CATEGORIAL FEATURES

Proposing a novel theory of parts of speech, this book discusses categoriza-

tion from a methodological and theoretical point of view. It draws on
discoveries and insights from a number of approaches – typology, cognitive
grammar, notional approaches and generative grammar – and presents a
generative, feature-based theory.
Building on up-to-date research and the latest findings and ideas in
categorization and word-building, Panagiotidis combines the primacy of
categorial features with a syntactic categorization approach, addressing the
fundamental, but often overlooked, questions in grammatical theory.
Designed for graduate students and researchers studying grammar and
syntax, this book is richly illustrated with examples from a variety of lan-
guages and explains elements and phenomena central to the nature of human
language.

phoevos panagiotidis is Associate Professor of Linguistics in the

Department of English Studies at the University of Cyprus.
 In this series
107. SUSAN EDWARDS: Fluent Aphasia
108. BARBARA DANCYGIER and EVE SWEETSER: Mental Spaces in
Grammar: Conditional Constructions
109. HEW BAERMAN, DUNSTAN BROWN and GREVILLE G. CORBETT:
The Syntax–Morphology Interface: A Study of Syncretism
110. MARCUS TOMALIN: Linguistics and the Formal Sciences: The Origins of
Generative Grammar
111. SAMUEL D. EPSTEIN and T. DANIEL SEELY: Derivations in
Minimalism
112. PAUL DE LACY: Markedness: Reduction and Preservation in Phonology
113. YEHUDA N. FALK: Subjects and Their Properties
114. P. H. MATTHEWS: Syntactic Relations: A Critical Survey
115. MARK C. BAKER: The Syntax of Agreement and Concord
116. GILLIAN CATRIONA RAMCHAND: Verb Meaning and the Lexicon:
A First Phase Syntax
117. PIETER MUYSKEN: Functional Categories
118. JUAN URIAGEREKA: Syntactic Anchors: On Semantic Structuring
119. D.ROBERT LADD: Intonational Phonology, Second Edition
120. LEONARD H. BABBY: The Syntax of Argument Structure
121. B. ELAN DRESHER: The Contrastive Hierarchy in Phonology
122. DAVID ADGER, DANIEL HARBOUR and LAUREL J. WATKINS:
Mirrors and Microparameters: Phrase Structure beyond Free Word Order
123. NIINA NING ZHANG: Coordination in Syntax
124. NEIL SMITH: Acquiring Phonology
125. NINA TOPINTZI: Onsets: Suprasegmental and Prosodic Behaviour
126. CEDRIC BOECKX, NORBERT HORNSTEIN and JAIRO NUNES: Control
as Movement
127. MICHAEL ISRAEL: The Grammar of Polarity: Pragmatics, Sensitivity,
and the Logic of Scales
128. M. RITA MANZINI and LEONARDO M. SAVOIA: Grammatical
Categories: Variation in Romance Languages
129. BARBARA CITKO: Symmetry in Syntax: Merge, Move and Labels
130. RACHEL WALKER: Vowel Patterns in Language
131. MARY DALRYMPLE and IRINA NIKOLAEVA: Objects and Information
Structure
132. JERROLD M. SADOCK: The Modular Architecture of Grammar
133. DUNSTAN BROWN and ANDREW HIPPISLEY: Network Morphology:
A Defaults-Based Theory of Word Structure
134. BETTELOU LOS, CORRIEN BLOM, GEERT BOOIJ, MARION
ELENBAAS and ANS VAN KEMENADE: Morphosyntactic Change:
A Comparative Study of Particles and Prefixes
135. STEPHEN CRAIN: The Emergence of Meaning
136. HUBERT HAIDER: Symmetry Breaking in Syntax
137. JOSE´ A. CAMACHO: Null Subjects
138. GREGORY STUMP and RAPHAEL A. FINKEL: Morphological Typology:
From Word to Paradigm
139. BRUCE TESAR: Output-Driven Phonology: Theory and Learning
140. ASIER ALCA ´ ZAR AND MARIO SALTARELLI: The Syntax of Imperatives
141. MISHA BECKER: The Acquisition of Syntactic Structure: Animacy and
Thematic Alignment
142. MARTINA WILTSCHKO: The Universal Structure of Categories: Towards
a Formal Typology
143. FAHAD RASHED AL-MUTAIRI: The Minimalist Program: The Nature
and Plausibility of Chomsky’s Biolinguistics
144. CEDRIC BOECKX: Elementary Syntactic Structures: Prospects of a
Feature-Free Syntax
145. PHOEVOS PANAGIOTIDIS: Categorial Features: A Generative Theory of
Word Class Categories
Earlier issues not listed are also available
CAMBRIDGE STUDIES IN LINGUISTICS
General Editors: p. austin, j. bresnan, b. comrie,
s. crain, w. dressler, c. j. ewen, r. lass, d. lightfoot,
k. rice, i. roberts, s. romaine, n. v. smith

CATEGORIAL FEATURES
A Generative Theory of Word Class Categories
C A T E G O R I A L FE A T U R E S
A GENERATIVE THEORY OF WORD
CLASS CATEGORIES

PHOEVOS PANAGIOTIDIS
University of Cyprus
University Printing House, Cambridge CB2 8BS, United Kingdom

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107038110
© Phoevos Panagiotidis 2015
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2015
Printed in the United Kingdom by Clays, St lves plc
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data
Panagiotidis, Phoevos.
Categorial features : a generative theory of word class categories / Phoevos Panagiotidis.
pages cm – (Cambridge studies in linguistics ; 145)
ISBN 978-1-107-03811-0 (Hardback)
1. Grammar, Comparative and general–Grammaticalization. 2. Categorial grammar.
3. Language, Universal. I. Title.
P299.G73P36 2014
415–dc23 2014020939
ISBN 978-1-107-03811-0 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
 Contents

Preface page xiii

1 Theories of grammatical category 1

1.1 Introduction 1
1.2 Preliminaries to a theory: approaching the part-of-speech problem 1
1.2.1 On syntactic categories and word classes: some clarifications 3
1.2.2 Parts of speech: the naïve notional approach 4
1.2.3 Parts of speech: morphological criteria 6
1.2.4 Parts of speech: syntactic criteria 7
1.2.5 An interesting correlation 8
1.2.6 Prototype theory 9
1.2.7 Summarizing: necessary ingredients of a theory of category 11
1.3 Categories in the lexicon 12
1.4 Deconstructing categories 17
1.4.1 Distributed Morphology 17
1.4.2 Radical categorylessness 18
1.5 The notional approach revisited: Langacker (1987)
and Anderson (1997) 19
1.6 The present approach: LF-interpretable categorial features
make categorizers 21

2 Are word class categories universal? 24

2.1 Introduction 24
2.2 Do all languages have nouns and verbs? How can we tell? 25
2.3 Two caveats: when we talk about ‘verb’ and ‘noun’ 26
2.3.1 Verbs, not their entourage 26
2.3.2 Misled by morphological criteria: nouns
and verbs looking alike 27
2.3.3 What criterion, then? 28
2.4 Identical (?) behaviours 29
2.5 The Nootka debate (is probably pointless) 32
2.6 Verbs can be found everywhere, but not necessarily as a word class 37
2.7 An interim summary: verbs, nouns, roots 40

ix
x Contents

2.8 What about adjectives (and adverbs)? 41

2.8.1 Adjectives are unlike nouns and verbs 41
2.8.2 Adjectives are not unmarked 42
2.8.3 Adverbs are not a simplex category 48
2.9 The trouble with adpositions 49
2.10 Conclusion 51

3 Syntactic decomposition and categorizers 53

3.1 Introduction 53
3.2 Where are words made? 54
3.3 Fewer idiosyncrasies: argument structure is syntactic structure 58
3.4 There are still idiosyncrasies, however 60
3.5 Conversions 62
3.6 Phases 65
3.7 Roots and phases 67
3.8 On the limited productivity (?) of first phases 70
3.9 Are roots truly acategorial? Dutch restrictions 72
3.10 Conclusion 77

4 Categorial features 78
4.1 Introduction 78
4.2 Answering the old questions 78
4.3 Categorial features: a matter of perspective 82
4.4 The Categorization Assumption and roots 89
4.4.1 The Categorization Assumption 89
4.4.2 The interpretation of free roots 93
4.4.3 The role of categorization 95
4.4.4 nPs and vPs as idioms 97
4.5 Categorizers are not functional 98
4.6 Nouns and verbs 100
4.6.1 Keeping [N] and [V] separate? 101
4.6.2 Do Farsi verbs always contain nouns? 103

5 Functional categories 106

5.1 Introduction 106
5.2 The category of functional categories 106
5.3 Functional categories as ‘satellites’ of lexical ones 110
5.4 Biuniqueness 111
5.5 Too many categorial features 116
5.6 Categorial Deficiency 117
5.7 Categorial Deficiency 6¼ c-selection 120
5.8 Categorial Deficiency and roots (and categorizers) 122
5.9 Categorial Deficiency and Agree 124
 Contents xi

5.9.1 On Agree 124

5.9.2 Biuniqueness as a product of categorial Agree 126
5.9.3 Why there are no mid-projection lexical heads 127
5.9.4 How projection lines begin 128
5.9.5 Deciding the label: no uninterpretable Goals 130
5.10 Conclusion 133

6 Mixed projections and functional categorizers 134

6.1 Introduction 134
6.2 Mixed projections 134
6.3 Two generalizations on mixed projections 136
6.4 Free-mixing mixed projections? 140
6.5 Switches as functional categorizers 142
6.6 Morphologically overt Switches 148
6.7 Switches and their complements 152
6.7.1 Locating the Switch: the size of its complement 153
6.7.2 Phases and Switches 159
6.8 Are all mixed projections externally nominal? 161
6.8.1 Verbal nouns 162
6.9 The properties of mixed projections 165
6.9.1 Similarities: Nominalized Aspect Phrases in English
and Dutch 166
6.9.2 Differences: two types of Dutch ‘plain’ nominalized infinitives 166
6.9.3 Fine-grained differences: different features in
nominalized Tense Phrases 169
6.10 Why functional categorizers? 170
6.11 Conclusion 172

7 A summary and the bigger picture 173

7.1 A summary 173
7.2 Loose ends 175
7.3 Extensions and consequences 176

8 Appendix: notes on Baker (2003) 179

8.1 Introduction 179
8.2 Are nouns referential? 180
8.3 Syntactic predication, semantic predication and specifiers 181
8.4 Are adjectives the unmarked lexical category? Are they roots? 183
8.5 Pred and other functional categories 184
8.6 Two details: co-ordination and syntactic categorization 185

References 189
Index 204
 Preface

The project resulting in this monograph began in 1999, when I realized that
I had to answer the question of why pronouns cannot possibly be ‘intransitive
determiners’, why it is impossible for Determiner Phrases (DPs) consisting of a
‘dangling D head’ (a turn of phrase my then PhD supervisor, Roger Hawkins,
used) – that is, made of a Determiner without a nominal complement – to exist.
The first answer I came up with was Categorial Deficiency, extensively argued
for in Chapter 5. Back then, however, Categorial Deficiency of functional
heads was just an idea, which was expounded in my (2000) paper. The case
for it was limited to arguments from biuniqueness and the hope was that it
would eventually capture Head Movement. The paper was delivered at the
April 2000 Spring Meeting of LAGB, in the front yard of UCL, in the open:
the fire alarm, this almost indispensable element of British identity and social
life, went off seconds after the talk started. It did not look good. However,
Categorial Deficiency did find its way into my thesis and the (2002) book
version thereof.
There were more serious problems, though: I quickly realized that ‘uninter-
pretable [N]’ and ‘uninterpretable [V]’ mean nothing if we have no inkling
of the actual interpretation of ‘interpretable [N]’ and ‘interpretable [V]’.
This inevitably brought me to the question of the nature of categorial features
and what it means to be a noun, a verb and an adjective. Surprisingly, this was
an issue very few people found of any interest, so for a couple of years or
so I thought I should forget about the whole thing. This outlook changed
dramatically in 2003, when Mark Baker’s book was published: a generative
theory of lexical categories with precise predictions about the function and
interpretation of categorial features. On the one hand, I was elated: it was about
time; on the other, I was disappointed: what else was there to say on lexical
categories and categorial features?
Quite a lot, as it turned out. Soon after my (2005) paper against syntactic
categorization, I had extensive discussions with Alan Bale and, later, Heidi
Harley. These were the impetus of my conversion to a syntactic decomposition

xiii
xiv Preface

approach. At around the same time, Kleanthes Grohmann and I thought it

would be a good idea to see if his Prolific Domains could be shown to be co-
extensive to the categorially uniform subtrees making up mixed projections
(Bresnan 1997).
It is easy to figure out that I have incurred enormous intellectual debts to a
number of people; this is to be expected when working on a project stretching
for well over a decade. Before naming names, however, I have to gratefully
acknowledge that parts of this project were generously funded by Cyprus
College (now European University Cyprus) through three successive faculty
research grants, between 2003 and 2006.
Moving on to people now: Paolo Acquaviva, whom I met in 2009 at the
Roots workshop in Stuttgart, made me regain faith in my project and provided
me with priceless insight on where we could go after we finished with
categories and how roots really mattered. I owe to David Adger some pertinent
and sharp questions on Extended Projections, feature (un)interpretability and
mixed projections. Relentless and detailed commentary and criticism by Elena
Anagnostopoulou go a long way, and they proved valuable in my sharpening
the tools and rethinking all sorts of ‘facts’. Thanks to Karlos Arregi I had to
seriously consider adpositions and roots inside them. Mark Baker, talking to
me in Utrecht in 2001 about the book he was preparing, and discussing nouns
and verbs in later correspondence, has been an inspiration and an indispensable
source of encouragement. Thank you, Hagit Borer, for asking all those tough
questions on idiomaticity. I am truly indebted to Annabel Cormack, who
significantly deepened (or tried to deepen) my understanding of the founda-
tional issues behind lexical categories and their interpretation. Discussing roots
and categorizers with David Embick in Philadelphia in 2010 served as a one-
to-one masterclass for me. Kleanthes Grohmann – enough said: a valuable
interlocutor, a source of critical remarks, a true collega. Heidi Harley, well,
what can I say: patience and more patience and eagerness to discuss pretty
much everything, even when I would approach it from an outlandish (I cannot
really write ‘absurd’, can I?) angle, even when I would be annoyingly ignorant
about things; and encouragement; and feedback. Most of what I know about
Russian adjectives I owe to Svetlana Karpava and her translations. Richie
Kayne has been supportive and the most wonderful person to discuss all those
‘ideas’ of mine with throughout the years. Richard Larson, thank you for
inviting me to Stony Brook and for all the stimulating discussions that
followed. Winnie Lechner helped me immensely in investigating the basic
questions behind categorization and category and his contribution to my
thinking about mixed projections was momentous and far-ranging. Alec
 Preface xv

Marantz took the time and the effort when I needed his sobering feedback
most, when I was trying to answer too many questions on idiomaticity and root
interpretation. Discussions with Sandeep Prasada, and his kindly sharing his
unpublished work on sortality with me, provided a much-needed push and the
opportunity to step back and reconsider nominality. Gratitude also goes to
Marc Richards, the man with the phases and with even more patience. Luigi
Rizzi has been a constant source of support and insight, through both gentle
nudges and detailed discussions. David Willis’ comments on categorial Agree
and its relation to movement gave me the impetus to make the related discus-
sion in Chapter 5 bolder and, I hope, more coherent.
I also wish to thank the following for comments and discussion, although
I am sure I must have left too many people out: Mark Aronoff, Adriana
Belletti, Theresa Biberauer, Lisa Cheng, Harald Clahsen, Marijke De Belder,
Carlos de Cuba, Marcel den Dikken, Jan Don, Edit Doron, Joe Emonds,
Claudia Felser, Anastasia Giannakidou, Liliane Haegeman, Roger Hawkins,
Norbert Hornstein, Gholamhosein Karimi-Doostan, Peter Kosta, Olga Kva-
sova, Lisa Levinson, Pino Longobardi, Jean Lowenstamm, Rita Manzini, Ora
Matushansky, Jason Merchant, Dimitris Michelioudakis, Ad Neeleman, Rolf
Noyer, David Pesetsky, Andrew Radford, Ian Roberts, Peter Svenonius,
George Tsoulas, Peyman Vahdati, Hans van de Koot, Henk van Riemsdijk.
I also wish to thank for their comments and feedback the audiences in Cyprus
(on various occasions), Utrecht, Pisa, Potsdam, Jerusalem, Patras, Paris, Athens
and Salonica (again, on various occasions), Cambridge (twice, the second time
when I was kindly invited by Theresa Biberauer to teach a mini course on
categories), Chicago, Stony Brook, NYU and CUNY, Florence, Siena, Essex,
Amsterdam, Leiden, York, Trondheim, Lisbon and London.
Needless to say, this book would have never been completed without
Joanna’s constant patience and support.
My sincere gratitude goes out to the reviewers and referees who have looked
at pieces of this work: from the editor and the referees at Language who
compiled the long and extensive rejection report, a piece of writing that
perhaps influenced the course of this research project as significantly as key
bibliography on the topic, to anonymous referees in other journals, and to the
reviewers of Cambridge University Press. Last but not least, I wish to express
my gratitude to the Editorial Board of the Cambridge Studies in Linguistics
for their trust, encouragement and comments.
Finally, I wish to dedicate this book with sincere and most profound
gratitude to my teacher, mentor and friend Neil V. Smith.
1 Theories of grammatical category

1.1 Introduction
In this first chapter, we will review some preliminaries of our discussion on
parts of speech and on the word classes they define. As in the rest of this
monograph, our focus will be on lexical categories, more specifically nouns
and verbs. Then I will present a number of approaches in different theoretical
frameworks and from a variety of viewpoints. At the same time we will discuss
the generalizations that shed light on the nature of parts of speech, as well as
some necessary conceptual commitments that need to inform our building a
feature-based theory of lexical categories.
First of all, in Section 1.2 the distinction between ‘word class’ and ‘syntactic
category’ is drawn. The criteria used pre-theoretically, or otherwise, to distin-
guish between lexical categories are examined: notional, morphological and
syntactic; a brief review of prototype-based approaches is also included.
Section 1.3 looks at formal approaches and at theories positing that nouns
and verbs are specified in the lexicon as such, that categorial specification is
learned as a feature of words belonging to lexical categories. Section 1.4
introduces the formal analyses according to which categorization is a syntactic
process operating on category-less root material: nouns and verbs are ‘made’
in the syntax according to this view. Section 1.5 takes a look at two notional
approaches to lexical word classes and raises the question of how their insights
and generalizations could be incorporated into a generative approach. Section
1.6 briefly presents such an approach, the one to be discussed and argued for in
this book, an account that places at centre stage the claim that categorial
features are interpretable features.

1.2 Preliminaries to a theory: approaching the part-of-speech problem

As aptly put in the opening pages of Baker (2003), the obvious and funda-
mental question of how we define parts of speech – nouns, verbs and

1
2 Theories of grammatical category

adjectives – remains largely unresolved. Moreover, it is a question that is rarely

addressed in a thorough or satisfactory manner, although there is a lot of
stimulating work on the matter and although there is no shortage of both
typological and theoretical approaches to lexical categories. In this book
I am going to argue that we can successfully define nouns and verbs (I will
put aside adjectives for reasons to be discussed and clarified in Chapter 2) if
we shift away from viewing them as broad taxonomic categories. More
specifically, I am going to make a case for word class categories as encoding
what I call interpretive perspective: nouns and verbs represent different
viewpoints on concepts; they are not boxes of some kind into which different
concepts fall in order to get sorted. I am furthermore arguing that nouns
and verbs are ultimately reflexes of two distinctive features, [N] and [V], the
LF-interpretable features that actually encode these different interpretive
perspectives.
The theory advanced here gives priority to grammatical features, to categor-
ial features more precisely. As mentioned, it will be argued that two unary
categorial features exist, [N] and [V], and that the distinct behaviour of nouns
and verbs, of functional elements and of categorially mixed projections result
from the syntactic operations these features participate in and from their
interpretation at the interface between the Faculty of Language in the Narrow
sense (FLN) and the Conceptual–Intentional systems. The feature-driven char-
acter of this account is in part the result of a commitment to fleshing out better
the role of features in grammar. Generally speaking, I am convinced that our
understanding of the human Language Faculty will advance further only if we
pay as much attention to features as we (rightly and expectedly) do to
structural relations. True, grammatical features, conceived as instructions to
the interfaces after Chomsky (1995), will ultimately have to be motivated
externally – namely, by properties of the interfaces. However, we know very
little about these interfaces and much less about the Conceptual–Intentional
systems that language interfaces with. So, we cannot be confident about what
aspects of the Conceptual–Intentional systems might motivate a particular
feature or its specific values, or even its general behaviour. To wit, consider
the relatively straightforward case of Number: we can hardly know how many
number features are motivated by the Conceptual–Intentional systems to form
part of the Universal Grammar (UG) repertory of features – that is, without
looking at language first. More broadly speaking, it is almost a truism that most
of the things we know about the interface between language and the
Conceptual–Intentional systems, we do via our studying language, not via
studying the Conceptual–Intentional systems themselves.
 1.2 Approaching the part-of-speech problem 3

However, having thus mused, this monograph, a restrictive theory of

categorial features, sets itself somewhat humbler aims. In a nutshell,
I believe that a conception of categorial features as setting interpretive
perspectives, a view that can be traced back at least to Baker (2003), combined
with a syntactic decomposition approach to categories, as in Marantz (1997,
2000) and elsewhere, can achieve a very broad empirical coverage. This is
more so when such a theory incorporates valuable insights into parts of speech
from the functionalist-typological literature and from cognitive linguistics. The
theory here captures not only the basic semantics of nouns and verbs, but also
their position in syntactic structures, the nature of functional categories and the
existence and behaviour of mixed projections. It also makes concrete predic-
tions as to how labels are decided after Merge applies – that is, which of the
merged items projects, the workings of recategorization and conversion, and
the properties of mixed projections.

1.2.1 On syntactic categories and word classes: some clarifications

Rauh (2010) is a meticulous and very detailed survey of approaches to
syntactic categories from a number of theoretical viewpoints. In addition
to the sheer amount of information contained in her book and the wealth of
valuable insights for anyone interested in categories and linguistic theory in
general, Rauh (2010, 209–14, 325–39, 389–400) makes an important termino-
logical distinction between parts of speech (or what we could call ‘word
categories’) and syntactic categories.1 Roughly speaking, syntactic categories
are supposed to define the distribution of their members in a syntactic deriv-
ation. On the other hand, parts of speech correspond to the quasi-intuitively
identified classes into which words fall. In this sense, members of a part-of-
speech category/word class may belong to different syntactic categories;
consequently, syntactic categories are significantly finer-grained than parts of
speech. As this is a study of a theory of word class categories, I think it is
necessary to elaborate by supplying two examples illustrating the difference
between parts of speech and syntactic categories.
Since the late 1980s Tense has been identified in theoretical linguistics as a
part of speech, more specifically a functional category. However, finite Tense
has a very different syntactic behaviour, and distribution, to those of to, the
infinitival/defective Tense head. Hence, infinitival/defective to can take PRO
subjects, cannot assign nominative Case to subjects, and so on. Thus, although

1
A distinction already made in Anderson (1997, 12).
4 Theories of grammatical category

both future will and infinitival to belong to the same part of speech, the
category Tense, they belong to different syntactic categories, if syntactic
categories are to be defined on the grounds of distribution and distinct syntactic
behaviour.
Of course, one may (not without basis) object to applying distinctions
such as ‘part of speech’ versus ‘syntactic category’ to functional elements.
However, similar considerations apply to nouns – for example, proper nouns
as opposed to common ones, as discussed already in Chomsky (1965). Proper
and common nouns belong to the same part of speech, the same word class;
however, their syntactic behaviour (e.g., towards modification by adjectives,
relative clauses and so on) and their distribution (e.g., whether they may merge
with quantifiers and determiners . . .) are distinct, making them two separate
syntactic categories. This state is, perhaps, even more vividly illustrated by the
difference between count and mass nouns: although they belong to the same
word class, Noun, they display distinct syntactic behaviours (e.g., when
pluralized) and differences in distribution (e.g., regarding their compatibility
with numerals), as a result of marking distinct formal features.2
The stand I am going to take here is pretty straightforward: any formal
feature may (and in fact does) define a syntactic category, if syntactic categor-
ies are to be defined on the grounds of syntactic behaviour and if syntactic
behaviour is the result of interactions and relations (exclusively Agree rela-
tions, according to a probable hypothesis) among formal features. At the same
time, only categorial features define word classes – that is, parts of speech.
This will turn out to hold not only for lexical categories like noun and verb, as
expected, but for functional categories as well.
Henceforth, when using the term ‘category’ or ‘categories’, I will refer to
word class(es) and part(s) of speech, unless otherwise specified.

1.2.2 Parts of speech: the naïve notional approach

Most of us are already familiar with the notional criteria used in some school
grammars in order to define parts of speech. Although these are typically
relatively unsophisticated, notional criteria are not without interest. Further-
more, there are cognitive approaches that do employ notional criteria with
interesting results, Langacker (1987) and Anderson (1997) being the most
prominent among them. Indeed, contemporary notional approaches can turn
out to be germane to the project laid out here, as they foreground salient

2
An anonymous reviewer’s comments are gratefully acknowledged here.
 1.2 Approaching the part-of-speech problem 5

criteria of semantic interpretation in their attempt to define parts of speech;

such criteria are central to any approach seeking to define parts of speech in
terms of their interpretive properties as classes.
Let us now rehearse some more familiar and mainly pre-theoretical notional
criteria employed to define nouns and verbs and to distinguish between them.
So, typically, notional criteria distinguish between nouns and verbs as follows:
(1) NOUN VERB

‘object’ concept action concept

‘place’ concept ‘state’ concept
abstract concept
Counterarguments are not hard to come up with and criticism of something
like (1) is too easy, the stuff of ‘Introduction to Linguistics’ courses. Let us,
however, first of all observe that the state of affairs in (1) reflects both a
notional and (crucially) a taxonomic approach to categories. This notional and
taxonomic definition of categories – that is, deciding if a word goes into the
‘noun’ box or the ‘verb’ box on the basis of its meaning – is indeed deeply
flawed and possibly totally misguided. Consequently, yes, there are nouns and
verbs that do not fall under either of the above types: there are nouns that
denote ‘action’ concepts, such as handshake, race, construction and so on.
And we can, of course, also say that some verbs ‘denote abstract concepts’,
such as exist, emanate or consist (of).
Still, as already mentioned, we need to make a crucial point before dispara-
ging notional approaches: the table in (1) employs notional criteria to create a
rigid taxonomy; it therefore creates two boxes, one for a ‘Noun’ and one for a
‘Verb’, and it sorts concepts according to notional criteria. Which of the two
decisions, using notional criteria to sort concepts or creating a rigid taxonomy,
is the problem with the classification above? The answer is not always clear.
Research work and textbooks alike seem to suggest that the problem lies with
employing notional criteria: they generally tacitly put up with the rigid taxo-
nomic approach. An example of this is Robins (1964, 228 et seq.) who advises
against using ‘extra-linguistic’ criteria, like meaning, in our assigning words to
word classes. However, the notional criteria are anything but useless: Lan-
gacker (1987) and Anderson (1997), for instance, return to them to build a
theory of parts of speech – we will look at them in more detail in Section 1.5.
Equally importantly, when considering notional conceptions of categories,
we need to bring up the observation in Baker (2003, 293–4) that concepts
of particular types get canonically mapped onto nouns or verbs cross-
linguistically; see also Acquaviva (2009a) on nominal concepts. Two
6 Theories of grammatical category

representatives of types of concepts that canonically get mapped onto a

category are object concepts, which are mapped onto ‘prototypical’ nouns
(e.g., rock or tree), and dynamic event concepts, which are mapped onto
‘prototypical’ verbs (e.g., buy, hit, walk, fall), an observation made in Stowell
(1981, 26–7). Contrary to actual or possible claims that have been made in
relation to the so-called ‘Nootka debate’, no natural language expresses the
concept of rock, for instance, by using a simplex verb. Put otherwise, not all
nouns denote objects but object concepts are encoded as nouns (David
Pesetsky, personal communication, September 2005). So, maybe it is neces-
sary to either sharpen the notional criteria for category membership or recast
them in a different theoretical environment, instead of summarily
discarding them.

1.2.3 Parts of speech: morphological criteria

Pedagogical grammars informed by 100 years of structural linguistics typically
propose that the noun–verb difference is primarily a morphological one, a
difference internal to the linguistic system itself. In a sense, this is the exact
opposite of notional approaches and of all attempts to link category member-
ship to ontological or, even, modest semantic criteria. This is a point of view
that many formal linguists share (cf. Robins 1964, 228 et seq.), at least in
practice if not in principle. However, this approach to parts of speech goes
much further back, to Tēkhnē Grammatikē by Dionysius Thrax and to De
Lingua Latina, by Marcus Terentius Varro, who was Dionysius’ contempor-
ary. In both works, ‘division into parts of speech is first and foremost based on
morphological properties . . . the parts of speech introduced . . . are primarily
defended on the basis of inflectional properties’ (Rauh 2010, 17–20).
A contemporary implementation of these old ideas is illustrated in the table
in (2), where the distinction between noun and verb is made on the basis of
inflectional properties.
(2) NOUN VERB

number tense
case aspect
gender agreement3
Of course, here too, some semantic interpretation is involved, albeit indirectly:
for instance, the correlation of nouns with number, on the one hand, and of
verbs with tense, on the other, does not appear to be accidental – or, at least, it

3
Agreement with arguments, subjects most typically.
 1.2 Approaching the part-of-speech problem 7

should not be accidental, if important generalizations are not to be missed.

Both number morphology and tense morphology, characteristic of nouns and
verbs respectively, have specific and important semantic content: they are
unlike declension or conjugation class morphology, which are arbitrary,
Morphology-internal and completely irrelevant to meaning.4 We have also to
set grammatical case aside, which appears to be the result of processes between
grammar-internal features, and agreement with arguments, which is a property
of the Tense head or of a related functional element. Having done thus, the
interesting task underlying a (simplified) picture like the one in (2) is to
understand why the remaining generalizations hold:
a. Nouns exclusively pair up with Number, a category about individu-
ation and quantity.
b. Verbs exclusively pair up with Tense, a category about anchoring
events in time.5
I think that the above generalizations are strongly indicative of deeper rela-
tionships between the lexical categories of noun and verb and the functional
categories of Number and Tense respectively, relationships that go beyond
Morphology. Moreover, I will argue that these are relationships (noun–
Number and verb–Tense) which actually reveal the true nature of the semantic
interpretation of lexical categories.

1.2.4 Parts of speech: syntactic criteria

As implied above, an assumption tacitly (‘in practice if not in principle’)
underlying a lot of work involving some treatment of categories is that the
noun–verb difference is one concerning purely the linguistic system itself. One
way to express this intuition is by claiming that the noun–verb difference is
exclusively and narrowly syntactic, in a fashion similar to the difference
between nominative Case and accusative Case. For instance, we could claim
that the fundamental difference between nouns and verbs is that nouns project
no argument structure, whereas verbs do (Grimshaw 1990). Given the compli-
cations that such an approach would incur with respect to process nominals,
one could alternatively appeal to a similar, or even related, intuition and

4
Gender systems typically fall somewhere in between (Corbett 1991).
5
Nordlinger and Sadler (2004) and Lecarme (2004) argue that nominals (certainly encased inside
a functional shell) can be marked for independent tense – that is, bear a time specification
independent from that of the main event (and its verb). However, Tonhauser (2005, 2007)
convincingly argues against the existence of nominal Tense, taking it to be nominal Aspect
instead.
8 Theories of grammatical category

rephrase the noun–verb distinction along the terms of whether the expression
of their argument structures is obligatory or not:6
(3) NOUN VERB

Obligatory expression of argument structure? no yes

Of course, there are serious complications regarding a generalization like the

one in (3), and we will review some of these complications in Chapter 6 when
we investigate mixed projections and nominalizations more closely. However,
(3) has the look of a nice concrete difference, readily expressible and
sufficiently fundamental. Having said that, in relatively recent approaches to
argument structure, beginning with Hale and Keyser (1993, 2002), through
Kratzer (1996) and all the way to Ramchand (2008), Pylkkänen (2008) and
elsewhere, argument structure has no longer been viewed as the direct unmedi-
ated projection of lexical properties of the verb, as the result of the celebrated
Projection Principle. On the contrary, the growing trend is to have arguments
hosted by functional categories: for instance, Agents, as in Carla built a shed,
are by now commonly understood as the specifiers of a Voice category (Kratzer
1996). In other words, argument structure is currently understood as functional
structure that somehow reflects or translates lexical properties of the verb.
The above and other complications notwithstanding, the obligatory expres-
sion of argument structure is something that characterizes the projections
containing a verb, unlike those that contain a noun. Having said that, it would
be desirable if this difference could in turn be somehow derived, instead of
standing as an irreducible axiom. One motivation for this is that the (non-)
obligatory expression of argument structure also plays a very significant role
in our discussion of adjectives and, even more so, of adpositions: adjectives
seem to possess some kind of argument structure, especially when used
predicatively, whereas adpositions seem to be pure argument structures of
some sort – matters we will come back to in Chapter 2.

1.2.5 An interesting correlation

Setting up a broad framework of assumptions in which a theory of categorial
features will be developed, I have reaffirmed the understanding that, in its

6
Fu, Roeper and Borer (2001) influentially explain away such ‘complications’ by claiming that
process nominals contain verb phrases (VPs). Certainly, the expression of argument structure in
nominals can be a more intricate affair than Indo-European facts suggest: Stiebels (1999)
discusses Nahuatl, a language where all sorts of derived nominals, not just those with an event
reading, express their argument structure via affixes common with their base verbs.
 1.2 Approaching the part-of-speech problem 9

naïve version, the notional–taxonomic definition of parts of speech is

fallacious, with an emphasis on the problematic character of parts of speech
as taxonomies. At the same time, it has also been suggested, albeit in a
tentative fashion, that we would nevertheless have to vigorously seek criteria
of a conceptual/semantic nature in our endeavour to capture the noun–verb
distinction, as opposed to purely morphological or syntactic – that is,
grammar-internal – ones. This desideratum makes a lot of sense, at least
intuitively speaking, given that the distinction between noun and verb seems
to matter for interpretive reasons. It also appears that the noun–verb distinction
would reflect some sort of conceptually significant difference regarding the
very elements in the clause that are nouns or verbs – something that can hardly
be claimed about, say, the difference between Nominative and Accusative.
I think that we must regard the noun–verb distinction as one reflecting
conceptually significant differences, if important generalizations are not to be
missed: recall that the vast majority of words for physical objects are nouns
cross-linguistically; object concepts (tree, rock, stick etc.) are mapped onto
nouns. Of course, not all nouns denote concepts of physical objects. Baker
(2003, 290–5) discusses this generalization in an insightful way, crucially
adding that the nouns rock and theory cannot belong together in any concep-
tual taxonomic category, despite their both being nouns, following here the
discussion in Newmeyer (1998, chap. 4). However, what Baker does not
mention is this: the fact that rock and theory are both nouns is an argument
against the taxonomic aspect of the naïve notional approach, not against using
notional–semantic criteria to define categories – compare Acquaviva (2009a),
to which we will return in Chapter 4.
So, there appears to exist a correlation, after all, between object concepts
and nouns, as well as dynamic action concepts (hit, run, jump, eat etc.) and
verbs. How can such a correlation be captured?

1.2.6 Prototype theory

In the functional–typological methodological tradition, categories are viewed
as prototypes. In work by Givón (1984, chap. 3) and Croft (1991) categories
are conceived as prototypes occupying fuzzy areas along a continuum of
temporal stability, after Ross (1973). In this line of research, lexical categories
like nouns, adjectives and verbs are understood to differ with respect to their
protypical time stability. Hence, prototypical nouns are the most time-stable,
whereas prototypical verbs are the least time-stable; prototypical adjectives lie
somewhere in between. Put slightly differently: nouns are the most time-stable
category, verbs the least time-stable one, with adjectives in between. Baker
10 Theories of grammatical category

(2003, secs. 1.1–1.3) elaborates on the issues with this approach and with the
prototypical approach in general, principally along the lines of prototypes
predicting very little. Thus, a verb like persist encodes time-stability by
definition, whereas a noun like tachyon has time-instability encoded in its
meaning. Of course, the existence of nouns like tachyon, which express non-
time-stable concepts does not contradict protypicality: tachyon would qualify
as a non-protypical noun. Similar facts hold for non-prototypical verbs
expressing more or less time-stable concepts. This is precisely the problem
of what prototype-based theories of word classes actually predict. Consider,
for instance, the mid-section of the time stability continuum, where non-
prototypical relatively time-stable ‘verbal’ concepts, non-prototypical rela-
tively non-time-stable ‘nominal’ concepts and ‘adjectival’ ones (between
nouns and verbs, by definition) co-exist: the question is what conceptual
mechanism decides which category concepts populating that middle area are
assigned to? Is category-assignment performed at random? This is a matter that
Rauh (2010, 313–21) also raises, although departing from a slightly different
set of theoretical concerns; she goes on to argue for discrete boundaries
between categories.
A more interesting issue is one mentioned above: prototypical (like rock)
and less prototypical (like theory) nouns and prototypical (like buy) and less
prototypical (like instantiate) verbs all behave in the same fashion as far as
grammar itself is concerned (Newmeyer 1998, chap. 4). Clearly, to the extent
that prototypicality matters for the mechanism of the Language Faculty per se,
and to the extent that prototypicality is reflected on the grammatical behaviour
of nouns, verbs and adjectives, prototype effects spring from factors external to
the syntax.7
The limited role of prototypicality as far as the grammar-internal behaviour
of more prototypical or less prototypical members of a category is concerned
is acknowledged in Croft (1991, 2001), who argues that prototypicality correl-
ates with two kinds of markedness patterns across languages. First, prototypi-
cality correlates with structural markedness, in that items deviating from the
semantic prototype (e.g., referential expressions that denote events, like hand-
shake or wedding, or object-denoting words used as predicates, like ice in The
water became ice) tend to occur with additional morphemes. Interestingly, this
is a generalization about the functional layer around an event-denoting noun or
an object-denoting predicate, not about the lexical elements themselves.

7
I am grateful to a reviewer for this discussion.
 1.2 Approaching the part-of-speech problem 11

A second markedness pattern that prototypicality correlates with, again

according to Croft (1991, 2001), is behavioural markedness: namely, that
items deviating from the semantic prototype tend to have a more limited
distribution than prototypical items. Again, distribution is a question of the
functional layer within which a lexical category is embedded. In this respect,
distribution is a matter of syntactic category, as opposed to a matter of word
class/part of speech. This distinction is also rehearsed in detail and with
reference to both empirical evidence and conceptual arguments, in Newmeyer
(1998, chap. 4) and Rauh (2010, 325–39). Concluding this very brief
overview, Croft’s theory of course allows for the possibility that prototypical
and non-prototypical items behave in exactly the same way (like stone and
theory in English); moreover, his theory makes the weak prediction that the
non-prototypical members of a category will not be less marked than the
prototypical ones.
Despite the criticism above, work on defining categories along the lines
of prototypes brings an extremely important insight to the discussion of
categories: namely, the relevance of time stability in defining parts of speech
as well as what their interpretation would be. The significance of this
contribution will be revisited in Chapter 4, when the interpretation of verbs
will be addressed.

1.2.7 Summarizing: necessary ingredients of a theory of category

The table in (4) outlines some empirical differences between nouns and verbs:
(4) NOUNS VERBS

number tense (aspect)

Case-marked Case-assigning
gender agreement with arguments
argument structure covert argument structure overt
determiners particles

The task at hand, the one that the theory of categorial features to be presented
here will take up, is to explain these differences in a principled manner. We
have already discussed the shortcomings of prototype-based analyses: they
neither predict the identical grammatical behaviour of prototypical nouns like
stone to that of non-prototypical ones like theory, nor do they explain the
differences between nouns and verbs. Still, most generative theories of
category are not faring any better (Baker 2003, chap. 1): they are also descrip-
tive part-of-speech systems that make no predictions about either the syntactic
behaviour or the semantic interpretation of an element x belonging to a
12 Theories of grammatical category

category N or V. In other words, they set up categorial feature systems which

(occasionally) cross-classify, but generative theories of category do not tell us
what these categories do and in what way they are different from each other.
As Baker (2003, chap. 1) correctly acknowledges, a theory of category must be
explanatory or, at least, predictive.
In the rest of this chapter, I will present some generative theories of word
classes, along with Langacker’s (1987) and Anderson’s (1997) versions of a
notional part-of-speech theory.

