,
metonymy, 106-108 17-19, 105
middle construction, 111, 125-147, network
263 schematic, 14, 15,45
agency, 131 system, 4, 7, 13, 30, 33-34, 37-
agnation, 7, 114, 117-118, 150, 39
159-161 nominal
conduciveness, 137, 141, 143- class, 25, 92
144, 147, 154 correspondences with the
destiny-oriented, 134-135, 136 clause, 6, 70, 71, 81, 92-93
ergative approach, 126-127, functional approach to the, 6,
263 20-21, 25, 61-70
facility-oriented, 132-133, 136 vs. noun, 25-26
feasibility, 132-134, 136 non-agentive -er, 100, 103-104,107,
intransitive middles, 128-130, 115-118, 149, 153-158
263 agnation with the middle, 159-
modal value, 137-142, 159 161
passive approach, 127-128
Index 285
puipose-designed status of, 107, quantifier

154-155 absolute, 64—66
semantic typology, 154-158 relative, 64-66
noun
nominal, 25-26, 62 rank scale
non-plural mass noun, 63, 234, downranking or rank shift, 5,
260 22,49-51, 95, 186, 208, 227-
plural noun, 63, 234, 260 228, 239, 249-250,251
proper noun, 26, 61-62, 69 radically functional approach
singular count noun, 63 to, 4, 5,19, 22-23, 28,49-51
structural, 21-23, 257
oblique participants reclassification
definition, 115 external/synthetic vs. inter-
-er nominalization, 115, 149- nal/analytic 4, 5, 7,49-51,208-
150,153-154, 262 210,249-252
functional approach to, 4, 50-
paradigmatic relations 51, 94-96, 208,227-228, 249-
of components, 33, 53, 55, 228 250
of constructions (agnation, ena- structural approach to, 49-50
tion), 6, 7, 8, 33-40, 52-58 reference mass, 64, 66-68
parataxis, 186-187,213-217 representational
Patient metafiinction, 6, 30-31, 38-39,
Goal or Medium, 108-110 52, 95, 110, 128-132, 150-154
person deixis, 85-86 modality, 83-84
polarity, 79-80 Residue, 76-77, 93
productivity, 44-45, 149-150 responsibility
projection, 185, 187-190, 194, 211- causal or agentive, 143
217 middle Subject, 126-127, 131—
pronouns, 21, 69 132,143
progressive -ing, 75-76, 260 objective, 91-92, 143
proper name speech-functional or modal, 87-
strategy, 61-62, 65, 69, 95, 92, 143
209-210, 233-238, 239-240,
241-244,246, 258 salience, 102-106
prototype sanctioning, 10, 12, 27,43
criteria for identification of the, scanning
27-28, 257-258 serial or sequential, 75, 208
extension of, 15, 27-28,43-44, summary or holistic, 75, 208,
149 224-225
reanalysis of -er, 102-104 schema
theory, 14 constructional, 13,43, 106
-er nominalization, 100, 103,
quantification 106,110,174-176
clausal, 6, 79-81 high-level vs. low-level, 43-45,
nominal, 6, 25-26, 63 100, 106,149
286 Index
schematic characterization of the psychological ability of