1.3 Categories in the lexicon

In order to make solid the desideratum expressed by Baker (2003) for an
explanatory theory of category, let us consider the system in Chomsky’s
(1970) Remarks on Nominalization, also known as the ‘Amherst system’
(Rauh 2010, 94), so-called because it was revised to include prepositions in
a series of lectures by Chomsky in Amherst, Massachusetts. The reason for
doing so is twofold. First, this system is still quite popular and (at least)
influential. By ‘popular’, I mean that, in its near-original version, it is still
tacitly presupposed in a substantial body of work touching upon the question
of parts of speech and grammatical categories. Moreover, it has given rise to
some theories of category vying for explanatory adequacy, such as the ones in
Stowell (1981), Déchaine (1993) and van Riemsdijk (1998a). Second, the
Amherst system is – crucially – feature based: it therefore avails itself of a
theoretical and methodological machinery which, potentially, makes it suitable
for capturing important generalizations and for making predictions of an
explanatory nature – again, like its version in Stowell (1981). Furthermore,
its being expressed in terms of features makes it compatible with both lexicalist
and non-lexicalist (including weak lexicalist) views on grammatical structure.
In these respects, it informs the theory presented in this monograph in a
significant and substantial way.
Having said that, the Chomskian categorial feature system has the major
handicap of making no predictions, as it is purely taxonomic. It is summarized
in tabulated form in (5):
(5) N V

Noun þ 
Verb  þ
Adjective þ þ
Preposition  
 1.3 Categories in the lexicon 13

So, let us comment on the system represented in (5): we have the cross-
categorization of four categories by means of two binary features. Hence,
instead of three (or four) primitive lexical categories, we have two primitive
binary features: [N] and [V] with their (expected)  values. The attributes [N]
and [V] are not identified with any sort of interpretation, at least not in the
original Amherst system and not in any clear-cut way, and the  values
apparently mark the positive versus negative value of such an attribute,
whatever it may represent. The picture, however, certainly looks elegant, as
now it is possible to cross-categorize lexical categories. Having said that, this
cross-categorizing cannot become truly significant, or even useful, until we
resolve the question of what these features, and their values, stand for. To
make this clearer, it is very difficult to get the batteries of properties character-
istic of nouns and verbs in (4) to result from the system in (5). Following Baker
(2003), I agree that this is the main problem with the Amherst feature system:
it is a purely taxonomic system that does not predict much. Should we take up
a more clement attitude to it, the Amherst system establishes cross-
categorizations at an abstract level: nouns share a [þN] feature with adjectives,
verbs a [þV], again with adjectives, and the two appear to be completely
dissociated. Although these are hardly trivial predictions, they are clearly not
robust enough to capture the picture in (4) or, even, part thereof. One reason
for this problem is exactly that we do not know what the attributes [N] and [V]
stand for. All we have at this point is nouns and adjectives forming a natural
class, both being [þN], and verbs and adjectives forming a natural class
because of their [þV] value; finally, prepositions also belong to a natural
class with nouns (both categories being [V]) and verbs, too (due to a
common [N] value) – but not with adjectives (cf. Stowell 1981, 21–2).
Departing from the natural classes defined by the Amherst system of
categorial features [N] and [V], Stowell sets out in the first chapter of his
(1981) thesis to show that these are classes to which syntactic and morpho-
logical rules actually refer. These classes are given in (6).
(6) Natural classes according to the Amherst system

[þN] nouns, adjectives

[N] verbs, prepositions
[þV] verbs, adjectives
[V] prepositions, nouns
Simultaneously, he argues that syntactic rules do not refer to ‘unnatural’
classes like {noun, verb} or {adjective, preposition}. He goes on to spell out
some of the rules that refer to the natural classes above:
14 Theories of grammatical category

• the [þN] categories, nouns and adjectives, project phrases where

of-insertion applies in English, as in destruction of the city and fearful
of ghosts;
• the [N] categories, verbs and prepositions, are the Case-assigning
ones: only they can take ‘bare NP objects’ (Stowell 1981, 23);
• the [þV] categories, verbs and adjectives (and participles), can be
prefixed with un– in English (a word formation rule); in German they
can function as prenominal modifiers, as in the examples in (7):
(7) [þV] prenominal modifiers in German
a. der [seinen Freundin überdrüssige] Student
the his girlfriend weary student
‘The student weary of his girlfriend.’
b. die [mit ewigen Snee bedeckten] Berge
the with eternal snow covered mountains
‘The mountains covered with eternal snow.’

• Finally, XPs headed by the [V] categories, nouns and prepositions,

can be clefted.
A second argument for the syntactic and morphological significance of the
classes defined by the categorial features [N] and [V] is that in many
languages these classes collapse together into a single category – for example,
nouns and adjectives, or verbs and adjectives. This is arguably not possible
between nouns and verbs, or adjectives and prepositions.
Generally speaking, Stowell (1981) attempts to show that classes not
defined by the two categorial features [N] and [V] do not form natural
classes to which syntactic or morphological rules are sensitive. Hence, Stow-
ell’s attempt to make categorial features relevant for syntax genuinely informs
the account developed here as a desideratum. Déchaine (1993) is another
important analysis in this line of thinking.
Déchaine (1993, 25–36) begins with a review and criticism of previous
accounts of categorial features. She then introduces her own system of cat-
egorial features, one that aspires to capture the natural classes that categories
form and to which grammatical rules refer (as in Stowell), the lexical–
functional distinction and biuniqueness between functional elements and some
lexical categories.8 She therefore proposes, for the first time to the best of my
knowledge, three privative/unary/monovalent categorial features: [Functional],

8
The lexical–functional distinction and biuniqueness are examined in Chapter 5 of this book.
 1.3 Categories in the lexicon 15

[Nominal] and [Referential]. The cross-classification of these features yields

the following category system (Déchaine 1993, 38):
(8) The Déchaine (1993) categorial system
[Nominal] Adjective
[Nominal] [Referential] Noun
[Referential] Verb
– Preposition
[Functional] [Nominal] Kase
[Functional] [Nominal] [Referential] Determiner
[Functional] [Referential] Tense
[Functional] Comp

Some comments are in order: the primacy of [Nominal], as opposed to a

purported feature [Verbal], is argued for on the basis that some processes
select only for nouns (and adjectives) and on the basis of reinterpreting
acquisition facts. Prepositions are labelled (via the absence of any categorial
feature specification) as a sort of ‘elsewhere’ lexical category that no deriv-
ational process can target (we return to this in Chapter 2, Section 2.9). The
general idea is that derivational affixes attach to and derive only categories that
are specified as [Nominal] or [Referential], which is to say that these two are
the features defining lexical categories. The noun–Determiner connection,
Extended Projections (Grimshaw 1991), the (non-)identity between adjectives
and adverbs, the modifying role or non-referential categories are all examined
in turn as evidence for (8). A nice summary of the system and its rationale, and
also of the desire to establish categorial features on the basis of fundamental
conceptual categories, is given in Déchaine (1993, 71): ‘[Nominal] is to be
preferred over [Verbal] as the basic feature which distinguishes categories;
[Referential] provides a means of characterizing the notion of core extended
projection;9 and [Functional] provides a principled distinction between open-
class and closed-class items.’ A final innovation in Déchaine’s system is, to my
mind, her manifest effort to reconceptualize categorial features as semantically
significant; this becomes almost evident from the passage above and by the
way she names her features, especially the abstract, but crucial to her system,
[Referential] feature.
Moving on, the above three generative systems of categorial features and –
to a significant extent – Baker (2003) unambiguously incorporate the thesis
that lexical items come from the lexicon specified for category, by virtue of
features like [N] and [V]. So, dog is a noun and write is a verb because we have

9
That is, the projection of V and that of N.
16 Theories of grammatical category

memorized them as such and because they are part of the lexicon – the lexicon
containing everything in language that must be memorized: ‘a list of excep-
tions, whatever does not follow from general principles’ (Chomsky 1995,
234). In other words, lexical elements come from the lexicon labelled for
a category: noun, verb and so on. The lexicon contains entries like dog
[þN, V] and do [N, þV]. Of course, this is roughly the way most
traditional grammars view the matter, as well.
Baker (2003) explicitly announces that he aims to give content to the
features [N] and [V], rather than view them as convenient labels that create
taxonomies. He revises the system in (5) by positing two privative features,
like Déchaine (1993), instead of the received binary ones. Importantly,
these are expressly hypothesized to be LF-interpretable features as well
as to trigger particular syntactic behaviours. The table in (9) summarizes
this:
(9) Semantic interpretation Syntactic behaviour

[N] sortality referential index

[V] predication specifier

Sortality is what makes nouns nominal. Baker (2003, 290) essentially treats
a sortal concept as one that canonically complies with the principle of
identity: a sortal concept is such that it can be said about it that it is the
same as or different from X.10 Furthermore, sortality is understood as the
very property that enables nouns to bear referential indices. At the same
time, [V], which makes verbs, is taken to encode predication. Baker also
argues that verbs are the only lexical categories that can stand as predicates
without the mediation of a functional category Pred. The predicative nature
of verbs is correlated with them arguably being the only lexical category
that projects a specifier.
Baker’s (2003) system yields the following lexical categories:
(10) Nouns [N] ! sortal concepts, referential indices in syntax
Verbs [V] ! predicates, with (subject) specifiers in syntax
Adjectives  ! ‘other’ concepts, pure properties

Baker’s account makes a wide-ranging set of predictions about verbal and

other predicates in co-ordination, the behaviour of causatives, and the noun–
individuation relation (namely, the elective affinities between nouns, Number

10
We will return to a more detailed discussion of sortality in Chapter 4.
 1.4 Deconstructing categories 17

and quantifiers). His account, which forms the basis from which our theory of
categorial features departs, is revised in Chapter 4, extended in Chapters 5 and
6 and further pored over in the Appendix.

1.4 Deconstructing categories

1.4.1 Distributed Morphology
An alternative and very dynamic approach to word classes and the nature
of parts of speech emerged in the 1990s within the framework of Distrib-
uted Morphology. The basic idea is non-lexicalist: the syntactic
deconstruction of words. Therefore, categories like nouns and verbs are
products of syntactic operations and do not come marked on lexical items.
Nouns and verbs are not pre-packaged as such in the lexicon, they are
‘made’ so in the course of the syntactic derivation. The empirical conse-
quences of syntactic categorization have been explored in detail in a
significant body of work, including – but not restricted to – Harley and
Noyer (1998), Embick (2000), Alexiadou (2001), Folli, Harley and Karimi
(2003), Arad (2003, 2005), Folli and Harley (2005), Harley (2005a, 2005b,
2007, 2009, 2012a), Marantz (2005, 2006), Basilico (2008), Embick and
Marantz (2008), Lowenstamm (2008), Acquaviva (2009b), Volpe (2009),
Acquaviva and Panagiotidis (2012) and, in a slightly different framework
but in considerable detail, Borer (2005, 2009) and De Belder (2011). I will
not attempt to summarize the diverse and insightful findings of this line of
research and inquiry; I will only present the way it works and return to it in
Chapter 3.
So, suppose that lexical elements (roots) do not come pre-packaged with
categorial features from the lexicon. In other words, and very roughly
speaking, the lexicon (or its equivalent) contains entries like dog and do
without them bearing any categorial features. Words of the ‘lexical categor-
ies’ N, V and A are created in the syntax via the combination of at least a
categorizing head and a root: roots themselves are category-less or ‘acat-
egorial’. Thus, ‘noun’, ‘verb’ and ‘adjective’ are not categories specified on
lexical items in a pre-syntactic lexicon: categorization is a syntactic
process.
More precisely, according to the syntactic decomposition of categories,
acategorial roots are inserted in syntax. Dedicated syntactic heads, categorizers –
a nominalizer (n), a verbalizer (v) and an adjectivizer (a) – make them nouns,
verbs or adjectives. Of course, the projections of n and v (and a) may contain
more than just themselves and a root:
18 Theories of grammatical category

(11)

See Marantz (2000) – also Halle and Marantz (1993) – for the background
regarding the categorial decomposition in Distributed Morphology. We will
return to this framework, in which the theory presented here is couched, in
Chapter 3.

1.4.2 Radical categorylessness

A radical way to do syntactic decomposition is to posit not only that lexical
elements (roots) do not come pre-packaged with categorial features from the
lexicon but also that there are no dedicated categorizers in Syntax, either. This
is what is argued for in Borer (2003, 2009), De Belder (2011), and also
Alexiadou (2001): that the functional environment dominating a root actually
defines the category of the structure containing the root. An obvious conceptual
consequence of such an approach is that it is not Comp and Tense that ‘go with’
a verb, let alone that are selected by a verb, or vice versa. Instead, Comp and
Tense make a root a verb. Similarly, Det and Num do not ‘go with’ a noun: Det
and Num make a root a noun. This is illustrated in the simplified trees below:
(12)

Expanding upon the rationale of this approach, we can say that the noun–
verb distinction is purely configurational. Nouns and verbs are essentially
structures containing a root and being characterized purely and exclusively
by the functional environment around the said root. So, what matters is the
syntactic positioning of the root. To illustrate, if a root (say SLEEP) is inserted
within a functional complex like the following, then it becomes a verb:
(13)
 1.5 Notional approach: Langacker and Anderson 19

Similarly, if the selfsame root SLEEP is inserted within a functional com-

plex like the following, then it becomes a noun:
(14)

As part of a programme of radical constructionalism, where any root can be

slotted into any functional superstructure, as happens above, syntactic categor-
ization without categorizers also informs Borer (2005) but is significantly
revised in Borer (2009). The radical categorylessness approach will become
especially cogent for the discussion in Chapter 6, when mixed projections are
examined.

1.5 The notional approach revisited: Langacker (1987)

and Anderson (1997)
Cognitive Grammar (Langacker 1987) is a non-formal theory of grammar.
It argues that grammar is ‘inherently meaningful’ and that language constitutes
‘an integral facet of cognition’ (Langacker 1987, 590). There is no syntactic
mechanism independent of semantics, and there are no levels of representation,
at least not in the way these are understood in modular approaches to the
Language Faculty. To wit, Cognitive Grammar does use diagrams to represent
grammatical meanings and structures, but differences in grammatical behaviour
are directly attributed to conceptual/semantic differences.11 Thus, it is no
wonder that Langacker’s approach to categories is a notional one. However,
and remembering the distinction I drew between notional and taxonomic
treatments of category, Cognitive Grammar does not treat categories as big
boxes in which concepts (or rather, the words expressing them) are sorted
according to ontological criteria. Instead, different categories in Cognitive
Grammar are understood as different conceptualizations. That is an extremely
important point in understanding the semantics of categories like noun and verb
by themselves: shifting away from naïve ontological taxonomies, Langacker
(1987) understands categories as a particular way in which concepts are ‘seen’.
Rauh (2010, 238) provides an example of how ‘noun’ is understood; Langacker
(1987) describes nouns as nominal predications that designate a THING.

11
I gratefully acknowledge comments by an anonymous reviewer on Cognitive Grammar.
20 Theories of grammatical category

However, THING does not stand for a physical object (this would bring us back
to naïve ontologically informed taxonomies of school grammars) but, rather, as
a ‘region in some domain’ (Langacker 1987, 189). Immediately it becomes
evident, at least intuitively, how rock, theory and wedding can all be nouns:
they define areas in different conceptual domains. Of course, a lot depends on
what these conceptual domains are and, most crucially, how exactly they are
mapped on words – see Acquaviva (2009a) for criticism and Acquaviva and
Panagiotidis (2012) for discussion. However, understanding categories as con-
ceptualizations instead of ontologies constitutes a major step in understanding
what they mean and how they work.
As far as the other ‘basic’ categories are concerned, adjectives and adverbs
conceptualize ATEMPORAL RELATIONS (Langacker 1987, 248), whereas
verbs are understood as conceptualizing PROCESSES. Regarding the gener-
alization in Section 2.5 – namely, that the majority of words meaning physical
objects are nouns cross-linguistically – Langacker (2000, 10) captures it by
appealing to the prototypicality of THING when it comes to object concepts
and to the protypicality of PROCESS when it comes to ‘an asymmetrical
energetic interaction involving an agent and a patient’. The details are very
interesting and certainly intricate; however, they are not readily translatable to
a formal framework like the one employed here.
Turning now to Anderson (1997), the picture is similar to that presented by
Langacker’s approach, with three important differences: Anderson’s frame-
work is a formal one and expresses generalizations by encoding them as
features. He understands categories as ‘grammaticalisations of cognitive – or
notional – constructs’ (Anderson 1997, 1), thus making his theory more
compatible with the underpinnings of a generative theory: he essentially claims
that categories are ways in which grammar, which forms a separate module
from general cognition, translates or imports (‘grammaticalizes’) concepts.
Third, he formalizes the different categories, the different grammaticalizations
of concepts, using two features and a relation of preponderance between them:
a feature P, standing for predicativity or, rather, predicability, and a feature N,
standing for the ability to function as an argument. The resulting categories for
English are the following:

(15) {P} auxiliary

{P;N} verb
{P:N} adjective
{N;P} noun
{N} name
{} functor
 1.6 Categorial features make categorizers 21

Explicating, auxiliaries {P} are only usable as predicates, names {N} only as
arguments, and functors { } as neither. Adjectives {P:N} are the result of a
balanced relationship between predicability and the ability to function as
arguments, whereas in verbs {P;N} predicability takes preponderance over
possible argumenthood and vice versa in nouns {N;P}. To my mind it is
precisely this ‘preponderance’ factor that makes the relation between two
features very hard to formalize any further. Moreover, in Anderson’s system
there is again a conflation between the category by itself, say ‘verb’, and the
functional layer around it. For instance, saying that verbs by themselves may
function as arguments is misleading, as verbs can only function as arguments
when they are embedded within a nominalizing functional shell, in which case
they form part of a mixed projection, such as gerunds, nominalized infinitives
and the like.

1.6 The present approach: LF-interpretable categorial features

make categorizers
The theory to be presented in this monograph is one of categorial features. In
this it is in the spirit first of all of the original Amherst system and of its
refinements, developments and reconceptions – namely, Stowell (1981),
Déchaine (1993) and Baker (2003). It is inevitably also in the spirit of the
way Head-Driven Phrase Structure Grammar (HPSG) and Lexical Function
Grammar (LFG) treat grammatical category: as a feature or as the result of
feature combinations; see Rauh (2010, chap. 6) for a survey of how category is
treated in these frameworks in terms of features. Finally, it displays affinities
with Anderson (1997), in that it seeks to capture parts of speech as the
phenotypes of different categorial features. Moreover, as in Baker’s and
Anderson’s systems, I will seriously explore the idea that categorial features
encode conceptual (if not notional) content – namely, that they encode differ-
ent conceptualizations: what I will call different interpretive perspectives on
concepts. In this, we will follow the lead of Langacker (1987) and seek to
understand categorial features as setting up different perspectives on concepts.
In specifying the interpretive perspectives which [N] and [V] encode, we will
capitalize both on insights in Baker (2003) as well as those regarding time
stability in the functional–typological literature. Finally, once we take categor-
ial features to encode conceptual content (technically speaking: to be LF-
interpretable), then we will be able both to explore their participation in
syntactic relations, most notably Agree, and to establish them as the very
features that make categorizers, n and v, necessary in frameworks like
22 Theories of grammatical category

Distributed Morphology, which do categorization syntactically. In other

words, we will weave together a seemingly disparate number of yarns in order
to hopefully produce a coherent theory of lexical, functional and mixed
categories.
As is already evident, the role of categorial features in defining grammatical
category is taken very seriously in this monograph, especially in the light of its
theoretical and methodological commitment to syntactic decomposition and to
viewing categorization as a syntactic process. I understand categorial features
as LF-interpretable and as no different from other syntactic features: in other
words, I take [N] and [V] to be instructions to an interface, viz. the interface
between the Faculty of Language in the Narrow sense (FLN) and the
Conceptual–Intentional systems (which I will still informally call ‘LF’). At
the same time, contra Panagiotidis (2005), I delegate all true categorial dis-
tinctions to categorizers. I consider them to be true syntactic heads, and I will
go as far as claiming that the nominalizer n and the verbalizer v are the only
lexical heads in Syntax, serving as the elements that enable roots to structurally
combine with bundles (or, perhaps, batteries) of FLN-intrinsic features,
selected from a pool made available by UG.
The resulting theory proposes that each categorizer is the locus of a categor-
ial feature. Categorial features introduce fundamental interpretive perspectives
in which the categorizers’ complements are to be interpreted. These funda-
mental perspectives are necessary when the categorizers’ complements consist
of roots and/or root projections: this subsumes Embick and Marantz’s (2008, 6)
Categorization Assumption.
The theory also predicts the existence of uninterpretable categorial fea-
tures; these are taken to flag bundles/batteries of features (i.e., functional
heads) that must have lexical material in their complement. This thesis
provides a simple solution to how we can state the relationship between
functional projections and the lexical material they dominate. Under the
straightforward assumption that uninterpretable categorial features act as
Probes for Agree relations, the theory can also capture the biunique relation-
ship between particular functional and lexical elements – for example, the
biunique relation between Number and nouns. Appealing to Agree relations
between uninterpretable and interpretable categorial features – that is, cat-
egorial Agree – also explains why lexical material is always merged first, ‘at
the bottom of the tree’. Finally, categorial Agree can explain how the label is
decided in most instances of First Merge, possibly all of them: for instance,
when a head – an ‘LI’, for ‘lexical item’, in Chomsky (2000, 2004) – merges
with another head.
 1.6 Categorial features make categorizers 23

The ordinary character of categorial features as LF-interpretable will serve

to solve the recurring problem in grammatical theory of how best to analyse
mixed projections. The existence of functional categorizers, Switches, con-
sisting of one uninterpretable (hence ‘functional’) and one interpretable (hence
categorizing) categorial feature is posited and their position and role are
scrutinized – namely, what size and type their complements can be.
2 Are word class categories
universal?

2.1 Introduction
A theory of categorial features cannot overlook the question of whether the
lexical classes we recognize as nouns, verbs and adjectives are universal or
particular to a subset of languages. This task becomes more important if one
considers the frequent charge that categories like ‘noun’ and ‘verb’ have been
analytically imposed by Eurocentric grammarians and linguists (mostly) onto
non-Indo-European languages: this has been a matter of debate for at least the
past 80 years.1 This chapter critically reviews some of the points of contention
in the literature on the universality of lexical categories, beginning with the
need to make sure we know what we are talking about when we talk about
nouns, verbs and adjectives. The reason such a caveat is necessary, followed
by some preliminary clarifications and statements on methods and termin-
ology, is because different scholars tend to mean different things when they
talk about nouns, verbs and adjectives – especially, but not exclusively, when
they use them in a mainly pre-theoretical fashion. After these necessary
refinements, the universality itself is discussed, as well as a detailed justifica-
tion of why it was decided not to include adjectives (and adpositions) among
lexical categories in this book.
The universality question is introduced in Section 2.2, and two methodo-
logical guidelines – that is, to distinguish verbs from their functional super-
structure and to use morphological criteria cautiously – are introduced and
examined in Section 2.3. Section 2.4 reviews evidence from Tagalog and Riau
Indonesian that has been claimed to demonstrate grounds for a single undiffer-
entiated lexical category in these languages. Section 2.5 revisits the Nootka
debate and Baker’s (2003) treatment of it, further suggesting that the debate is

1
This debate was recently vigorously revived, or rather ‘updated’, in the wake of Everett’s (2005)
claims that Pirahã is radically and profoundly different from other languages. See Nevins,
Pesetsky and Rodrigues (2009) for a careful assessment.

24
 2.2 Are nouns and verbs universal? How can we tell? 25

possibly moot in a syntactic categorization approach. The resulting picture is

reviewed in Section 2.7. Section 2.8 justifies the non-inclusion of adjectives in
the discussion, and Section 2.9 does the same for adpositions. The final section
concludes the chapter.

2.2 Do all languages have nouns and verbs? How can we tell?
We set out in the previous chapter to develop a UG theory of grammatical
category. This entails making claims about the Language Faculty. More
specifically, claims about the existence of (particular) categorial features
within the purported ‘pool’ of UG-available features will be raised; we will
then have to go beyond typological generalizations that could reflect
(a) extralinguistic factors (e.g., all languages have a word for ‘mother’ or
‘sun’ for obvious reasons) or, worse, (b) a methodological bias that seeks
to impose Indo-European categories on languages working in distinctly
non-Indo-European ways.
If our desideratum of a theory of category founded on LF-interpretable
syntactic features is to have any substance at all, we need to ensure first that
differences between nouns and verbs are universal and subsequently to
inquire as to whether the two lexical categories are indeed manifested in all
natural languages.2 This question is not a matter of pure data collection,
however. If the noun–verb distinction is so persistent and fundamental, it
needs to be shown why this is the case and what kind of conceptual distinc-
tion it reflects and encodes. Moreover, on the methodological side of things,
we need to move away from the situation realized by languages like English:
in languages like English, the noun–verb distinction is also one between two
broad syntactic categories – that is, between two categories that regulate or,
at least, affect syntactic distribution: whether X will appear in such and such
a position depends (also) on its word category membership – see Rauh (2010,
325–39). However, as will be seen, this is not the case cross-linguistically: in
some languages, verbs and nouns have a seemingly free distribution; in other
languages nouns and verbs are embedded within layer upon layer of func-
tional material, material which, typically, is morphologically attached to the
lexical element. Finally, a great number of languages possess mixed projec-
tions. In this section we will examine cases of allegedly free distribution
between nouns and verbs, as well as verbs and nouns embedded inside a lot

2
This second statement would be highly desirable but it is not a necessary condition for the
universality of categorial features.
26 Are word class categories universal?

of functional material. We shall leave a more detailed discussion of mixed

projections for Chapter 6.

2.3 Two caveats: when we talk about ‘verb’ and ‘noun’

Given the above, we first need to decide how we can tell whether a language
has nouns and verbs in a quasi-pre-theoretical way. This is going to be useful
for the discussion that will follow.

2.3.1 Verbs, not their entourage

A first point to watch is the following: it is always important and necessary to
distinguish between a verb and the larger constituent within which verbs are
typically contained. In other words, we need to separate the verbal element –
let us call it V for the sake of exposition – from the verbal phrasal constituent it
is embedded within: for example, Voice phrases, aspectual phrases, Tense
phrases and even complete Complementizer phrases. Note also that this
embedding is customarily morphologically mediated, in that Morphology
rarely (if at all) expresses a V as a stand-alone word. This is not the case only
in inflectionally rich languages, but also in English.
Let us begin with what is considered a periphrastic verb form in most
pedagogical descriptions of English, have been watching (the ‘present perfect
continuous form of the verb watch’). Of course, have been watching is not a
‘verb’; it is not even a constituent. The most appropriate way to describe have
been watching would be to acknowledge that it is part of a Tense Phrase (TP)
containing three Verb Phrases (VPs) embedded within each other – see, for
example, Haegeman (2006, sec. 3.4). However, the object of watch, which
would complete the TP constituent together with the subject, is left out.
(1)

Although the above example certainly feels trivial, it does so because it is

drawn from a familiar language. Moreover, it is indicative of the fact that
similar considerations apply, and must be kept in mind, when we examine
heavily inflected languages, even of the familiar sort, like Greek, Russian or
Turkish: what we call a ‘verb’ in such languages usually contains the verb plus
 2.3 Two caveats 27

a lot more structure. Illustrating this, let us have a look at an equally straight-
forward example from Turkish, a token of what traditionally is described as a
‘verb’; notice that, as expected from an agglutinating language, it consists of a
single word:
(2) unut- ma- y- acağ- ımız
forget Neg Fut 1 st .Pl

Again, the ‘verb form’ packages together a part of a TP constituent. More

accurately, the single morphological word unutmayacağımız packages together
the heads of the TP projection line. So, as with the English example in (1)
above, we must be very careful to tease apart the V from its functional
environment, a functional superstructure that includes auxiliaries, modals,
Mood, Tense, Aspect and Voice markers, agreement markers and arguments
themselves, before we make any claims about V itself.

2.3.2 Misled by morphological criteria: nouns and verbs looking alike

A second point to watch, equally elementary but hardly trivial, is that the
sharing of identical morphological forms – for example, affixes – and the fact
that some inflectional morphology may be common to both nouns and verbs
do not entail that there is no noun–verb distinction in a language. The sharing
of, say, suffixes between nouns and verbs could be a sign of them being
embedded within identical functional shells, when of course we are not simply
dealing with a historical accident resulting in (extensive) morphological
syncretism. Similarly, the lack of morphological distinction between nouns
and verbs certainly does not entail that there is no verb–noun distinction in the
grammar. An obvious example of this is the fact that English ‘zero suffixation’,
whichever is the proper way to analyse it, does not neutralize the noun–verb
distinction in English. Moreover, as Pesetsky (1995), Marantz (2000), Arad
(2005) and Borer (2005) show, ‘zero suffixation’ effectively belies the wealth
of structure present in examples like certificate (as a denominal verb) below.
(3) work do
sleep find
fight test
play poke
water certificate (vs certify)

To wit, observe English ‘–ing’, to which we will return in Chapter 6. No matter

what analysis we pick for it, we have hardly one category-neutral instance of
painting in examples (4), (5), (6), (7) and (8), where –ing belies the varied
28 Are word class categories universal?

roles the word painting can have in each of the sentences, the different
categories it can belong to and the diverse structures it can participate in.
Thus, whether there are two instances of –ing, an aspectual one and a nom-
inalizer, or just one –ing, an aspectual marker, painting in (4) is a prototypical
noun denoting an object concept, while in (8) it is a prototypical verb denoting
a dynamic event, an activity. Even more interestingly, in the examples
in-between, each instance of painting is embedded in rich (mixed) projections
and involves complex and differentiated structures.3 So, we have a process
nominal, still a noun, in (5), a gerund with verbal and nominal properties in (6)
and a verbal element in (7).
(4) The painting is on the wall.

(5) [Helena’s painting of horses in her free time] proved more than a hobby.

(6) [Helena(’s) painting horses in her free time] proved more than a hobby.

(7) I watched [her painting horse paintings].

(8) She was painting horses.

2.3.3 What criterion, then?

Using well-described languages like English and Turkish to illustrate the two
points above serves the purpose of highlighting not only the perils of
exoticization on behalf of too eager field linguists and/or enthusiastic local
informants, but primarily of relying on surface criteria to discuss category
membership, especially when it comes to talking about less well-described
languages. Moreover, the two points raised in this section are methodological
mementos and cannot be used to stipulate noun–verb differences where there
would be none. Indeed, a safe criterion that a language does not distinguish
between nouns and verbs would be the following, from Baker (2003,
sec. 3.9):
(9) If a language can use all of its lexical elements interchangeably in all
contexts, then it does not distinguish between nouns and verbs.

This means that in such languages we should be able to freely insert lexical
elements in both ‘nominal’ and ‘verbal/clausal’ functional projections. We will
see that this is most probably not attested in any natural language. Interest-
ingly, however, we will argue that it appears that a slightly different state of

3
Again, these matters will be looked at in detail in Chapter 6.
 2.4 Identical (?) behaviours 29

affairs could well be instantiated in some languages – that is, that we should be
able to freely insert roots in both nominal and verbal lexical projections.
Summarizing, when we talk about the noun–verb distinction in a language
and, consequently, the universality of nouns and verbs, we may discuss three
very different empirical matters:
(10) Noun–verb distinction in a given language
a. Whether there is extensive nominalization of verbs (denominal verbs and
mixed projections) and extensive verbalization of nouns (deverbal nouns).
This is a red herring: verbs and nouns are real in this language, although
they can be morphosyntactically recategorized by a suitable functional
entourage.
b. Whether all roots can become both nouns and verbs. If yes, then
categorization is certainly a grammatical process: nouns and verbs are
grammatical constructs, syntactic categories, not specifications on lexical
items (i.e., word classes). Essentially, the noun–verb distinction still
stands.
c. Whether (9) holds – that is, whether all lexical elements can be used
interchangeably in all contexts. This would be an example of a language
where a noun–verb distinction, and lexical categorization, would be
pointless.

2.4 Identical (?) behaviours

Given the discussion in the previous section, the most radical challenge to the
universality of nouns and verbs and to the noun–verb distinction would come
from a natural language as described in (9), where any lexical element would
be freely inserted in any grammatical context. Such languages would undeni-
ably constitute genuine examples of the absence of a noun–verb distinction,
languages with no noun–verb distinction either as word classes or as syntactic
categories. Two languages have been famously claimed to fit the bill: Tagalog
and Riau Indonesian.
Tagalog has been variously presented as a language whose roots ‘are always
[category-]neutral or “precategorial” and receive categorial specification via
morphological specification’ (Rauh 2010, 343). Even if this were true, it would
not entail that Tagalog has no verb–noun distinction. It would only mean that
categorization in Tagalog would be a purely grammatical process and that there
would be no nouns and verbs as distinct word classes coming ‘pre-packaged as
nouns and verbs’ from the lexicon. In other words, Tagalog would be like
Nootka and Lillooet Salish (or St’át’imcets) (see Section 2.5 for details).
However, Tagalog is not even a language that illustrates the b. case in (10).
Himmelmann (2008), cited in Rauh (2010, 343–5), observes that in Tagalog,
30 Are word class categories universal?

as in Germanic languages like English or Dutch (Don 2004), some roots make
better nouns or only nouns. In Tagalog, the version of this is slightly different:
not all roots can be inserted in any morphological environment. Himmelmann
(2008) illustrates this by discussing ma–, a polysemous prefix. The marker
ma– expresses states (e.g., magandá ‘beautiful’ < gandá ‘beauty’) when
affixed to a specific set of roots; however, it expresses accomplishments
(e.g., magalit ‘become angry’ < galit ‘anger’) when affixed to a different set
of roots. So, magandá cannot mean ‘become beautiful’. The distinction
between the two classes of roots appears to be arbitrary – that is, it is not the
case that roots like gandá (‘beauty’) can only denote states. To wit, using a
different marker, the infix –um–, gandá may also express an accomplishment:
g-um-andá indeed means ‘become beautiful’. Finally, using a different mor-
phological process – stress shift – roots like galit can express states: galit
(‘anger’) versus galít (‘angry’). The above show that Tagalog roots cannot be
freely inserted in any grammatical context. But what about categories like
nouns and verbs? Is it meaningful to talk about such ‘Eurocentric’ (Gil 2000)
categories in Tagalog? The answer seems to be affirmative. Himmelmann
(2008), again as cited in Rauh (2010, 343–4), makes a very interesting point
with respect to what he calls ‘V-words’. Tagalog roots expressing things,
animate beings and actions are nouns when used in isolation, and they become
verbs, ‘V-words’, when they undergo Voice affixation. 4 Although it is unclear
whether the resulting words are simplex or denominal verbs, I think that
processes like Voice affixation make talking about a noun–verb distinction
in Tagalog meaningful. Baker (2003, 185–9) follows a very similar path to
show that a noun–verb distinction is meaningful in a number of languages,
including Tukang Besi – whose behaviour closely resembles that of Tagalog.
There exists, however, an example of a language that has been claimed to
exactly fit the bill in (9). Gil (1994, 2000, 2005, 2013) has consistently claimed
that Riau Indonesian is a case of a language as described under c. in (10): all
lexical material, all lexical elements, can be used interchangeably in all
contexts; they can be freely inserted pretty much everywhere.5 This purported
radical state of affairs would render the noun–verb distinction in the particular
language completely unnecessary and spurious, reducing it to a sort of analyt-
ical and methodological straightjacket. Consider the most famous example of
this lack of lexical categories, as presented in Gil (2005, 246):

4
The apparent ‘default’ nature of nouns will be examined in Chapter 4, Section 4.6.
5
See Gil (n.d.) for a thorough and well-rounded presentation of the Riau Indonesian variety from a
sociolinguistic viewpoint, a matter that has raised some controversy.
 2.4 Identical (?) behaviours 31

(11) makan ayam/ ayam makan

eat chicken chicken eat
‘The chicken is eating.’
‘The chickens are eating.’
‘A chicken is eating.’
‘The chicken was eating.’
‘The chicken will be eating.’
‘The chicken eats.’
‘The chicken has eaten.’
‘Someone is eating the chicken.’
‘Someone is eating for the chicken.’
‘Someone is eating with the chicken.’
‘The chicken that is eating.’
‘Where the chicken is eating.’
‘When the chicken is eating.’

Gil (2005) claims that in Riau Indonesian any of the glosses above is an
appropriate translation of the two-word sentence makan ayam/ayam makan –
depending on the context, of course. So, indeed, Riau Indonesian is presented
as a category-less language in the most thorough and dramatic fashion.6
Yoder (2010) put this claim to the test, going back to the original naturalistic
data in Gil’s publications, data originating from ‘recorded spontaneous speech’
(Yoder 2010, 2).7 He went on to review the 154 examples presented in Gil’s
published work on Riau Indonesian until 2010 with a very simple goal: to
detect whether there is any correlation between nouns, verbs and adjectives, on
the one hand, and the syntactic functions argument, predicate and modifier, on
the other (Yoder 2010: 7).
His methodology is simple but solid: departing from a comparative view-
point, he took all the words in Gil’s 154 examples and checked them against
the category to which these words belong to in Standard Indonesian, an
extremely similar and much better described variety, which does make cat-
egorial distinctions. Matching the Riau words to the Standard equivalents, he
‘found a total of 271 nouns, 161 verbs, and 55 adjectives’ (Yoder 2010: 7). He
then revisited each of the 154 examples and examined whether items

6
We must be very cautious with what could be called the ‘Hopi fallacy’ – that is, claims of
exceptionality stemming from an exoticized (re)presentation of a language – on behalf of field
linguists studying it. See, for instance, Nevins, Pesetsky and Rodrigues (2009) on the progres-
sively exoticized grammatical descriptions of Pirahã.
7
Gil does not seem to systematically ask for native speaker intuitions or to elicit responses. This
is, to my mind, the major methodological obstacle in substantiating claims of Riau Indonesian
being category-less.
32 Are word class categories universal?

categorized as nouns in Standard Indonesian functioned as arguments in Riau

Indonesian; whether items categorized as verbs in Standard Indonesian func-
tioned as Riau predicates; and whether adjectives in the Standard variety
functioned as Riau Indonesian modifiers. If a verb, noun or adjective in the
Standard variety had a different syntactic function in Riau Indonesian than the
default one, he counted it as an exception.
His findings are pretty solid, as summarized in the table in (12), adapted
from Yoder (2010: 7):
(12) Members of lexical categories and their functions in the Gil corpus
Default Non-default

noun (default ¼ argument) 244 (90%) 27 (10%)

verb (default ¼ predicate) 160 (99%) 1 (>1%)
adjective (default ¼ modifier) 27 (49%) 28 (51%)

Yoder proceeded to tackle the non-default cases and found that the majority
thereof fall under five morphosyntactic patterns, which are also attested in
Standard Indonesian: (i) nominal modifiers (belonging to all three categories)
following the noun they modify; (ii) equative clauses with a nominal or an
adjectival predicate; (iii) three instances of noun phrases headed by a phonolo-
gically null noun (like English four reds and a yellow); (iv) use of a verbalizing
prefix N-; (v) use of a verbalizing suffix –kan attaching on adjectives.8
In the light of the above, I think we can safely consider Riau Indonesian as a
run-of-the-mill language, at least as far as its lexical categories are concerned,
pending a more in-depth investigation of its grammar, based on more empirical
evidence, compounded by elicited responses and – crucially – speaker intu-
itions. Until then, Yoder (2010) makes a very convincing point that this variety
has nouns, verbs and adjectives both as syntactic categories and as word
classes, just like Standard Indonesian.

2.5 The Nootka debate (is probably pointless)

It has been extensively argued that there are languages where no distinction can
be made between verbs and nouns. The most famous ones are Nootka and
Lillooet Salish (also known as St’át’imcets), to the extent that the discussion on
whether nouns and verbs are universal is termed ‘the Nootka debate’. Very

8
The three functions that do not fall under either the default cases or the five non-default patterns
involve a noun in predicate position, a noun in verbal modifier position and a verb in argument
position. None of these is typologically outlandish, let alone unheard of.
 2.5 The Nootka debate (is probably pointless) 33

broadly speaking, most analyses arguing that languages like Nootka and
Lillooet Salish do not distinguish between nouns and verbs are usually informed
by the fact that lexical elements can all function as predicates in isolation and
they can be differentiated with respect to their syntactic role only via particular
functional elements. Evans and Levinson (2009, 434–5) eloquently summarize
the case for the non-existence of a noun–verb distinction as follows:
A feeling for what a language without a noun-verb distinction is like comes
from Straits Salish. Here, on the analysis by Jelinek (1995), all major-class
lexical items simply function as predicates, of the type ‘run,’ ‘be_big,’ or
‘be_a_man.’ They then slot into various clausal roles, such as argument (‘the
one such that he runs’), predicate (‘run[s]’), and modifier (‘the running
[one]’), according to the syntactic slots they are placed in. The single open
syntactic class of predicate includes words for events, entities, and qualities.
When used directly as predicates, all appear in clause-initial position,
followed by subject and/or object clitics. When used as arguments, all lexical
stems are effectively converted into relative clauses through the use of a
determiner, which must be employed whether the predicate-word refers to an
event (‘the [ones who] sing’), an entity (‘the [one which is a] fish’), or even a
proper name (‘the [one which] is Eloise’). The square-bracketed material
shows what we need to add to the English translation to convert the reading in
the way the Straits Salish structure lays out.

There are two matters to review in the summary above, which echoes Jelinek’s
extensive work on Salish.
First, we have the fact that in most Salish languages arguments need a
Determiner, as illustrated in (16) for Lillooet Salish and in (17) for Nootka.9
This holds whether they are doubled by argument clitics on the predicate or
not – that is, by whether they are true arguments or DP adjuncts (Jelinek 1996).
However, does this necessarily entail that ‘when used as arguments, all lexical
stems are effectively converted into relative clauses through the use of a
determiner’?10 Should this be so, then it would be a strong argument for all
lexical stems in Salish being inherently predicative. Not necessarily, however.
In Romance, Determiners are generally necessary for nouns to function as
arguments and predicate nouns may surface without a Determiner – this is
what happens in Salish, too.11 Moreover, we do not know what lexical stems

9
In Makah, nouns like wa:q’it (‘frog’) can function as arguments with or without a Determiner
suffix –oiq; however, verbs must be suffixed with –oiq in order to do so (Baker 2003, 179).
10
Emphasis added.
11
Of course, Salish is not Romance! All that is highlighted here is the obvious fact, also reviewed
in Chierchia (1998) and elsewhere, that obligatory Determiners in argument positions, even
with proper names, cannot be taken as direct evidence for the inexistence of nouns.
34 Are word class categories universal?

surfacing as nouns in other languages would do in ‘other’ Salish environments,

because Salish seems to allow DPs only in adjunct positions. This restriction of
Salish DPs to adjunct positions has profound consequences like the lack of wh-
movement and A-dependencies (Jelinek 1996); it also forces these languages
to resort to non-DP mechanisms of quantification (Matthewson 2001). At the
end of the day, the in the dog is just a Determiner; in *(the) Paris of my youth it
introduces something like a reduced relative – Salish could be a little like that,
as well.
The second matter is whether what Evans and Levinson (2009) call ‘major-
class lexical items’ is actually anyhing more than bare roots. Should this be so,
then Salish would eventually provide strong evidence for acategorial roots
being categorized in syntax by the structure dominating them; in other words,
we would have evidence for Distributed Morphology and Borer-style syntactic
categorization: for the syntactic decomposition of nouns and verbs (see Chap-
ter 3). In this case, Lillooet Salish would be a prominent example of a language
where any root can be freely inserted inside any lexical environment to become
a noun, a verb and so on – that is, a language falling under the b. case of (10).
Harking back to the caveats raised in Section 2.3.2 and the working
principle in (9), it is precarious to assume that if a lexical element X takes
the same type of affix as lexical item Y this makes X and Y identical. Again
turning to languages more familiar for sociocultural and historical reasons, this
point is nicely illustrated by the following situation in Turkish. Although few,
if any, would argue that Turkish does not distinguish between nouns and verbs,
nevertheless the suffix –dI can be found after verbal elements as well as after
nominal ones:
(13) muhtar-dı bayıl-dı
village headman-Past faint-Past
‘S/he was village headman.’ ‘S/he fainted.’

However, this is possibly a case of allomorphy of sorts. First, as evident from

the glosses, the interpretation of the noun above is distinctly different to that of
a verb: the suffix –dI attaching to a noun functions as a past tense copula, but
when it attaches to a noun it is simply a past tense marker. Second, stress
patterns differ: the suffix –dI is stressed (marked in boldface above) only when
attached to a verb. Finally, the picture becomes very different once we move
away from the past tense:
(14) Copula versus Tense in Turkish
a. muhtar-dır bayıl-ıyor
village headman-Cop faint-Present
 2.5 The Nootka debate (is probably pointless) 35

b. muhtar ol-acak bayıl-acak

village headman be.fut faint-Fut

Once we look at the present and the future tenses, it becomes obvious that the
suffix –dI is the exponence of two different elements: of a past tense copula
and of past tense. The conclusion is that we must be very careful before
making claims of non-distinctness, given that the morphological exponence
of functional categories within which nouns and verbs are embedded can
mislead us. As repeatedly pointed out above, the essential task is threefold:
first, to ensure that homophony of affixes is just that – a matter of exponence
and forms; second, to correctly distinguish nouns and verbs from their func-
tional entourage; and, third, to correctly tell whether the root embedded within
the various functional structures is a categorized root (i.e., lexical noun or
verb) or an uncategorized root.
Let us, however, turn to some real data from Lillooet Salish and Nootka
(Baker 2003, 173–89). Lillooet Salish can use any lexical category as a
predicate, something to be expected if all lexical categories are indeed
predicates (Higginbotham 1985), using the same (?) agreement-like
suffix:12
(15) Lillooet Salish: predicates
a. Qwatsáts-kacw
leave-2ndsg
‘You leave/left.’
b. Smúlhats-kacw
woman-2ndsg
‘You are a woman.’
c. Xzúm-lhkacw
big-2ndsg
‘You are big.’

So far, the situation could be similar to what happens with Turkish in (13).
Nevertheless, as already mentioned, in Salish languages there is a determiner-
like element that can nominalize anything, while whatever appears in the first
position of a sentence can act as a predicate (Baker 2003, 175).
(16) Lillooet Salish: arguments
a. Qwatsáts-Ø ti smúlhats-a
leave-3rdabs the woman-the
‘The woman left.’

12
The examples are from Baker (2003, 175).
36 Are word class categories universal?

b. Smúlhats-Ø ti qwatsáts-a
woman-3rdabs the leave-the
‘The one who left is a woman.’

The picture is similar in Nootka:

(17) Nootka: arguments
a. Mamu:k-ma qu:?as-?i
work-indic man-the
‘The man is working.’
b. Qu:?as-ma mamu:k-i
man-indic work-the
‘The working one is a man.’

What is worth observing above is that in (16), the predicate – that is, the first
word in the sentence – takes a null subject agreement marker (third-person
absolutive) that links it with a constituent nominalized by a determiner(-like
element). Once more, the situation in (13) springs to mind, where –dI can be
either a past tense marker (with the verbal stem bayıl) or a past tense copula
(with the noun muhtar). Nootka in (17) is, on the other hand, much more
interesting, as the suffix attaching on both ‘work’ and ‘man’ in the first
position is one encoding indicative Mood.
Abstracting away, the workings of roots illustrated above could in reality
be no different to the ones in English: the category of lexical elements
becomes visible through the morphology surrounding the roots and by
these elements’ syntactic position. If one goes for the view that some roots
can make only nouns or only verbs, the true question here is whether there
are Salish roots that cannot be made into first-position predicates, as, for
example, boy and – for most speakers – cat cannot be made into verbs in
English. Baker (2003, 177 et seq.) proceeds to ‘isolat[e] the lexical heads
from their functional support systems, to see if the noun-nonnoun contrast
reemerges in those environments’. Revisiting candidate languages for a
lack of noun–verb distinction, he shows that in Greenlandic and Nahuatl
only nouns incorporate, despite the wide use of substantivized adjectives
as nouns in other contexts; that in Kambera, Samoan, Tongan and
Niuean (Austronesian) only true nouns can appear as arguments without
a determiner (see also footnote 9). Revisiting Lillooet Salish, Baker
(2003, 182) follows Davis (1999) in foregrounding a very important
observation: while the determiner ti–a can nominalize anything, as shown
in (16), only true nominals can take the demonstrative ti7 plus the deter-
miner ku:
 2.6 Verbs: everywhere, not always as a word class 37

(18) a. Áts’x-en¼lhkan ti7 ku-sqaycw

see-directional¼1 stsg.subject dem det-man
‘I saw that man.’
b. *Áts’x-en¼lhkan ti7 ku-qwatsáts / ku-tayt
see-directional¼1 stsg.subject dem det-leave / det-hungry
‘I saw that leaving one/that hungry one.’

So, even in Lillooet Salish there are lexical elements that display an unam-
biguously nominal behaviour. This confirms our suspicions that in the b.
example of (16) there is more structure than meets the eye; after all, subjects
like ti qwatsáts-a a (det leave det) ‘the one leaving’ has been standardly
analysed as a reduced relative.13

2.6 Verbs can be found everywhere, but not necessarily

as a word class
Let us now turn to verbs. As will be extensively shown in the following two
chapters, verbs are grammatically more complex than nouns, irrespective of
whether one subscribes to the view that argument structure is syntactic struc-
ture: see Hale and Keyser (1993, 2002), Pylkkänen (2008), Ramchand (2008)
and elsewhere. This pre-theoretical observation – namely, that verbs have
more structure than nouns – apparently holds even if we abstract away from
periphrastic forms that include more than just the lexical verb – recall the
discussion of (1) – or affixes belonging to the functional entourage of the verb,
as in (2). Beginning with Germanic, particle verbs are very common and
particles are an integral part of the verbal predicate (Harley 2005b). Thus,
the verb cook displays different selectional restrictions to the particle verb cook
up (Basilico 2008):
(19) a. the criminals cooked a meal/#an evil scheme
b. the criminals cooked up an evil scheme

Complex verbs of a different type are Light Verb Constructions, in languages

like Japanese, where verbal predicates are expressed by the combination of a
light verb suru with an appropriate predicate. In Iranian languages, the
picture is similar, with two additional factors that make them particularly
interesting. Using Farsi to illustrate the above points, there are very few
lexical verbs; Complex Predicates (to be discussed in Chapter 4) express all

13
Or, perhaps, a mixed projection (see Chapter 6).
38 Are word class categories universal?

other verbal concepts, via the combination of a non-verbal element (also

known as ‘preverb’) and one of the following fourteen light verbs, from
Family (2008):
(20) Farsi light verbs Ð
kærdæn
Ð ‘do’ ke idæn ‘pull’
odæn ‘become’ amædæn ‘come’
zædæn ‘hit’ aværdæn ‘bring’
xordæn ‘eat’ bordæn ‘take’
dadæn ‘give’ ræftæn ‘go’
gereftæn
Ð ‘get’ ændaxtæn ‘throw’
da tæn ‘have’ oftadæn ‘fall’

The most exciting case of a language with complex verbs would be Jingulu,
an Australian language discussed in Pensalfini (1997) and Baker (2003,
90–4). Jingulu possesses only three verbal elements – ‘come’, ‘go’ and
‘do/be’ – and there are no (other) lexical verbs, unlike Japanese and Farsi,
which do possess a few lexical verbs. ‘Come’, ‘go’ and ‘do/be’ in Jingulu
may stand as the only verbal predicates in a sentence and they are strictly
obligatory; additionally, they are the only necessary elements in a sentence.
In order for other verbal concepts to be expressed, one of these three
elements must be combined with what looks like a bare root. Here are some
examples from Baker (2003, 90–1):
(21) Jingulu verbs
a. Ya-angku
3rdsg-will.come
‘He will come.’
b. Jirrkiji-mindu-wa
run-1stincl.du-will.go
‘You and me will run off.’
c. Ngaruk baka-nga-rriyi
dive 1stsg-will.go
‘I’ll dive down.’
d. Ngaruk baka-ngayi arduku
dive 1stsg-will.do carefully
‘I’ll submerge (something) carefully.’

Pensalfini (1997) argues that ‘come’, ‘go’ and ‘do/be’ are light verbs.
I understand that ‘light verbs’ in this instance can possibly be equated to
semi-lexical verbs (see Chapter 4). More concretely, they could be the lexi-
calization of verbalizers themselves, in the manner of Folli and Harley (2005),
who claim that three flavours of v exist, vCAUS, vDO, and vBECOME, each
 2.6 Verbs: everywhere, not always as a word class 39

making different types of verbs.14 This receives support by the differences in

transitivity between examples c. and d. in (21), induced by the choice of ‘go’
versus ‘do’, the latter inducing transitivity. Baker (2003, 91–4), on the con-
trary, takes ‘come’, ‘go’ and ‘do/be’ to be inflected auxiliary verbs (like have
and be in English) with roots like ngaruk (‘dive’) in (21) to be the actual verbs
themselves. However, I am not convinced that something like ngaruk (‘dive’)
in (21) is a verb instead of an acategorial root. Baker’s (2003, 91–4) principal
argument for the verbal nature of Jingulu roots like ngaruk (‘dive’) is that
‘come’, ‘go’ and ‘do/be’ do not necessarily affect transitivity and thematic
properties and that the root, rather than ‘come’, ‘go’ and ‘do/be’, determines
the verbal predicate’s theta grid. This kind of debate is familiar from discus-
sions on Farsi. The opening lines of Folli, Harley and Karimi (2003) are
particularly enlightening about how to deal with the question of what part of
the complex verb determines what:
It has been argued in the literature that the argument and event structures of
Persian complex predicates (CPr), as well as syntactic properties such as
control, cannot be simply derived from the lexical specifications of the
nonverbal element (NV) or the light verb (LV) . . . In this paper, we show
that the event structure of LV is not always the same as the event structure of
its heavy counterpart. Furthermore, although LV determines the agentivity
(xordan ‘collide’ versus zadan ‘hit’) and the eventiveness of the CPr, it fails
to completely determine the event structure and the telicity of the CPr. Thus,
depending on the NV element, the same LV may occur in different types of
event structure. (Folli, Harley, and Karimi 2003, 100)

Appealing to the Farsi case is valuable, because there is consensus that its
complex predicates are made of light verbs and a non-verbal element. Neverthe-
less, both elements contribute to the properties of the overall complex predicate.
In other words, although Jingulu roots affect the type of the complex predicate’s
thematic grid, this is no evidence that these roots are verbal. I therefore believe
we can adhere to an analysis in the spirit of Pensalfini (1997); after all, even in
inflected languages with solid lexical verb paradigms, like Greek, verbs may
display an articulated structure in which the verbal or verbalizing element is
kept distinct from the acategorial root material (Panagiotidis, Revithiadou and
Spyropoulos 2013). At the same time, languages like Jingulu, Japanese and
Farsi and many others lend support to the syntactic decomposition approach to
lexical categories, which will be described and presented in Chapter 3.

14
More precisely, Jingulu ‘come’, ‘go’ and ‘do/be’ would be the exponence of verbalizers fused
with Tense heads, as glossed in (21).
40 Are word class categories universal?

2.7 An interim summary: verbs, nouns, roots

After the extensive overview of different flavours of the noun–verb distinction
cross-linguistically, we can now revisit the hypotheses laid out in (10).15
Tagalog and Riau Indonesian do not seem to constitute cases of grammars
where all lexical elements can be used interchangeably in all contexts. Lan-
guages claimed to possess a single undifferentiated lexical category and no
noun–verb distinction cannot be claimed to have been discovered, at least
not yet.
Nootka (pending more in-depth investigation) seems to be a language where
all roots can be slotted in all category-defining syntactic contexts. In languages
like this, evidence is very strong for roots being categorized by their syntactic
context, and it makes no sense talking about nouns and verbs in Nootka as
distinct word classes: lexical categories are syntactic categories. At this point it
is worth remembering Baker’s finding that where there is extensive nominal-
ization of verbs (yielding deverbal nouns and mixed projections) and extensive
verbalization of nouns (yielding denominal verbs), the verb–noun distinction
may appear blurred. Still, this is a red herring: verbs and nouns are real in this
language, although they are easy to morphosyntactically recategorize with the
mediation of a suitable functional entourage.
In other Salish languages, and many more languages, some roots make
better nouns than verbs (the reverse does not seem to hold – see also
Chapter 3). Consider the Lillooet Salish example in (18), where ‘man’ is only
a noun, and English roots like boy, ball and (perhaps) cat, all of which give
nouns but no underived verbs, or, at least, not very good ones. These are
empirical facts to be explained, possibly invoking grammar-external factors.
What about languages where nouns and verbs appear to form solid word
classes? One possibility to explore is the following: in grammars where nouns
and verbs form solid word classes, this is an illusion of rich morphology. This
could be especially true in fusional languages like Romance. Consequently, it
is not conceptually necessary for nouns and verbs to belong to different word
classes if categorization is a syntactic process. More precisely, word classes
like noun and verb could be the result of syntactic derivations and morpho-
logical processes and not necessarily constitute two classes of words distin-
guished by the root that these words are built around. We will look at these
topics in detail in the next chapter.

15
The interested reader is referred to the works cited in Baker’s (2003) survey for further details
and a more in-depth scrutiny of the matter.
 2.8 What about adjectives (and adverbs)? 41

2.8 What about adjectives (and adverbs)?

This monograph has so far exclusively discussed nouns and verbs and will
continue to do so. Hence, the obvious category missing from the picture is
adjectives. The question at this point is what adjectives are and why they are
excluded from the present discussion.
I believe that adjectives are not a basic or fundamental word class, as I claim
nouns and verbs to be. However, we have not looked at the mechanics of
categorization yet, something we will do in Chapter 3. Therefore, in this
section I will restrict myself to empirical and quasi-theoretical arguments in
order to claim something less ambitious: first, that adjectives do not fall into
the same categorial type as nouns and verbs; second, that adjectives are neither
the unmarked category nor universal. I will tentatively conclude that adjectives
involve more structure than nouns and verbs and that this structure is of quite a
different nature to that ‘making’ nouns and verbs

2.8.1 Adjectives are unlike nouns and verbs

The idea that adjectives are not of the same ilk as nouns and verbs already exists
as a sort of tacit understanding in a large part of the literature. This understand-
ing stems from a number of descriptions and analyses, according to which there
are languages that actually lack the category of adjective, with Dixon (1982)
and Schachter (1985) figuring most prominently among them. In the generative
literature, Amritavalli and Jayaseelan (2004, 19–20, 29–31) discuss the non-
existence of adjectives in Kannada and Malayalam; Hale and Keyser (2002,
13–14) mention that Navajo ‘adjectives’ are verbs, whereas those in Warlpiri
are nouns. Notably, in Baker (2003, 238–63), to which we will return, a whole
section is dedicated to arguing for the universality of adjectives.
But before looking into the (non-)universality of adjectives, let us first
review evidence with respect to a more modest goal – namely, showing that
adjectives are qualitatively different from nouns and verbs, that they do not
belong to the same class. Surprisingly, given his commitment to the opposite
thesis, some grounds indicating this difference can be found in Baker (2003,
chap. 4). In this chapter of his Lexical Categories monograph, Baker boldly
argues that adjectives are the unmarked lexical category – that is, the word
class category used whenever nouns and verbs cannot be used. As reviewed in
the previous chapter, Baker (2003) considers nouns to bear referential indices,
making them suitable to stand as arguments, and verbs as the only lexical
categories that can stand as predicates by themselves – that is, without the
mediation of a functional predicator. In this respect, adjectives, being
42 Are word class categories universal?

unmarked, are restricted to two positions: (i) predicate positions, with the
mediation of a functional predicator Pred, just like nouns and other syntactic
constituents, yielding examples like The flag is red; I painted the wall green,
and (ii) modifying positions: The red flag; A house with green walls.
In order to formalize adjectives as the unmarked lexical category, Baker
(2003, chap. 4) is restricted by his two-membered privative feature system
(i.e., [N] and [V]), so he has to argue that adjectives do not bear any categorial
specification – that is, no categorial feature whatsoever. This claim has at least
two consequences:
a. They have no ‘special characteristics’ of their own as a category
(Baker 2003, chap. 4), unlike nouns and their relationship with
determiners and ‘counting’ or verbs and their argument structure
and their relation with tense;16
b. As practically admitted in Baker (2003, chap. 5), adjectives are
essentially identical to roots: their categorial specification is precisely
the lack of any categorial specification, if category is encoded by
features like [N] and [V].
The above entail at least that adjectives are different from nouns and verbs. Of
course, in Baker’s theory they are different because they are the unmarked,
basic lexical category. Even if this basic intuition is wrong, as I will propose,
the fact remains that adjectives are different: either because they are catego-
rially featureless roots (which I will disagree with) or because they are not a
fundamental and universal lexical category like nouns and verbs (which seems
to be going in the right direction).

2.8.2 Adjectives are not unmarked

Baker makes strong claims about adjectives which, as I postulated above, set
them apart from the other two categories, nouns and verbs, one way or another.
However, can it be the case that adjectives are unmarked, actually identical to
free roots – despite the claim in Baker (2003, 269) that free uncategorized roots
do not exist?17
Baker’s claims, as we have seen, have two consequences: that adjectives are
the unmarked lexical category and that they are identical to roots. Baker (2003,

16
This decision is of course reminiscent of Déchaine’s (1993) treatment of prepositions. See
Section 1.3 in Chapter 1.
17
After all, what would categorize a root into an adjective if no categorial features for
adjectives exist?
 2.8 What about adjectives (and adverbs)? 43

230) calls them the ‘elsewhere case’ among lexical categories. Now, if
adjectives are indeed the unmarked lexical category, expressing, say, pure
properties, then we would expect
a. adjectives to be universal: if categorially unmarked roots surface as
the category adjective, then all languages should possess this
category, even more so than verbs: after all, in Baker’s system, a
language without [V] on lexical elements could potentially exist,
having all predication mediated by the functional predicator Pred;
b. adjectives not to have any ‘special’ characteristics of their own
whatsoever.
However, a lot of work on adjectives seems to be pointing in the opposite
direction and evidence that adjectives are not a universal category is abundant.
Besides those discussed in Dixon (1982), Schachter (1985), Hale and Keyser
(2002, 13–14) and Amritavalli and Jayaseelan (2004), more languages seem
not to possess an adjective category, let alone one that is the elsewhere case.
If adjectives are indeed the unmarked category, what are we to make of
Chichewa which has six adjectives (Baker 2003, 246–9) or of Hausa which
is reported to possess no more than a dozen? And there also exist languages
without an adjective category; in these languages, three scenarios are typically
attested.
First, in languages like Japanese, the ‘adjective category’ is split in two
(Miyagawa 1987) and adjective-like words behave as if they belong to two
distinct classes with precious little in common. See also the discussion of
Japanese ‘adjectival nouns’ in Iwasaki (1999, chap. 4).
In Korean (and Navajo) adjectives behave as a subclass of the verb category.
Citing Haspelmath (2001, 16542), ‘in Korean, property concepts inflect for
tense and mood like verbs in predication structures [see (22) below], and they
require a relative suffix . . . when they modify a noun [see (23) below], again
like verbs’.18 The examples below (Haspelmath 2001, 16542) illustrate the
situation:
(22) a. salam-i mek-ess-ta
person-nom eat-past-declarative
‘The person ate.’

18
It is true that Korean property verbs do not take the present-tense suffix –nun, but I am not sure
this is enough to make them a lexical category on a par with nouns and verbs, as opposed to a
subclass of verbs or a verbal category.
44 Are word class categories universal?

b. san-i noph-ess-ta
hill-nom high-past-declarative
‘The hill was high.’

(23) a. mek-un salam

eat-relative person
‘A person who ate.’
b. noph-un san
high-relative hill
‘A high hill.’

Turning to Dravidian languages, they are also analysed as not possessing

adjectives distinct from nouns incorporated into Kase, which typically surfaces
as dative (Amritavalli and Jayaseelan 2004, 19–20, 29–31); this is illustrated
below for Kannada and Malayalam:19
(24) a. raama udda-kke idd-aane Kannada
Rama.nom height-dat be-3sg
b. raaman-@ uyaram uND@ Malayalam
Raman- dat height is
‘Rama(n) is tall.’

Chichewa also seems to overwhelmingly use nouns in the way other languages
use adjectives.
Turning to languages with a distinct adjectival category that constitutes
an open class, this does not possess any of the hallmarks of an unmarked
lexical category, let alone of a category that is posited to be the direct
manifestation of a root without any categorial features. In Romance,
Germanic and Greek a large number of adjectives, if not the majority, are
derived from nouns and verbs, as opposed to being derived directly from
roots. A good case in point is that of Slavic, which makes extensive and
productive derivation of adjectives via suffixes for a number of purposes –
for example, for the resulting adjectives to function as possessives in lieu of
simple genitives. This is elegantly exemplified in Russian, as described in
Valgina, Rosental and Fomina (2002, sec. 145), which the review below
follows closely.20

19
Both examples are from Amritavalli and Jayaseelan (2004, 29), who also make the conjecture
that ‘there is . . . an implicational relationship between the absence of adjectives and the
prevalence of the dative experiencer construction’. As in Chichewa, there are in Kannada ‘a
few indisputable underived adjectives, such as oLLeya “good”’.
20
I am grateful to Svetlana Karpava and Olga Kvasova for discussing the Russian facts with me
and for lending me their native speaker intuitions. Karpava kindly translated Valgina, Rosental
 2.8 What about adjectives (and adverbs)? 45

Russian has ‘possessive adjectives’ which indicate a possessor; ‘typically,

possessive adjectives are formed from animate nouns, with the help of the
suffixes –in, –nin, –n-ij, –ov, –ev, –sk-ij’ (Valgina, Rosental and Fomina 2002,
sec. 145):
(25) Some Russian possessive adjectives
Noun Possessive adjective Meaning

Liza Lizin Liza’s

brat bratnin brother’s
doch dochernin daughter’s
otec otcov father’s
Vladislav Vladislavlev Vladislav’s

What is important here is that these adjectives indicate referential possessors,

not properties. In other words, ‘Ljuba’s book’ can be expressed either with a
genitive possessor, as kniga Ljubij (book Ljuba.gen), or with a possessive
adjective derived from the name ‘Ljuba’: ljubina kniga (Svetlana Karpava,
personal communication, April 2013). In a restriction reminiscent of those by
which Saxon genitives in English abide, Russian possessive adjectives ‘that
are formed from inanimate nouns are very rare’ (Valgina, Rosental and Fomina
2002, sec. 145). Moreover, they are better with some first names, odd or
awkward with others, impossible with non-Russian names or surnames – and
so on (Svetlana Karpava, personal communication, April 2013; Olga Kvasova,
personal communication, April 2013). Finally, adjectives can be ambiguous
between a possessive and an attributive reading.
(26) Ambiguous interpretation
volchij hvost volchij appetit
wolf.adj tail wolf.adj appetite
‘The/a wolf’s tail’ ‘Wolfish appetite’
(cf. ‘Hungry like a wolf’)

Generally speaking, it seems that adjectives in most languages involve more

rather than fewer structural layers. For instance, in Romance, Greek and Slavic
modifying (and also predicative) adjectives are obligatorily embedded within
the functional structures that trigger concord and nominal agreement morph-
ology. What is also worth noting here is that the other elements, besides

and Fomina (2002, sec. 145), from which examples are drawn, as well as from Karpava
(personal communication, April 2013) and from Kvasova (personal communication,
April 2013).
46 Are word class categories universal?

adjectives, that display concord with the noun inside a DP are all functional
elements: quantifiers, demonstratives, possessives and articles. This is starkly
illustrated in Modern Greek.
(27) Concord
a. olus aftus tus neus typus
all.m.acc.pl these.m.acc.pl the.m.acc.pl new.m.acc.pl type.m.acc.pl
‘All these new types.’
b. oli afti i nea sodia
all.f.acc.sg this.f.acc.sg the.f.acc.sg new.f.acc.sg crop.f.acc.sg
‘All of this new harvest.’

Adjectives inside a DP are both dependent, agreement-wise, on nouns and

actually often resemble quantifiers, demonstratives and possessives. I think
that the above indicates that even in languages where adjectives look like a
bona fide lexical category, they are indeed very hard to be taken to belong to
the same type of lexical category as nouns and verbs.
Returning to the matter of the morphological complexity of adjectives
themselves, naturally, morphological complexity on its own does not neces-
sarily entail that adjectives as a category are non-basic: after all, deriving nouns
from verbs (e.g., place-ment) and adjectives (e.g., red-ness) and deriving verbs
from nouns (e.g., en-shrine) and adjectives (e.g., trivial-ize) is standard in a
large number of languages, without this making ‘noun’ and ‘verb’ non-basic.
However, there are two important consequences of morphological complexity
when it comes to adjectives.
First, and as mentioned above, very few adjectives in Romance, Germanic
and Greek are underived, leading us to the implication that Adjective is
definitely not the unmarked category, especially one co-extensive with roots.
Second, and far more problematic for Baker’s (2003) approach: if adjectives
are not marked for any categorial feature whatsoever, how are adjective-
making derivational affixes marked?21 For instance, if –y attaches to nouns
to make them adjectives (e.g., cloud-y, milk-y, water-y etc.), what categorial
feature is borne by –y? If none, how does the category change from noun
(marked with [N]) to adjective? I think this is a very serious problem for any
approach that takes adjectives to be categorial feature-free and, by extension,
to approaches arguing adjectives to be the unmarked lexical category, of the
same ilk as nouns and verbs.

21
This is precisely Déchaine’s (1993) argument for treating prepositions as not encoding any
categorial features; see Section 2.9.
 2.8 What about adjectives (and adverbs)? 47

On top of their customary morphological ‘heaviness’, adjectives are also

hardly unmarked in all other ways. Crucially, adjectives do seem to have
distinctive characteristics as a category, but they look as if they possess their
own functional category – namely, Degree. Work by Corver (1997) and
Neeleman, van de Koot and Doetjes (2004) is particularly enlightening in this
respect. Neeleman, van de Koot and Doetjes (2004) claim that English
possesses two classes of degree expressions. Class 1 degree expressions are
specific to adjectives:
(28) Class 1 Degree expressions

Class 1 Degree expressions (very, as, too, that, how) c-select an Adjectival
Phrase (AP), which they must precede: they take the AP as a complement. This
suggests that Class 1 Degree heads are members of the adjectival projection
line, possibly of the so-called Extended Projection of A.22 Class 1 Degree
expressions need much (a dummy A) in order to combine with non-APs: [[DegP
Deg much] [XP]]; they tolerate neither topicalization nor freer linearization
options – something to be expected if they take APs as complements qua heads
which are members of the projection line of A. Class 1 Degree expressions can
be claimed to be to the category A what D is to N and T to V.
In contrast, Class 2 Degree expressions (more, less, enough, a little, a good
deal) have no c-selectional requirements and they also combine with Prepos-
ition Phrases (PPs), Verb Phrases (VPs) and so on, apparently as adjunct
modifiers:
(29) Class 2 Degree expressions

Class 2 Degree expressions, being phrasal adjuncts, may possess internal

structure and they do not need any gradable predicate in order to modify the
XP they are adjoined to. As we might expect, topicalization of Class 2 Degree
expressions is possible, as well as their freer linearization.
A final point is this: basic nouns seem to be relatively simplex in structure.
Verbs are usually not, but verbal predicates can be expressed – in languages
like Japanese, Farsi and even Jingulu – by using a light verb plus something

22
Grimshaw (1991, 2003).
48 Are word class categories universal?

else. On the contrary, light adjectives, making adjectival predicates and/or

modifiers, do not exist. I think that this additional fact also seriously points to
adjectives being not of the same ilk as nouns and verbs.
Concluding this short preliminary discussion on adjectives: they do not
seem to be either identical to roots or universal (as plain roots, at least); they
tend to be morphologically complex and they possess their own dedicated
functional head, (Class 1) Degree, as part of their projection line.23 When they
appear as modifiers, they undergo concord precisely like functional elements
such as quantifiers, demonstratives, possessives and articles. In any event, they
stand apart from N and V.

2.8.3 Adverbs are not a simplex category

If adjectives are not of the noun and verb ilk, say because they are simplex
uncategorized roots or because they are parts of speech complexly derived
from nouns and verbs with the mediation of functional structure, it goes
without saying that adverbs are certainly not candidates for a basic lexical
category either. Of course, most languages do not visibly distinguish between
adjectives and adverbs. In any case, the answer to what adverbs are depends on
our story for adjectives, as even languages that distinguish adverbs from
adjectives allow both to be modified by Class 1 Degree expressions – allegedly
heads of the adjectival projection line/Extended Projection. This is acknow-
ledged by Baker (2003, 230–7), who takes adverbs to be adjectives incorpor-
ated into an abstract noun, one whose exponence in English is the suffix –ly.24
Recall also the well-known fact from Romance that adjectives becoming
adverbs are in the feminine, as if the adverbial suffix –ment(e) is (still) a
feminine noun with which they agree: for example, Spanish guapa-mente not
*guapo-mente. In other words, the morphologically simplex and apparently
basic character of adverbs like well should not misguide us to consider the
whole category on a par with nouns and verbs. Again, the fact that in some
Indo-European languages adverbs are a word class tells us something about
how Morphology works in these languages and not about adverbs as an
independent lexical category. This is a lesson we can extend to adjectives
and, almost certainly, to adpositions.

23
The view that Class 1 Degree expressions are specific to adjectives will be qualified and revised
in Chapter 5.
24
Corver (2005), more reasonably in my view, takes –ly to be a copular element. This perhaps ties
in nicely with the Davidsonian insights in Larson (1998).
 2.9 The trouble with adpositions 49

2.9 The trouble with adpositions

‘Adposition’ is the cover term used to include prepositions and postpositions
(typically in head-final languages like Korean, Turkish etc.). Adpositions are
absent from the original feature system in Chomsky (1970) and were added later,
to form the so-called ‘Amherst system’, filling in the [N][V] gap in it (see
Chapter 1). Adpositions have been the focus of a lot of work and of varying and
broad-ranging analyses. Besides having been treated as lexical items (Déchaine
1993), they have been assimilated to Complementizers (Emonds 1985, 156–7) or
to instances of Kase (Baker 2003, 303–25). Other detailed treatments argue for
the Adposition category to involve complex structure (actually, to be complex
structures), akin to argument structure, as in Svenonius (2007, 2008) and, in a
different way, Botwinik-Rotem and Terzi (2008) and Terzi (2010).
Although in most treatments adpositions are usually understood either as a
lexical category or as functional elements, there are three reasonable ways to
go. The first is to split them into two categories, a functional one and a lexical
one; the second is to treat them as a hybrid category; and the third is to
hypothesize that they are phrasally complex structures, involving both func-
tional elements and lexical, possibly nominal, material.
As far as splitting the ‘category’ Adposition in two, looking at English data
we can easily consider elements like on, at, to, in, of, by and so on to behave
like functional elements. Perhaps equally easily, over, above, front, behind,
while, during and so on can be understood to behave like lexical elements.
An approach acknowledging the Janus-like behaviour of adpositions, based
on their hybrid behaviour in West Germanic, is that of van Riemsdijk (1998a,
31), where he is led to propose that (i) ‘adpositions be characterized as
[N, V]’, as in the Amherst system, while (ii) ‘they should . . . be considered
(extended) projections of nouns, at least when they are transitive’; (iii) the
possibility that ‘there are also functional prepositions remain[s] fully valid’,
but (iv) in many cases adpositions are semi-lexical heads.
The complex structure solution consists of attributing each of the distinct
properties of adpositions to a separate functional or lexical component partici-
pating in a complex adpositional structure. Such a solution is, for instance, the
only one that makes sense in the case of Greek ‘complex prepositions’, where
lexical and functional components thereof and the restrictions ruling their
relative positions are plainly visible: see Botwinik-Rotem and Terzi (2008)
and Terzi (2010).
However, there exists an added dimension of complexity – namely, that the
selfsame form can sometimes act as functional and sometimes as lexical.
50 Are word class categories universal?

Consider the following examples from Panagiotidis (2003a), where the form of
can behave both as lexical (a. examples) and as functional (b. examples):
(30) a. My favourite picture is [of the Vice-Chancellor].
b. * My favourite student is [of Chemistry].

(31) a. [A photo _] was found [of the Vice-Chancellor drinking absinthe].

b. *[A student _] was jailed [of Chemistry].

(32) a. A (*Vice-Chancellor) picture (of the Vice Chancellor).

b. A (Chemistry) student (of Chemistry).

Lexical of behaves like a θ-assigning lexical head and its phrasal projection
does not have to be adjacent to the N, as shown in (30)a.; it can also extrapose,
as in (31)a.25 The reverse is true for functional of and its projection: its
licensing depends on whether the N student in the b. examples can assign a
θ role to it, as the alternation between Chemistry student and student of
Chemistry in (32) also suggests. It is unclear how this state of affairs can be
captured in a satisfactory and non-stipulative way in an analysis of adpositions
as complex structures. However, both the solution of splitting the category
Adposition in two and the one assuming adpositions to be a hybrid category in
the best of cases give the impression of merely restating the facts at a more
abstract level.
Déchaine (1993, 32–6), on her way to arguing for adpositions as the
elsewhere lexical category, observes that there is no English compound headed
by an adposition. So, although adpositions (or, perhaps, particles) may be part
of a compound like up-root, under-dog or over-cast, they can never be the
head of a compound, unlike nouns, verbs and adjectives. Her second general-
ization is twofold: on the one hand, there is no affix that derives adpositions;
on the other hand, very few derivational affixes attach to adpositions. Déchaine
(1993, 33) lists English words derived from an ‘adpositional’ stem: off-ing,
about-ness, under-ling, upp-ity. However, she argues adpositions to be a
lexical category, and not a closed-class either. At this point, we could say that
adpositions are indeed not word classes but complex and articulated syntactic
structures in which certain, one or more, lexical items are visible.
In this monograph, we will assume without any further discussion that,
despite all the complications, adpositions are complex structures à la Sveno-
nius (2007, 2008), which include both clearly defined functional heads like

25
Oga (2001) independently reached the conclusion that there are two types of of. Tremblay
(1996) shows exactly the same to be the case with with.
 2.10 Conclusion 51

Kase, Figure, Ground, Path and so on and lexical material, possibly of a

nominal nature, in the respective specifiers. So, oversimplifying, particles such
as at, in and the like could be the realizations of a projection that relates two
spaces and/or time intervals; such projections would be part of a larger
functional complex fleshing out this and similar relations: ‘heavy’ prepositions
of a nominal and/or adverbial flavour, such as front, under, after and so on,
would be sitting in dedicated specifiers as arguments or modifiers. If this
sketch is on the right track, adpositional projections would be akin to the
system of functional structure supporting and realizing the expression of verbal
arguments in Hale and Keyser (2002), Pylkkänen (2008), Ramchand (2008)
and elsewhere. Alternatively, adpositional projections could be relational, like
Tense structure as envisaged by Stowell (2007), a claim that is epigrammatic-
ally stated in Pesetsky and Torrego (2004, 506): ‘the category P[reposition] is
actually a kind of T[ense]’.

2.10 Conclusion
First of all, the noun–verb distinction is indeed universal: nouns and verbs are
distinct syntactic categories and, in a large number of languages, they are also
distinct word classes. Two examples of languages that have been claimed to
possess a single undifferentiated lexical category, Tagalog and Riau Indones-
ian, turn out to actually make a distinction, at least between nouns and verbs.
On top of this, and with the possible exception of Nootka, there are roots that
can only become nouns, not verbs. A syntactic categorization approach,
correctly calibrated, should capture this, together with the universal presence
of nouns and verbs.
When we say that verbs are universal we do not necessarily refer to those
exemplified by latinate verbs in English – for example, donate. Single-word
verbs in inflecting languages already contain a lot of structure, structure that is
actually visible in such languages as Japanese, Farsi and Jingulu: these and
other languages combine a light verb with some lexical material to yield
complex verbal predicates. The essential point here is that verbs, including
light verbs, are universal. In Chapter 4 I will actually argue that it is v heads,
verbalizers, that are universal.
Despite what surface descriptions may suggest (confounded and misled
either by morphological complexity or by the radical absence of familiar
morphological distinctions), verbs and nouns cannot be used interchangeably.
Repeating the initial point here: roots may be inserted in a number of gram-
matical contexts categorizing them, but categorized nouns and verbs are not
52 Are word class categories universal?

freely distributed, neither do they collapse into a single undifferentiated ‘predi-

cate’ or ‘lexical’ category. In the following chapter I will argue that roots are
categorized syntactically and in Chapter 4 that categorization is essential for
roots to participate in syntactic derivations – explaining why categorized roots,
as nouns or verbs, are universal.
We also looked at adjectives, which will be largely excluded from the
discussion that follows, and we reached two tentative conclusions: adjectives
are not the unmarked, elsewhere lexical category and they are not of the same
ilk as nouns and verbs – and nowhere are they as fundamental and/or universal.
On the contrary, they tend to be derived, they share a lot of characteristics with
nominal functional heads, they can sometimes even stand in for possessive
genitives, they tend to go either the way of verbs (Korean), of nouns (Chi-
chewa, Dravidian), or both. Adverbs are possibly the result of adjectives being
associated with even more syntactic structure.
Adpositions are still a kind of mystery. Although nouns, verbs and adjec-
tives form word classes in a number of languages, adpositions usually resist
inclusion in a single class. Moreover, when not manifestly pure functional
heads, adpositions appear to be the (partial) lexicalization of complex func-
tional structures that may host lexical material. The least we can say about
adpositions is that they are not part of a [N] [V] system, not all of them at
any rate.
Having established the fundamental character of nouns and verbs, which we
will interpret in Chapter 4, we first need to discuss syntactic decomposition
and syntactic categorization.
3 Syntactic decomposition
and categorizers

3.1 Introduction
As already stated in the first chapter, the approach to category in this monograph
is committed to conceiving of word classes as encoding different categorial
features. These features are to be understood not as taxonomic ones creating
word classes of a morphological nature – for example, like declension and
conjugation classes – but as genuine LF-interpretable features, as instructions
to the Conceptual–Intentional Systems. Put differently, we will be very serious
in taking [N] and [V] to be ordinary syntactic features: interpretable, with
uninterpretable versions, triggering syntactic operations and imposing syntactic
constraints. In order to do this, we will embrace an approach to categorization
which takes it to be a syntactic process. Arguing for syntactic categorization will
enable us to work with categorial features as bona fide syntactic features and also
to account for cross-linguistic typological data – for example, categorization
behaviours such as the ones outlined in the previous chapter.
The following section poses the question of how words are made, of where
morphology is located. The point of view assumed is that of a syntactician
seeking to understand the workings of categorial features, and lexical and
functional categories more specifically. After a brief and very sketchy critique
of lexicalism the Distributed Morphology take on word-making is introduced.
The following two sections, 3.3 and 3.4, zoom in on some details of this
approach – that is, how Distributed Morphology can explain idiosyncrasies in
the meaning of morphologically complex words, if all structure-building,
including word-building, is done in syntax, a system typically conceived of
as yielding compositional interpretations of the items it combines. Section 3.5
examines conversions in order to reveal how zero-derived verbs can have very
different derivational histories, being either the product of root categorization,
like the verb hammer, or of recategorization, like the denominal verb tape.
Section 3.6 introduces the very basics of Phase theory in a simplified form and
relates its implementation of cyclicity in grammar with respect to how we can

53
54 Syntactic decomposition and categorizers

explain the idiosyncratic character of some morphologically complex words.

Roots are identified as the source of idiosyncrasy, and they are discussed
in detail in Section 3.7, while in Section 3.8 we deal with the question of
why ‘word level’ syntax does not appear to be anywhere near as productive
as ‘phrasal’ syntax. In Section 3.9, we look at a different problem: we
closely examine possibly the most convincing case for roots actually carrying
categorial specifications. A concluding section finishes the chapter.

3.2 Where are words made?

The approach to syntactic categorization and syntactic decomposition of the
main lexical categories ‘noun’ and ‘verb’ that we follow here is framed within
Distributed Morphology: Halle and Marantz (1993), Marantz (1997, 2000,
2006), Harley and Noyer (1999, 2000) and elsewhere.1 More specifically,
Hale and Keyser (1993, 2002) introduced a syntactic approach to the con-
struction of lexical categories, with an emphasis on verbs; the distinct version
thereof developed in Marantz (1997, 2000) has gained considerable currency
in the past ten years or so. The general outline of the Marantzian approach
is that lexical categories such as ‘noun’, ‘verb’ and ‘adjective’ are not products
of the combination of categorial features with roots in a lexicon: categories are
not specified on lexical items in a pre-syntactic lexicon. On the contrary, roots
are inserted bare in syntax, where the assignment of roots to categories takes
place as a process of embedding the latter within categorizing projections:
thus, categorization is a syntactic process; it is the syntactic environment that
turns roots into ‘nouns’, ‘verbs’ or ‘adjectives’. This is achieved by inserting
roots inside the complement of categorizers – a nominalizer (n), a verbalizer (v)
and an adjectivizer (a). On top of this, a categorizer may recategorize – that is,
change the category of – an already categorized element: for example, in
the cases of denominal verbs and deverbal nouns (e.g., colony ! colonize,
colonize ! colonization).