class, 19, 23-25 symbolization, 11
schematic modality, 81-85, 141 unit, 11, 12, 13,17, 105, 257
schematic nature of agnation, system
56-58, 150 interaction with usage, 4, 9-15
schematic network, 14, 15,45, systemic motivation, 10, 12, 42
149 system network, 4, 7, 13, 30,
schematization 33-34, 37-39
the psychological ability of, 11, systemic (or paradigmatic) rela-
12 tions, 4, 19, 33-40, 52-58
in the language system, 13, 14- Systemic-functional Grammar, 4-5,
15,43 15, 17, 29, 33, 258, 259
Setting-Subject construction, 163
Subject textual metafunction, 6, 31, 52, 93,
gerundive nomináis, 241-244 151
instantiating function, 78-79, íAaí-nominalization, 49, 69-70
86, 175, 204-205, 241-244, embedded or downranked pro-
245-248, 261 jection, 185-195, 195-198, 208
locus of potency, 140-144, factive, 6, 8, 181-205,207-210
159-163, 176 factive prototype, 181-182
middle, 126-127,131-132, hypotactic, 186-187, 189, 191-
143-144 194, 195-198
person deixis, 85-86,175, 262 internal functioning, 200-203,
primary focal participant, 91- 209-210
92, 258 locutions/ideas, 193, 265
profile of -er, 118-120, 149- proper name strategy, 209-210,
150, 159-164, 174-176 220-221
speech-functional role, 87-92, Theme
175-176 Government-and-Binding, 111
systemic-functional view of, Systemic-Functional grammar,
29-32 29-32
Subject-Finite unit the fact //¡«/-nominalization, 6, 8,
-er nominalization, 149, 159, 181-182, 183, 190, 195, 197-
174 198, 202,211-221
grounding, 76-87, 93 apposition vs. postmodification,
interpersonal nature, 37-38, 81- 211-217
93 doubly-grounded nominal, 217-
mood options, 37-38, 87-92 221
speech-functional value, 87-92 vs. embedded locutions/ideas,
syntagmatic relations, 7, 12, 13, 19- 211-212,215-217, 220
23, 33 vs. relative clauses, 211-212,
symbolic 215,217, 266
relation lexicogrammar- transcategorization, 41, 227
semantics, 4, 5, 7, 17-19,41, transitivity, 108-110, 262
46-47, 105,257 Actor and Goal, 108-110
Index 287
unergatives or intransitives, symbolic, 11, 13, 17, 19, 257

113, 115-117 schematic, 13,45,165
type specification semantic, 13, 17
clausal, 6, 72-77 phonological, 13, 17
definite vs. indefinite reference, usage
68, 168-169 usage-based approach, 11,42-46
nominal, 6, 25-26, 62-63, 66- event, 27
68
verb group
unaccusatives, see ergativity internal complexity, 70-72, 261
unergatives, see transitivity processual profile, 72, 74
units atemporal, 73-76
conventional, 5, 10, 12,44
Cognitive Linguistics Research
Edited by René Dirven, Ronald W. Langacker and
John R. Taylor
Mouton de Gruyter · Berlin · New York
1 Ronald W. Langacker, Concept, Image, and Symbol. The Cognitive

Basis of Grammar. 1990.
2 Paul D. Deane, Grammar in Mind and Brain. Explorations in Cogni-
tive Syntax. 1992.
3 Conceptualizations and Mental Processing in Language. Edited by
Richard A. Geiger and Brygida Rudzka-Ostyn. 1993.
4 Laura A. Janda, A Geography of Case Semantics. The Czech Dative
and the Russian Instrumental. 1993.
5 Dirk Geeraerts, Stefan Grondelaers and Peter Bakema, The Structure
of Lexical Variation. Meaning, Naming, and Context. 1994.
6 Cognitive Linguistics in the Redwoods. The Expansion of a New Para-
digm in Linguistics. Edited by Eugene H. Casad. 1996.
7 John Newman, Give. A Cognitive Linguistic Study. 1996.
8 The Construal of Space in Language and Thought. Edited by Martin
Pütz and René Dirven. 1996.
9 Ewa D^browska, Cognitive Semantics and the Polish Dative. 1997.
10 Speaking of Emotions: Conceptualisation and Expression. Edited by
Angeliki Athanasiadou and Elzbieta Tabakowska. 1998.
11 Michel Achard, Representation of Cognitive Structures. 1998.
12 Issues in Cognitive Linguistics. 1993 Proceedings of the International
Cognitive Linguistics Conference. Edited by Leon de Stadler and
Christoph Eyrich. 1999.
13 Historical Semantics and Cognition. Edited by Andreas Blank and
Peter Koch. 1999.
14 Ronald W. Langacker, Grammar and Conceptualization. 1999.
15 Cognitive Linguistics: Foundations, Scope, and Methodology. Edited
by Theo Janssen and Gisela Redeker. 1999.
16 A Cognitive Approach to the Verb. Morphological and Constructional
Perspectives. Edited by Hanne Gram Simonsen and Rolf Theil Endre-
sen. 2001.
17 Emotions in Crosslinguistic Perspective. Edited by Jean Harkins and
Anna Wierzbicka. 2001.
18 Cognitive Linguistics and Non-Indo-European Languages. Edited by
Eugene Casad and Gary B. Palmer. 2003.
19.1 Applied Cognitive Linguistics I: Theory and Language Acquisition.
Edited by Martin Pütz, Susanne Niemeier and René Dirven. 2001.
19.2 Applied Cognitive Linguistics II: Language Pedagogy. Edited by Mar-
tin Pütz, Susanne Niemeier and René Dirven. 2001.
20 Metaphor and Metonymy in Comparison and Contrast. Edited by
René Dirven and Ralf Pörings. 2002.
21 Grounding. The Epistemic Footing of Deixis and Reference. Edited
by Frank Brisard. 2002.
22 Cristiano Broccias, The English Change Network. 2003.
23 Cognitive Approaches to Lexical Semantics. Edited by Hubert Cuyck-
ens, René Dirven and John Taylor. 2003.
24 Cognitive Models in Language and Thought. Ideology, Metaphors,
and Meanings. Edited by René Dirven, Roslyn Frank and Martin Pütz.
2003.
25 Rong Chen, English Inversion. A Ground-before-Figure Construction.
2003.
26 Liesbet Heyvaert, A Cognitive-Functional Approach to Nominaliza-
tion in English. 2003.