1
Although the analysis here is couched within Distributed Morphology, I believe that any
consistently realizational morphological framework can be employed equally well, as long as
it incorporates
a. a separationist distinction (and/or a dissociation) between syntactic feature structures
and their morphological exponence (Beard 1995) and,
b. syntax-all-the-way-down, as in Marantz (1997) and Harley and Noyer (2000) – that
is, taking the same combinatorial mechanism to lie behind both word-building and
sentence-building.
 3.2 Where are words made? 55

As already noted in Chapter 1 (Section 1.4.1), where some references

are cited, there exists an extensive body of research within the Distributed
Morphology framework refining the mechanisms and categories at work
as well as the empirical consequences of syntactic categorization, also explor-
ing their implications from a cross-linguistic point of view. In the following
chapter we will proceed to enrich the Distributed Morphology mechanism
of syntactic categorization through categorizers by incorporating concepts
and findings from Baker (2003), being indebted at the same time to the
insights expressed in Langacker (1987) and Anderson (1997). More specific-
ally, we will try to see what makes the categorizer n different from the
categorizer v (we have already argued that adjectives are different from both,
anyway), how n and v work and how they are interpreted. As should be clear
by now, in order to do this, emphasis will be placed on categorial features,
viewed as LF-interpretable features.
Before we can discuss the syntactic decomposition of category or, alterna-
tively, syntactic categorization, we need to have a theory of word-making. If
one abides by lexicalism and argues that (some or all) words come ready-made
from a lexicon, then the behaviour of categorial features as syntactic ones is
hardly evident or necessary: the noun–verb distinction could as well be one
like that between hyponyms and basic category terms. Therefore, if we need
a precise and falsifiable theoretical account of how and where nouns and
verbs are made, we need first to answer the questions of how and where words
are made.
It is conceded even in certain lexicalist models, the so-called ‘weak
lexicalist’ models, that some word-making must be done in the syntax;
see Baker (2003, sec. 5.2) and the discussion of Wasow (1977) and
Dubinsky and Simango (1996) in Marantz (2000). This would mean that
derivational morphology is not handled exclusively by the lexicon or by a
morphological component separate from syntax. Indeed, some derivational
morphemes must be attached on their host via a syntactic process. This
renders word-making a non-unitary process and essentially falsifies most
versions of the Lexical Integrity Principle, in that syntax is needed for the
creation of at least some word structure.2 The way Baker (2003, sec. 5.2)
summarizes the resulting state of affairs, his version of weak lexicalism, is
given in (1):

2
Lieber and Scalise (2007) provide a very informative and thorough overview of the Lexical
Integrity Principle and the different flavours of lexicalism.
56 Syntactic decomposition and categorizers

(1) Word-making
Lexical/
‘morphological’ Syntactic
word-making word-making

‘Raw material’ any category or a particular category

acategorial roots
Meaning idiosyncratic predictable meaning
meaning
Productivity limited full (allomorphy aside)
Examples –ful attachment, –ness attachment
compounding

In the picture adumbrated in the table in (1) words can be created in three
distinct ways (or, using a more Y-model-oriented metaphor, in three different
places). First, there are roots already coming from the lexicon with a categorial
specification. Thus, some nouns and verbs are roots specified for [N] and [V],
as nouns and verbs, and their category is learned; it is memorized. Examples of
such roots categorized in the lexicon would include cat, dog, coffee, bake, run,
kill and so on. Second, some nouns and verbs are created via lexical word-
making, via derivational morphology of a lexical nature – say, by combining
roots with nominal ([N]) and verbal ([V]) affixes through non-syntactic, mor-
phological processes: for example, quant-ify, cert-ify, cert-ifi-cate, free-dom,
geek-dom, the denominal zero conversion giving the verb water and so on.
Finally, syntactic word-making is also possible. Therefore, some nouns and
verbs are assembled during the syntactic derivation by means of syntactic
processes that combine nouns, verbs and adjectives with category-changing
syntactic heads (which also may surface as affixes). Examples of this would be
gerunds (train-ing) and adjectives like kind-ness, bleak-ness and so on, where
–ness behaves like a nominalizing syntactic head according to the criteria in (1).
Weak lexicalism, exemplified by the picture presented in the table above, is
a response to empirical necessity: strong lexicalism fails to capture important
generalizations because it assigns all word-making to the lexicon or to the
purported morphological combinatorial component. At the same time, weak
lexicalism is not particularly desirable from a conceptual viewpoint, as there
are now three distinct ways of making words; it is also empirically problematic
as Halle and Marantz (1993) and Marantz (1997) discuss at some length.
Until the mid-nineties, most syntactic analyses in the major generative frame-
works would assume that syntax manipulates words and/or morphemes. For
instance, in most of Chomsky’s work, such as in Chomsky (1957, 1995) (to name
but two landmark cases), the syntactic derivation manipulates words and
 3.2 Where are words made? 57

inflectional morphemes. Words would be constructed in some pre-syntactic

morphological component and then be handed over to the syntactic derivation.
However, in the nineties, syntactic theory took a turn in a new direction: first, it
began to seriously consider the type of features and the requirements imposed
on syntactic structures by the interfaces between the Language Faculty and the
Articulatory–Perceptual systems, and between the Language Faculty and
the Conceptual–Intentional systems. At the same time, syntactic theory paid
close and serious attention to Economy of derivation considerations, with mile-
stones such as Chomsky (1995, chap. 2), Collins (1997) and Fox (2000). The
convergence of the above two research paths led a number of formal linguists to
reconceive syntax as a system that optimally links two extra-linguistic systems:
Articulatory–Perceptual and Conceptual–Intentional; Brody (1995) is an exem-
plar of both this drive towards Economy and of the endeavour to drastically
eliminate superfluous levels of representation in the Language Faculty.
The positing of principles such as Full Interpretation (Chomsky 1995) and the
elaboration of the possible ways in which interface needs drive and/or shape
syntax are indicative of this reconceptualization, nowadays termed ‘minimalist’.
This turn towards ‘virtual conceptual necessity’ (Chomsky 1995, 169) made
syntacticians more mindful of the place and the role of Morphology within the
Language Faculty and of its precise relation and interaction with syntax. Beard
(1995) and Halle and Marantz (1993) proved very influential in this context, as
more and more syntacticians had to admit that syntax per se (or ‘narrow
syntax’) cannot possibly manipulate fully formed words, or even morphemes,
as pairs of sound and meaning. After all, words and morphemes contain both
phonological and morphological features (such as declension and conjugation
features) that are irrelevant for the syntactic derivation. In addition, lexical
words carry descriptive content (‘denotation’), which is irrelevant for narrow
syntax. Given the well-established thesis known as the Autonomy of Syntax
(Newmeyer 1998, chap. 2 and 3), the syntactic derivation is oblivious both to
the morphophonological properties and to the meaning of the lexical items it
manipulates. For instance, whether a verb in its past tense is regular (play-ed)
or irregular (ate) is certainly of no consequence to syntax: either way it will
syntactically behave in exactly the same way. Similarly, what matters when
it comes to a noun is a [number:plural] feature, not whether it is expressed as
a /-z/ plural or as an irregular form, like children.
One way to capture the above state of affairs is to posit that syntax manipulates
roots and morphosyntactic features, rather than fully formed words or mor-
phemes. Phonological features and, indeed, morphological forms would then
be inserted after syntax, by a morphological component on the way to the
58 Syntactic decomposition and categorizers

interface with the Articulatory–Perceptual systems. The theories advocating the

above state of affairs are realizational or separational morphological theories
(see also footnote 1). One of these realizational frameworks, one tailored to the
understanding of syntax as a derivational system, too, is Distributed
Morphology.
According to Distributed Morphology, there is no such thing as lexical word-
building: all forms (called ‘Vocabulary Items’) are inserted after syntax. Indeed,
it is a syntax-all-the-way-down (see footnote 1) framework: the same combina-
torial mechanism, syntax, responsible for phrase- and sentence-building, is also
responsible for building words. Hence, all words are created syntactically.
At first sight, relegating all word-building to syntax is equally problematic
with strong lexicalism: if strong lexicalism fails to capture important general-
izations such as the ones outlined in (1), then perhaps syntax-all-the-way-down
cannot explain the non-compositional and idiosyncratic character of (very
many) morphologically complex words. The dichotomy between composition-
ality and productivity vis-à-vis idiosyncrasy and non-productivity can be
captured in (1) because (at least) two distinct mechanisms are involved in
word-making: a disciplined, compositional, predictable and fully productive
syntactic one for the former and an arbitrary, idiosyncratic, non-productive
lexical–morphological one for the latter. The kinds of words each one yields
according to the table in (1) are therefore quite different: the relation of
rest-less-ness to rest-less and rest is very different from that of necess-ity to
necess-ary (and *necess does not even exist as a word).
However, different ‘ways’ of making words do not necessarily entail differ-
ent mechanisms (or different ‘places’) of word-making. In order to explain
the differences between ‘lexical’ and ‘syntactic’ word-making within a syntax-
all-the-way-down framework such as Distributed Morphology, first we need to
focus on the meaning idiosyncrasies of ‘lexical’ word-making.

3.3 Fewer idiosyncrasies: argument structure is syntactic structure

A key argument against having syntax create all words, including simplex
nouns like cat, is idiosyncrasy of meaning. It is too often the case that two
words – say, a noun and a verb – deriving from the same root have completely
different meanings and, certainly, these meanings cannot be related to each
other through compositional processes, as one typically expects from the
interpretation of syntactic output. Idiosyncrasy of meaning of words deriving
from the same word is actually one of the arguments in Panagiotidis (2005)
against the syntactic decomposition of nouns and verbs.
 3.3 Fewer idiosyncrasies: argument structure 59

Let us look at a sample of verbs which are seemingly idiosyncratically

related with corresponding nouns and adjectives, basing this on the line of
inquiry inaugurated by Levin (1993), and loosely following Levin and
Rappaport-Hovav (2005, 68–75):
(2) Verb Verb meaning Type

sweat make/exude sweat

push x give x a push
clear x make x clear change of state verb
open x make x open change of state verb
box x put x in a box Location verb
butter x put butter on x Locatum verb
brush y put x on y with a brush instrumental verb

The picture is quite telling: there is no single relation between a noun and a
verb. So, the verb sweat means ‘make sweat’ but the verb butter does not mean
‘make butter’; the verb box means ‘put in a box’ but in the verb brush the root
brush is about the instrument of applying x on a surface y – and so on. In
other words: there is a diverse number of types of verbs when it comes to the
relation they have with co-radical nouns and adjectives. As Levin (1993)
points out, these different types of verbs are systematic; they are most likely
the result of where the verbal root appears in the argument structure of the verb
(Hale and Keyser 2002). Thus, once we have a closer look, we realize that
there is more regularity in the possible meanings of noun–verb pairs than is
initially apparent. Broadly speaking, the root in the verbs in (2) behaves as a
direct object in sweat and push and as a predicate designating an end state in
change of state verbs like clear and open. Equally interestingly, in Location
verbs (e.g., box) the root behaves something like the object of a Preposition
Phrase, designating the background. In Locatum verbs (e.g., butter), however,
the root behaves like the subject of a PP, the figure to be placed against
a background. Finally, in an instrumental verb like brush the root acts as a
manner adjunct, an instrument adjunct, more precisely.
One issue that can be noted already is the complexity of some verbs’
argument structure: if we are to generate such structures ‘morphologically’,
we must make sure we do not gratuitously duplicate the workings of syntax
or, at least, those mechanisms familiar from the workings of phrasal syntax.
For instance, it is very interesting that change of state verbs seem to embed a
small clause consisting of the object x of the verb and the end state predicate,
and that the end state predicate is the root that ‘names’ – to echo Harley
(2005b) – the verb. The argument structure of an instrument verb such as brush
60 Syntactic decomposition and categorizers

seems to be even more complex: there, the root naming the verb is the
instrument, the modifying adjunct. Moreover, the surface direct object of the
verb brush sets the background: for example, we brush the roast thoroughly
(with a honey and mustard mix) but we do not #brush a honey and mustard
mix thoroughly although we can brush a honey and mustard mix on the roast
thoroughly; such structural alterations strongly suggest that displacement of
the syntactic sort is at play.
A second, and even more vital, point is that syntactic terminology like
‘subjects’, ‘objects’, ‘adjuncts’, ‘predicates’ turns out to be quite useful when
talking about the position of a root within the verb’s argument structure.
Briefly, and following the lead in Hale and Keyser (1993, 2002), we can claim
that different argument structures are actually different syntactic structures.
Idiosyncrasy is only apparent, as the different types of verbs like those in (2) –
and many more – boil down to where the root is to be found within the
argument structure of the verb: that is, inside a bona fide syntactic structure.
In this respect, we are justified in using syntactic terminology to discuss
argument structure because it looks like (such) verbs contain some hidden
syntax, what Hale and Keyser (1993) call L-syntax and Ramchand (2008) calls
First Phase Syntax. Harley (2005b) and Marantz (2005) explain in consider-
able detail how to analyse verbs in this way, within a syntactic decomposition
framework that embraces a realizational–separationist view on the syntax–
morphology relation – that is, Distributed Morphology.

3.4 There are still idiosyncrasies, however

Showing that some lexical differences can be reduced to systematic argument
structure patterns, which, in turn, can be conceived of as syntactic structures
explains away a great deal of idiosyncrasy. However, this move hardly elim-
inates idiosyncratic readings from co-radical noun–verb pairs. Consider some
well-known noun–verb pairs like the following:3
(3) deed–do, trial–try, action–act, revolution–revolve

(4) chair–chair, ship–ship, egg–egg, book–book, castle–castle4

3
We are not looking at the whole panorama of lexical idiosyncrasy here – for example, the non-
compositional interpretation of compounds – and we just focus on these aspects of the problem
that are more pertinent to matters of syntactic categorization.
4
Examples in (3) are from Chomsky (1970) and those in (4) are from Panagiotidis (2005) –
originally from Clark and Clark (1977).
 3.4 There are still idiosyncrasies, however 61

The above pairs (and many more) share a root, but it is extremely improbable
that the meaning of the verb could be guessed – let alone be derived compos-
itionally – by that of the noun, or vice versa, in the way things go with sweat,
push, box, brush and so on above. Incidentally, as argued in Acquaviva and
Panagiotidis (2012), examples like those in (3) and (4) render very problematic
the idea that roots are underspecified but still meaningful elements which give
rise to distinct interpretations depending on their immediate syntactic context
(Arad 2005): ‘even if we argue for impoverished and semantically under-
specified roots, we are still left with the empirical problem . . . that roots too
often do not capture a coherent meaning . . .. This renders unlearnable the
purported “common semantic denominator” roots are supposed to express’
(Acquaviva and Panagiotidis 2012, 109). To wit, what could be the ‘common
semantic denominator’ between revolution (as ‘uprising’ or ‘regime change’)
and revolve? Something as vague as ‘change’? Surely this is hardly coherent,
and the interested reader is again referred to Acquaviva and Panagiotidis
(2012) and, also, Borer (2009).
The stand taken in Marantz (1997, 2000) and elsewhere on pairs like (3) and
(4) is that they involve underlying syntactic structures which receive idiosyn-
cratic interpretations. Now, idiosyncratic interpretations of variously-sized
syntactic objects are not really an oddity: they are everywhere and they are
called idioms. Trivially, break a leg, kick the bucket, raining cats and dogs
are all syntactic phrases, vPs according to Svenonius (2005). However, they
have idiosyncratic meanings.
The idiom idea is definitely useful: lexical idiosyncrasies are not the result of
a ‘lexical/morphological’ way of making words, but the result of idiomatic
interpretation of otherwise ordinary structures. There are two questions, how-
ever: first, why can the meanings of words (nouns and verbs) made of the same
root be so distinctly, even crazily, different and to such a considerable extent?
Second, as Hagit Borer (personal communication, June 2009) pointed out,
phrasal idioms can typically also receive a fully compositional interpretation
when properly contextualized: kick the bucket may also simply describe a
bucket-kicking event. However, the verb castle is impossible to interpret
compositionally – for example, as a Location verb – and can only mean the
chess move it idiosyncratically names. Why should such a difference between
phrasal and ‘lexical’ idioms exist?
Panagiotidis (2011) seeks to answer both questions above by arguing the
following points. If it is indeed the case that roots by themselves are
characterized by impoverished or, perhaps, inexistent semantic content, then
non-compositional and idiosyncratic interpretations of syntactic material
62 Syntactic decomposition and categorizers

directly involving roots are the only interpretive option: after all, how could
compositional interpretation deal with the un- or under-specified meaning of
roots? In other words, idiomaticity is the only way to interpret constituents
containing roots, exactly because of the semantic impoverishment/deficiency
of roots. This point is already alluded to in Arad (2005) and fleshes out
the conception of the categorizing phrase in Marantz (1997, 2000) as the
limit of (compulsory) idiomaticity. However, systematic idiomaticity of
the categorizer projections is not due to the categorizer acting as some sort
of a limit, below which interpretation is/can be/must be non-compositional.
Rather, idiomaticity – that is, matching a structure with a memorized mean-
ing stored in the Encyclopedia (Harley and Noyer 1998) – stems from the
fact that the first phase (an nP or a vP)5 contains a root, an LF-deficient
element, which would resist any compositional treatment anyway. In other
words, the semantically deficient character of the root blocks the application
of a rule-based compositional interpretation. Therefore, inner versus outer
morphology phenomena (Marantz 2006) are due to the semantic impoverish-
ment of roots: once roots have been dispatched to the interfaces with the rest
of the complement of the categorizer, compositional interpretation may
canonically apply in the next phase up – see also the next chapter and
Panagiotidis (2011, 378–9).
Keeping these points in mind, we can now continue with the rest of this
chapter, which will be dedicated to reviewing the literature on syntactic
categorization. This will set the context in which the theory of categorial
features to be posited applies.

3.5 Conversions
Syntactic decomposition can advance our understanding of why some verbs
have a predictable, compositional relationship with their corresponding
nouns (as discussed in Section 3.3 above) whereas others display only an
arbitrary and idiosyncratic one (as reviewed in Section 3.4). The very short
answer, which follows Kiparsky’s (1982) analysis of the two levels of
morphological processes, is that the former are denominal verbs, whereas
the latter are root-derived. Staying with Arad (2003), whose account we
will closely follow in this section, we will call (true) denominal verbs tape-
type verbs and those verbs that are directly derived from a root hammer-type
verbs. The structural differences between them will be argued to account

5
See Chapter 4 for details.
 3.5 Conversions 63

for the idiosyncratic meaning of the latter, as opposed to the predictable

and systematic meaning of the former.
Let us begin with root-derived, hammer-type verbs. In Arad (2003), verbs
like hammer are directly formed from the root, just like corresponding nouns.
Consider the schematic phrasal markers below.
(5)

In the tree to the left, the nominalizer head n takes a root complement,
nominalizing it syntactically. In the tree to the right, the root hammer is a
manner adjunct to an xP (schematically rendered) inside the vP, more or less
along the lines of what was said about brush in Section 3.3 above, a configur-
ation responsible for the manner interpretation of the root hammer within the
verb hammer. Crucially, the verb does not mean ‘hit with a hammer’ but it
takes up a broader interpretation, roughly ‘hit in a hammer-like fashion’ –
maybe using a stone, a shoe, a fist, or similar.6
On the other hand, verbs like tape behave differently. These seem to be truly
denominal, formed by converting a noun into a verb, by recategorizing the
noun and not by categorizing a root. By hypothesis, the verbalizing head takes
as its complement a structure that already contains a noun – that is, an nP in
which the root tape has already been nominalized.
(6)

Again, in the tree to the left, a nominalizer n takes a root complement,

nominalizing it syntactically. However, the verb tape is unlike the verb
hammer: although it is also a manner/instrument verb, tape only means ‘fasten

6
Once more, this raises the question of roots’ inherent content, if any, as an anonymous reviewer
points out: what is the content of hammer? The answer I am partial to is ‘none’; more on this
line of reasoning and on ways the semantics could be executed can be found in Acquaviva
(2009b), Borer (2009), Panagiotidis (2011) and, in detail, Harley (2012a).
64 Syntactic decomposition and categorizers

with tape’ – not with rope, pushpins, paperclips or even bandages. In other
words, the meaning of the verb tape is not idiosyncratic but it wholly and
predictably relies on the meaning of the noun tape, which it contains embed-
ded in an adjunct position. This difference cannot be due to the verbalizer v,
which is possibly the same in both cases. It therefore must be due to the fact
that the verbalizer makes a verb out of root material in (5), yielding the verb
hammer, whereas in (6) it makes a verb out of a ready-made noun, tape.
This entails that hammer-type verbs can receive special meanings, whereas
tape-type verbs do not.7
The general question here concerns the specifics of how these differently
derived verbs, those categorizing a root and those recategorizing a noun, come
to be. This question reduces in part to the whole issue of what is the limit of
‘special meaning’ in a syntactic tree. Inspecting the trees here, and as is the
consensus in research, the limit of special, idiosyncratic meaning in a syntactic
tree is the first categorizer projection as in Arad (2003, 2005) and Marantz
(1997, 2000, 2006). Having said that, good arguments of an empirical nature
are levelled against this thesis in Anagnostopoulou and Samioti (2009), Borer
(2009), Harley (2012a) and Acquaviva and Panagiotidis (2012): the above
seem to point towards the direction that, at least in the verbal/clausal domain,
it is the Voice Phrase (or an even higher projection) that constitutes the ‘limit’
of idiosyncrasy.
If the first categorizer projection is the limit of idiosyncratic interpret-
ation, then the verb hammer has a special meaning because categorization,
brought about by the v head in (5), includes the x projection with its
hammer root adjunct. On the contrary, the verb tape in (6) includes an x
projection with an nP adjunct – that is, an already formed noun, an already
categorized constituent. The nP has already received an interpretation of
‘(sticky) tape’, so the range of possible interpretations of the vP containing
it is significantly limited: ‘stick x on y with tape’. The obvious problem that
emerges now is why the first categorizer projection should be the limit of
‘lexical’ idiosyncrasy. Why is it not the first projection containing the root,
like those indicated as xP in (5) and (6)? What is special about the first
categorizer projection?

7
I have deliberately chosen not to represent the position of the internal argument of hammer and
tape in the diagrams (5) and (6) purely for reasons of exposition. I would, however, think they
are merged as the specifier of x or v. The choice depends on the nature of the argument, as
discussed in Marantz (2005).
 3.6 Phases 65

3.6 Phases
In order to answer the question of why the first categorizer projection is the
limit of ‘lexical’ idiosyncrasy, we need to appeal to cyclicity in syntax.
The first syntactic cycle, which typically, but not necessarily, coincides
with the word domain, manipulates impoverished roots; hence it is impossible
for this first cycle to yield compositional interpretations. Once the interpret-
ation of the first cycle is ‘idiomatically’ fixed, then its interpretive output may
be used by higher cycles to yield compositional interpretations.
Let us follow the consensus in syntactic literature that syntactic operations
are local and cyclical. A specific way to conceive cyclicity, one that makes
precise and falsifiable empirical predictions, too, is through Phase Theory, as
proposed and developed in Chomsky (2000, 2001, 2008) and a lot of subse-
quent work. Let us see how thinking in terms of phases may enable us to
conceive the ‘character’ of the categorizer projection. According to Marantz
(2006), a categorizer phrase (vP or nP) is a Phase. The other two phases are
VoiceP8 and Complementizer Phrase (CP). So, phases are both interpretive
(meaning) units and phonological–prosodic ones – but see Hicks (2009) for
solid arguments from binding that we need to separate LF-phases from
PF-phases.
Let us turn to an example of a derivation with phases. Suppose we start
building a tree – say, a clausal one:
(7)

Given that vP is a phase, it is despatched to Morphology/PF and LF in order

to be interpreted accordingly.

8
There exists some terminological confusion on what the label v stands for. For the work cited and
followed here, it is the verbalizing categorizer, the ‘verbalizer’. However, in the work mainly
concentrating on Phase Theory, beginning with Chomsky (2000), et seq., v essentially stands for
Kratzer’s (1996) Voice: a causative–transitive or passive head which hosts the external argument
and of which the transitive version may assign accusative case, as per Burzio’s Generalization.
Now, both approaches take v (the categorizer or the Voice head) to be a phase head. Things
become slightly more confusing in that Chomsky, and others, seem to consider the two elements,
v and Voice, as one unitary head. I will here be concerned with the phase status of the
categorizing v (and n) and will remain agnostic about that of Voice. However, see Anagnosto-
poulou and Samioti (2009) and Harley (2012b) for arguments on why v and Voice should be
kept distinct.
66 Syntactic decomposition and categorizers

(8)

The syntactic derivation will from this point on only be able to see the edge
of the phase: the topmost head – that is, v in (8) – and its specifier (none in the
example above). The rest of the phase (shaded) will not be visible:

(9)

The derivation continues building structure until VoiceP is merged, which is

the next phase; this in turn is also sent to Morphology/PF and LF:

(10)

Once the VoiceP phase is interpreted, again, only its edge, the Voice
head and the specifier hosting the Agent argument are visible for further
syntactic computations. The derivation continues on to the next phase, CP,
and so on.
 3.7 Roots and phases 67

In order to summarize how syntactic categorization interacts with the cyclic/

phasal character of syntactic derivations, here are some essential points to
retain from this very sketchy adumbration of Phase Theory:
1. Once a phase is completed, it is interpreted: therefore, its morpho-
logical form and its semantic interpretation become fixed.
2. The internal structure of a completed phase, except its edge, is
invisible for the structure to be created higher up – that is, for the
rest of the derivation.
3. The categorizer phrase (nP and vP) is the First Phase.

3.7 Roots and phases

Let us now apply the above to show how the ‘lexical versus syntactic’
difference in word formation can be explained away. In order to do this, we
will revisit the denominal verbs hammer and tape (minus their internal argu-
ment; see also footnote 7) from Section 3.5:
(11) hammer versus tape

The verb hammer is made from root material (including some syntactic head x,
as the root HAMMER behaves like a manner adjunct) and the verbalizer v; no
interpretation or form is matched with the syntactic structure until vP, the first
Phase, is completed. This is why hammer has an idiosyncratic meaning,
exactly like any other word made directly out of a root, as opposed to words
containing already categorized constituents in their structure. In the case of the
verb tape, the root and the nominalizer n combine first (making a noun). This is
a phase: it is sent to LF and Morphology/PF, where the meaning and the form
of the nP tape are fixed. This nP phase can then participate in more syntactic
structure: in (11) this nP merges as an adjunct with structure that is eventually
verbalized.9

9
The question here is whether we can embed the root TAPE directly under a v projection. The
answer is that syntactically this is possible, and ‘looser’ usages of a verb tape are reported by
some speakers; the whole thing boils down to whether Morphology makes available the form(s)
68 Syntactic decomposition and categorizers

So, now we can explain why vP and nP are the limit of ‘special meaning’
in a syntactic tree: vP and nP are the first Phases. They are the first cycle
after the derivation begins, provided of course that they categorize root
material and not already categorized constituents – which of course
would already be phasal constituents. Thus, far from acting as limits of
idiosyncrasy, categorizer projections may receive a compositional interpret-
ation if they recategorize already categorized material, just as is the case
of denominal tape with its [vP v [xP nP x]] structure. What causes idiosyn-
crasy is the semantic deficiency/impoverishment of roots within their
first phase.
Further illustrating this point – namely, that verbs like hammer consist of a
single phase – compare also the morphophonology and meaning of the near-
homophones dígest (noun), digést (verb) and dígest (verb). The noun dígest
and the verb digést are each derived directly by a categorizing head, an n and a
v respectively, taking the projection that contains the root as its complement.
Meanings of the respective noun (nP) and verb (vP) are distinct, as expected
from the discussion above, each of them directly embedding the impoverished/
semantically deficient root digest. Stress, a morphophonological property,
is also distinct (N dígest vs V digést), a property decided and fixed at phase
level, as well:10

(12) A root-derived noun and a root-derived verb

Furthermore, the noun dígest can be converted into a verb, yielding the
denominal verb dígest; this is a process identical to that giving the verb tape.
Notice in the phrase marker below that the meaning of denominal dígest is
again compositionally derived from the meaning of the noun and the contribu-
tion of y. The denominal verb dígest is therefore interpreted as ‘make a dígest’;
it also preserves the stress pattern, fixed after the nP phase was despatched to
the interfaces:

necessary to express this direct verbalization of the root TAPE as a word. For discussion of
blocking and gaps, see Embick and Marantz (2008) and references therein.
10
Again, in (12) and (13) I have deliberately left out the internal argument. See also footnote 7.
 3.7 Roots and phases 69

(13) A denominal verb

Moving away from English and zero-derived conversions, let us observe the
behaviour of two more elements as inner and outer morphemes and let us see
how root deficiency and phases can capture their non-compositional behaviour
when they are directly associated with roots below phase level. We begin with
the Japanese causativizer –(s)ase (Marantz 1997). Next to compositional
causatives, like suw-ase (‘make somebody smoke’) below, Japanese causati-
vizers may also yield idiomatic, non-compositional interpretations, like tob-
ase (‘demote someone to a remote post’) below. As illustrated below, the
causativizing element –(s)ase in the case of suw-ase (‘make somebody
smoke’) combines with a vP phase, whose interpretation has already been
fixed as ‘smoke’, and the result is predictable and compositional. In the case of
tob-ase (‘demote someone to a remote post’), the causativizer merges with a
root tob- (which also yields the verb ‘fly’) and derives an unpredictable and
specialized verb tob-ase (‘demote someone to a remote post’) when it eventu-
ally becomes categorized by a v head.11
(14) Outer and inner causativizers

A similar picture emerges with Greek syn- (Drachman 2005). The interpret-
ation of syn- (as an outer morpheme) can be described as comitative and it is

11
As Marijke De Belder (personal communication, 2011) notes, even in idiomatic tob-ase, there is
clear causation involved. This is interesting but, as will be illustrated for Greek syn-, inner
morphemes need not preserve any of the meaning they have when they are used as outer,
‘compositional’, morphemes.
70 Syntactic decomposition and categorizers

productively used to yield verbs and nouns that have a ‘with X’ interpretation.
So, the verb syn-trog-o means ‘eat with somebody’, from the verb tro(g)-o
(‘eat’), and so on. Still, syn- is also used as an inner morpheme, with no
discernible semantic contribution whatsoever. Hence, syn-tax-i, which means
‘syntax’ or ‘pension’. Words like syn-tax-i belong to words that ‘have mor-
phological structure even when they are not compositionally derived’ (Aronoff
2007, 819). Another example of this non-contribution is the noun syn-graf-eas
(‘author’), whose morphological structure is transparently related to that of
graf-eas (‘scribe’) – still, syn-graf-eas cannot mean ‘someone who acts as a
scribe together with someone else’. To wit, ‘co-author’ in Greek would be
something like syn-[syn-graf-eas].12 The diagram below illustrates the differ-
ence between syn- as an outer morpheme, with a compositional comitative
interpretation, combining with a phasal vP in syn-trogo (‘eat with sb’), and
syn- as an inner morpheme in syn-graf-eas (‘author’), where it combines with a
root awaiting the phasal nP in order to receive its idiosyncratic interpretation.
(15) Outer syn- and inner syn- in Greek

We have captured the differences between ‘lexical’ and ‘syntactic’ word-

making as involving the merging of material inside the categorizing first phase
versus the merging of material with already categorized material, with phasal
nPs and vPs, which have had their interpretation and form already fixed.
Before we look at a serious challenge towards acategorial roots, we can round
up the discussion in this chapter by addressing the question of why syntactic
derivations at the categorizer phase are not fully productive.

3.8 On the limited productivity (?) of first phases

If word-making, including what lexicalism understands as ‘lexical’ or ‘mor-
phological’ word-making, is fully syntactic, if there is a single combinatorial

12
When I was writing those lines I was confident that syn-[syn-graf-eas] was nothing but a
slightly contrived coinage. While revising the chapter, I ran an exact Google search of συν-
συγγραφέας out of curiosity: it yielded about 21,500 results.
 3.8 On the limited productivity (?) of first phases 71

mechanism that makes words and phrases, then we would expect word-making
to be significantly more productive, whereas it is nowhere near so, compared
with phrasal productivity. The short answer to this problem is the role of stored
forms, Vocabulary Items in Distributed Morphology, associated (or not) with
particular structures.
Indeed, syntactic composition at the first, categorial phase is unlike ‘phrasal
syntax’ – that is, syntactic operations above the ‘word level’, involving higher
phases. For instance, there are no verbs cat, hair or tree, for most speakers of
English at least; there is a difference between father (which is only a ‘becom-
ing’ verb) and mother (which is only a manner/instrument verb) and there is no
verb #father meaning ‘to treat as a father would’ – see Clark and Clark (1977)
for scores of similar examples; not all roots can be used to derive Location/
Locatum verbs; and so on.
Of course, productivity, or lack thereof, is a morphological matter:13
whether Vocabulary contains an appropriate form to insert in a given structural
environment. So, perhaps syntax is after all able to construct a Location
verb ‘butter’, next to the well-known Locatum one, or even to construct a
manner verb ‘butter’. The problem in these cases would be whether Vocabu-
lary (or Morphology, more generally) can make available the form(s) that will
express such syntactic structures (see also footnote 9). Syntax is always free
and always combinatorial; at the same time, a lot of Morphology is inevitably
about a repository of memorized forms. It is very characteristic that in lan-
guages with few inflectional restrictions, such as English, spontaneous coinage
of words – that is, novel morphological expressions of syntactic structures at
the level of the word or whereabouts – is easier than in Romance or Greek,
where all sorts of inflectional criteria must be met and where zero derivation is
not available: in English, in the right context, even a manner verb butter is
possible – for example, in a coinage like #The vegetable oil was bad quality;
it buttered (‘coagulated like butter’).
Invoking morphological restrictions – that is, what is listed – is far from
ad hoc. Indeed, there are gaps and morphological idiosyncrasies every-
where, even in the bona fide syntactic inflectional morphology: two well-
known conundrums of this sort, hardly unique, from my native Greek are

13
Aronoff and Anshen (1998, 243) raise a crucial point with respect to productivity – namely, that
it is far from being an absolute notion: ‘Some linguists treat morphological productivity as an
absolute notion – a pattern is either productive or unproductive – but there is a good deal of
evidence for the existence and utility of intermediate cases, . . . so we will assume . . . that affixes
may differ continuously in productivity, rather than falling only into the polar categories of
completely productive and completely unproductive . . .’.
72 Syntactic decomposition and categorizers

the general unavailability of a genitive plural form for Greek feminine

nouns in –a, even of high-frequency ones like mana (‘mother’), patata
(‘potato’) and kota (‘hen’), or the awkwardness of the majority of genitive
singular forms for Greek diminutives in –aki. Once more, I refer the reader
to (at least) Aronoff and Anshen (1998), Marantz (2000) and Embick and
Marantz (2008).
Having said that, there is an additional factor that must be considered, one
discussed amply and in careful detail throughout Ackema and Neeleman
(2004). Recasting this factor in the framework employed here, we must
remember that categorizing first phases (nPs and vPs) are typically matched
with Vocabulary Items, with stored forms, but not necessarily so. To wit,
there are few morphologically inexpressible VoicePs and probably no CPs;
however, there are morphologically inexpressible nPs and vPs because
these are typically matched with stored, word-like forms – or they are expected
to be matched with word-like forms. Having said that, this is not the case in
languages like Jinggulu, Farsi and, to a large extent, Japanese, where verbs are
not expected to be co-extensive with a word – in languages possessing
periphrastic verbal forms, light verbs and/or particle verbs, expressing verbal
projections is much easier.

3.9 Are roots truly acategorial? Dutch restrictions

One of the assumptions underlying the workings of syntactic categorization is
that free roots, or ‘roots by themselves’ so to speak, are category-less – a
different term being ‘acategorial’. This assumption goes part and parcel with
roots being underspecified (or impoverished) semantically and/or bearing no
UG features.14 However, the assumption that roots bear no category is neither
obvious nor uncontested. Already in the previous chapter we have mentioned
the casual observation that some roots make better nouns than verbs. For
instance, the roots cat, dog and boy make good nouns in English but not
great verbs, if at all. Moreover, verbs like dog and cat appear to be denominal
verbs – that is, like tape – with the noun functioning like a manner modifier:
to dog means ‘to follow like a dog’ and similar facts hold for cat, at least
for those speakers that accept it as a lexicalized verb and not as a spontaneous
coinage (Barner and Bale 2002). Moreover, as will be reviewed in the
following chapter, languages like Farsi seem unable to verbalize most roots,

14
These matters are discussed in more detail in the following chapter.
 3.9 Are roots truly acategorial? Dutch restrictions 73

with Hindi/Urdu apparently working the same way (Rajesh Bhatt and David
Embick, personal communication, October 2010). That is, the lack of verbs
like boy and #cat in English is an empirical fact to be explained, possibly
invoking grammar-external factors.15
More systematic and solid criticism of acategorial roots comes from Don
(2004). He looks at conversions in Dutch to argue that ‘the lexical category of
roots should be lexically stored’ (Don 2004, 933). More precisely, he claims
that hammer-type derivations – that is, with a root directly yielding a noun
and a verb – are not to be found and that all conversions involve denominal
derivations (like tape) or deverbal ones (like the noun throw). Don makes two
arguments, a morphological one and a phonological one. We will review them
in turn.
Don’s morphological argument against category-less roots (in Dutch) is
very simple and comes from examining two properties of zero-derived
noun–verb pairs in Dutch: irregular morphology on verbs and gender on nouns
(Don 2004, 939–42). He first states that four logical possibilities present
themselves when it comes to cross-classifying zero-derived noun–verb pairs,
as follows:16

(16) Four possible verb–noun conversion pairs

a. regular verb – common gender noun
b. regular verb – neuter gender noun
c. irregular verb – common gender noun
d. irregular verb – neuter gender noun

Interestingly, only the first three options in (16) are attested in the language:

(17) Regular verb – common gender noun:

fiets – de fiets ‘cycle’
ren – de ren ‘run’
tel – de tel ‘count’
twijfel – de twijfel ‘doubt’

15
This state of affairs is possibly different to the Romance and Greek cases, where the existence of
‘nominal’ and ‘verbal’ roots looks like an illusion of rich morphology, like the result of roots
being embedded within ‘their functional support systems’, their morphological entourage. This
is also suggested by Panagiotidis, Revithiadou and Spyropoulos (2013), where it is argued that
all but a handful of verbal stems in Greek contain an overt verbalizing morpheme whose
exponence depends on complex (sub-)regularities.
16
Dutch has two genders: common (non-neuter), taking a de definite article, and neuter, taking a
het definite article.
74 Syntactic decomposition and categorizers

(18) Regular verb – neuter gender noun:

werk – het werk ‘work’
deel – het deel ‘part’
feest – het feest ‘party’
slijm – het slijm ‘slime’
(19) Irregular verb – common gender noun
val – de val ‘fall’
wijk – de wijk ‘flee’
loop – de loop ‘walk’
kijk – de kijk ‘look’
There are no verb–noun conversion pairs where the verb is irregular and the
noun is neuter. Don (2004, 941) explains this gap in the following way:
1. V-to-N conversion renders common gender nouns;17
2. N-to-V conversion renders regular verbs;
3. There is no such thing as direct zero-derivation from acategorial roots.
From the above assumptions, it turns out that in (18) N-to-V conversion takes
place, since V-to-N conversion would yield common gender nouns. In (19),
V-to-N conversion takes place, since N-to-V conversion would yield regular
verbs. A conversion pair ‘irregular verb – neuter gender noun’ is impossible: an
N-to-V conversion would yield regular verbs; a V-to-N would yield common
gender nouns.18 Hence, according to Don, roots are stored with a category: if
roots were category-less, then it should be possible to derive an irregular verb and
a neuter noun directly from the same root, the hammer way, so to speak.
Looking to reinterpret Don’s morphological facts from a point of view
where roots are indeed category-less, one conclusion we can draw is that we
are dealing with facts regarding the morphological exponence of categorizers
in Dutch. Recall that the morphological realization of syntactic heads is
decided post-syntactically (Late Insertion): morphological forms, Vocabulary
items, are inserted late – that is, after syntax – to match the features on
syntactic heads. Put succinctly, it seems that we are most likely dealing with
morphological constraints on null nominalizers in Dutch. Reinterpreting the
facts above, we can hypothesize that

17
It is quite common for morphological processes to determine the gender of their output. Don
(2004) cites Beard (1995) for examples of this in different languages.
18
Crucially, no statement can be made on the pairs in (17): they could be V-to-N or N-to-V
conversion pairs, verbs being regular and nouns bearing common gender. They could, for that
matter, be directly derived from an acategorial root.
 3.9 Are roots truly acategorial? Dutch restrictions 75

(20) An n in an [nP n vP] syntactic environment cannot surface both as Ø

and neuter;19

So, in (19), a vP (an irregular verb) with a Ø v is in the complement of a n

that is Ø and [common] gender – as per (20). Illustrating:
(21) A null nominalizer converting a verb surfaces with common gender

Note that there is no blanket ban on neuter gender nominalizers of verbs.

Don (2004: 941–2) discusses the neuter gender element –sel, which nominal-
izes verbs like zaag (‘to saw’), giving zaagsel (‘sawdust’). So, (20) is a
morphological constraint on null nominalizers with a vP complement.
However, and returning to the unattested conversion pair ‘irregular verb –
neuter gender noun’ in (16), what would prevent a Ø-exponence n bearing a
[gender:neuter] specification from taking a ROOT complement that also
derives an ‘irregular’ v? I think that, again, instead of invoking category
encoded on roots themselves, we had better examine the phenomenon from a
morphological and morphophonological point of view.
‘Irregular’ verbs in Dutch are not just verbs with deviant inflectional
paradigms, they are actually Germanic strong verbs – that is, verbs displaying
synchronically opaque stem allomorphy, what I would claim to be genuine
root allomorphy, resulting from Late Insertion of root forms, as in Galani
(2005, chaps. 5 and 6), Siddiqi (2006, chap. 3) and Haugen (2009). Here are
some examples from Don (2004, 940):
(22) Stem/root allomorphy in Dutch: irregular/strong verbs
Present Past Past participle

spijt(-t)(-en) speet ge-speet-en ‘to regret’

val(-t)(-en) viel(-en) ge-val-en ‘to fall’
bind(-t)(-en) bond(-en) ge-bond-en ‘to bind’
sla(-t)(-en) sloeg(-en) ge-slag-en ‘to beat’
koop(-t)(-en) kocht(en) ge-kocht-t ‘to buy’

19
See Lowenstamm (2008) for how gender defines different flavours of n.
76 Syntactic decomposition and categorizers

A second generalization then seems to be that

(23) A Ø-exponence nominalizer cannot bear the neuter gender when taking as its
direct complement a root displaying allomorphy.

Putting (20) and (23) together, it turns out that

(24) A Ø n head bearing a neuter gender specification can only take non-
allomorphic root complements.

The above looks like the kind of morphological constraints we find in language
after language, as opposed to being the offshoot of the purported categorial
specification of roots. Additional evidence that the gap Don discusses is of a
morphological and morphophonological nature comes from Don’s (2004,
942–5) phonological argument for categorially specified roots. Reviewing
evidence in Trommelen (1989), he observes an interesting constraint on Dutch
verbal stems, summarized in (25):
(25) Phonological properties of verbs in Dutch
Simple syllable Complex syllable
structure structure

With identical noun numerous: bal, lepel, some: oogst, feest, fiets
kat
Without identical numerous: win, kom, No examples
noun vang
Apparently, verbs with complex syllable structure are denominal. Looking
at the gap – that is, verbs with complex syllable structure but no correspond-
ing nouns – we can perceive a picture where verbal stems can only have
simple syllable structure – unless they are denominal conversions. Without
getting too deeply into matters of Dutch morphology, two options present
themselves.
a. The Dutch lexicon assigns a verbal category only to roots with a
simple syllable structure. This claim is odd even within a lexicalist
framework, because it regulates the category membership of a root, a
matter ultimately tied to concepts (Acquaviva and Panagiotidis 2012),
according to syllable structure.
b. Root forms are inserted late: roots in the complement of v are spelled
out as forms with a simple syllable structure. Restrictions on the form
of roots categorized as verbs are the result of Dutch grammar-internal
requirements on forms.
 3.10 Conclusion 77

If the latter is the case, then the Dutch restrictions, which have to do with the
exponence of roots and categorizers in all cases, are not deep generalizations
about the nature of roots but constraints that are morphological in nature: they
are surface requirements, rather than the result of a universal principle that
roots be lexically categorized.

3.10 Conclusion
Now that we have carefully separated the workings of syntax from those of
morphology, we have a complete picture of how lexical decomposition and,
more relevant to our topic of category here, syntactic categorization work: the
picture is more or less complete. Thus, we can replace the outline in (26):
(26) Lexical/‘morphological’ Syntactic
word-making word-making

‘Raw any category or a particular category

material’ acategorial roots
Meaning idiosyncratic meaning predictable meaning
Productivity limited full (allomorphy
aside)
Examples –ful attachment, compounding –ness attachment
with something like the outline in (27):
(27) Working with roots
Syntax: (inside nP/vP) Working with nP/vP

‘Raw Roots nP/vP

material’
Meaning idiomatic usually compositional
Productivity depending on Vocabulary almost complete
Examples –ful attachment, compounding, –ness attachment, tape,
verbs hammer and digést, verbs tape and digést,
lexical causatives syntactic causatives
Now it is time to turn to what ‘makes’ n and v, to closely examine categorial
features themselves.
4 Categorial features

4.1 Introduction
This chapter contains the core of the proposal presented here, one that weds
syntactic categorization with a new explanatory theory of categorial features
as LF-interpretable entities. Interpretable categorial features, borne by
categorizing heads, define the fundamental interpretive perspective of their
complement, thus licensing root material. The discussion begins in Section 4.2
with a recasting of the question regarding the difference between nouns and
verbs as one about the difference between n (the nominalizer) and v (the
verbalizer). Section 4.3 introduces categorial features [N] and [V] as
perspective-setting features and in Section 4.4 Embick and Marantz’s (2008)
Categorization Assumption is used as a guide in order to explain the role of
categorization induced by categorial features, its nature as well as why it
is necessary to categorize roots. In Section 4.5, calling upon evidence from
semi-lexical heads, it is argued that categorizers are not functional heads but,
on the contrary, the only lexical heads in a grammar, as they are the only
elements that can categorize root material, setting a perspective on it. Perhaps
more crucially, it is also argued that categorizers are the only necessary
elements, even in the absence of any root, on which functional superstructures
can be built. Section 4.6 examines empirical evidence from languages like
Farsi in order to put to the test the view that perhaps [N] and [V] could be more
closely related to each other, with [N] possibly being the default value of a
perspective-setting feature.

4.2 Answering the old questions

In the previous chapter, we saw that the syntactic decomposition approach can
answer a number of questions regarding lexical categories: the source of
meaning idiosyncrasies in noun–verb pairs, the (im)possibility of conversion,
the kind of interpretations that nouns or verbs derived from already categorized

78
 4.2 Answering the old questions 79

elements can and cannot have, the syntactic role of roots within the First
Phase – and so on. We also saw that syntactic decomposition elevates
categorizers such as n and v to a pivotal role in the derivation of nouns and
verbs respectively. Consequently, what we call nouns and verbs are essentially
(subconstituents of) nPs and vPs. Before moving on, let us illustrate this point,
based on Marantz (1997, 2000, 2006) and as discussed in the previous chapter.
Take a verb like bake. Essentially, the verb is a subconstituent of a vP.
(1)

However, saying that nouns and verbs are really (parts of) nPs and vPs does
not answer a fundamental question, the one we set out to answer in the first
chapter. More precisely: suppose that a noun is a morphological unit created
around the head of nP, an n constituent typically, and a verb a morphological
unit created around the head of vP, a v node in typical cases as in (1) above.
If the above is the case, then the problem of what distinguishes nouns from
verbs now becomes one of what distinguishes n from v.
(2) What distinguishes n from v? What is their difference at LF?

In other words, the long-standing conundrum of how lexical categories are

defined and in what ways they differ from each other (Baker 2003, chap. 1) is
now recast as what makes categorizers different from each other. Although
this question is still one that has been largely overlooked, answering it is
essential: otherwise we are merely rephrasing the problem of category as
one of the n–v distinction. Plainly put, even if we conceive nouns and verbs
as syntactically decomposable entities within nPs and vPs, the syntactic
difference between ‘nouns’ and ‘verbs’ remains real and cross-linguistically
persistent and apparently rooted in some essential property of language;
therefore, this syntactic difference now needs to be expressed as a (fundamental)
difference between categorizers n and v.
At this very point, we can obviously follow the received formalist approach
to this kind of problem. Instead of looking for distinctive features of nouns and
verbs, now that we have adopted categorizers and we have outlined a process
80 Categorial features

of syntactic categorization, the simplest hypothesis would be to look for

distinctive features that differentiate n from v. The most straightforward
version of this approach would be that each categorizer (n and v) bears a
distinctive feature (or a distinctive feature value). Therefore, we need to
identify
a. the distinctive feature or feature values on categorizers and
b. its (or their) interpretation.
Keeping with the established notation in the research on categories, let us call
these features [N] and [V] and let us take categorial features as the distinctive
features on categorizers. Regarding this matter, an alternative methodological
path to take would be to first seek the precise interpretation for each of the
features on n and v – as we will set out to do in this chapter anyway – and then
notate them accordingly, discarding the old [N] and [V] labels. However, I will
adhere to [N] and [V] for mnemonic purposes, if nothing else.
Now, if categorial features [N] and [V] truly exist and act as the features
distinguishing between n and v, they must also be interpretable at one of the
two interfaces, according to a version of Chomsky’s (1995) Full Interpret-
ation and following current assumptions on features as instructions to the
interfaces. Moreover, these features most probably somehow affect the syn-
tactic derivation, given that features are understood not only as properties and
instructions to the interfaces, but also as that which triggers (most) syntactic
operations (Chomsky 1995). What we had better not argue for is [N] and [V]
being purely taxonomic and otherwise syntactically inert ‘features’: this
would bring us back to approaches criticized in Chapter 1, which simply take
categorial features to be class-marking flags of no interpretive significance.
This would also be empirically inadequate, as nPs and vPs do appear to have
syntactic differences. Considering the above, the following options present
themselves:
(3) Approaches to categorial features
a. Categorial features are PF-interpretable: the noun–verb distinction
would be utterly superficial. This possibility can be easily discarded as,
cross-linguistically speaking, there is no unambiguous, across-the-board
phonological marking of the noun–verb distinction.
b. [N] and [V] are purely morphological and/or post-syntactic features. In
this case, categorial features would be like grammatical gender (under
some approaches to gender). Appealing to better-studied examples, if
categorial features were morphological and/or post-syntactic, they would
be germane to morphological class (e.g., noun declension and verb
conjugation) features, as these are analysed in Marantz (1991), Aronoff
 4.2 Answering the old questions 81

(1994), Embick and Noyer (2001), Arad (2005, chap. 2).1 However, the
universal relevance and cross-linguistic significance of the noun–verb
distinction extend well beyond both morphological exponence and –
crucially – word-class membership. Nouns and verbs are not mere
morphological epiphenomena: thus, a claim that [N] and [V] are
morphological and/or post-syntactic features is a highly implausible one.
c. All categorial features are uninterpretable, in a fashion similar to that in
which Case features are analysed in Chomsky (1995). If this is so, then
grammatical category would accordingly be a grammar-internal
mechanism with no direct interpretive effect. In the same way that there is
no uniform semantic interpretation for all nominative or accusative DPs,
similarly there would be no uniform interpretive identity for all nouns or
all verbs. As I will sketch below, and as implied in Chapter 1 already, this
would be undesirable given the research on the semantic differences
between (bare) nouns and (bare) verbs.
d. [N] and [V] are LF-interpretable, as suggested in Déchaine (1993) and
explicitly argued in Baker (2003). This looks like the option worth
exploring, perhaps the only one. Of course, this is a take on word classes
that has been more or less embraced by Anderson (1997) and linguists
working in functionalist frameworks, as reviewed in Chapter 1.

Concluding, even more than in a model where categorial information is

decided in the lexicon, it is necessary for categorial features to be syntactic
LF-interpretable features, acting (at least) as distinctive features on categor-
izers n and v. Thus, there is no escape from categorial features (or from syntax,
for that matter), a result already anticipated in the influential Chomsky (1970).
A second question that has to be answered – especially once we have
committed ourselves to a syntactic decomposition approach to categories,
and word-making in general – is why we cannot have bare roots inserted
directly in the syntax; that is, why the Categorization Assumption (Embick and
Marantz 2008, 6) is valid. Generally speaking, roots need to be assigned to a
category and they cannot be freely inserted in syntax, as observed in Baker
(2003, 268) and – from a slightly different perspective – Acquaviva (2009b).
Why is it the case that roots can only appear embedded within the complement
of a categorizer? An answer to why the Categorization Assumption holds,
especially a principled one, is a basic desideratum of a theory of category.
Continuing the discussion begun in Section 3.7 of the previous chapter, I will
try to show here that such an answer is not only possible, but that it can be
made to emerge from a theory of categorial features.

1
See Alexiadou and Müller (2007) for an account of such features as uninterpretable syntactic
features.
82 Categorial features

Before concluding this section, I must make a reference, albeit a brief one
and without doing justice to it, to the system of lexical categories put forth in
Pesetsky and Torrego (2004). More or less in the spirit of Stowell (1981) and
Déchaine (1993), they conceive lexical categories noun, verb and adjective as
grammar-internal entities that can be completely defined in a contextual way,
bringing together the most interesting ingredients of approaches b. and c. in (3)
above: so, nouns, verbs and adjectives are morphological epiphenomena and,
consequently, a grammar-internal business. There are no categorial features; in
which of the three forms a predicate manifests itself is regulated contextually –
that is, by the superimposed syntactic structure, but also locally, according to
the features and selectional requirements of a very low tense head, a TO:
(4) Contextual determination of lexical categories (in Pesetsky and Torrego
2004, 525):
Predicates are morphological verbs when associated with a TO that seeks
uninterpretable Tense features (i.e., Case, in the Pesetsky–Torrego system);
predicates are morphological nouns when associated with a TO that seeks
interpretable Tense features;
otherwise, predicates are morphologically adjectives.

The approach is very elegant and simple; it also fleshes out the programmatic
desire we briefly mentioned in Chapter 1 to strip lexical categories of all kinds
of interpretive characteristics. However, I think that a fully contextual defin-
ition of lexical categories, to the extent that it can be implemented in the way
of (4) or otherwise, is undesirable for exactly this reason: it strips lexical
categories of all interpretive characteristics, which – let me repeat this – results
in missing generalizations of paramount importance. So, I will pursue the goal
set out here, to discover the LF-interpretation of categorial features.

4.3 Categorial features: a matter of perspective

Let us now substantiate our hypothesis that [N] is the distinctive feature on the
categorizer n and [V] the distinctive feature on the categorizer v.
We can start doing so by siding with Déchaine (1993) and Baker (2003) in
stating that the features [N] and [V] must be LF-interpretable. Of course, if [N]
and [V] are truly LF-interpretable features, then we have to consider their LF
interpretation. So, we are finally facing the really tough question: what makes a
noun a noun and a verb a verb?
Let us approach the problem first from the level of categorizers themselves.
Apparently, an [N] feature is the one common to all different types of n head
and a [V] feature the one common to all different types of v head. Hence,
 4.3 Categorial features: a matter of perspective 83

by hypothesis, different n heads, bearing [N], and different v heads, bearing

[V], would also bear different additional LF-interpretable features. To wit,
Folli and Harley (2005) claim that three ‘flavours’ of v exist – that is, vCAUS,
vDO and vBECOME – making different types of verbs; these would all share a
[V] categorial feature, but each entry would bear different additional features.
A similar state of affairs would hold for Lowenstamm’s (2008) different
flavours of n: they would all share an [N] feature but differ in their other,
additional features. Therefore, [N] and [V] would encode at LF some inter-
pretive instruction common to all n and v heads respectively: exemplifying
on v, the interpretation of [V] is what remains when the differences among
the purported vCAUS, vDO and vBECOME are abstracted away from. Similarly,
the interpretation of [N] would be what remains when the differences among the
various types of n are abstracted away from.
Following the received path towards attacking the problem of the
LF-interpretation of [N] and [V], we are brought back to the point of asking
about the semantics of grammatical category: what it means to be a noun, what
it means to be a verb. Quickly revising the discussion in Chapter 1, the table in
(5) summarizes the differences between nouns and verbs:

(5) Nouns versus verbs

NOUNS VERBS

Number Tense
Case-marked Case-assigning
gender and so on agreement with arguments
argument structure covert argument structure overt
determiners particles

From the discussion in Chapter 1 it is also worth recalling two simple but crucial
points. First, ‘prototypical’ members of each lexical category (e.g., rock or tree
for nouns) share exactly the same grammatical properties as ‘non-prototypical’
ones (e.g., theory, liberty and game for nouns); this is treated in length in
Newmeyer (1998, chap. 4). Second, the same concept, as sometimes expressed
by an identical root, like sleep, can appear both as a noun and as a verb.
As also announced in Chapter 1, and following Baker (2003, 296–7), we
have to bear in mind that category distinctions must correspond to perspectives
on the concepts which roots and associated material are employed in order to
express: category distinctions are certainly not ontological distinctions,
whether clear-cut or fuzzy. Crudely put:
(6) conceptual categorization 6¼ linguistic categorization
84 Categorial features

Thus, although all physical objects are nouns cross-linguistically, not all
nouns denote concepts of physical objects (David Pesetsky, personal com-
munication, September 2005): thus, rock and theory cannot belong together
in any useful, or even coherent, conceptual category. The concepts expressed
by rock and theory can, however, be viewed by the Language Faculty in the
same way. This would entail that grammatical categories, such as ‘noun’ and
‘verb’, are particular interpretive perspectives on concepts, that there is a way
in which rock and theory are treated the same by grammar, even if they share
no significant common properties notionally. This stance is essentially taken
in Langacker (1987), Uriagereka (1999), Baker (2003, 293–4) and Acqua-
viva (2009a, 2009b). Finally, understanding categorization as grammar
imposing interpretive perspectives on concepts, we can tackle the question
I raised in Section 1.2.6 of Chapter 1 – namely, what conceptual mechanism
decides which category concepts are assigned to. The reply is ‘grammar does
the categorization’, giving us sleep the noun and sleep the verb, built from
the same concept (and root), but encoding different interpretive perspectives,
or even cross-category near-synonyms like fond the adjective and like
the verb.2
Building on the above points, we can now turn to the question of what the
actual interpretations of [N] and [V] are. The question is of course now recast
as one regarding the different interpretive perspectives that categorial features
impose on the material in their complement. In principle, we could adopt
Baker’s (2003) interpretations of [N] as sortal and [V] as predicative.
However, as discussed in the Appendix, there are several issues with the
way Baker imports sortality into his system and links it with referentiality.
Even more damagingly, [V] as a feature encoding predicativity that also forces
the projection of a specifier is also multiply problematic. I am therefore
departing from Baker’s interpretation for the two categorial features and
I am proposing the following alternative interpretations for [N] and [V]:
(7) LF-interpretation of categorial features
An [N] feature imposes a sortal perspective on the categorizer’s
complement at LF.
A [V] feature imposes an extending-into-time perspective on the categorizer’s
complement at LF.

The statements above will now be explained and discussed thoroughly. To

begin with, here I will diverge from Baker’s interpretation of sortality, which

2
I wish to thank an anonymous reviewer for comments on categorization.
 4.3 Categorial features: a matter of perspective 85

principally revolves around the criterion of identity, and I will explore instead
the notion of sortality as implemented in Prasada (2008) and Acquaviva (2009a).
Prasada (2008) notes that sortality incorporates three criteria: application,
identity and individuation. The criterion of application ‘means that the repre-
sentation is understood to apply to things of a certain kind, but not others.
Thus, the sortal dog allows us to think about dogs, but not tables, trees, wood
or any other kind of thing’ (Prasada 2008, 6). In this respect, the criterion of
application differentiates (sortal, but not exclusively) predicates from
indexicals like this and from elements with similar functions. The criterion
of application also incorporates the received understanding of bare nominal
expressions as kinds in Chierchia (1998), as it ‘provides the basis for thoughts
like dogs, [which] by virtue of being dogs, remain dogs throughout their
existence’ (Prasada 2008, 7), for as long as external conditions permit them
to maintain their existence (for a short time, like puppy, or for a long one, like
water and universe). Very interestingly, this is precisely the meeting point with
the intuitions in prototype theory and in functionalist literature (see Chapter 1)
as far as the ‘time stability’ of nouns is concerned: while it turns out that
concepts denoted by nouns are not themselves necessarily time-stable – as
cogently pointed out in Baker (2003, 292–4) – nouns are, however, viewed by
the Language Faculty as time-stable, irrespective of the actual time stability of
the concepts they denote. This is where the notion of interpretive perspective
becomes crucial.3
Regarding the criterion of identity, I will adopt its reinterpretation in
Acquaviva (2009a). The discussion in Acquaviva (2009a, 4) goes like this: if
we take a kind (e.g., the kind person), it has instances (i.e., persons) which are
particulars and which do not themselves have instances. In this way, being a
person is different from being tall: only the property person identifies a type of
entity. At the same time, the property of being tall is characteristic of all the
entities it is true of, but it does not define a category of being. This distinction
leads us back to Baker’s (2003, 101–9) discussion of the criterion of identity:
the criterion of identity essentially defines something which may replace A in
the relative identity statement ‘x is the same A as y’. Acquaviva (2009a, 4)
continues by pointing out the following: a concept that defines what it means

3
Acquaviva (2009a, 1–5) contains more detailed and in-depth discussion of nominal concepts as
such, which goes beyond this sketch of the two criteria of application and identity. Here we will
be satisfied with Prasada’s (2008) two criteria of application and identity as being enough to
define sortality for our purposes: namely, exploring what ‘nominality’ means in grammar and –
crucially – what the interpretation of a nominal feature [N] is.
86 Categorial features

to be an entity of a particular kind functions as a condition of identity for the

corresponding kind.4 I would therefore argue that this is how the feature [N]
imposes a sortal interpretive perspective: it enables a concept – say, rock or
even sleep – to act as a condition of identity.5 Summarizing and simplifying,
the criteria of application and identity taken together adequately characterize
the sortal interpretive perspective that [N] features impose on the concept they
associate with, an association mediated by syntax.6
Let us now turn to the interpretation of [V]. We will once more be calling
upon the intuitions framed within prototype theory and incorporated in the
functionalist literature on word classes. In work by Givón (1984, chap. 3), who
conceives categories as prototypes along a continuum of temporal stability
after Ross (1973), verbs are placed at the least time-stable end of the spectrum
(see also Chapter 1). The intuition that the temporal perspective seriously
matters for the interpretation of verbs is also echoed in Ramchand (2008, 40;
emphasis added): ‘procP is the heart of the dynamic predicate, since it repre-
sents change through time, and it is present in every dynamic verb’.7 The very
same intuition is explored in more detail in Uriagereka (1999). His approach to
nouns and verbs can be encapsulated along the lines of both nouns and verbs
corresponding to mathematical spaces of various dimensions; the difference
between them is conceived as whether those spaces are seen as permanent or
mutable. Uriagereka’s approach incorporates three crucial ingredients: first, the
temporal dimension argued for by functionalists as a crucial factor in the
distinction between nouns and verbs. Second, Uriagereka also upholds and
expands Langacker’s (1987) introduction of spaces in our understanding of
lexical categories – recall that he defines nouns as uninterrupted ‘region[s] in
some domain’ (Langacker 1987, 189) and verbs as processes. Third, as the use
of seen stresses in the précis above, Uriagereka conceives lexical categories as

4
Acquaviva distinguishes between a criterion, which is a necessary and sufficient condition, and a
condition. I am here using both terms loosely and interchangeably.
5
When not associated with a concept, [N] would trigger a pronominal interpretation; see Pana-
giotidis (2003b, 423–6). As an anonymous reviewer points out, if [N] is about identity and if it is
present inside pronouns, then it is hard to explain the workings of expletive pronouns, such as it
in it rains. I have no coherent answer to this problem.
6
The criterion of individuation – namely, that ‘two instances of a kind are distinct because they
are the kinds of things they are’ (Prasada 2008, 8) – does not apply to mass nouns. However, it
may play a role in the object bias in the acquisition of nouns (Bloom 2000, chap. 4) and the
perceived prototypical character of objects over substances in terms of nominality.
7
There is no one-to-one correspondence between Ramchand’s (2008, 38–42) process projection
and the verbalizer projection here. However, it is crucial that procP, which ‘specifies the nature
of the change or process’ (40), is understood as the essential ingredient of every dynamic verb.
 4.3 Categorial features: a matter of perspective 87

being different perspectives on concepts, as different ‘grammaticalisations’ of

concepts, to recall Anderson (1997). Zooming in on verbs, Uriagereka begins
his treatment with the more or less received lore that ‘themes are standardly
nouns’ and that ‘verbal elements [are] functions over nouns’. He brings these
two statements together by claiming the following:
(8) A verb expresses the derivative of its theme’s space over time.

Here, then, we have two essential ingredients for the interpretation of categor-
ial features: perspective-setting, and the relevance of temporality (as opposed
to predicativity) for verbs and their distinctive feature.8 In a similar vein,
Acquaviva (2009a, 2) notes that because ‘verbal meaning is based on event
structure [. . .], it has a temporal dimension built in. Nominal meaning, by
contrast, does not have a temporal dimension built in.’ If we replace ‘meaning’
with ‘perspective’ in Acquaviva’s quote, we can make the claim that [V]
encodes an actual perspective over the concept with which it is associated
and that this perspective is of the said concept as extending into time: this is
why verbs and their projections are the basic ingredients of events. From this
point of view, we can actually call Vs and VPs subevents, with the feature [V]
contributing the temporal perspective to event structures.
Some consequences of the way features [N] and [V] are interpreted, if we
go by (7), include the following: first of all, we can now explain why objects –
but, also, substances – are typically conceived as sortal concepts in the way
sketched above; they smoothly satisfy both criteria of application and iden-
tity. This is compatible with the canonical mapping of such concepts onto
nouns cross-linguistically and – as already pointed out – object and substance
concepts are typically expressed as nouns. At the same time, together with
Uriagereka (1999) and Baker (2003, 290–5), we expect dynamic events
(activities, achievements, accomplishments) to be conceived typically as
extending into time, hence the canonical mapping of such concepts onto
verbs. If, on top of everything, dynamic events are compositionally derived
from states and states are, very roughly, equivalent to VPs, then a theory of
event structure, such as the ones in Rappaport Hovav and Levin (1998),
Borer (2005), Levin and Rappaport Hovav (2005) and Ramchand (2008),
receives added justification along the lines of verbal constituents being

8
I have long thought that Langacker (1987) and Uriagereka’s (1999) understanding of nouns as
regions and spaces could be unified with Acquaviva’s (2009a) treatment of nominal concepts as
‘unbroken’ – that is, as having no temporal parts (as discussed below). However, I presently can
offer no true insight on this prospect.
88 Categorial features

inherently (sub-)eventive by virtue of the temporal perspective contributed

by the categorial feature [V].
Furthermore, Tense and Voice exclusively combine with verbal projections
(Tonhauser 2007). This generalization can now easily be associated exactly
with the kind of perspective a [V] feature imposes on the concept it is
associated with, through syntactic categorization of the root material. The root,
along with everything else within the complement of v, will be interpreted as
extending into time. In a similar fashion, Number combines with nouns. This
can now be linked with the perspective an [N] feature imposes on the comple-
ment (typically containing a root) of an n: that of a sortal predicate satisfying
both the criteria of application and identity. Interestingly, the association with a
Num head, or similar, guarantees that the sortal concept in question – for
example, tree – will also fulfil the third sortal criterion, that of individuation, as
individuation can be coerced upon a wide range of ‘inherently’ non-countable
concepts by classifiers and Number (Borer 2005).
An informal example illustrating how this difference in interpretive per-
spective works can be glimpsed by looking at the pair sleep the noun and
sleep the verb. Both words can be said to encode the same concept. However,
the noun sleep forces the viewing of this concept to be a sortal one, which is
to say that the perspective over the concept of sleep that the nominal category
imposes is of sleep as some type of virtual object or substance. We can
therefore lose our sleep (like we lose our keys or lose blood), get more sleep
(like we can get more air, food or water); we can also talk about morning
sleep being different or sweeter than early evening sleep, and so on. On the
other hand, the verb sleep forces us to view the concept of sleep as a
subevent, as extending into time, which readily offers itself to temporal
modification and so on. Similarly, the perspective over the concept of sleep
that the verbal category imposes is of sleep being expressible as a time
interval, as something potentially having duration (long, short etc.). This
nicely ties in with the following long quote from Acquaviva (2009a, 2–3)
(emphases added):
. . . nouns allow speakers to describe events unfolding in time as if they were
complete entities that acquire and lose properties over time. We know that
argument or wedding describe events, but we can speak of them as if they
could undergo changes in time, as in (1):
(1) a. the argument was calm at first, then it became heated (Simons
1987, 134)
b. the wedding moved from the church to the bride’s parents’
house
 4.4 The Categorization Assumption and roots 89

Speakers know that only a part of the event had a property and another part
had the other property, but this is disregarded in a structure which predicates
two contradictory properties of the same subject. The nature of the nouns’
referents as occurrents is only disregarded, not changed; this becomes obvi-
ous when we explicitly state that the whole subject is there at a given time:

(2) a. the iron became heated (all of it) 6¼ the argument became heated
(all of it)
b. the wedding moved from A to B 6¼ they married first in A then
in B

In (2a), all of it may not refer to all of the event’s temporal parts . . . In (2b),
the right-hand side is not a good paraphrase of the left-hand side because
marry is a telic verb and so the sentence entails that the event is completed,
first in A, then in B. To explicitly describe the referents of argument and
wedding as lacking temporal parts, thus, conflicts with their lexical semantics.
Yet the sentences in (1) are natural even though they sideline temporal
constitution. Reference to temporal constitution is thus inessential for nouns
referring to occurrents. This, then, is a clear difference: verbal reference has a
temporal dimension built in (in terms of actionality, not tense; this applies to
permanent states as well as to bounded events); nominal reference does not,
and can do without such a dimension even when referring to occurrents.

In any case, the truly important part of (7) is that [N] or [V] on n and v encode
different perspectives, rather than different inherent properties of the concept
itself as expressed by the root. This is actually a most welcome consequence of
syntactically decomposing lexical categories in the light of pairs like N work –
V work, which brings us to the answer of what distinguishes n from v:

(9) Categorial features [N] and [V] are interpretable on the n and v categorizers
respectively.

As we have seen, categorial features provide the fundamental interpretive

perspective through which the meaning of the categorizer’s complement will
be negotiated. This will become especially relevant in the context of discussing
the behaviour of roots below and in deriving Embick and Marantz’s (2008, 6)
Categorization Assumption, a discussion that began in Section 3.7 of the
previous chapter.

4.4 The Categorization Assumption and roots

4.4.1 The Categorization Assumption
We turn next to the Categorization Assumption and examine it from the angle
of how categorizers – their categorial features actually – enable the inclusion of
90 Categorial features

roots in syntactic derivations that can be interpreted at the interfaces. Illustrat-

ing the Categorization Assumption below, we see that a root cannot be directly
merged with a functional head – say, a Number or a Voice head: such NumPs
or VoicePs are illicit.
(10)

This generalization is supported by a cursory look at languages where

roots are morphologically bound morphemes, like Italian and Spanish: they
can never combine with any functional material in the absence of categor-
izing heads.9 Consider the situation in Greek. Roots emerge as stems like
aer– (‘air’), which are bound morphemes in the language. The impossi-
bility of a structure like *[NumP Num root] is very difficult to demonstrate,
because Greek is a fusional language and nominal endings are typically the
exponence of fusion between n, Number and Case (or ‘Kase’). However,
there are two facts to consider with respect to (10). The first is this: Greek
possesses a linking form, typically expressed as –o–, which satisfies the
morphological requirement for Greek stems to be bound but encodes no
grammatical (or other) content. It is typically used in root–root
compounding, where two roots merge together and are subsequently
categorized:10
(11) aer-o-vol-a aer-o-vol- n.neut.pl ‘airguns’
aer-o-por-os aer-o-por-n.masc.sg.nom ‘aviator’
aer-o-gram-i aer-o-gram-n.fem.sg ‘airline’

This –o– morph is an elsewhere form and its sole purpose is to satisfy the
requirement that Greek stems be morphologically bound. Interestingly, if
that was all that there was – that is, a morphological restriction on stems –
acategorial roots could easily be part of syntactic structures by getting –o–
attached onto them as stems: we would expect –o– to have a wider
distribution. However, –o– is illicit in all environments where the root
is expected to be categorized – that is, where the roots will then have
to combine with functional structure. Thus we have pairs like the
following:

9
Perhaps it might be possible to correlate the cross-linguistic avoidance of roots as free
morphemes with the Categorization Assumption itself; however, this is not a claim I would
make here.
10
N here stands for ‘nominal’.
 4.4 The Categorization Assumption and roots 91

(12) vrox-o-lastix-a
vrox-o-lastix-n.neut.pl
‘rain tyres’

lastix-a vrox-is
lastix- nom.neut.pl vrox-n.fem.sg.gen
‘rain tyres’

Before protesting that (12) simply adumbrates the morphology–syntax

competition (Ackema and Neeleman 2004), we must consider work by
Panagiotidis, Revithiadou and Spyropoulos (2013), which discusses Greek
verbalizers: these are overwhelmingly expressed morphologically, and they
make distinct verbal stems out of roots.11 In the case of the root aer, the
verbalizer is expressed as –iz– and combines with the root, yielding the
unambiguously verbal stem aeriz– (‘to air’). So, although the fact that roots
in Greek must emerge as morphologically bound stems can be bypassed by
using –o–, roots still also need to be categorized as nominal and – visibly – as
verbal.
Hence, the descriptive observation that roots must be categorized is already
made in the literature but neither Embick and Marantz (2008) nor Baker (2003)
before them provide an attempt to derive it from more general principles – as
Embick and Marantz claim should be done. Now, at the beginning of this
chapter we noted that the Categorization Assumption is equivalent to an
understanding that roots cannot be interpreted unless inside the complement
of a categorizer. Having decomposed lexical categories ‘noun’ and ‘verb’ to
parts of nP and vP projections, we are left with at least the following elements
for syntax to build structures with:12
a. Roots
b. Categorizers (e.g., n and v)
c. Functional heads (e.g., Num, D, T, C . . .)
Therefore the statement that ‘roots need category’ can be expanded to the
descriptive generalization below:
(13) Roots can only be merged inside the complement of a categorizer, never
inside the complement of a functional head in the absence of a categorizer.

11
The caveat ‘overwhelmingly’ is inserted because I am not sure light verbs exist in Greek.
12
The list is incomplete and will be revised and expanded: we need also to include inner,
subcategorial, morphemes, which merge directly with root material, as seen in the previous
chapter. See Chapter 5, footnote 12.
92 Categorial features

A more traditional way to express (13) is of course that functional heads,

unlike categorizers, cannot support real descriptive content, as in Abney’s
(1987, 64–5) oft-quoted criteria. However, we must still address the real
question of why configurations like those in (10) are impossible and why
(13) holds.
In order to reach a principled answer, we need to weave together three
threads:
(14) the zero hypothesis that roots are syntactically unexceptional;

(15) an examination of the conceptual content and of the interpretation of roots as

objects – that is, by themselves;

(16) the role of categorial features [N] and [V] in the licensing of roots.

First, we take seriously the hypothesis that roots are unexceptional syntac-
tically, by being merged inside the complement of material below the
categorizers or, perhaps, by projecting their own phrases, too, as in Marantz
(2006) and Harley (2007, 2009) – but see De Belder (2011) for argumenta-
tion against roots projecting phrases. This is the zero hypothesis, as stated
in (14): if roots are manipulated by syntax, then they should behave like
all other LIs (‘lexical items’), unless there is evidence to the contrary.
Having said that, (14) goes against what we have in (10) and (13), unless
of course we try to stipulate (10) and (13) via some kind of c-selectional
restrictions. Still, it is not the case that all roots are directly selected by n
and v: as seen in Chapter 3, some roots may be deeply embedded inside
more material.
Second, the thread in (15) is about carefully considering the LF interpret-
ation of roots in isolation, so to speak, about examining their own conceptual
content; in this I will assume that the semantic content of the root is seriously
underspecified/impoverished.
Third, (16) announces that we will employ our newly developed under-
standing of categorial features on categorizers as providing the necessary
perspective (e.g., sortal or temporal) through which the root will be interpreted;
we will also explain why the absence of such an interpretive perspective
prevents roots from participating in legitimate syntactic objects. In other
words, I argue that categorizers exclusively provide the grammatical ‘context’
of Marantz (2000) for the root to be assigned an interpretation and/or a
matching entry at the interface with the Conceptual–Intentional/SEM systems.
Let us now look at the second and third threads, (14) and (15), in more
detail.
 4.4 The Categorization Assumption and roots 93

4.4.2 The interpretation of free roots

Bare roots are sometimes thought of as standing for stative predicates
(Alexiadou 2001), whereas Doron (2003) and Levin and Rappaport Hovav
(2005) assume them to contain argument structure information. This would
make roots the semantic core of verbs; for instance, a root like open would be
the ‘core element’ of the verbs in The window opened and Tim opened the
window. However, Acquaviva (2009a, 2009b) argues against any semantic
content for roots along the following lines, already familiar from Aronoff
(2007):13 as the same root can underlie a number of nouns, verbs and adjec-
tives with significantly different meanings, lexical information must necessar-
ily be root-external (Acquaviva 2009b, 5); this is something to be expected if
lexical information is assigned to grammatical structures at the interface with
the Conceptual–Intentional systems. Moreover, it is not the case that the
‘meaning’ of the root can be a (proper) subset of that of the nouns, verbs
and adjectives in the derivation of which it participates. To wit, Aronoff (2007,
819) observes that ‘words have morphological structure even when they are
not compositionally derived, and roots are morphologically important entities,
[even] though not particularly characterized by lexical meaning.’ He supplies
the example of the Hebrew root kbš (typically rendered as bearing a meaning
like ‘press’):
(17) kbš
Nouns:
keveš (‘gangway’, ‘step’, ‘degree’, ‘pickled fruit’)
kviš (‘paved road’, ‘highway’)
kviša (‘compression’)
kivšon (‘furnace’, ‘kiln’)

Verbs:
kavaš (‘conquer’, ‘subdue’, ‘press’, ‘pave’, ‘pickle’, ‘preserve’, ‘store’, ‘hide’)
kibeš (‘conquer’, ‘subdue’, ‘press’, ‘pave’, ‘pickle’, ‘preserve’)

We realize that there is no way in which ‘step’, ‘degree’, ‘furnace’, ‘pave’,

‘highway’, ‘pickle’ and ‘conquer’ can be compositionally derived from a root
with conceptual content along the lines of ‘press’, or similar. If the words in
(17) are all ‘lexical idioms’, then we are better off arguing, as Marantz (2000),
Aronoff (2007) and Acquaviva (2009a, 2009b) essentially claim, that all
lexemes are idioms. In Levinson’s (2007) analysis of roots as polysemous,

13
Acquaviva also discusses the issue of morphological class membership information and the
problems arising if we argue that it is (not) encoded on the root.
94 Categorial features

all these words with their diverse and unrelated meanings should be listed as
possible interpretations of the root KBŠ. However, if roots themselves are
polysemous, then a root (say, KBŠ) must also lexically encode information on
which of its many meanings is available in which syntactic environment.
Whatever the precise amount of semantic content that characterizes roots,
what is important for the discussion here is that roots on their own have
minimal semantic content or, as Arad (2003; 2005, chap. 3) proposes, that
they are severely underspecified. This could be understood to result in free
roots not being adequately specified to stand on their own as legitimate LF
objects.14 Consider that syntactic objects at LF – say, a vP phase – consist
purely of interpretable and/or valued UG features and roots. Now, by hypoth-
esis, syntax uses the operation Agree in order to eliminate uninterpretable and/
or unvalued UG features before the phase is completed (i.e., before the
derivation is sent off to the interface). But what about roots? Roots are possibly
UG-extraneous elements. Even if this claim is too strong, roots do not form
part of UG, as they can be borrowed or even coined, pre-theoretically speak-
ing, at least. Roots are essentially ‘imported’ into the syntactic derivation.
However, FLN (the Faculty of Language in the Narrow sense) must somehow
manipulate roots, apparently in order to be able to express a variety of concepts.
If no roots are manipulated by FLN in a particular derivation, we get expres-
sions made up exclusively of UG features like ‘This is her’, ‘I got that’, ‘It is
here’ and so on (cf. Emonds 1985, chap. 4). Therefore, the ability of FLN to
manipulate roots enables it to denote concepts and, ultimately, to be used to
‘refer’ (Acquaviva and Panagiotidis 2012).
However, roots being possibly extraneous to FLN, and given that they
probably do not contain any UG features (as is the emerging consensus), they
need to be somehow dealt with by FLN: categorization is exactly the way this
is achieved. In a nutshell: uncategorized roots are FLN-extraneous; either just
because of this or also because they are semantically underspecified them-
selves, uncategorized roots would not be recognized at the interface between
syntax and the Conceptual–Intentional/SEM systems.

14
The semantic underspecification of (uncategorized) roots can be understood as the reason
they are not legitimate objects at LF, but more needs to be said on the matter. For instance,
Horst Lohnstein (personal communication, 2010) points out that a root like spit could still
be interpretable in some sense even if it is underspecified, unlike kbš; the ‘basic meaning’ of
spit as a root is more clearly circumscribed. However, even a relatively straightforward root
like spit means much less when it is uncategorized than, for example, the verb it names. See
Rappaport Hovav and Levin (1998) for discussion; see also Acquaviva and Panagiotidis
(2012).
 4.4 The Categorization Assumption and roots 95

The specific claim made here is that syntax does not use a special operation
to ‘acclimatize’ roots but embeds them within a categorizer projection, whose
categorial features provide an interpretive perspective in which Conceptual–
Intentional systems will associate the root with conceptual content. The short
of it is that
(18) (Free/uncategorized) roots are not readable by the Conceptual–Intentional/
SEM systems.