A Cognitive-Functional Approach To Nominalization in English

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

A Cognitive-Functional Approach To Nominalization in English

Uploaded by

Copyright:

Available Formats

A Cognitive-Functional Approach

© Printed on acid-free paper

Bibliographic information published by Die Deutsche Bibliothek

© Copyright 2003 by Walter de Gruyter G m b H & Co. K G , D-10785 Berlin

Part I Towards a theoretical-descriptive approach to

Chapter 2: Theoretical assumptions 9

1. Language: The relation between system and usage 9

6. Theoretical assumptions: Conclusion 40

2. Nominalizations as non-arbitrary symbolic units 46

Chapter 4: The functional organization of nominal and clause 61

1. From noun type specification to nominal 61

3. Nominalization as functional reclassification 94

Part Π Deverbal -er nominalization

1.1.1. -er nominalizations as natural symbolic units 104

Chapter 6: The middle construction 125

1. Middle formation: A state of the art 126

Chapter 7: A multifunctional approach to -er nominalization 149

Part ΙΠ Factive nominalization

Chapter 8: Factive nominalization: Towards a descriptive po- 181

Chapter 9: A functional analysis of factive nominalizations as 207

1. ί/ζα/-facti ves: From finite clause to proper name strategy 207

3.2.3. The nominal functioning of gerundive nominalization:

Chapter 10: Conclusion 249

The focus of the present study is on deverbal nominalization in English.

proach can account for the lexicogrammatical properties and meanings of

Fourthly, nominalizations are best described from a multifunctional per-

(presented in Chapter 6) sheds a new light on the system of -er nominaliza-

Nominalization represents a theoretical challenge and has to be situated in

1. Language: The relation between system and usage

1.1. A usage-based approach to language

The analysis of nominalization set out in this study is rooted in a view of

functional level of organization in language will be discussed in more de-

guistic structure" and the "source of language change" (Langacker 1987a:

1.2. Cognitive Grammar: The interface between system and usage

Langacker considers the following psychological abilities of the language

1.3. The language system as network

Crucially, the relationships of symbolization and composition, comparison

schemata, which capture generalizations and serve as 'templates' for the

Figure 1. A schematic network

In including not only the categorizing relationship of extension, which is

elude schematization among the categorizing relationships because a cate-

2. The natural symbolic relation between lexicogrammar and

2.1. A natural description of language

The idea that a description of language, apart from being usage-based,

nomena" (Langacker 1987a: 13). A natural model of language is a model

2.2. The natural or non-arbitrary relation between grammar and meaning

Language is essentially symbolic in nature in that it consists of an open-

nological form, seems unfit to capture the non-arbitrary nature of gram-

3. Constructions: Composition, classification and function3

... the components of a complex expression should not be thought of as

Owing to the structural diversity of nomináis, a universally valid schematic

In this section, a radically functional approach to the analysis of construc-

3.1. A functional approach to composition

In Langacker's (1987a) model of language, composition is said to involve

3.2. Levels of functional analysis

In Haas (1954), it is claimed that a functional analysis of composition is

labels a fixed number of layers in the hierarchy of constituents, such that

the rank scale to so-called minimal or functional bracketing, that is "brack-

3.3. A functional approach to classification

Symbolic units are typically assigned to particular classes because of their

3.3.1. An internal functional analysis of constructions

A radically functional approach to classification identifies classes not only

3.3.2. Classes as complex categories: Schema and prototype