4.4.3 The role of categorization

Let us now take up the third thread, that in (16), and look at how categorial
features license roots. In (18) we concluded that free roots are unreadable by
the Conceptual–Intentional/SEM systems and in (13) we had rephrased
Embick and Marantz’s Categorization Assumption as a need for roots to
always appear inside the complement of a categorizer. We now must clarify
how categorization – that is, embedding roots inside the complement of a
categorizer – cancels out the roots’ LF-deficient character and allows them to
participate in LF-convergent derivations.
By (7) and (9), the way this is achieved is that the categorial feature [N] or
[V], on n or v, provides the interpretive perspective in which to interpret the root.
At the same time, (13) is trivially satisfied because no functional heads bear
interpretable categorial features (more on this in Chapter 5).15 This picture is
completed by what is pointed out in the very definition of the Categorization
Assumption (Embick and Marantz 2008, 6): ‘If all category-defining heads are
phase heads . . . the categorization assumption would follow from the general
architecture of the grammar . . .’.Given that roots cannot be directly embedded in
the context of a functional phase head like C – compare the discussion of (10) –
I would be tempted to think that the phasehood of categorizers (‘category-
defining heads’) is not a sufficient condition for the categorization of roots.16
The argument goes as follows: roots are semantically impoverished when
they are free. However, we have already noted that they are syntactically
unexceptional objects: they may merge with elements such as low applicatives,

15
This is already stated in Marantz (2000): ‘To use a root in the syntax, one must ‘merge’ it
(combine it syntactically) with a node containing category information.’ I of course take [N] and
[V] to be the said ‘category information’ here.
16
Non-phasal heads could also provide a context for the interpretation of roots; categorizers could
in principle be non-phasal heads. Note that if the objections in Anagnostopoulou and Samioti
(2009) are correct, this is exactly the way things are: categorizer phrases are not phases but they
are (inside) the complements of phase heads.
96 Categorial features

low causatives, abstract prepositions, particles and what we call ‘inner

morphemes’. Roots can also participate in small clause structures and, perhaps,
they can project their own phrases, according to Marantz (2006) and Harley
(2007, 2009). Nevertheless, the resulting structures would still contain an
offending item, the LF-deficient root itself. In this sense, a ‘free’ root is simply
an uncategorized root. In order to supply a concrete example, consider the
configuration in (19), adapted from Marantz (2000, 27):
(19)

Merge creates a syntactic object from the root and its object ‘tomatoes’, with
the root projecting: a syntactically unexceptional object. However, if this is
embedded under functional structure without the licensing ‘intervention’ of a
categorizer, then the resulting syntactic object will lack interpretive
perspective, because of the SEM-deficient root Grow. This is why the
categorial feature on the categorizer, [V] on v or [N] on n, is necessary: it
assigns an interpretive perspective to the object as extending into time or as
sortal, therefore enabling the resulting vP or nP – the so-called First Phase – to
be interpreted. At the same time, the root-categorizer object, associated with
an interpretive perspective, can be matched with a vocabulary item (grow or
growth) and an appropriate ‘lexical’ concept, a ‘meaning’ (cf. Aronoff 2007).
(20)

Summarizing, categorial features on the categorizer close off material associ-

ated with the root; they do so by providing this material with a fundamental
perspective for the conceptual systems to view it in. No such perspective can
be supplied by a (phase) head without categorial features: it is [N] or [V] on n
or v that provide the necessary perspective, the ‘context’, for the root to be
interpreted. Therefore, the association of root material with categorial features
[N] and [V] enables the former to be interpreted at the interfaces and beyond.
The categorization of roots is consequently not a narrow-syntactic require-
ment, but one of the interface between syntax and the conceptual–intentional/
SEM systems. It indeed follows “from the general architecture of the gram-
mar” (Embick and Marantz 2008, 6).
 4.4 The Categorization Assumption and roots 97

4.4.4 nPs and vPs as idioms

Before turning to the status of categorizers – that is, whether they are
functional heads – a note on the systematic idiomaticity of nPs and vPs is in
order. To begin with, the interpretation of the First Phase is canonically
non-compositional: in Section 3.4 of Chapter 3 we had a cursory look at
well-known pairs such as N water – V water – A watery, N dog – V dog, N
castle – V castle, N deed – V do and so on. Moreover, as argued in that same
section, meanings associated with material such as root-v and root-n in (20) are
invariably listed and almost always idiosyncratic. This is a well-known and
widely examined fact, and recall that it is one of the key arguments supporting
the analysis of word-formation in ways different from phrase-building.17 This
canonical idiosyncrasy/non-compositionality is what tempts one to think of the
First Phase as a somehow privileged domain for non-compositional interpret-
ation and to correlate idiomaticity of material with the fact that this material
appears below the categorizer, within the categorizer’s complement. However,
we need to consider two factors, extensively discussed in Chapter 3:
a. Idiomaticity, non-predictability and non-compositionality are in part
explained away once the role of subcategorial material (‘inner
morphemes’ and the like) and argument structure are spelled out in
more detail.
b. Given the impoverished or (even) inexistent semantic import of roots
themselves, having non-compositional and idiosyncratic interpret-
ations of material in nP and vP is the only option: no function yielding
compositional interpretations could take a root as one of its operands.
Furthermore, as has been argued in the literature, idioms larger than the First
Phase do exist: see, for instance, the discussion in Nunberg, Sag and Wasow
(1994) and McGinnis (2002) – pace Svenonius (2005). So, although an
idiomatic interpretation may be associated with syntactic constituents (phases
perhaps) of various sizes, idiomaticity is the only option for first phases,
exactly because of the semantic impoverishment/deficiency of roots.
Let us then summarize the conclusions of the previous chapter in the light of
this chapter’s analysis of categorizers. The systematic idiomaticity of first
phases – that is, of categorizer phrases containing root material – is not due
to the categorizer acting as a limit, below which interpretation is or must be
non-compositional. On the contrary, idiomaticity is due to this projection

17
For overviews, see Ackema and Neeleman (2004) and Marantz (1997, 2000, 2006).
98 Categorial features

(an nP or a vP) containing a root, an LF-deficient element by (18), that would

resist acting as suitable input for any compositional function anyway. The
categorial features on categorizers supply the complement of n or v with an
interpretive perspective, within which the whole projection will be matched
with a concept when the phase is interpreted. The perspective-setting categor-
ial features are necessary for the matching of root material with a concept at
LF, as no other phase head can take a free (i.e., ‘uncategorized’) root as its
complement. Therefore, inner versus outer morphology phenomena (Marantz
2006) are due to the semantic impoverishment of roots: once roots have been
dispatched to the interfaces, along with the rest of the complement of the
categorizer, compositional interpretation may canonically apply in the next
phase up.

4.5 Categorizers are not functional

Based on the above discussion, we can distil our analysis of categorial features
on categorizers as follows:
(21) Categorial features
a. contribute the interpretive perspective phase-internally, and
b. identify the whole phase externally (as ‘nominal’, ‘verbal’ etc.).

So, categorial features on categorizers enable roots to participate in the deriv-

ation, essentially empowering FLN to manipulate concepts not encoded by
UG-features – that is, the vast majority of them. Furthermore, categorial
features form the basis on which functional structure is built on, as will be
demonstrated in the following chapter.
Although they are phase heads, categorizers are by no means functional
heads, like Complementizer or Determiner. At the first level of empirical
inquiry, it already becomes evident through the fact that no functional head
can categorize roots and root material. This is true both for ‘major’ functional
categories such as Voice, Aspect, Tense, Complementizer, Number or Deter-
miner (recall (10) above) and for those subcategorial elements that Marantz
(2000, 2006) terms ‘inner morphemes’. I think that precisely here lies the
difference between ‘lexical’ and functional’: only the former can categorize a
free root in its complement. So, effectively
(22) There is only one class of ‘lexical elements’ that qualify as atomic
‘verbs’: v heads.

(23) There is only one class of ‘lexical elements’ that qualify as atomic
‘nouns’: n heads.
 4.5 Categorizers are not functional 99

Hence, contrary to the received perception of them, categorizers are not

functional heads; they are the only lexical heads in existence. A way to
systematically capture the above observations is to adopt Categorial
Deficiency for ‘major’ functional categories (Panagiotidis 2002, 170–83),
something to be scrutinized in the next chapter. However, the distinction
between lexical and functional can already be recast as follows.
(24) Lexical heads can support roots, by categorizing them. Functional heads
cannot categorize roots; therefore they cannot be directly associated
with roots.18

The fact that what we used to classify as ‘lexical heads’ are in reality the
categorizers themselves can be glimpsed from a very privileged angle once
we consider semi-lexical categories (Corver and van Riemsdijk 2001).
Semi-lexical elements are lexical nouns and verbs that do not carry any
descriptive content, including English one (as in the right one), ‘empty
nouns’ (Panagiotidis 2003b) and at least some types of light verbs. Emonds
(1985, chap. 4) already analyses semi-lexical elements – which he calls
grammatical nouns and grammatical verbs – as instances of N and V heads
without any descriptive, concept-denoting features. This line of analysing
semi-lexical heads is taken up and developed in van Riemsdijk (1998a), Haider
(2001), Schütze (2001) and Panagiotidis (2003b). I think it is very interesting
that Emonds’ definition, dating from 1985, of such elements as lexical elem-
ents that only bear formal features is exactly what we would think categorizers
themselves are in a lexical decomposition framework: consider, for instance,
Folli and Harley’s (2005) postulation of three v heads caus, do and become.
Indeed, Emonds (1985, 159–68) claims that different lexical entries of gram-
matical nouns and grammatical verbs can be distinguished from each other by
virtue of their formal features only, as they are completely devoid of any
descriptive content.
It seems, then, that when we deal with semi-lexical elements, the root
supplying the descriptive content – whichever way it does it – is absent. So,
a first straightforward conclusion would be that in order to have a ‘noun’ (nP)

18
As a consequence, adpositions and functional elements such as determiners should not contain
roots. Received lore within Distributed Morphology would indeed take all functional elements
to be bundles of features realized by Vocabulary Items. As for adpositions, the lexical material
they contain – that is, the ‘heavy’ nominal and adverbial material briefly mentioned in Section
2.9 of Chapter 2 – must already be categorized. I am grateful to Carlos de Cuba (personal
communication, April 2008) who first raised this question in 2008, and to an anonymous
reviewer.
100 Categorial features

or a ‘verb’ (vP) a root is not necessary, whereas categorizers always are. This is
a conclusion that Harley (2005a) arrives at, in her treatment of one: she
convincingly shows that one is, precisely, the Vocabulary Item inserted when
an n head is not associated with a root, under certain morphosyntactic
conditions.
The above strongly suggests that (category-less) roots and subcategorial
material are – syntactically speaking – optional and that a well-formed syntac-
tic representation can be constructed using just categorizers and a functional
structure superimposed on them: we can build syntactic structures using only a
categorizer as its ‘seed’; roots are not necessary. We will return to this
observation in the next chapter.

4.6 Nouns and verbs

According to the account presented in this chapter, a nominal feature [N] is
completely different from a verbal feature [V]. [N] and [V] are presented as
different attributes, as also suggested by the different interpretive perspectives
they impose, rather than different values of the same feature. But is this really
the case? Is there a principled reason to prevent them from being the two
values of a categorial feature, a [perspective] feature, more concretely? Alter-
natively, what prevents nominality from being the default perspective,
resulting from a [V] feature or from the absence of a [V] feature?19
Kayne (2009) makes a very interesting proposal along the lines of nouns
being the unmarked category. He applies antisymmetry (Kayne 1994) to the
lexicon to argue that there are lexical items of category x and lexical items of
the category y. Lexical items of the category x form an open class, they only
contain valued features and are not the locus of parametric variation – these are
the nouns. Nouns, by virtue of having no unvalued features, are also the only
lexical elements that can denote. Lexical items of the category y form a closed
class, they enter the derivation with unvalued features and are the locus of
parametric variation – these are ‘non-nouns’: that is, a category containing at
least verbs and aspectual heads. Effectively, Kayne (2009) takes nouns to be
the only (open) lexical category, whereas verbs are made of light verbs
(a closed class, akin to v) with nominal complements, as in Hale and Keyser’s
(1993, 55) proposal concerning laugh and similar items: ‘laugh is a noun that
in some sentences co-occurs with a light verb that is unpronounced, giving the

19
Some relevant options had already been explored by Déchaine (1993) in her three-feature
system for lexical and functional categories.
 4.6 Nouns and verbs 101

(misleading) impression that laugh in English can also be a verb. Strictly

speaking, though, laugh is invariably a noun, even when it incorporates (in
some sense of the term) into a (silent) light verb’ (Kayne 2009, 7). In a
nutshell, English ‘verbs’ are all denominal, nouns are the only lexical category
and something like v (Kayne’s y) verbalizes them. Of course, we saw in the
previous chapter that deriving all verbs from nouns is not the right way to go
and adopting such a thesis would both entail missing a number of generaliza-
tions (beginning with the hammer–tape contrast) and blurring the very useful
(if not absolutely vital) distinction between nouns and acategorial roots.
Finally, I think it would not be far removed from the spirit of Kayne’s proposal
to claim that roots are nominal by default. This is, however, a point Acquaviva
(2009a) extensively argues against.
The questions of whether [N] and [V] are simply values of a single [per-
spective] feature and whether one of them is unmarked and/or default present
a rather intricate conceptual matter, and relevant empirical evidence can
be very hard to come by. What I will do in the remainder of this chapter is
(a) provide arguments for keeping the [N] and the [V] features separate and
(b) debate to what extent [N] shows signs of being the unmarked categorial
feature.

4.6.1 Keeping [N] and [V] separate?

Although both [N] and [V] impose interpretive perspectives, perhaps sup-
plying the grounds for treating them as merely different values of a [per-
spective] feature, a sortal perspective and a temporal one are quite different
from each other. Moreover, once we have a proper theory of syntactic
categorization from roots in place, we do not need to automatically think
of nouns as more basic/elementary than verbs or to think of verbs as more
marked than nouns. Finally, the impairment of nouns and that of verbs in
aphasia are doubly dissociated (Caramazza and Hillis 1991), as would be
expected if the distinction underlying the noun–verb distinction involved
different features.
However, there are some facts that could point towards understanding [V]
and [N] as different values of the same feature, with [V] – or whatever its
proper notation would be – appearing to be more marked than [N]. An initial
piece of data pointing in this direction comes from Arad (2005, chap. 3), who
discusses the fact that nominal and verbal morphology seem to be of a different
nature in languages such as Hebrew and Russian, both inflectionally rich
languages. I will exemplify the phenomenon she discusses using Greek
examples, as the facts seem to be identical in this language.
102 Categorial features

In Hebrew, Russian and Greek nouns can be borrowed exactly as they are,
modulo phonological adaptation. Observe, for instance, the following loan-
words into Greek:
(25) Loanwords into Greek: nouns
rok ‘rock music’
solo ‘solo’
zum ‘zoom’
indriga ‘intrigue’ – from Italian intriga

At the same time, when borrowing roots to make verbs, special verbalizing
morphology must be added to these root forms in order to turn them into
legitimate verbal stems – recall the discussion in Section 4.4.1:
(26) rok-ar-o ‘I rock.’
sol-ar-o ‘I do a solo.’
zum-ar-o ‘I zoom in.’
indrig-ar-o ‘I intrigue sb.’

On the basis of similar evidence from Hebrew and Russian, Arad argues that
this asymmetric behaviour is the result of a VoiceP (as opposed to a bare vP)
being the minimum verb. She goes on to argue that although a nominalized
root is a noun, a verbalized root is not a verb: this results in the paradox of a
verbalized root being smaller than a verb. Corroborating this is the fact that, in
Hebrew, ‘roots make nouns more easily than verbs’ (Arad 2005, 56), a fact that
also seems to hold true for Greek, even beyond the domain of borrowing.20
However, taking a better look at (26) above, it turns out that the extra
morphology, the –ar– form, attached to (borrowed) roots to make verbs is
not an exponence of Voice but, rather, of v itself. Panagiotidis, Revithiadou
and Spyropoulos (2013) observe that the –ar–-type elements that make verbal
stems out of roots co-exist with Voice, Aspect, Tense and Agreement morph-
ology, and they are obligatory, even when combining with native roots: it is
impossible to form verbs from roots without one of these pieces.21 Even so, the
fact that verbalizers, v heads, overwhelmingly tend to be morphologically
expressed in Greek (and, possibly, in Hebrew and Russian, too), unlike the n
head, is something that in turn must be explained: the issue of the markedness
of verbs re-emerges.

20
Nouns vastly outnumber verbs in any given dictionary, after all.
21
There is actually a very limited number of verbs, all of Ancient Greek origin, that do not show
an overt piece of verbalizing morphology; these include graf-o (‘I write’), vaf-o (‘I paint/dye’),
trex-o (‘I run’), idri-o (‘I found/establish’).
 4.6 Nouns and verbs 103

Given this apparent markedness of verbs, would we perhaps have to express

this asymmetry by keeping the [V] feature while replacing [N] as [V], the
unmarked ‘elsewhere’ value? If this is the case, then nouns would be the
default category, exactly as we saw Kayne (2009) predict, taking nouns to be
the default category. A language that looks ideally suited to test the claim that
nouns are the unmarked lexical category is Farsi.

4.6.2 Do Farsi verbs always contain nouns?

Farsi is known for having only 100 simplex verbs (Karimi-Doostan 2008a),
although far fewer are reported to be commonly used (Peyman Vahdati,
personal communication, 2005). All other verbal concepts are expressed
via periphrastic constructions made of preverbs, predicative elements, which
combine with fourteen light verbs (Family 2008) to give Light Verb
Constructions known in the literature as Complex Predicates (Folli, Harley
and Karimi 2003). This is illustrated in the examples below (from Family
2008):22
(27) Farsi Complex Predicates with nominal preverbs
hærf zædæn word hit ‘talk’
xun ændaxtæn
Ð blood throw ‘cause to bleed’
kæm odænÐ few become ‘be diminished’
qæbulÐda tæn assent have ‘maintain as true’
arame
Ð dadæn calmness give ‘calm’
jæ
Ð n gereftæn
Ð party obtain ‘celebrate’
t opoq keÐ idæn pipe pull ‘smoke a pipe’
qorbæt ke idæn remoteness pull ‘long for home’

Farsi Complex Predicates, substituting for verbs, are multiply exciting. What
we are going to focus on here, however, is the categorial status of the preverbs.
The preverbs in (27) all look like nouns, making Farsi Complex Predicates
apparently similar to English periphrases like have/take a shower, have/take a
look, make a call/a statement/a mistake, take a picture and the like. The Farsi
situation is much more intricate. Karimi-Doostan (2008a, emphasis added)
explains why:

Some of the [preverbs] are adverbs, adjectives, prepositions or nouns but

some of them lack any lexical category and will be called Classless Words
(CWs) . . .. The adverbs and adjectives can be used in superlative and
comparative forms and they can be modified by intensifier adverbs like

22
Farsi is a subject–object–verb (SOV) language.
104 Categorial features

‘very’. The nouns can be pluralized and selected by D, demonstrative adjec-

tives and prepositions and they can function as subject and object. However,
CWs do not have any of the properties of adverbs, adjectives and nouns
referred to here. CWs are different from prepositions and simple verbs too.
Prepositions can directly select DPs, but CWs cannot. Simple verbs may co-
occur with verbal inflectional items, but CWs may not.

Karimi-Doostan (2008a) goes on to show that Classless Words are listed as

independent lexemes, behave unlike nouns with respect to ezafe – that is, a
type of nominal agreement (see Larson and Yamakido 2008), cannot be the
input to nominalization and cannot function as subjects or objects or – of
course – verbs:23
(28) Farsi Complex Predicates with Classless Word preverbs
faraamush kardæn forgetting do ‘forget’
mahsub kardæn taking into account do ‘take into account’
haali kardæn understood do ‘cause to understand’
moraxas kardæn releasing do ‘let leave, release’
vaadaar kardæn persuading do ‘persuade’
kansel kardæn cancel do ‘cancel’
gom kardæn losing do ‘lose’

There is an interesting issue here. Complex predicates in (28), involving the

purported Classless Words, seem to mostly have predictable, possibly
compositional, meanings. This would discourage us from pursuing an
analysis in which they are subcategorial uncategorized roots, merged in
the derivation as the complements of a categorizing v head. After all, as
Karimi-Doostan (2008b, 3) shows in detail, all other preverbs in Farsi
Complex Predicates, whether they yield compositional or idiosyncratic
interpretations, are fully categorized: Preposition Phrases, adjectives,
adverbs, nouns. If this observation is on the right track, then in Farsi there
also exists a default lexical category that is neither verbal nor nominal. Its
distribution is restricted inside Complex Predicates, by virtue of this ‘Class-
less’ default lexical category encoding neither a sortal perspective, a contri-
bution of [N], nor a temporal one, a contribution of [V] – Classless Words
lacking either categorial feature. At the same time, I think that the existence
of Classless Words constitutes evidence that nouns are not the default
lexical category.
Where would the above leave us with respect to whether [N] and [V] are
different features or different values of a [perspective] feature? An initial

23
All examples are from Karimi-Doostan (2008b).
 4.6 Nouns and verbs 105

answer would be that [N] should perhaps be recast as [perspective:sortal], [V]

as [perspective:temporal], while Classless Words would perhaps be introduced
by a categorizer with an unvalued [perspective:] feature. [N] and [V] will still
be used, as mnemonically convenient flags, but the question of unvalued and,
crucially, uninterpretable features takes us into the next chapter and the status
of functional categories.
5 Functional categories

5.1 Introduction
Arguing for two privative/unary categorial features, [N] and [V], as setting
particular interpretive perspectives marks a significant step in our understanding
of the workings and status of categorizers n and v respectively. However, the
theory developed here goes well beyond providing an interpretation for
the behaviour of categorizers. In this chapter I will put forward the hypothesis
that uninterpretable categorial features are borne by functional heads. In
Section 5.2 the case against dedicated ‘category’ features for functional elem-
ents is made and in Section 5.3 the idea that functional heads are satellites of
lexical ones, members of a lexical category’s functional entourage, is rehearsed,
an idea familiar from such analytical concepts as Grimshaw’s (1991) Extended
Projections, van Riemsdijk’s (1998a) M-projections and Chomsky’s
(2001) supercategories. Section 5.4 correlates this idea with the notion of
biuniqueness – namely, that functional elements may appear in the superstruc-
ture of only one lexical head and in Section 5.5 I argue against the claim that
biuniqueness entails that functional heads bear the categorial specification of the
lexical head in their ‘Extended Projection’. Categorial Deficiency, the hypoth-
esis that uninterpretable categorial features flag functional heads, is introduced
in Section 5.6 and is refined in Sections 5.7 and 5.8. In Section 5.9 we review
the crucial consequences of Categorial Deficiency: how uninterpretable categor-
ial features triggering Agree operations can account for some well-established
characteristics of phrase structure and how uninterpretable categorial features
are the Agree probes that regulate labelling of (some) projections after the
application of Merge. The final section concludes the chapter.

5.2 The category of functional categories

In the previous chapter a theory of category was introduced in which categorial
features

106
 5.2 The category of functional categories 107

(1) a. are LF-interpretable;

b. are borne by categorizers;
c. define the interpretive perspective of the categorizers’ complement.

The above represents an attempt to make clear predictions with respect to

resolving the question of what makes nouns nouns and verbs verbs, suggesting
what their difference is, clarifying the interpretive role of categorizers and,
more precisely, the interpretive function of the features on categorizers that are
responsible for the categorization of roots or the recategorization of already
categorized constituents. If this is the picture regarding lexical categories,
the evident thing to ask next is ‘what about functional categories’?
Functional categories are certainly peculiar beasts. The very detailed survey
in Muysken (2008) is particularly revealing in this respect. Muysken exten-
sively examines the theoretical status of functional categories and their role in
grammar; he also studies their behaviour in diachronic processes (e.g., gram-
maticalization), their acquisition and their processing, as well as their vulner-
ability in language breakdown and language death, and in a variety of contact
situations. Repeatedly throughout the book he observes that there exist no
unambiguous criteria for functional category membership. Hence, he con-
cludes that ‘indeed there is an overall correspondence between the functional
status of an element and its form, but that this correspondence cannot be
captured by structural principles’ (Muysken 2008, 41). To this Muysken adds
that ‘very few semantic features, if any, unambiguously characterise a class of
elements that may be reasonably termed functional categories’ (2008, 52).
A central problem with any pre-theoretical definitions of what functional
categories are stems from the lack of any isomorphism between the phono-
logical, the morphosyntactic and, say, the parsing behaviour of the purported
class of functional elements. Muysken (2008, chap. 18) eventually opts for a
multi-factorial model in order to capture what he conceives as a symmetrical
lexical–functional schema, heavily based on formal criteria nevertheless; the
multi-factorial model turns out to be useful in teasing apart the different
entwined strands characterizing phenomena like grammaticalization and
borrowing. The overall conclusion is that the lexical–functional distinction is
real but, ultimately, one that must be made theory-internally, as happens with
most of the important analytical distinctions in scientific enquiry.
One can, therefore, take a variety of approaches towards establishing this
distinction formally. The first is to argue that next to [N] or [V], there should
also exist dedicated categorial features for functional categories such as [T] for
Tense, [C] for Complementizer or Chomsky’s (1995) [D] for Determiner.
These features would define functional categories as word classes, whereas
108 Functional categories

the features themselves (e.g., [T], [C], [D] etc.) would be interpretable at LF.
Again, this appears to form among theoretical syntacticians the received, albeit
largely unexamined, view, a view made explicit, for example, in textbooks
such as Adger (2003). Still, in the spirit of looking for an explanatory theory of
category, as opposed to settling for a purely taxonomic grouping of grammat-
ical elements, we must seriously ask the question of what kind of interpretation
features like [T], [C] or [D] would have.
(2) What is the interpretation of categorial features on functional heads?

Answering this question is both easy and difficult: easy because one can
straightforwardly come up with specific ideas on the interpretation of the [T]
feature – for example, anchoring in time; [C] would most likely encode
illocutionary force and [D] would encode referentiality and, possibly, also
deixis. However, if we are guided by a desire for thoroughness, we will soon
stumble upon the fact that answering (2) can be a rather complex affair,
punctuated by a number of complications.
An initial complication involves the categorial features of functional cat-
egories beyond Tense, Complementizer and Determiner – that is, of categories
such as Focus, Topic, Mood, Aspect, Voice (in the clausal/verbal projection
line), and Quantifier, Number, Classifier (in the nominal projection line).
Surely, these are all natural classes with identifiable LF interpretations.
Adopting a coherent theory of features where features consist of an attribute
and a range of values, we could easily capture functional categories as natural
classes defined by their features. In the case of Complementizer, for instance,
the [C] feature could be reformulated as an ‘illocutionary force’ attribute with
declarative, interrogative and so on being different values thereof. Adger
(2010) offers a cogent and detailed discussion on the details of implementing
a system like that. Hence, we could treat categorial features for functional
categories as ordinary LF-interpretable features consisting of an attribute
and possible values. Different attributes – for example, ‘illocutionary force’,
‘reference/deixis’, ‘anchoring in time’ – would define different functional
categories, Complementizer, Determiner and Tense respectively. The above
would give us the following: (a) the LF-interpretable content of functional
categories, and (b) why, for example, a [Q](uestion) head and the Comple-
mentizer that belong to the same natural class: because they bear the same
illocutionary force feature, albeit differently valued.
However, would these features truly be our categorial features? If we answer
this in the affirmative, then we actually commit ourselves to a view, one with
some currency, that a functional head bears at most one interpretable feature:
 5.2 The category of functional categories 109

this hypothesis is formulated as Kayne’s (2005, chaps. 8 and 12) Principle of

Decompositionality. A version of this principle informs Cinque (1999) and the
cartographic approach in general, and forms the impetus of Starke’s (2009,
2011) nanosyntax, where ‘features are terminals’ (Starke 2009, 2). If this were
the case, then in a given grammar there would be as many functional heads as
the interpretable features that can be encoded on them. Essentially, in this
view, each interpretable feature makes its own functional head. A second
matter is whether or not all UG-features are available in all grammars. If all
UG-features are active in every natural language grammar, then the repertory
of functional heads is also universal under these assumptions – making
functional categories such as Classifier, Topic and Mood universal. This is
presupposed in the influential Cinque (1999) and much subsequent work in
Cartography, with the added assumption that there is a fixed order of these
functional heads.
Here I wish to take a very different stand. Building on work by Thráinsson
(1996), Bobaljik and Thráinsson (1998), Hegarty (2005) and Panagiotidis
(2008), I wish to claim that the repertory of functional categories may vary
from grammar to grammar.1 According to this approach, the learner builds
grammar-specific functional categories drawing from a subset of the feature
pool that UG makes available: Focus, Topic, Force, Mood, Aspect, Voice
Quantifier, Referentiality, Deixis, Number features and so on. To illustrate the
consequences of this view, let us take Referentiality and Deixis features as an
example. Whereas Grammar A may possess a functional head Det bearing
Referentiality features and a separate functional head Dem bearing Deixis
features, Grammar B may possess a single, unitary head bearing both Referen-
tiality and Deixis features: English Determiner seems to be one of them
(for discussion, see Panagiotidis 2008). If a functional head may host two or
more LF-interpretable features – say, Referentiality and Deixis – it automatic-
ally follows that such features cannot act as categorial features. This becomes
a very serious issue if one considers the possibility that (at least) some
languages conflate the Complementizer Field (Rizzi 1997) into two Comp-
type heads (Preminger 2010) or, even, a single, unitary Complementizer head
(cf. Newmeyer 2004): would Focus, Topic or Force features together act as the

1
There is some debate even about which would be the zero hypothesis, especially from an
acquisition perspective: a universal fixed repertory of functional categories which are encoded
in UG or forced by the Conceptual–Intentional systems, or a pool of UG features for the learner
to assemble her own repertory of functional categories. See also Borer (2005) for insightful
discussion.
110 Functional categories

categorial features of such a unitary Comp-head? Similar problems arise when,

for example, we have languages with a conflated Quantifier–Determiner head –
and so on.
A second reason for choosing this path – that is, positing grammar-
specific functional categories assembled from a subset of the UG feature
pool – has to do with the functional–lexical distinction and with ‘biuniqueness’
in the special sense that the term is used in Felix (1990). Both are
examined below.

5.3 Functional categories as ‘satellites’ of lexical ones

A common way to deal with the difference between lexical and functional
categories is to think of functional elements as bearing no descriptive, concept-
denoting features; alternatively – and very similarly – functional elements are
conceptualized as unable to assign θ-roles. This is an idea that has been around
for some time, although it has hardly ever been defended explicitly. Still, it has
gained some currency; for instance, Haegeman (2006) invokes it exactly with
the purpose of distinguishing lexical from functional categories: the latter are
unable to assign thematic roles.
However, distinguishing lexical from functional on the basis of θ-assignment
and/or semantic content becomes much less convincing once semi-lexical
elements are considered. Recall from the previous chapter these are typically
lexical heads, lexical nouns and verbs, which do not seem to carry any descrip-
tive content: still, they are not functional elements; see also the treatment in the
following section.
The functional–lexical conundrum is extensively and insightfully
discussed in Chametzky (2000, 22–32; 2003, 213–19) and Hinzen (2006,
174). Both Chametzky and Hinzen essentially take functional elements to
be satellites of lexical ones; functional elements are understood to belong
to the same ‘supercategory’ (Chomsky 2001, 47) as the lexical categories of
which they form the functional entourage: hence, no matter how many
functional categories are hypothesized, motivated and discovered, no actual
proliferation of the number of stricto sensu parts of speech is necessary.
An extension of this view of functional elements as ‘dependencies’ of lexical
ones, as members of a lexical ‘supercategory’, is the even more radical and
exciting approach that functional categories do not exist as primitives of the
grammar at all. In this view, functional elements are perhaps just collections
of features (either feature bundles or feature structures) in the Numeration
which are flagged by a feature, just like Hegarty (2005) claims, in order to
 5.4 Biuniqueness 111

stand as a functional head during the derivation. This view essentially echoes
the insights in Chametzky (2003, 213–19).

5.4 Biuniqueness
Besides their acting as satellites of lexical heads, there is a second property
of functional elements that a proper account of their categorial features
should capture – namely, what Felix (1990) calls biuniqueness, harking back
to at least Martinet. Biuniqueness, according to Felix, is a general require-
ment that functional elements merge in the projection line of only one kind of
lexical head. For instance, D can only merge in the projection line of N, T in
that of V, and so on. For mixed projections, which constitute a principled
exception to this, see Bresnan (1997), Alexiadou (2001) and the following
chapter.
There is a recurrent theme in the generative literature with respect to how to
best capture biuniqueness, while also successfully addressing the question of
the category of functional categories. Ouhalla (1991), Grimshaw (1991) as
well as van Riemsdijk’s (1998a) Categorial Identity Thesis all converge on the
following hypothesis:
(3) Functional categories bear the categorial specification of the lexical head in
their projection line.

Illustrating, Aspect and Tense heads will all bear a [V] feature, whereas
Number heads and Determiners must also bear an [N] feature. Any version
of feature matching consequently guarantees that D (an [N] category) will
never select V; T (a [V] category) will never select N – and so on.
But why only talk about nouns and verbs in (3)? We have of course
discussed a number of objections to the uniformly lexical status of adpositions
and even to their forming a coherent category in Chapter 2, Section 2.9; we
have also argued in Section 2.8 of the same chapter that adjectives, even if they
are a uniform lexical category, are not of the same ilk as nouns and verbs.
However, these objections are not enough if we wish to exclude adpositions
and adjectives from the discussion of biuniqueness: we would now need to
show that there are no functional categories biuniquely associated with P and
A. Interestingly, it has already been argued that what makes biuniqueness
essentially a relation between nouns and the nominal functional layer and
between verbs and the verbal/clausal functional layer is the fact that ‘while
V[erb]/N[oun] are uniquely selected by Functional heads, P[reposition]/A
[djective] aren’t’ (Déchaine 1993, 32).
112 Functional categories

Let me begin with adpositions, which are relatively easy to deal with.
Déchaine (1993, 34) puts it curtly: ‘I know of no examples of P c-selected
by a Functional head.’ Assuming that some adpositions are indeed lexical –
say, behind, aboard, underneath, opposite, regarding and the like – these are
typically related to nominal functional material; they are probably specifiers of
functional elements in a nominal superstructure, therefore of satellites of N.
This is one of the main topics in van Riemsdijk (1998a), who examines the
behaviour of Germanic ‘lexical P-DP-functional P’ structures like auf den Berg
hinauf (‘on the mountain upon’), the c-selection relations of prepositions with
both the verbs selecting them (e.g., depend on) and the nouns inside their
complement (e.g., in 1996, on Sunday). He concludes that prepositions
‘should, in a sense yet to be made precise, be considered (extended) projec-
tions of nouns, at least when they are transitive. One aspect of this decision is
that a nominal projection embedded in a prepositional shell does not constitute
a maximal projection DP; instead, there is a transition from D’ to P’, induced
by the prepositional head’ (van Riemsdijk 1998a, 31). Sticking with this
conclusion and not going into any more detail here, as in Chapter 2 the reader
is referred to Emonds (1985, 156–7), where Prepositions are equated to
Complementizers, to Baker (2003, 303–25), who considers Adpositions
to be instances of Kase, to Svenonius (2007, 2008), Botwinik-Rotem and
Terzi (2008) and Terzi (2010), for whom adpositions are complex structures
parallel to argument structures.
Turning to adjectives, and in order not to repeat the discussion in
Chapter 2, Section 2.8, on Degree Phrases and the question of the extent
to which they are part of an Extended Projection of adjectives, I think a
point made by Déchaine (1993, 35) actually sheds a lot of light on Degree
heads, as opposed to Degree adjuncts:2 ‘any stative predicate can take a
degree modifier, as long as it is gradable’. This generalization is already
captured in Maling (1983) and becomes a reasonable working hypothesis
when one considers Greek Degree expressions. Recall from the discussion
of Neeleman, Doetjes and van de Koot (2004) that we have to distinguish
between two classes of Degree expressions. On the one hand, we have what
they call ‘Class 2’ Degree expressions – Degree adjunct modifiers like
more, less, enough, a little, a good deal – which are promiscuous in that
they adjoin to PPs, VPs and so on. Contrasting with them, we also have
‘Class 1’ Degree heads like very, as, too, that, how, which appear to belong

2
See Section 2.8.2 in Chapter 2.
 5.4 Biuniqueness 113

to the Extended Projection of an Adjectival Phrase, to the adjectival

projection line, taking APs as their complements, as sketched below:
(4) Class 1 Degree expressions

Let us turn to Class 1 Degree expressions in Modern Greek and see if we

can indeed make the case for them being functional heads in a biunique
relationship with adjectives, as is the claim in Neeleman, Doetjes and van
de Koot (2004). Like English, Greek also seems to possess at least one
Class 1 Degree expression: pio (‘more’). Like very in English, pio needs
to be embedded in a structure with a dummy adjective like poly (‘much’)
in its complement, in order to combine with most non-APs: [[DegP Deg
poly] [XP]]. Compare the behaviours of Class 1 expression pio (‘more’)
and Class 2 expression toso (‘that much/little’), which does not need a
dummy adjective poly (‘much) when it modifies non-adjectival
expressions:3
(5) Class 1 (‘pio’) and Class 2 (‘toso’) Degree expressions in Greek
a. With adjectives and adverbs:
pio kryo (more cold) ‘colder’
pio kato (more down) ‘further down’
toso kryo (that cold) ‘that cold’
toso kato (that down) ‘that far down’
b. With verb phrases:
efaye pio *(poly) (ate more much) ‘s/he ate more’
efaye toso (poly) (ate that much) ‘s/he that much’
c. With preposition phrases:
pio *(poly) sto kryo (more much in.the cold) ‘more in the cold’
toso (poly) sto kryo (that much in.the cold) ‘that much in the cold’
d. With nominal phrases:
pio (poly) anthropos (that much human) ‘more of a human being’
toso (poly) anthropos (that more human) ‘that much of a human being’
pio #(poly) spiti (more much home) ‘more of a home’
toso (poly) spiti (that much home) ‘that much of a home’

3
Although it translates as ‘more’, pio is not a morpheme purely for the formation of periphrastic
comparatives: it combines with indeclinable adverbs like kato (‘down’), mesa (‘inside’), pera
(‘far away’) and ektos (‘outside’) which are not derived from adjectives. Moreover, its behaviour
is identical to that of a number of Degree prefixes intensifying the adjective: kata- (‘over’), olo-
(‘all’), pan- (‘total’), theo- (‘god’), yper- (‘super’), tris- (‘thrice’), tetra- (‘fourfold’), penta-
(‘fivefold’) and so on – which I will not discuss here for the sake of brevity.
114 Functional categories

The behaviour of the two expressions with VPs and PPs is as expected in
Neeleman, Doetjes and van de Koot (2004). However, when it comes to noun
phrases, the picture changes. If the noun is inherently gradable, then pio is
felicitous on its own, without a dummy adjective poly (‘much’) as its comple-
ment, suggesting that this Class 1 Degree expression can actually take a
nominal complement, as long as it is plausible to perceive it as gradable
(Déchaine 1993, 35). Anthropos (‘human being’) is such a gradable noun, so
‘pio anthropos’ (‘more of a human being’) in (5)d. can be felicitous either in
the context of discussing evolution (e.g., homo erectus is more of a human than
homo habilis) or moral qualities and/or being humane. Even more interest-
ingly, just as Number can coerce the nominal in its complement into a kind
and/or countable reading (cf. the ambiguous ‘we need three coffees here’), the
Degree expression pio can coerce an inherently non-gradable noun to behave
as such, just like spiti (‘home/house’) in (5)d. above, which can be uttered
comparing the appearance of a particular residence (a converted warehouse,
a hut etc.) to a prototypical home or house.4 So, at least in Greek,
Class 1 Degree expressions cannot be claimed to be to the category A what
D is to N and T to V. In other words, the relation between Degree and
Adjective is not a categorial one; better put, to the extent that the Degree–
Adjective relation (somehow) refers to category, it is not a biunique one
between Degree and Adjective. Putting this together with the brief review on
adpositions, we can argue for the following.
(6) Biuniqueness is a relationship only between nouns and the nominal
functional heads (D, Num etc.) and between verbs and the verbal/clausal
functional heads (Voice, Asp, T, Mood etc.).

Now, the thesis in (3), a more articulated version of the intuition about
‘supercategories’, effectively captures the insights in Chametzky (2000),
Hegarty (2005) and Hinzen (2006) that functional elements are satellites of
lexical ones, and that functional categories do not exist as primitives of the
grammar. However, a solution to the problem of ‘the category of functional
categories’ like the one in (3) leads to a rather undesirable result – namely, that
we cannot distinguish functional from lexical categories on the basis of
categorial features only: we hardly want to say that, for example, T or Asp
are verbs by virtue of their [V] feature. In other words, the very same feature

4
Identical facts hold for the intensifying prefixes in footnote 3, with a lot of room for innovation
and speaker variation – morphophonological matters, allomorphy and the usual morphological
idiosyncrasies notwithstanding.
 5.4 Biuniqueness 115

that is supposed to distinguish verbs from other categories, a categorial feature,

fails to discern verbs from functional elements inside the clausal/verbal pro-
jection line, in their own ‘Extended Projection’ (Grimshaw 1991). Identical
facts hold for [N] as a feature of ‘nominal’ functional elements.
As already mentioned, attempts to distinguish lexical from functional elem-
ents on the basis of the latter not supporting conceptual content and/or θ-roles
stumble upon the existence of semi-lexical elements, these being lexical heads,
typically lexical nouns and verbs, which do not carry any descriptive content
(Panagiotidis 2003b). At the same time, the claim that functional heads
cannot assign thematic roles made good sense in frameworks where lexically
listed θ-roles were assigned by lexical verbs (and, perhaps, nouns, adjectives
and prepositions, too) as a projection of their lexical properties. However,
this makes little sense in a more syntacto-centric approach to argument
structure, where, for instance, Voice and applicative heads (both definitely
functional) also assign θ-roles. Whether this approach to θ-assignment is
correct or not, semi-lexical elements, clearly not functional heads, also assign
no thematic roles.5
Suppose, then, that we attempt to overcome this problem via trying to
capture the lexical–functional distinction in a more up-to-date fashion, and
that we state that functional elements cannot bear any descriptive, concept-
denoting features. Clearly, this makes C, D, T or Asp different from roots,
commonly assumed to bear descriptive content, but not different from n and v
themselves. Finally, a formulation along the lines of ‘functional elements
cannot support (uncategorized) root material in their complement’ seems
accurate (see the discussion on the Categorization Assumption in the previous
chapter). However, as already detailed there, this generalization is an expli-
candum not an explicans: it is merely a generalization and not an explanation.
Thus, capturing biuniqueness as the sharing of a categorial feature between
lexical and functional elements in a projection line effectively blurs the
lexical–functional distinction.6

5
Panagiotidis (2003a) links the semi-lexical elements’ lack of denotation to their inability to θ-mark.
6
In Ouhalla’s (1991) analysis lexical and functional categories can only be distinguished according
to their position: lexical categories occupy the bottom of a projection line – an important matter
we will return to. Similarly, in Grimshaw (1991), a numerically valued feature F stipulates the
different position of the head within an Extended Projection: a value of 0 places the head at
the bottom, rendering it both lexical by definition and the head of the Extended Projection, a value
of 1 places it higher and a 2 value places it at the top of the Extended Projection. The role of this
F0. . .Fn feature is then purely one of deciding the order of con-categorial heads within an
Extended Projection, while its interpretation at LF is at best unclear.
116 Functional categories

5.5 Too many categorial features

Analyses according to which functional heads bear an [N] or a [V] feature
proliferate the number of identical categorial features in a projection line,
inside an ‘Extended Projection’, to use Grimshaw’s term. Consider a concrete
example like the one below:
(7) They will probably not finish it.

Taking will to be a Tense head here, probably to be an adjunct, not to be the

head of a Neg projection and finish to be a verb head-moved into a Voice head,
and by both Ouhalla’s and Grimshaw’s analyses, we can count at least four [V]
features in the projection line. Let us visualize this state of affairs in the form of
a schematic phrase marker:

(8) Too many [V] features?

According to the idea that functional categories bear the categorial specifi-
cation of the lexical head in their projection line, the above tree should contain
four [V] features, and we are now facing the puzzle of how all of them are to
be interpreted at LF. In other words, if we take (3) at face value, we end up
treating categorial features exactly like taxonomic markers that are there purely
to guarantee biuniqueness, which is nothing but a special case of c-selection
(more on this below). Put differently, what is important is the fact that,
according to (3), all three functional elements and the lexical verb in (7) are
specified for a categorial [V] feature. If [V] is not a taxonomic feature and, in
accordance with the discussion in the previous chapter, if it encodes an
extending-into-time perspective, what happens at the interface with the
Conceptual–Intentional/SEM systems, where this interpretive perspective is
encoded four times inside a TP? To what end would this happen?
Summarizing, if we follow the path indicated in (3), then we must also
explain what the role of categorial features on functional heads is; furthermore,
we also need to suggest a general enough and language-independent way in
which to capture the fundamental lexical–functional distinction. Hence,
accounts invoking identity between the categorial features of lexical and
 5.6 Categorial Deficiency 117

functional elements in order to account for (biunique) selectional relations

between them, as in (3), suffer from two drawbacks: they proliferate the
instances of a categorial feature within a projection, and they come with the
tacit entailment that such features are purely selectional, rather than
LF-interpretable. This second matter is quite serious, and arises from the
discussion of what possible interpretation at LF a feature repeated on virtually
every head of the projection can have. Evidently, something more needs
to be said.

5.6 Categorial Deficiency

A way to resolve the problem in the previous section is the following:
(9) Categorial Deficiency: functional elements bear the uninterpretable version
of the categorial feature of the lexical head at the bottom of their projection
line (cf. Panagiotidis 2002, chap. 5).

Let us clarify the above: if v, the categorizing head, bears a [V] feature, then Asp,
T and so on will all bear an uninterpretable [uV] feature. Identical facts hold
for the nominal functional heads Num and D: they must also bear an uninterpret-
able [uN] feature. It follows that not all categorial features are interpretable:
those on functional heads are not. The following diagram, a reinterpretation
of the hypothetical state of affairs in (8), illustrates Categorial Deficiency:
(10) One [V] feature, three [uV] features

Essentially, Categorial Deficiency is in the spirit of Chomsky (1995), but

takes further his discussion of (i) categorial feature strength as driving move-
ment and (ii) ‘affixal features’ (269) on functional heads. Both these points
will be discussed below but let us just indicate here that we are using ‘affixal’
features, uninterpretable categorial features, in order to guarantee the
uniformity of ‘supercategories’ (Chomsky 2001, 47): such nominal and verbal
supercategories consist of a lexical element, bearing an interpretable categorial
feature, and the functional elements merged in its projection, bearing uninter-
pretable categorial features. Adger (2003, chap. 3) and Matushansky (2006)
118 Functional categories

have also called upon the uninterpretability of categorial features, but in

different ways, which I will discuss in the following section.
Let us now review some welcome consequences of Categorial Deficiency,
beginning at the conceptual level. A first consequence has to do with func-
tional category membership. More specifically, no matter how many functional
categories are hypothesized, motivated and discovered, no proliferation of the
number of parts of speech stricto sensu, is necessary – of course, this is already,
a consequence of (3). In other words, perhaps only Complementizer, Tense,
Voice and Determiner exist, or maybe more functional categories, or maybe
even the functional category forest of Cinque (1999) and cartographic
approaches. At the end of the day, this does not matter: we do not need dozens
of categorial features for functional categories, depending on our theory:
verbal/clausal functional heads will all be marked as [uV], nominal ones as
[uN]; they will be distinguished from each other by virtue of their interpretable
features, such as [voice], [tense], [aspect], [(illocutionary) force] and the like.
On top of this, we may now implement, at the level of features, the radical
and exciting proposal that functional categories do not exist as primitives of
the/a grammar at all. Functional heads are bundles of features assembled in the
Numeration, each flagged by an uninterpretable categorial feature. Again, this
is in accordance with the logic of Chametzky (2003, 213–19) and, as already
mentioned, brings us back to an analysis of functional heads developed in
Thráinsson (1996), Bobaljik and Thráinsson (1998) and Hegarty (2005). For
the sake of clarity, I will present below how this analysis is reprised in
Panagiotidis (2008): the core idea is that Universal Grammar (UG) provides
the learner with a fixed repertoire, not of categories, but of features from which
to choose in order to build functional heads. What Categorial Deficiency in (9)
would add to this is that each head would bear either [uN] or [uV]. Let us
sketch how this would work by way of an example:
Suppose, for the sake of argument, that UG makes available three feature
attributes for the nominal domain – say, [foc(us)], [ref(erentiality)] and [ind
(ividuation)]. Given the above, it follows that the possible number and types of
nominal functional heads in grammars would be the following:

• a single ‘hybrid’ [uN] [foc] [ref] [ind] functional category;

• two heads: an independent nominal focus category [uN] [foc] and a
‘hybrid’ [uN] [ref] [ind] Determiner–Number category;
• two nominal functional categories, again, but this time a ‘hybrid’
[uN] [foc] [ref] Focus–Determiner and an independent [uN] [ind]
(a Number category);
 5.6 Categorial Deficiency 119

• three nominal functional categories, with each head hosting one

feature: nominal focus [uN] [foc], Determiner [uN] [ref], and Number
[uN] [ind];
and so on. We may therefore speculate that, although all languages contain
nouns and verbs, the repertory of functional heads can in principle be varied
from grammar to grammar – contra Cinque (1999) and Borer (2005). Whether
this is true or not is ultimately an empirical matter.
Moving on, according to categorial Deficiency, the functional–lexical
distinction is captured as one of interpretability of categorial features:
(11) Lexical heads bear interpretable categorial features, either [N] or [V].
Functional heads bear uninterpretable categorial features, [uN] or [uV].

As a consequence of this, we do away with all ad hoc systems stipulating

the lexical–functional distinction on the basis of dedicated ‘lexical’ and
‘functional’ features – for example, the system supporting Extended Projection
(see also footnote 6). We also do away with the necessity for any notational
variants of the Extended Projection system, at the same time capturing its main
intuition: namely, that lexical heads are the prominent elements in their
respective Complete Functional Complexes (Chomsky 1986), the elements
around which functional structure is organized. Functional heads are indeed
the entourage of lexical categories, to use a turn of phrase from Chapter 2.
Less theory-internally, by invoking Categorial Deficiency for functional
elements we can reaffirm the intuition that the lexical–functional distinction
is real and sharp, and that there is no such thing as a lexical–functional
gradient. Now, there are claims in the literature that a more or less smooth
continuum between lexical and functional may exist – for instance, in the
process of language change and grammaticalization, as in Traugott (2007) and
Francis and Yuasa (2008); I believe that the illusory effect of a lexical–
functional continuum is the result of the existence of semi-lexical elements.7
All in all, a principled way to uphold the existence of a sharp lexical–
functional divide must certainly be good news: in the study of language
acquisition it has been observed that functional categories tend to be acquired
later in L1 – among others, see Radford (1990, 1996), Guilfoyle and Noonan
(1992), Vainikka (1994). Functional categories are also harder to acquire
during the course of L2 acquisition, while experimental studies in Psycholin-
guistics and Neurolinguistics reveal that functional elements may be

7
I wish to thank an anonymous reviewer for raising this matter.
120 Functional categories

selectively impaired in Specific Language Impairment or in cases of trauma –

see Muysken (2008, chap. 8–11) for a survey of the above two matters.
Explanations for this lexical–functional dissociation can now be framed as
explanations regarding the differential acquisition or impairment of uninter-
pretable (categorial) features: uninterpretable (categorial) features would be
both tough(er) to acquire and more vulnerable. This is interesting, because it
links the acquisition and breakdown of functional elements to that of uninter-
pretable features; the importance of feature interpretability in language acqui-
sition and language breakdown has been discussed extensively: see Hawkins
and Chan (1997), Tsimpli (2003), Hawkins and Hattori (2006) and Tsimpli
and Dimitrakopoulou (2007) for L2 acquisition; Friedmann (2006) and
Fyndanis (2009) for language disorders resulting from trauma; Clahsen, Bartke
and Göllner (1997), Tsimpli and Stavrakaki (1999) and Tsimpli and Mastro-
pavlou (2007) for Specific Language Impairment; Tsimpli, Sorace, Heycock
and Filiaci (2004) for language attrition.
Finally, the long-standing intuition of considering functional heads as
grammar-internal elements can now be tangibly captured. In a model of grammar
where uninterpretable features are conceived as derivation-internal atoms, we
expect whatever involves them to be a grammar-internal affair and not some kind
of interface-induced requirement. A very sketchy example would be the contrast
between Focus movement, Quantifier Raising and Binding, on the one hand, all
of which satisfy requirements of the interface with Conceptual–Intentional/SEM
systems, and Case, Verb Second and types of scrambling, on the other, all of
which seem to be purely syntax-internal. In the former operations, an interpret-
able feature is the trigger; in the latter we probably have an interplay between
uninterpretable features. Functional heads, essentially expressing nominal and
verbal/clausal properties, are also now understood to be flagged by a grammar-
internal atom: uninterpretable categorial features.

5.7 Categorial Deficiency 6¼ c-selection

Ideally, Categorial Deficiency captures something fundamental, if not elemen-
tal, about functional heads: that they do not exist as members of independent
categories or as independent elements but exclusively as satellites of lexical
material, as subconstituents of a nominal or a verbal supercategory. Put
differently, according to (9), functional heads are outlying feature bundles or
feature structures – that is, heads – necessary for the interpretation of lexical
material: not only do they contribute interpretable features to the Conceptual–
Intentional/SEM systems but they also can host specifiers, supporting intricate
 5.7 Categorial Deficiency 6¼ c-selection 121

structural relations. Whatever the ontology of functional elements is, Categor-

ial Deficiency has as its consequence that a functional head such as an imper-
fective Asp head belongs to the same syntactic category as a perfective Asp
head not because they belong to the same ‘functional category’ Asp, or due
to some ‘categorial [Asp] feature’, but because both bundles of features
(the imperfective Asp head and the perfective Asp head) bear interpretable
features with the same attribute, [asp:], albeit with different values. Cate-
gorially speaking, both aspectual heads bear a [uV], an uninterpretable verbal
categorial feature, which guarantees that the aspectual heads will be part of a
verbal supercategory and not of a nominal one.
If this picture is close to accurate, then uninterpretable categorial features are
only incidentally responsible for a type of c-selection – namely, the biunique
relation between lexical and functional elements. The main role of uninterpret-
able categorial features [uN] and [uV] is to mark (functional) bundles of
features (¼ heads) as belonging to one or the other supercategory (nominal
or verbal); their uninterpretability has as its principal consequence that heads
bearing them may not categorize roots, as discussed in the previous chapter.8
Now, Adger (2003, chap. 3) employs uninterpretable categorial features to
do all c-selection. So, for instance, in letters to Peter, the noun letters is argued
to contain a [uP], an uninterpretable prepositional category feature, which will
be checked against a [P] feature on the prepositional head to of to Peter; to
will in turn contain [uN], an uninterpretable nominal category feature,
which will be checked against the [N] feature of Peter, and so on. Adger
extends this mechanism so that he can capture the selectional restrictions of
wonder versus ask: unlike ask, wonder would bear [uC], an uninterpretable
Complementizer category feature, which would force wonder to select clausal
but not nominal complements. At the same time, ask has no such feature,
hence no similar restrictions, and it can take both clausal (I am asking what
happened) and nominal (I am asking the time) complements.9
In a similar vein, Matushansky (2006) uses the same mechanism to capture
c-selection and head movement, as already discussed in Panagiotidis (2000,
2004). The problem in invoking Categorial Deficiency in order to cast head
movement as the overt checking of uninterpretable categorial features is that
in this manner only lexical-to-functional X0 movement can explained: lexical-
to-lexical head movement, where no categorially deficient heads are involved,
inevitably remains unaccounted for. This entails that head movement

8
See also Section 5.8 below.
9
I am grateful to David Willis for raising and discussing this matter.
122 Functional categories

operations such as N-to-V incorporation would remain a mystery – see Baker

(1988; 2003, 305–11) and Li (1990) on lexical-to-lexical head movement.
Additionally, and perhaps far more seriously, if V-to-T movement is driven by
the [uV] feature on T, there is no principled way to prevent XP movement
of vP to T, with the [V] feature of the vP head pied-piping the whole projection
with it.10
Returning to Adger’s view of something like Categorial Deficiency under-
lying all c-selection, I think it is problematic on at least two levels: first, it
requires a full menagerie of categorial features of the type and taxonomic
shallowness we have been arguing against – for example, a [P] feature for
prepositions, a [C] one for Complementizers, and many more; why this is
undesirable should have been made entirely clear by now. On a second level,
suppose that the basic idea behind Adger’s analysis is correct and that instead
of uninterpretable categorial features, like [uP] or [uC], it is the uninterpretable
versions of exactly the features that a functional head bears – say, [force] for
C or [relation] for P – that drive c-selection. Elaborating, suppose that wonder
bears not [uC], an uninterpretable Complementizer category feature, but an
uninterpretable (or unvalued) illocutionary force feature; this would suffice to
capture the c-selectional requirements of a verb like wonder, which c-selects
CPs, as clausal and not nominal constituents are headed by a Force head, a
Complementizer.
I think that even this version is tough to support, although it both looks far
more attractive and leaves Categorial Deficiency as a property solely of
functional elements. The reason is that it requires lexical elements to encode
uninterpretable features, which is a rather unattractive option for reasons
discussed in Chomsky (1995), reasons that range from the nature of parametric
variation to that of uninterpretable features themselves.

5.8 Categorial Deficiency and roots (and categorizers)

In the previous chapter we had a look at the fact that only categorizers can have
root material in their complements. Elaborating on Embick and Marantz’s
Categorization Assumption, we also spelled out the germane generalization
that a root cannot be directly merged with a functional head – say, a Number or
a Voice head. We looked at an explanation for why roots must be categorized;

10
Of course, this is not a problem in analyses such as Jouitteau (2005) and Richards and Biberauer
(2005), where the whole vP might be pied-piped to SpecTP, at least for EPP purposes – but
much more would need be said on the conditions under which this may happen, if at all.
 5.8 Categorial Deficiency, roots, categorizers 123

we also explained the categorizers’ exclusive privilege of hosting root material

in their complements as a result of only categorizers having a perspective-
setting role, due to the interpretable categorial features they bear.
Functional heads, by (9), bear uninterpretable [uN] and [uV] features and,
consequently, they cannot impose interpretive perspectives on their comple-
ments. This by hypothesis prevents them from hosting root material in their
complements, unless embedded within a categorizer projection. So, Categorial
Deficiency can explain away the Categorization Assumption, and answer the
original question in Baker (2003, 269): free roots, and their projections, cannot
appear in the complement of anything but a head bearing interpretable cat-
egorial features. Therefore, roots cannot be merged directly with functional
heads, which are categorially deficient and bear uninterpretable [uN] and [uV]
features.
In view of the above, there may arise at this point a valid conceptual
objection to Categorial Deficiency, as a consequence of applying Occam’s
Razor: why have something – that is, an uninterpretable categorial feature – if
one can have nothing? Expanding this objection: even though there are good
arguments for the existence of LF-interpretable categorial features on categor-
izers, why can the Categorization Assumption not be captured along the lines of
functional heads bearing no categorial features at all? If functional heads
encode no categorial features, as opposed to encoding uninterpretable ones,
the prediction that roots must be embedded within a categorizing projection still
stands: functional heads, being devoid of categorial features, would fail to
provide an interpretive perspective for the roots. Simply put, in order to show
that the following structures (repeated from Chapter 4) are impossible, one does
not necessarily need uninterpretable categorial features on Num and Voice:
having no categorial features on Num and Voice would yield the same results.
(12)

Hence, if Num, Voice and all other functional heads lack categorial features
altogether, they would be unable to set interpretive perspectives on root
material. Everything else we said about functional heads as satellites of lexical
ones, as parts of a supercategory, would most probably still hold, as well.
Given these considerations, and the purported explanatory benefits of Categor-
ial Deficiency, the question is framed thus: is biuniqueness, which potentially
could be worked out on purely semantic and/or conceptual terms, a sufficient
condition to justify positing uninterpretable categorial features on functional
124 Functional categories

heads?11 More epigrammatically: when it comes to functional heads, do we

need Categorial Deficiency or is it enough to say that they are acategorial?12
Addressing this objection, in the section below I will provide some argu-
ments for the usefulness of Categorial Deficiency that go well beyond the
questions of biuniqueness and categorization.

5.9 Categorial Deficiency and Agree

5.9.1 On Agree
If categorial features are ordinary interpretable features, as claimed in Chapter 4,
and if uninterpretable versions thereof exist on functional elements, as claimed in
this chapter, then we would expect these features to enter into Agree relations.
I am going to show below that Agree relations between uninterpretable categorial
features as Probes and interpretable categorial features as Goals can capture a
number of facts about phrase structure. This hopefully will render the Categorial
Deficiency hypothesis a tool that possesses explanatory value that goes beyond
what it was devised to address – that is, the lexical–functional distinction,
biuniqueness and categorization restrictions.
We begin with an informal working definition of Agree, purely in order to
illustrate categorial Agree, based on Chomsky (2000, 2001, 2004) as well as
the summary in Baker (2008, 48):
(13) An element F (the ‘Probe’) agrees with F’ (a ‘Goal’) if:
a. F c-commands F’ – the c-command condition (Chomsky 2000, 122);
b. there is no X such that F c-commands X, X c-commands F’, and X has
φ-features – the intervention condition (Chomsky 2000, 122);
c. F’ has not been rendered inactive by the Phase Impenetrability
Condition – the phase condition (Chomsky 2000, 108).

11
Emphasis here should be placed on ‘worked out’: on a more personal note, I have heard this
objection raised several times since Panagiotidis (2000). Fleshing out exactly how biuniqueness
can be reduced to semantic requirements and/or conceptual restrictions remains at a program-
matic stage in most cases, with the exception of Borer (2005). However, in order to embrace
Borer’s execution, one would also have to fully embrace her system on the workings of
functional material in its entirety – no grafting of her solutions onto other analyses is possible.
12
There is a potentially important distinction to be made here; however, I will not pursue it in
detail. If functional heads are bundles of UG-features flagged by [uX], an uninterpretable
categorial feature, then we could hypothesize that inner morphemes (see Section 3.7 of
Chapter 3) – that is, subcategorial elements like low applicatives, low causativizers, particles
and the like – are precisely acategorial bundles of UG-features not flagged by any categorial
features, interpretable or uninterpretable. There is a lot more to be said on this conjecture, so
I will leave it as is in this work.
 5.9 Categorial Deficiency and Agree 125

Further clarifying for the purposes of the discussion here, ‘Probe’ and ‘Goal’
can be understood either as bundles of features or as the features themselves,
but we will take Probes and Goals to be features here. We will also assume that
we do not need an unvalued feature (Case or other) to flag the F’ (the Goal) as
active: simple matching between F and F’ should be enough (at least for our
purposes here) to guarantee the possibility of Agree between them, if all other
conditions are respected.13
At this point something needs to be said about the c-command condition
in (13). If we seriously subscribe to a bottom-up model of structure-building via
successive applications of Merge, the c-command condition must be adhered to:
(14) The Probe must always c-command the Goal – never the other way round:
Chomsky (2000, 122), Richards and Biberauer (2005, 121) and Donati
(2006).

In the reverse scenario, if Agree could occur with the Goal c-commanding the
Probe, we would have derivations where a potential Probe is merged and has
to ‘wait’ for a suitable Goal. Of course, as mentioned, ‘waiting’ is in principle
possible until phase level is reached: until the spell-out of a phase, all syntactic
operations are system-internal and not evaluable at either level of representa-
tion – that is, at either interface. Thus, phase-internally, the relative structural
configuration between F and F’ is of no consequence: see Richards (2004),
Baker (2008) and Hicks (2009, chap. 2). However, here we will adhere to a
strict version of the c-command condition – that is, for F to c-command F’ as in
(14). The reason we adhere to this stricter version is because we conceive it as
a more special case of the following requirements, already in Chomsky (2000,
133–4; 2004, 109), Hegarty (2005, 32) and Donati (2006):
a. that the Probe is always a head: a lexical item (LI) rather than a
syntactic object (SO);14
b. that the Probe (F) projects.
Thus, after the application of Merge:
(15) The Probe, a head, projects.

The above can resolve labelling in a large number of cases (and, if further
refined, possibly in all of them). Here the focus will be on how (15) works in

13
Although ‘matching’ should be clear, on an intuitive plane at least, it is true that a formalization
of the notion of matching is indeed needed: see Hegarty (2005), Adger (2006).
14
That is, an object already assembled by a previous application of Merge.
126 Functional categories

resolving the label in those cases where two LIs (lexical items) merge in order
to ‘begin a tree’.
Let us now turn to Agree relations between uninterpretable [uX] and
interpretable [X] categorial features. Let us informally call this instance of
Agree ‘categorial Agree’ and make it explicit:
(16) If [X] is an interpretable categorial feature, [uX] serves as a Probe for the
Goal [X], and not vice versa: [X] cannot ever act as a Probe for [uX] and
[uX] can never act as a Goal.

5.9.2 Biuniqueness as a product of categorial Agree

A first effect of categorial Agree is, of course, biuniqueness. Biuniqueness
reduces to the Probe–Goal matching requirement of (categorial) Agree: func-
tional heads marked for [uV] will only appear in the Extended Projection of
V (i.e., of v), which itself bears an interpretable [V] feature; at the same
time, those marked for [uN] will only appear in the Extended Projection of
N (i.e., of n), which itself bears an interpretable [N] feature. Hence, a derivation
like the one below would crash due to feature mismatch.
(17)

The above straightforwardly derives ‘Extended Projections’ in Grimshaw

(1991; 2003, chap. 1) as a result of (categorial) Agree, more specifically the
feature-matching requirement thereof.15
Conceiving biuniqueness as a result of categorial Agree raises the following
issue: in principle nothing rules out beginning a derivation by just merging all
the lexical heads together and afterwards adding the functional heads on top,
giving projection lines with heads arranged like this:
(18) [uN]. . .[uV]. . .[V]. . .[N].

This unattested, and presumably impossible, state of affairs can be ruled out if
it is independently impossible to merge all the lexical heads together in the
absence of functional structure (see also footnote 19). Perhaps even more
elegantly, it can be claimed that [uV] in (18) creates a defective intervention
effect by preventing [uN] from probing beyond it.16
A more general problem is that concerning Agree as a feature valuation
process. Given the situation in (17), we are forced to argue for uninterpretable

15
On the ordering of functional heads, see Hegarty (2005).
16
I wish to thank an anonymous reviewer for raising this matter.
 5.9 Categorial Deficiency and Agree 127

versions of [N] and [V] in order to ensure matching with their interpretable
versions on lexical heads and, consequently, biunique projection lines. Arguing
for an unvalued [ucat(egory)] feature on functional heads will not derive the
desired empirical results, as this [ucat] feature could be valued by either [N] or
[V], assuming these to really be [cat:N] and [cat:V]. I of course understand
that, unless we adopt the framework in Pesetsky and Torrego (2004, 2005),
the notion of valued uninterpretable features such as [uN] and [uV] sits uncom-
fortably with current notions on Agree as feature valuation (rather than feature
checking). Here, however, it seems necessary to argue for [uN] and [uV] in order
to preserve biuniqueness – that is, in order to prevent, for example, Tense heads
from appearing in the projection lines of nouns and the like.17

5.9.3 Why there are no mid-projection lexical heads

Now we move to a more intricate empirical question, one that the existence of
categorial Agree can resolve without any additional assumptions. This ques-
tion concerns the fact that lexical categories always appear at the bottom of a
projection line and never somewhere in the middle. Although this is a state of
affairs evident in tree diagram after tree diagram, it curiously has, to the best of
my knowledge, only attracted the attention of Baker (2003, 269).
Before addressing the absence of mid-projection lexical material, let us first
resolve a much easier issue: why we must have lexical material in a tree – that
is, why there are no purely functional projections, no syntactic structures
without a lexical head (i.e., without a categorizer) (see also the previous
chapter). Suppose we build a tree without any lexical material – for example,
something like the phrasal marker below:
(19)

By (16), this structure would contain no Goal for the three Probes, the [uV]
features, on each of the three functional heads. Although details and (im)

17
At this point, a serious empirical question emerges – namely, what we are to make of functional
elements like Aspect, which appear in the projection line of both verbs and nouns. The account
developed here can accommodate such marginal cases under the idea that particular grammars
allow the inclusion of a [uN] feature in bundles containing aspectual features. Alternatively,
whenever we encounter aspectual heads within a nominal, we can perhaps demonstrate that we
are dealing with a mixed projection (see the next chapter).
128 Functional categories

possible scenarios to salvage a derivation like the one in (19) are discussed
below, the bottom line is that there can be no syntactic structure without lexical
material, without a categorizer encoding an interpretable categorial feature.
We now turn to explaining how categorial Agree can also exclude mid-
projection lexical heads sandwiched by functional material. Let us construct
the beginning of a simplified sample derivation which would eventually yield a
mid-projection lexical verb, one sandwiched by functional material:
(20)

A first answer (to be revised below) to why (20) is impossible can be given
as follows. Both will (e.g., a Tense head) and the Aspect head are Probes for
categorial Agree, as they bear [uV], uninterpretable [V] features. However,
they do not c-command a suitable Goal – for example, a lexical verb drink
bearing the interpretable [V] feature. Consequently, there is a violation of
(14) – and (16): of the two potential Probes, neither will nor Asp c-command
their potential Goal drink; an Agree relation cannot be established and the
derivation crashes. A welcome result of an explanation along these lines is that
the impossibility of mid-projection lexical heads can now be expressed without
seeking recourse to a templatic schema, cartographic or similar, where lexical
heads ‘must’ be stipulated to appear at the bottom of the tree.

5.9.4 How projection lines begin

The above issue – that is, the position of lexical heads inside the tree – opens
up a broader issue, that of how projections begin. This is what we are now
going to have a better look at: the first application of Merge on a given
Numeration, in a given workspace. Let us begin with the scenario where this
first application involves merging two Lexical Items (LIs) and consider a
simplified instance of the beginning of a nominal constituent:
(21)

Assume that both D and N are LIs – that is, monadic elements drawn
from the Numeration and not previously constructed by an application of
Merge. By (15) above, either of them may project because they both qualify
as potential heads of a projection; thus, between the two, it is the Probe of an
 5.9 Categorial Deficiency and Agree 129

Agree relation that must project. Put otherwise, the label will be decided on the
basis of which of the two LIs, D or N, is a Probe for Agree. Of course, it has
been common knowledge since the formulation of the DP-hypothesis that in a
case like (21), it is the Determiner that projects, which in turn suggests that,
by (15), D is the Probe in some Agree relation. However, it is not self-evident
what kind of Agree relation could hold between a Determiner and a Noun. It
is at this point that categorial Agree turns out to have explanatory value.
Determiners, being functional, bear [uN] features; lexical nouns by definition
bear [N] features.18 So, the answer to the question of why D is a Probe – so as to
project – is because it hosts an uninterpretable categorial feature. This feature
makes it a Probe for categorial Agree and the noun, bearing [N], is its Goal.
Therefore, D, the Probe, projects after it is merged with N, because the
Determiner is functional, thus bearing a [uX] categorial feature, a [uN] more
specifically:
(22)

In a nutshell, a Determiner, like all functional LIs, is a Probe for (categorial)

Agree, as per (15) and (16). When merging with a lexical LI, its (categorial)
Goal, the functional LI will invariably project. The emerging picture is inter-
esting in that it potentially resolves the question of labelling when a functional
and lexical LI merge:
(23) When a functional [uX] LI and a lexical [X] LI merge, the functional LI, a
Probe for categorial Agree, projects.

The Categorial Deficiency of functional heads triggering categorial Agree can

automatically be extended to also resolve labelling when a functional LI and a
lexical Syntactic Object (SO) merge, too: the lexical SO will of course be
headed by an element bearing an interpretable categorial feature [X],
thus making it a Goal for categorial Agree.19 Furthermore, we can now also
predict the label in a (run of the mill) scenario where a functional LI merges

18
Well, nominalizers, actually.
19
However, categorial Agree obviously cannot predict the headedness of cases where two lexical
LIs merge. Could this be because directly merging two lexical LIs is an impossibility? The
ubiquity of applicative heads (Pylkkänen 2008), subordinating conjunctions (Cormack and
Smith 1996), relators (Den Dikken 2006) and the like might suggest so, but I wish to remain
agnostic here.
130 Functional categories

with a functional SO. Consider the following structure, where the LI the
merges with the nominal SO three little ducks:
(24)

Let us assume the following structure for (24):

(25)

Once more the Determiner the is a Probe for (categorial) Agree, because it
bears an uninterpretable categorial feature [uN]. The functional SO (e.g., a
Number Phrase) with which it merges contains only one instance of an
interpretable categorial feature, [N] on ducks. This is because the [uN] feature
on three has already been checked off and eliminated after the previous
application of Merge that created the SO three little ducks.20 Therefore, the,
a Probe for (categorial) Agree, projects. If this story is correct, then when a
functional LI and an SO merge, the functional LI will always project: this
straightforwardly predicts that functional LIs merging with a syntactic object
can never be specifiers. Epigrammatically put:
(26) When a functional [uX] LI and an SO merge, the functional LI, a probe for
categorial Agree, projects as the head of the new constituent;

(27) When a functional [uX] LI and an SO merge, the functional LI, a probe for
categorial Agree, cannot be a specifier.

At the same time, when two functional SOs merge, headedness of the resulting
object, its label, cannot be decided on the basis of categorial Agree. This is
because both SOs will, by hypothesis, each have had all their uninterpretable
categorial features already checked. Hence, a different instance of Agree will
be relevant for deciding the label of the resulting object.

5.9.5 Deciding the label: no uninterpretable Goals

The above account, capitalizing on Agree relations between categorial fea-
tures, has yet another important consequence. To illustrate, let us return to the
phrase marker in (20), repeated below:

20
Actually, in all probability three is the specifier of a Number Phrase with a null Num head, but
this is immaterial to the point made here.
 5.9 Categorial Deficiency and Agree 131

(20)

On the basis of (14), it was argued that (the [V] feature of) drink cannot act
as a Goal for (the probing [uV] features of) will and Asp. However, there may
be a way to account for the impossibility of structures like (20) in a more
principled way. First, consider the fact that the phrase marker above is the
result of two applications of Merge. Let us then backtrack to the first applica-
tion of Merge that putatively gave the SO [will Asp], with which drink is
supposed to merge.
(28)

Looking at (28) the following question emerges: the exclusion in (16) notwith-
standing, why is it actually the case that (the [uV] feature of) will cannot probe
(that of) Asp? Or vice versa? Both will and the aspectual head are probes for
Agree. What prevents the [uV] feature of Asp or that of will from serving as a
Goal for categorial Agree? The condition in (14) would not be violated, as will
and Asp c-command each other.
The short answer to the above question in a conception of Agree as a
valuation process is that unvalued (and, presumably, uninterpretable) features
cannot serve as Goals, as they cannot provide the Probe with a value; recall
that this is stipulated in (16) but stems in a principled way from Chomsky
(2001, 6; 2004, 113) and much subsequent work. However, it is perhaps useful
to examine what would happen in examples like (28) if it could indeed happen
to be the first application of Merge in a workspace.21
The first problem in a situation like that in (28) would be which of the [uV]
features on each LI is the Probe. This cannot be resolved, as, in principle, both
[uV] features can be Probes for categorial Agree. Therefore, in principle,
both will and Asp would be possible to project. This in turn would result in
(a) optional labelling for the resulting constituent, or (b) an intrinsic failure of
the system to determine the head, or (c) co-projection of both will and Asp.
I take it with Chomsky (1995, 244) that the last two options are impossible and

21
I am grateful to Marc Richards and an anonymous reviewer for discussing interpretability of
Goals with me.
132 Functional categories

I would think that the first one, that of optional labelling, is highly undesirable,
too. In other words, the above scenarios run afoul of the received understand-
ing of how labels are set after Merge: when merging X and Y, we expect either
X or Y to project, as discussed in detail in Chomsky (2000, 2001, 2004).
Moreover, if Hinzen (2006, 187–9) is correct in saying that the label is
determined by lexical properties of the elements involved and that label
determination is not part of the definition of Merge, then it is plausible to
think that such relevant properties would interact with Agree and that there
would exist no context-independent mechanisms to decide the label of a newly
merged constituent.
Therefore, the relevance of the Probe–Goal relationship for determining labels –
(15): the Probe, a head, projects – and the ubiquity of (categorial) Agree due to
categorial Deficiency in (9) lead us to the following empirical generalization:
(29) We cannot begin a tree by merging two functional LIs.22

Let us summarize the effect of Categorial Deficiency on deciding the label

after an application of Merge in the table in (30):
(30) Labelling predictions with categorial Probes and Goals
Predicted
result Example

[uX] LI merges with [uX] LI * *[T Asp]

[uX] LI merges with [X] LI [uX] LI [DP D n]
projects
[uX] LI merges with SO [uX] LI [DP D NumP]
(always [X]) projects
[uX] LI merges with root * *[Num cat]
[X] LI merges with [X] LI ? ?23
[X] LI merges with SO ? [vP v nP]
(always [X])
[X] LI merges with root ? [nP n cat]
SO merges with SO (both ? [TP [DP the cat]
always [X]) [TP meows]]

22
This has as a consequence that the claim in Abney (1987) – and much subsequent work – that
pronouns are bare – that is, nounless, Determiner constituents – must be false: a pronoun like
[DP D Num], in the absence of an n, will be impossible. However, following Panagiotidis (2002)
and Déchaine and Wiltschko (2002) in that all pronouns contain at least a minimally specified
nominal head, the tension is resolved in favour of the generalization here.
23
In the framework followed here I am not sure that two lexical LIs – that is, two categorizers –
can merge. What would something like an [n v] be? This is why no example is provided in the
‘[X] LI merges with [X] LI’ row. See also footnote 19.
 5.10 Conclusion 133

By the categorial Agree account, all these characteristics of syntactic structures

(the projection of functional heads after the application of Merge, the presence
of a lexical element at the bottom of a syntactic structure, the existence of
Extended Projections/M-Projections) derive from the fact that uninterpretable/
unvalued features cannot act as Goals for Agree.

5.10 Conclusion
In this chapter I have proposed that functional heads are essentially category-
less. On top of that, I have argued against categorial labels for functional heads
and against functional categories as word classes like nouns and verbs are,
claiming that functional heads are flagged by uninterpretable categorial fea-
tures, that they are categorially deficient. Building on this hypothesis I have
managed to derive the Categorization Assumption, the part where functional
heads cannot directly merge with roots, more specifically. Biunique relations
between functional and lexical elements as well as the nature of nominal super-
categories/Extended Projections/M-Projections also fall out from Categorial
Deficiency. Finally, the categorially deficient nature of functional heads – that
is, their encoding uninterpretable categorial features – was argued to make
them Probes for Agree relations with lexical heads; these Agree relations can
shed light on labelling and the position of lexical material inside a tree, as well
as on the impossibility of purely functional structures without any lexical
material.
6 Mixed projections and functional
categorizers

6.1 Introduction
The problem of mixed projections – that is, projections combining both
nominal and verbal/clausal functional subconstituents – is discussed in this
chapter. Our theory of categorial features combined with the existence of
uninterpretable categorial features can capture the existence and function of
mixed projections without novel theoretical assumptions. Section 6.2 intro-
duces the problem and Section 6.3 presents two empirical generalizations on
mixed projections: Categorial Uniformity and Nominal External Behaviour.
The following section reviews evidence against freely mixing nominal and
verbal functional elements – that is, without any concerns regarding categorial
uniformity or biuniqueness. Section 6.5 introduces functional categorizers,
Switch elements, as heads that both recategorize their complement, bearing
an interpretable categorial feature, and belong to the functional projection of
a lexical head. In Section 6.7 the type and size that the complement of a
functional categorizer can take is examined, as well as the phasal status of
Switch heads. Whether mixed projections all behave externally as nominals
is reviewed in Section 6.8, on the basis of the behaviour that verbal nouns in
Korean and Japanese display. Section 6.9 attempts to explain away differences
between mixed projections that consist of the same type of subconstituents –
say, Tense Phrases – by appealing to the different feature specification of the
heads within the verbal/clausal functional subtree. The need for functional
categorizers next to the better-known ones – that is, n and v – is addressed in
Section 6.10 and the last section concludes the chapter.

6.2 Mixed projections

So far I have proposed a theory of categorial features and I have examined some
of its consequences. I started off with a commitment to a theory according to
which categorial features are LF-interpretable features. These features ‘make’

134
 6.2 Mixed projections 135

categorizers (n and v), the same way one can argue that temporal relation
features make Tense heads. Categorial features are interpreted as fundamental
interpretive perspectives on the material in the categorizers’ complement.
Uninterpretable versions of categorial features flag functional heads and give
rise to a number of familiar phrase-structure phenomena: the biunique relation
between lexical and functional elements and the position of lexical material
at the bottom of a projection line, they also play a central role in labelling after
the application of Merge.
Given all of the above, we might expect syntactic projections to be either
nominal or verbal/clausal. In other words, we expect biuniqueness all the way
up the tree: essentially, the categorial feature [N] or [V] of the categorizer
should guarantee the categorial uniformity of the whole projection line. Graph-
ically expressed, we expect, for instance, that projections be exclusively made
up of functional elements with a [uV] specification, forming the verbal/clausal
entourage of a lexical verb.
(1)

The obvious question is how this system can accommodate mixed projec-
tions. Generally speaking, mixed projections are a problem for any system of
phrase structure that (implicitly or, as in this case, explicitly) admits total
categorial uniformity of projections. At the same time, most conceptions of
phrase structure would, for instance, prevent a D taking a TP complement or a
T taking a DP complement, and they would do so for very good reasons, pace
Alexiadou (2001) and (in a sense) Borer (2005). This is why the existence and
properties of mixed projections must be seriously addressed.
Of course, mixed projections have already been variously addressed, time
and time again; they have indeed posed a serious problem for syntactic
theories. In order to get a sample of the ways mixed projections have been
dealt with, let us briefly review the types of analyses proposed for Poss–ing
gerunds (after Hudson 2003), maybe the best-studied mixed projections.
A first analysis comes from Jackendoff (1977), who develops an insight in
Chomsky (1970): Poss–ing gerunds are exocentric NPs consisting of a VP.
This could be roughly understood as an NP without an N head but dominating
136 Mixed projections and functional categorizers

a VP, a state of affairs that was possible given the labelling and bar-level
conventions in the 1970s and, more importantly, something that captures a key
intuition: that Poss–ing gerunds, like most mixed projections, ostensibly
behave externally as nominal, despite their having a verb at their heart – this
is an intuition we will extensively scrutinize here. Baker (1985) argues that
gerunds are NPs headed by –ing, an affix which is then lowered onto the verb,
via a version of Affix-hopping; the underlying suggestion is that –ing is
nominal, which is a commonly shared assumption. Abney (1987), updated
and refined in Yoon (1996a), claims that in gerundive projections Det directly
selects IPs or VPs.
A more elaborate solution for the violation of biuniqueness in mixed
projections like gerunds is offered in Pullum (1991), who, within the HPSG
framework, proposes the weakening of the Head Feature Convention, thus
allowing the mother phrase and its head to have different values for
N and V. In a similar vein, in the HPSG account of Lapointe (1993), the NP
and VP nodes have ‘dual’ lexical categories <X|Y>, where X and
Y determine external and internal properties respectively: nominal externally
and verbal internally. Finally, in Bresnan (1997) – to which we will return – an
LFG account is offered, in which a single c-structure N (the gerund) maps to
both an N and a V position in f-structure.
Summarizing, the old chestnut of mixed projections has been attacked from
two viewpoints: either writing categorial duality into their head, as in Jackend-
off (1977), Pullum (1991), Lapointe (1993) and Bresnan (1997), or arguing for
a structure where an abstract nominal element selects a VP, as in Baker (1985),
Abney (1987) and Yoon (1996a). Here we will combine both lines of
reasoning in order to tackle the problem: thankfully, treating categorial fea-
tures as ordinary LF-interpretable features (and not as flags merely identifying
grammar-internal entities) combined with Categorial Deficiency enables us to
do exactly that.
Before presenting the analysis, let us first look at mixed projections in a
descriptive way and let us adumbrate the generalizations we can extract from
looking at empirical evidence.

6.3 Two generalizations on mixed projections

Mixed projections combine characteristics from more than one category. This
is what makes mixed projections different from Complementizer Phrases
(‘clauses’), which contain a series of ‘verbal’ functional categories and a verb
(at least a verbalizer), and from Determiner Phrases (‘nominal phrases’), which
 6.3 Two generalizations on mixed projections 137

contain a series of ‘nominal’ functional categories and a noun (at least a

nominalizer). Typically (and expectedly) mixed projections combine
a. verbal/clausal and nominal characteristics (like gerunds), or
b. verbal/clausal and adjectival characteristics (like participles)
Following the practice here to ignore adjectives (see Chapter 2), we will
concentrate only on mixed projections bringing together verbal/clausal and
nominal characteristics.1 Three examples (among many) of such mixed pro-
jections are English Poss–ing gerunds (2), Spanish nominalized infinitives (3)
and Greek DþCP constructions (4).
(2) [Bob’s insulting them all] annoyed us.

(3) [El cantar yo La Traviata] traerá malas consecuencias.

The sing.inf I La Traviata bring.fut bad consequences
‘My singing the Traviata will not end well.’

(4) [To oti fevyi i Niki] dhen ine provlima.

The that leaves the Niki not is problem
‘That Niki is leaving is not a problem.’

An immediate, very robust generalization that can be made about a number

of mixed projections comes from Bresnan (1997, 4), also informing Borsley
and Kornfilt (2000):
(5) Phrasal Coherence: the mixed projection ‘can be partitioned into two
categorially uniform subtrees such that one is embedded as a constituent of
the other’ (after Malouf 2000).

According to Bresnan, Phrasal Coherence holds for mixed projections in a

range of typologically unrelated languages: Hebrew, Arabic, Turkish, Kikuyu,
Italian, Dutch, German, Dagaare (a Gur language of the Niger–Congo family)
and others. What Phrasal Coherence essentially amounts to is that we never
have alternating nominal and verbal constituents making up a mixed projec-
tion. In other words, the nominal and the verbal chunks in a mixed projection
are distinct and occupy different ‘sides’ thereof; crucially, they never inter-
sperse. Hence, in mixed projections there must always exist a cut-off point

1
This decision is not made on purely methodological considerations regarding the status of
adjectives. Mixed projections that combine adjectival and nominal characteristics are, well,
adjectives (recall the discussion in Section 2.8 of Chapter 2); mixed projections combining
adjectival and verbal characteristics will have to be set aside for future research, once there is a
more concrete picture of the categorial status of adjectives.
138 Mixed projections and functional categorizers

where verbal/clausal characteristics end and nominal ones begin: verbal/

clausal and nominal characteristics and elements are located in different parts
of the tree. See also Lapointe (1993), Borsley and Kornfilt (2000), Malouf
(2000), Schoorlemmer (2001), Ackema and Neeleman (2004, 174). In order to
visualize the state of affairs described by Phrasal Coherence, consider the
following abstract phrase markers:
(6) A mixed projection abiding by Phrasal Coherence

(7) A mixed projection not abiding by Phrasal Coherence

If Phrasal Coherence is a valid generalization and if mixed projections like the

one portrayed in (7) are indeed impossible, then an important step has been
made towards analysing mixed projections not as oddities or as exceptional
constructions but as projections consisting of two phrasally coherent parts: two
subtrees, each of which displays Categorial Uniformity.
Let us now take a closer look at two examples of mixed projections
displaying Phrasal Coherence. First, in English Poss–ing gerunds the higher
part of the projection may assign Genitive, which is a nominal characteristic
(possibly the signature of a Determiner), whereas the lower part may assign
Accusative, a verbal characteristic (the signature of Voice). It is also worth
noting that the sequence of functional elements, the nominal above the verbal
ones, resembles the hypothetical schema in (6), a matter to which we will
return below. A second example of a mixed projection displaying Phrasal
Coherence would be Japanese verbal nouns; see Tsujimura (1992, 477–9) and
Manning (1993). In this case, the higher part of the projection may assign –ga
Nominative, which would be a verbal characteristic (the signature of Tense),
whereas the lower one may assign Genitive, a nominal characteristic (possibly
the signature of n). The apparent sequence here is the reverse of what we have
 6.3 Two generalizations on mixed projections 139

in the case of English Poss–ing gerunds, and (6), with verbal elements above
the nominal ones (a matter to which we will return in Section 6.8.1).
If Phrasal Coherence is a first generalization about mixed projections,
there is also a second generalization that can be surmised by surveying the
literature on mixed projections: Borsley and Kornfilt (2000), Malouf (2000),
Hudson (2003) more specifically: mixed projections externally behave as
nominals.
(8) Nominal External Behaviour: mixed projections externally behave as
nominal constituents.

Externally, mixed projections generally display straightforward nominal

behaviour; this behaviour is morphosyntactically manifested by the presence
of a nominalizing element and case-marking, like in the Turkish example of a
nominalized clause in (9), an article, as in Spanish nominalized infinitives
in (10), or the licensing of a possessor in English Poss–ing gerunds in (11).
Moreover, as is evident by looking at all the examples below, mixed projec-
tions can stand as run-of-the-mill arguments of verbs: for example, a direct
object in the Turkish example in (9), a subject in the Spanish example in (10),
and both in the English example in (11).
(9) Hasan [Ayşe-nin gel-me-sin]-i iştiyor.
Hasan Ayşe-gen come-nom-3sg-Acc wants
‘Hasan wants Ayse to come over.’

(10) [El cantar yo La Traviata] traerá malas consecuencias.

The sing.inf I La Traviata bring.fut bad consequences
‘My singing the Traviata will have dire consequences.’

(11) [Bob’s insulting us all] annoyed them.

Few took notice of [Bob’s insulting us all].

Complementing this observation there is also a lack of indisputable evidence

for bona fide mixed projections behaving externally as verbs or as clauses,
while containing a ‘real’ nominal element, a lexical noun. The apparent
exception of Japanese verbal nouns as adumbrated above will be discussed
in Section 6.8.1, so, for the time being, we assume (8) as a working hypothesis
which rests on a solid empirical basis.
So far we have hardly explained anything regarding the structure and the
syntactic behaviour of mixed projections: all we have done is adopt Phrasal
Coherence in (5) as a working hypothesis and acknowledge an empirical
generalization on the nominal external behaviour of mixed projections. So, it
is now that the truly tough question emerges – namely,
140 Mixed projections and functional categorizers

(12) How can two categorially different subtrees, a nominal one and a verbal one,
be combined to form a single projection?

Recall that we have captured the categorial uniformity of projections in terms

of a biunique relation between functional heads and a lexical category, which
is instantiated through categorial Agree between categorially deficient func-
tional heads and (lexical) categorizers bearing interpretable categorial features.
Consequently we have to answer (12) along the lines of where a second lexical
head would have to be merged in a projection line in order for the functional
part growing out of it to be supported. More precisely, consider the phrase
marker in (6), repeated below for convenience and enriched with categorial
features on every head, as predicted by Categorial Deficiency:
(13) Phrasal Coherence – but what about Categorial Deficiency?

According to the account in Chapter 5, in the phrase marker above (which

could be an abstract representation of Poss–ing gerunds and the like) the lower
verbal functional part of the projection is licensed, in the broad sense, by the
lexical verb, or more precisely by the feature [V] of v. But what about the two
functional ‘nominal’ heads c-commanding the verbal functional structure?
There is no head bearing an [N] feature for the [uN] feature on them to probe.
Should the derivation in (13), then, not be ruled out on the same grounds on
which functional structures without a categorizer/lexical head are banned, as
we saw in the previous chapter?

6.4 Free-mixing mixed projections?

One way, a radical one, to explain how mixed projections are possible is to
abandon any notion of categorial uniformity. So, (12) and the questions
surrounding it would not pose a problem. This is the path Alexiadou (2001)
takes, in the spirit of radical categorylessness (see Chapter 1). Recall that, like
Borer (2003, 2005) and De Belder (2011), Alexiadou claims that there are no
categorizers. Instead, the functional environment around a root actually defines
the root’s category. So, biuniqueness is superficial and, indeed, illusory as it is
 6.4 Free-mixing mixed projections? 141

not Tense that ‘goes with’ a verb. Instead, a T category makes a root a verb: a
root inside TP will surface as a verb (Alexiadou 2001, 19). Similarly, it is not
the case that a Determiner ‘goes with’ a noun. Instead, a D category makes a
root a noun: a root inside DP will surface as a noun (Alexiadou 2001, 19). This
hypothesis is illustrated in the simplified trees below:
(14)

A consequence of the above is that, if there are no categorizing heads, there is

no need for category-changing heads either, which would necessarily mediate
between categorially uniform subtrees in a mixed projection. As a conse-
quence, you can freely mix together any kind of functional heads – for
example, D with Asp, T with Num, D with Voice, T with D. In this view,
biuniqueness essentially does not exist, and Alexiadou’s account describes the
simplest state of affairs that would be conceptually possible. However, before
we are lured by this option, the question, as ever, is whether it is the correct
way of approaching the problem, empirically speaking.
In general, it seems that verbs behave as verbs and nouns as nouns even
when there is very little functional structure above them. Precisely this point is
convincingly elaborated upon in Baker (2003, 265–90) on the basis of inspect-
ing cross-linguistic evidence on incorporation and compounding, with the
conclusion summarized as follows: ‘category-specific behaviour can arise even
when there is no sign of any functional superstructure dominating the lexical
head . . . [E]xactly where there is less functional structure, we find more
categorial distinctiveness’ (Baker 2003, 268). I believe this is the correct
generalization, and it is tacitly incorporated in statements like Embick and
Marantz’s (2008) Categorization Assumption, which we discussed in length in
Chapter 4. Let me then just complement Baker’s observation by quickly
looking at verbs and nouns with very little functional superstructure
dominating them.
Beginning with verbs, they behave like verbs already at the vP level
and certainly at the VoiceP level, as nexus constructions (Svenonius 1994)
indicate – that is, structures like Me drink alcohol? Never, and the like.
Constituents containing a root certainly do not have to wait for a full verbal/
clausal functional shell to be merged before they can display verbal behaviour
in full as lexical verbs. Turning to nouns, there are of course Determiner-less
142 Mixed projections and functional categorizers

constituents that are clearly and fully nominal. Illustrating this with Greek
examples, consider kind readings of nouns in (15) and mass readings, as in
(16), where there is no need for any Determiner:
(15) Eyine papia
became duck
‘S/he became a duck.’
(16) Efaye papia
ate duck
‘S/he ate duck.’

In (15), which would be felicitous in the context of a fairy tale or similar, papia
(‘duck’) has a kind reading: the sentence is about something or someone
switching kind into the kind ‘duck’. In (16), papia (‘duck’) refers to duck
meat. In both cases there is no determiner layer present but papia is unam-
biguously a noun. What is even more interesting is that even NumP, another
functional projection which could be held accountable for categorizing roots,
is also most possibly perfunctory in the examples above, as neither example
involves individuation, to begin with; see Borer (2005, chap. 4, passim) for
very extensive discussion on the function and the semantics of individuation
and Number.
For the reasons reviewed above, the free-mixing version of how to capture
mixed projections will not be pursued here. Now, a second way to understand
how mixed projections are possible is to say that the two categorially uniform
subparts of a mixed projection are linked by a category-changing head, a
special type of categorizer. This is the path to be explored here, in a way that
will bring together the insights in two distinct schools of thought on mixed
projections. On the one hand, we will follow Jackendoff (1977), Pullum
(1991), Lapointe (1993) and Bresnan (1997), who encode categorial duality
into a head, our purported special categorizer; on the other hand, we will also
incorporate elements from analyses positing a structure where an abstract
nominal element selects a VP, as in Baker (1985), Abney (1987) and Yoon
(1996a): again, our special categorizer.

6.5 Switches as functional categorizers

The hypothesis proposed here aims to capture the existence and properties of
mixed projections in the following fashion: mixed projections, unlike ‘ordin-
ary’ categorially uniform ones, contain a ‘mixed’ category head between their
two parts, which I will call a Switch for mnemonic purposes (Panagiotidis
 6.5 Switches as functional categorizers 143

and Grohmann 2009).2 This Switch head ‘mediates’ between the nominal
and the verbal half, by virtue of its categorial feature makeup. More precisely,
Switches are categorizers, like n and v, hence they bear interpretable cat-
egorial features. What makes them special, and suitable as mediators between
categorially different functional layers of structure, is that they also bear
uninterpretable categorial features.
(17) Switches are categorizers that bear both interpretable [X] (i.e.,
‘categorizing’) and uninterpretable [uX] (i.e., ‘functional’) categorial
features.

Let us first consider how Switches would work and then discuss their nature.
I will exemplify on English Poss–ing gerunds.
Suppose, essentially following Reuland (1983) and Hazout (1994), that
gerundive projections contain a head Ger. This Ger head takes a verbal
complement but is selected by a Determiner or a Determiner-like element,
initiating a switch of categorial identity within the projection line. More
specifically, Ger takes a verbal complement, an AspP, as indicated by –ing.3
I take this Ger head not to be an ad hoc category or, even, a gerund-specific
element, but a functional categorizer as described in (17), a Switch. In
keeping with our analysis in the previous chapter, and in line with what was
claimed about the Categorial Deficiency of functional elements, we can now
simply claim that a Switch bears a [uV] feature: Switch behaves as a
verbal/clausal functional head that can participate in verbal/clausal projection
lines by virtue of its uninterpretable categorial feature, which probes its
complement for a [V] feature. Actually, van Hout and Roeper (1998) suggest
that the verb head overtly climbs up to Ger, but this is probably wrong, when
adverb placement is considered (Ad Neeleman, personal communication,
February 2007).
Now, this Ger head (our Switch) appears, like nouns do, in the comple-
ment of the possessive Determiner head of the null variety, which assigns
Genitive Case to SpecDP – as already claimed in Abney (1987), Borsley and
Kornfilt (2000, 105) and elsewhere. This Ger head will then contain an [N]
feature, given that Determiners bear a [uN] feature. Note that the intuition that

2
Lapointe (1999) was the first to talk about ‘category switchover points’. The term ‘switch’ is also
used in Schoorlemmer (2001) to describe the point in the structure of nominalized infinitives
where category changes.
3
On why –ing itself cannot be the nominalizing morpheme, contra Abney (1987), Milsark (1988)
and others, see Ackema and Neeleman (2004, 175–81). Johnson (1988) claims it to be a Tense
affix but here we side with Siegel (1998) in considering it to be just an aspectual marker.
144 Mixed projections and functional categorizers

Ger contains a nominal categorial feature is already expressed in Reuland

(1983, 113) and Hazout (1994), who actually claim Ger to be a noun.4
However, this cannot be correct: Ger cannot be a noun and, more generally,
no lexical noun can be responsible for the categorial shift within gerundive and
other mixed projections; there are actually good empirical reasons why nouns
as switching elements are impossible.
If a noun mediated between the nominal and the verbal/clausal part of a
mixed projection, this ‘noun’ would necessarily be both a phrasal affix and
a lexical noun – that is, a noun with an obligatory complement. This sounds
like a very bizarre type of noun and this point has already been made in
Borsley and Kornfilt (2000, 119), regarding Turkish nominalizations, which
we will revisit in Section 6.6. Supposing, for the sake of argument, that affixal
lexical nouns would be possible, the ones participating as Ger in the formation
of gerundive projections would invariably select for a verbal complement
which would, moreover, not be their argument. Furthermore, these nouns
would have a morphological exponence of zero: in the discussion of
category-changing elements in Ackema and Neeleman (2004, 175–81) it turns
out that in head-initial languages the category-changing affix must have a
zero exponence, a conclusion also arrived at in Siegel (1998). Once more, this
kind of morphophonological restriction on lexical nouns is unexpected.
Finally, Borsley and Kornfilt (2000, 119) present an elaborate argument
against Ger (and category-changing elements in general) as nouns: although
this matter will not be discussed here, it is reasonable to assume that subjects
of Poss–ing gerunds originate from within the verbal constituent. However,
if they do, then the analysis of Poss–ing gerunds with Ger as a lexical noun
would involve the extraction of a subject from the complement of a noun, Ger.
This kind of extraction is impossible in English, as illustrated in the examples
below, adapted from Borsley and Kornfilt (2000, 119).
(18) John appeared _ to be drunk.

(19) [John’s appearing _ to be drunk] surprised us.

(20) *[John’s appearance _ to be drunk] surprised us.

So, Ger bears an [N] feature but is not a noun. Naturally, given the discussion
in Chapter 4, the [N] feature on the Ger/Switch can be explained if the Ger
head is a categorizer and, more specifically, a nominalizer, an n head.

4
This is what Abney (1987) and van Hout and Roeper (1998) also seem to argue for.
 6.5 Switches as functional categorizers 145

However, nominalizers typically appear low: as repeatedly discussed, they

head the lowest phases, taking as their complements either root projections
(e.g., dog or truth) or vPs (e.g., destruction in the destruction of evidence); see
Chapter 3. At any rate, we would not expect a categorizer like n to take
functional constituents as complements, as seems to be the case with Ger
and its AspP complement. More precisely, we have so far not encountered
categorizers taking complements that contain any functional structure: this
suggests that Switches are no ordinary categorizers (no ordinary nominali-
zers, in the case of Poss–ing gerunds). We can therefore hypothesize that
Switches also contain an uninterpretable categorial [uV] feature, just like
Asp and other verbal/clausal functional heads, and that this is how they can
appear in the projection line of the verb: if this is true, then they take a verbal
projection as their complement, an AspP, in line with how categorial Agree
ensures categorial uniformity of the functional heads in a projection.
An illustration of the claims made so far in this section is given below in the
form of a simplified tree for Albert’s eating herring – see also Siegel (1998)
and Moulton (2004).
(21) A simplified tree for [Albert’s eating herring].

Thus, a nominal Switch, like ‘Ger’, contains a [uV] feature besides its [N]
one. It is a functional categorizer.
The hypothesis of Switch as a functional categorizer should be quite
straightforward by now; essentially, a Switch can be identified with the
category-changing abstract phrasal affix in Ackema and Neeleman (2004,
172–81). Such category-changing affixes are postulated to attach on
projections of various levels. More precisely, Ackema and Neeleman discuss
nominalizing affixes attaching on verbal projections of various sizes, a point
to which we will return below. As regards its exponence, the category-
changing affix is phonologically null in head-initial languages and con-
versely so in head-final ones; this is a prediction made by a principle of Input
146 Mixed projections and functional categorizers

Correspondence (Ackema and Neeleman 2004, 140) that maps structures onto
morphological forms, and is borne out in a number of languages (176–81).
Now, the above claims immediately beg the following questions:
a. whether it is possible for two categorial features to co-exist on a
single head;
b. how come this co-existence does not induce a categorial clash;
c. what it means (LF-wise) for a syntactic head to be specified as [N] [uV].
First, as insistently claimed here, categorial features are not taxonomic-
classificatory markers; they are LF-interpretable features. Rather than flags used
purely to classify words and delineate constituents, categorial features
are genuine instructions to be interpreted at the interfaces, as expected from
formal features under Chomsky’s (1995) Full Interpretation. A Switch like the
so-called Ger head is a nominal categorizer by virtue of its interpretable,
perspective-setting [N] feature. At the same time, Ger bears a [uV] feature,
eliminable via categorial Agree: it can therefore be part of a verbal projection,
behaving like a ‘functional category’. Thus, the complement of an [N][uV]
Switch element will be recategorized, in a way familiar from simplex categor-
izers (recall the denominal verb tape): the complement of Switch will be
interpreted in the sortal perspective that [N] imposes. Additionally, the [uV]
Probe on Switch will search for a [V] target. There is no categorial clash
whatsoever, given that the [uV] feature of the Switch, along with the ones on
Asp, Voice and so on, will be eliminated before Spell-Out; at the same time, this
feature guarantees that a Switch acts like a functional element: given its [uV]
feature it cannot take root material directly as its complement.
This state of affairs immediately derives Phrasal Coherence in (5). In order
to illustrate this point, let us follow the derivational history of a mixed
projection using abstract phrase markers like we did in (13).
(22)

First, a v (bearing [V]) is merged with the root material and a number of
[uV] functional heads (FHs) recursively merge, giving a categorially uniform
verbal/clausal subtree through successive applications of categorial Agree
between the [uV] features and [V].
Then an [N][uV] head, the functional categorizer or Switch, is merged; its
[uV] feature probes for a [V] and agrees with it.
 6.5 Switches as functional categorizers 147

(23)

However, the next head to be merged must be [uN], not [uV], on the grounds
of a version of minimality.5 We cannot merge a [uV] functional head with a
projection headed by a Switch because the Switch’s interpretable [N]
feature would intervene between a [uV] and the [V] on v, as illustrated below:
(24)

Consequently, the derivation will proceed as follows: the Switch head will
participate in the lower verbal/clausal subtree and effectively ‘begin’ the
nominal subtree dominating it, ‘switching’ the categorial identity of the deriv-
ation, it being a categorizer after all. The empirical result is that now Phrasal
Coherence is readily captured, with the functional categorizer acting both as
the Switching element (by [N]) and as the ‘glue’ (by [uV]) between the two
categorially distinct subtrees:
(25)

5
Uninterpretable [uV] creating an intervention effect for a probing [uN], and vice versa, is
independently necessary, as discussed in Chapter 5 in the context of deriving biuniqueness from
categorial Agree. This of course suggests that [N] and [V] are in fact values of an attribute
[perspective]. Thus, [N] must actually be [perspective:sortal] and [V] must be [perspective:
temporal], exactly as discussed in Chapter 5. I will, however, continue using [N] and [V] as
shorthand, for convenience.
148 Mixed projections and functional categorizers

Let us now briefly review how this proposal relates to the previous ones in
the literature. A primary virtue of the account presented here is that it does not
treat mixed projections as exceptional constructions resulting from weakened
labelling conditions – for example, allowing a V-headed constituent project as
an NP – or from the projection of categorially dual elements. Recall that this
second option is impossible, already by a version of Baker’s Reference–
Predication Constraint (2003, 165): the same syntactic node cannot be both
‘nominal’ and ‘verbal’. A Switch, being only nominal (as its [uV] will be
eliminated before it gets a chance to be interpreted), respects that.
Second, our analysis only utilizes the two categorial features independently
argued for and it invokes the concept of an uninterpretable feature as a Probe
for Agree to derive both ‘ordinary’ (i.e., categorially homogeneous) and
‘exceptional’ (i.e., mixed) projections. Focusing on the Switch head, it is
indeed a categorially dual head, as expressed by the intuition in Jackendoff
(1977), Pullum (1991), Lapointe (1993) and Bresnan (1997) – making the best
of the explanatory merits of the Agree system. Looking at the abstract phrase
marker in (25) or, more concretely, at the one of a Poss–ing gerund in (21), we
can observe that mixed projections are simply structures created when a
functional nominalizer, the Switch, selects a verbal projection, continuing
the line of reasoning of Baker (1985), Abney (1987) and Yoon (1996a).
However, in our case this nominalizer’s selection for a verbal projection comes
not as an ad hoc stipulation but due to Categorial Deficiency, precisely because
the nominalizer, Switch, is itself a functional element bearing a [uV]
feature.6
Summarizing, we capture the duality of mixed projections as the result of
the co-occurrence of an interpretable [N] and an uninterpretable [uV] feature
on the same syntactic node: mixed projections are not special in any way that
has to do with their phrase-structure status, and heads that are interpreted as
both nominal and verbal are not necessary in order to explain mixed
projections.

6.6 Morphologically overt Switches

Having argued for the existence of functional categorizers, it would be
interesting to review some instances of morphologically overt Switch heads
so as to examine their properties. Ackema and Neeleman’s (2004, 140)

6
See Section 6.10 of this chapter on why we need functional categorizers in mixed projections,
instead of ordinary ones such as n and v.
 6.6 Morphologically overt Switches 149

principle of Input Correspondence predicts that overtly realized Switch

heads, their category-changing affixes, will be found only in head-final
languages. This is a prediction that is borne out, at least in the small sample
of languages I have surveyed. Indeed, Korean, Turkish and Basque – all head-
final languages – each have overt functional categorizers. For reasons of
exposition clarified in footnote 19, I will only discuss Basque and Turkish
Switch heads here.
Basque contains a nominalizing element inserted among functional mor-
phemes which easily qualifies as a functional nominalizer according to Borsley
and Kornfilt (2000, 111–12). We therefore have examples like the following,
adapted from the same source:7
(26) [Jon-ek bere hitzak hain ozenki es-te-a-n] denok harritu
Jon-erg his words so loudly say-fn-d-iness all surprise
ginen.
aux
‘We were all surprised at John saying his words so loudly.’
(27) [Zu-k etxea prezio honetan hain errazki sal-tze-a-re-kin]
you-erg house price that-in so easily sell-fn-d-gen-with
ni-k ez dut ezer irabazten
I-erg neg aux anything win
‘I don’t get anything out of your selling the house so easily.’

The suffixal te/tze element, the functional nominalizer by hypothesis, nomin-

alizes Tense Phrases. The constituent it heads is the complement of a
Determiner –a, the same used with noun phrases, which in turn can be the
complement of a Kase element, as in (26), or of a postposition like with in
(27) – again, as happens with ordinary nominal constituents. The resulting
picture is therefore pretty straightforward.
Turkish presents a slightly more intriguing situation when it comes to the
morphological – or, rather, morphosyntactic – status of Switch heads.
Turkish, although it does possess a subordinating Complementizer, ki, is
famous for the fact that subordination in this language is done mainly via
nominalizing the phrasal complement. In (9), repeated below for convenience,
we saw an instance of a nominalized infinitive, known as ‘action
nominalization’ in the literature on Turkish:8

7
ERG ¼ Ergative Case, FN ¼ Functional Nominalizer, D ¼ Determiner, INESS ¼ Inessive Case.
8
Kornfilt (1997) is a classic description of the language in English.
150 Mixed projections and functional categorizers

(28) Hasan [Ayşe-nin gel-me-sin]-i iştiyor.

Hasan Ayşe-gen come-fn-3sg-Acc wants
‘Hasan wants Ayşe to come over.’
In the example above I provisionally gloss –me (whose underlying form is mA)
as the functional nominalizer, but this is a matter that we will not investigate
further. Suffice it to say for our purposes here that in ‘action nominalizations’,
like (28), the overt subject of the nominalized clause is invariably in the
genitive case; this corroborates an analysis thereof as nominalized infinitives.
The picture of nominalizations in Turkish becomes more stimulating
(and intriguing) when one turns to a different type of nominalized clause,
what Kornfilt (1997) and Borsley and Kornfilt (2000) term ‘factive
nominalizations’. The example below is from the latter:
(29) Ben [siz-in tatil-e çık-tığ-ınız-ı] duy-du-m.
I you-gen vacation-dat go.out-fn.past-2pl-acc hear-past-1sg
‘I heard that you had gone on vacation.’

In both (28) and (29), what follows the purported nominalizing suffixes, –me–
and –tığ– respectively, are purely nominal functional elements: nominal agree-
ment and (accusative) Case. Concentrating on (29), the important element in
examples like it is the underlying form of –tığ– (i.e., dIk) which Borsley and
Kornfilt (2000) gloss as ‘factive’. The interesting twist here is that the particu-
lar form also encodes past tense and that dIk is a form similar, although non-
identical, to the verbal past morpheme dI. Moreover, this ‘factive’ element also
comes in a future tense version, AcAk, which is identical to the verbal future
tense suffix. Hence, the example below forms a minimal pair with (29):
(30) Ben [siz-in tatil-e çık-acağ-ınız-ı] duy-du-m.
I you-gen vacation-dat go.out-fn.fut-2pl-acc hear-past-1sg
‘I heard that you will go on vacation.’

Calling an element that forces a future interpretation ‘factive’ is something of a

paradox. This is perhaps somehow reflected in Borsley and Kornfilt (2000,
108) taking both ‘action’ mA and the two versions of the ‘factive’ nominalizer,
dIk and AcAk, to be the realization of a nominal mood (MN) category. This
choice of term is quite telling because, I think, it reflects the inherent duality of
nominalizing dIk and AcAk, which are called (inevitably perhaps) nominal but
are also acknowledged as encoding ‘mood’, a verbal/clausal category which
here should be identified with Tense. If dIk and AcAk are indeed dual heads, as
the analysis goes, then their very duality reveals the kind of morphosyntactic
interactions a Switch can establish with other heads of the verbal/clausal
part of a mixed projection. Perhaps the [uV] feature on a Switch enables it
 6.6 Morphologically overt Switches 151

also to encode Tense features, making it both a genuine member of the verbal/
clausal projection line – and one carrying a temporal specification, too – and a
nominalizer at the same time. We could further speculate that this type of
Switch encoding temporal features is made possible in Turkish because
functional nominalizers, affixes in Ackema and Neeleman (2004, 172–81),
must be overt in head-final languages: it just happens that they additionally
carry temporal features.
However, unpublished work by Tosun (1999) offers a more elegant account
of how Turkish functional nominalizers of the ‘factive’ denomination end up
with past and future Tense specifications. First, she addresses analyses
according to which the ‘factive’ nominalizers, dIk and AcAk, are actually Tense
morphemes followed by a version of the Complementizer ki. In the case of dIk
this would entail analysing the form as dI (past) þ k (the Complementizer):
these forms would be the exponents of Tense þ Complementizer sequences.
However, evidence suggests that independent Tense/Aspect morphemes are
not allowed inside subordinate nominalized clauses:9
(31) gid-iyor-du-m in main clauses
go-imperf-past-1sg
‘I was going.’
(32) git-tiğ-im in embedded clauses
go-fn.past-1sg
(33) * gid-iyor-duğ-um in embedded clauses
go-imperf-fn.past-1sg
The above examples illuminate the following state of affairs: although two
Tense/Aspect morphemes may co-exist in Turkish main clauses, as in (31), no
such thing is possible in nominalized embedded ones (33), where an independ-
ent Tense head is not available. Actually, it is impossible to express aspectual
information in ‘factive’ nominalizations, as the ungrammaticality of (33)
demonstrates, and the correct form in (32) appears to encode Tense features
only via the nominalizer, which is dIk in this example. Tosun (1999, 7) claims
exactly this: that the ‘factive’ nominalizer – more precisely, a form like dIk
(i.e., –tığ–) in our (32) – ‘bears both tense and nominal features’. So, the above
examples take the same direction as Borsley and Kornfilt (2000, 108) in that
there is a single head encoding both nominal and temporal features.
Tosun (1999), however, argues against the existence of a single syntactic
head by looking at the availability of object shift and scope ambiguity with

9
I have adapted the glossing to reflect the working hypotheses here.
152 Mixed projections and functional categorizers

indefinites, ‘Diesing effects’ after Diesing (1992), in embedded nominalized

clauses. Tosun finds out that object shift is possible, when not obligatory, and
that Diesing effects available in all embedded nominalized clauses – that is,
both ‘action’ nominalizations with mA and ‘factive’ ones with dIk and AcAk
are available. According to Bobaljik (1995) and Bobaljik and Thráinsson
(1998), the availability of object shift and Diesing effects entails the presence
of two syntactic heads, each projecting a specifier. This in turns leads Tosun
to offer a Fusion analysis (Halle and Marantz 1993) within the Distributed
Morphology framework, according to which in nominalized embedded
clauses there are indeed two heads: a Tense head and a Gerundive head
(our functional nominalizer, the Switch), which provide the necessary
specifiers that enable or sanction object shift and permit scope ambiguities
regarding indefinites (the ‘Diesing effects’). These heads are then fused
together and subsequently either AcAk, a form identical to the verbal future
morpheme, or dIk, a form non-identical to the verbal past morpheme dI, is
inserted into this fused node.
The conclusion of this brief survey is that Switch heads may have
diverse morphological realizations. They are typically realized as null
morphemes in head-first languages. They may be realized as identifiable
morphemes in head-last languages like Korean (the –um element) and
Basque (the suffixal te/tze element). Finally, in Turkish they can fuse with
Tense heads, as both belong to the group of verbal/clausal functional heads by
virtue of their [uV] features.10

6.7 Switches and their complements

We have now reached a point where the issue regarding the position of
Switches must be addressed. I claimed that these heads are functional
categorizers, by virtue of their bearing an interpretable categorial feature.
However, in Chapters 3 and 4 I have sided with Marantz’s (2000) claim that
categorizers are phase heads – a prediction I have understood to stem precisely
from their bearing an interpretable categorial feature, which sets an interpretive
perspective for the categorizer’s complement and closes off root material.
Consequently, all other things being equal, projections headed by a Switch

10
In the spirit of van Riemsdijk (1998b) and Hegarty (2005), one could perhaps speculate that
fusion is only possible between heads bearing identical uninterpretable categorial features – that
is, if both heads are marked as [uN] or as [uV]. Unfortunately, in the context of this study I can
only offer this as mere speculation.
 6.7 Switches and their complements 153

should also be phases, as a Switch also bears an interpretable categorial

feature. Having said that, if we turn to the literature on mixed projections and
phasehood, to the best of my knowledge, there is nothing on whether subcon-
stituents within mixed projections constitute phases, let alone whether func-
tional categorizers constitute phase heads.11 As a consequence of this, the
discussion here will have a preliminary and exploratory character; we begin by
phrasing the question from an empirical point of view, in three versions:
(34) Phases within mixed projections
a. What types and sizes of nominal and verbal/clausal constituents can be
part of a mixed projection?
b. Where can we place a Switch within a derivation?
c. What are the possible types and sizes of Switch complements?

Let us now attempt to sketch an answer basing ourselves on theoretical

assumptions compatible with the theory of categorial features developed here.
If Switches contain categorial features, features which I have claimed in
Chapter 4 induce phasehood, then Switches themselves must trigger a
phase every time they are merged. This would in turn suggest that they can
never be merged mid-phase: phases are by definition interpretive units read-
able by the interfaces and inserting a Switch in the midst of a phase, would
induce something like an ‘incomplete phase’, a contradiction in terms. So, we
can begin by proposing that Switches can only be merged at the edge of
phases. Granting that, we can examine whether functional categorizers are
themselves heads inducing phases, phasal heads. We will now explore some
evidence that might lend support to this idea.

6.7.1 Locating the Switch: the size of its complement

Dutch nominalized infinitives are a type of mixed projection that has been
scrutinized and analysed in considerable detail, also with reference to the
cut-off point between the nominal and the verbal/clausal subconstituent.
Schoorlemmer (2001) has argued that there are two types of nominalized
infinitives: ‘expressive’ infinitives, which contain a large verbal/clausal sub-
constituent, a TP, and ‘plain’ infinitives, which only project an AspP.12

11
Most of the work on phases looks only at uniform verbal/clausal projections (CPs). The question
of whether or not (categorially uniform) DPs are phases has been addressed to a lesser extent,
with some representative discussion in Svenonius (2004) and Hicks (2009, chaps. 4 and 5).
12
This statement on ‘plain’ nominalized infinitives is refined and revised below, in Section 6.9.2.
Panagiotidis and Grohmann (2009) also address the matter; the discussion in this section revisits
some of their arguments and observations.
154 Mixed projections and functional categorizers

This state of affairs is exemplified below, with the verbal/clausal subtree

placed in brackets for convenience. The examples below are taken from
Ackema and Neeleman (2004, 173):
(35) Deze zanger is vervolgd voor dat [stiekem succesvolle liedjes
this singer is prosecuted for that [sneakily successful songs
jatten].
pinch.inf]

(36) Deze zanger is vervolgd voor dat stiekeme [succesvolle liedjes

this singer is prosecuted for that sneaky successful songs
jatten].
pinch.inf]
‘This singer is prosecuted for sneakily pinching successful songs.’13

In (35) the verbal/clausal constituent (‘expressive infinitive’) is large enough to

contain a projection hosting an adverb, stiekem (‘sneakily’). The Switch is
merged with this large verbal/clausal subtree and the superimposed nominal
part of the mixed projection seems to consist solely of the demonstrative dat.
(37) A verbal/clausal subtree with an adverb

Turning to the nominalized ‘plain’ infinitive in (36) the cut-off point

between the nominal and the verbal/clausal subtree of the mixed projection
is lower: the verbal/clausal projection is too small to contain an adverb (it is
nominalized below the position where adverbs attach) and this is why the
adjective stiekeme (‘sneaky’) is merged, instead of an adverb. In this case, the
superimposed nominal subtree is large enough for both the demonstrative dat
and for a position below it to host the adjective:

13
Schoorlemmer (2001) also explains away an apparent violation of Categorial Uniformity: in
nominalized infinitives nominal and verbal properties look like they can be interspersed, with
direct objects of the verb showing up as phrases headed by van (‘of’).

(i) nominal . . . verbal . . . nominal

She shows that low van phrases may merge inside the verbal phrase if they can later be checked
against a higher nominal functional structure, i.e. that dominating the verbal/clausal constituent.
 6.7 Switches and their complements 155

(38) An adverb-less verbal/clausal subtree

The validity of the structures outlined in (37) and (38) is corroborated by the
fact that in Dutch an adjective modifying the nominalized infinitive may
precede an adverb (Ackema and Neeleman 2004, 174):
(39) Deze zanger is vervolgd voor dat constante [stiekem succesvolle
liedjes jatten]
this singer is prosecuted for that constant [sneakily successful
songs pinch.inf]
The structure for the example above is given in the phrase marker below:
(40) Adjective plus a verbal/clausal subtree with an adverb

The reverse, an adverb preceding the modifying adjective, is not possible, however.
(41) *. . . dat constant stiekeme [succesvolle liedjes jatten]
that constantly sneaky successful songs pinch.inf
‘This singer is prosecuted for constantly sneakily pinching successful songs.’
Again, the structure for the ungrammatical adverb–adjective order is given in
the phrase marker below:
(42) An impossible state of affairs: an adverb within the nominal subtree

The impossibility of (42) is multiply interesting. First, it is yet another

example against the crude version of an analysis where functional elements
can mix for free and where there is no biuniqueness or categorial uniformity at
play. Second, it both illustrates Phrasal Coherence and is compatible with a
Switch analysis. Third, it also illustrates the impossibility of flip-flopping
156 Mixed projections and functional categorizers

(David Adger, personal communication, February 2007) – namely, a state of

affairs where we start building a verbal/clausal tree, then we switch (using a
Switch) to a nominal tree, only to switch back to a verbal/clausal one, and so
on. As to why flip-flopping is in principle impossible, we will try to work out
an answer in the following section.
Turning to Spanish, we observe a very similar set of possibilities: on the one
hand, nominalized infinitives with just a VoiceP-internal accusative object in
(43); on the other, nominalized infinitives containing a nominative TP-internal
subject in (45) (examples adapted from Ackema and Neeleman 2004, 178):
(43) El [tocar la guitarra] de Maria . . .
the play.inf the guitar of Maria
‘Maria’s playing the guitar . . .’

(44) [DP El [[SwitchP Switch [VoiceP tocar la guitarra]] de Maria]]

(45) El [cantar yo La Traviata] . . .

the sing.inf I La Traviata

(46) [DP El [SwitchP Switch [TP pro cantar yo La Traviata]]]]

In (44) a Voice Phrase includes the verb tocar (‘play’) and its object in
accusative Case, which is assigned by the Voice head. It must be that the
Switch takes that VoiceP as its complement, because the external argument
de Maria is expressed as an adjunct on the resulting nominal constituent, in a
manner reminiscent of that of ‘possessive’ subjects in English Poss-ing
gerunds. In (46) the verbal/clausal constituent is large enough to include a
Tense head, one that assigns nominative to a post-verbal subject yo (‘I’).14
A first conclusion is therefore that a Switch can appear in different positions
even in the same language: the complement of a Switch head in (36) is smaller
than the one in (35). Similarly, the complement of the Switch head in (43)
seems to be roughly the size of a VoiceP, excluding the nominative-assigning
structure, whereas in (45) the complement of the Switch seems to be the size
of (at least) a TP. Interestingly, in many languages, including Modern Greek,
a Determiner can appear in the syntactic nominalization of a full CP comple-
ment, a mixed projection:

14
At first glance, the nominative in the context of an infinitive is unexpected. Even if it is not a
complete TP that is nominalized here and even if the source of nominative is not T, it still
remains the case that the verbal/clausal constituent in (45) is large enough to contain a
postverbal subject. Yoon and Bonet-Farran (1991) discuss the Case-marking of Spanish infini-
tival subjects both in nominalized and in ‘sentential’ infinitives, arguing that Nominative is not a
default Case, but indeed is the result of Case-marking.
 6.7 Switches and their complements 157

(47) Ksero [to [poso sklira agonizeste]]

I.know the how.much hard you.are.fighting
‘I know how hard you are fighting.’
(48) [DP to [SwitchP Switch [ CP poso sklira agonizeste]]]

The preliminary conclusion is that it is indeed the case that verbal/clausal

constituents of various sizes can be the complement of Switch, even within
the same grammar, as Dutch and Spanish examples demonstrate.
Although we will return to this topic below, we need to say a few more
words about the nominalization of CP constituents before continuing: in their
discussion of Spanish nominalized infinitives, Yoon and Bonet-Farran (1991,
364–5) follow Plann (1981) in claiming that whenever a full CP is nominal-
ized, this is actually done via the mediation of an empty noun synonymous to
hecho (‘fact’). This would be quite plausible to the extent that such nominal-
izations would have a factive reading, which seems not always to be the
case with nominalized infinitives in Spanish, anyway (Rosemeyer 2012).
In the remainder of this subsection we will compare the two alternative
accounts for the nominalization of full CPs, which are schematically illustrated
in the phrasal markers below, with the empty noun (eN) analysis to the left and
the functional categorizer one to the right.
(49) Alternatives to [DP D CP]

Before proceeding, note that neither of the above accounts suffers from the
paradoxical state of affairs of an analysis where D directly selects a CP. A [DP
D CP] constituent would either violate biuniqueness, as it has D (a [uN] head)
select a CP, or would force us to concede that all Complementizers are
nominal – see Roussou (1990, sec. 4.1), Davies and Dubinsky (1998, 2001)
and Manzini (2010) for discussion of the (non-)nominal character of (all)
Complementizers.
So, let us now turn to languages with bona fide nominalizations of complete
CPs, such as Polish and Greek.15 Roussou (1990) already discusses the
implausibility of an account involving an empty noun, on the grounds that this

15
Borsley and Kornfilt (2000, 116–17) discuss Kabardian as one more language where full CPs
can be nominalized and be assigned Case.
158 Mixed projections and functional categorizers

noun cannot host an adjective or, indeed, any other element associated with
nouns. Expanding on this observation by Roussou, nothing can intervene
between the article to (the neuter gender and default form) and the CP in
structures like the one in (47). Moreover, the role of eN, the phonologically
empty non-descriptive noun of Panagiotidis (2003b), is very well understood in
languages like Greek: its lack of interpretation underlies a number of pronom-
inal and elliptical structures, and their functions.16 However, in the context of
nominalized CPs eN would have to mean ‘fact’, as in Plann (1981) and Yoon
and Bonet-Farran (1991, 364–5), ‘matter’, ‘question’ and the like. This range of
different interpretations for the empty noun eN does not look feasible.
This last point takes us to the relevant Polish and Greek structures. In both
languages, all subordinate CPs can be nominalized: declarative and interroga-
tive, in either indicative or subjunctive; in Polish demonstratives to, tego and
tym are used, whereas in Greek the default article to is used. Some data from
the two languages clearly illustrate this:17
(50) [To, że Maria zmienia pracę] Jan oznajmił. Polish
that comp Maria is.changing job Jan announced
‘Jan announced that Mary is changing her job.’
[To oti i Maria alazi dhulia] anakinose o Yanis. Greek
the comp the Maria is.changing job announced the John
‘John announced that Mary is changing her job.’
The nominalized declarative CPs have been fronted to show that
they form a single constituent together with the Det-element introducing them.
Note that the examples above are compatible with both of the alternatives
in (49), as a ‘fact’ empty noun between the demonstrative or the article and the
clause would be compatible with their interpretation – keeping in mind, of
course, the objections on the distribution, interpretation and compatibility with
adjectives that characterize true eN in Greek. These objections become even
more relevant in examples like the ones below:
(51) Jan rządał [(tego), żeby Maria zmieniła pracę]. Polish
Jan demanded that comp Maria changed job
‘Jan demanded that Maria change her job.’
[To na fiyis] ine efkolo. Greek
the irrealis leave.2sg is easy
‘For you to leave is easy.’

16
In the spirit of Harley’s (2005a) analysis, eN would have to be identified with an n head not
associated with any root material; see the discussion in Section 4.5 in Chapter 4.
17
Polish examples and discussion from Borsley and Kornfilt (2000, 113–14).
 6.7 Switches and their complements 159

(52) Jan zastanawia się nad [tym, czy kupić nowy samochód]. Polish
Jan wondered over that whether buy new car
‘Jan wondered whether to buy a new car.’
Eksetazis [to an tha fiyi]. Greek
examine.2sg the if will leave.3sg
‘You are considering whether s/he will go.’

(53) Jan zastanawia się nad [tym, kiedy kupić nowy samochód]. Polish
Jan wondered over that when buy new car
‘Jan wondered when to buy a new car.’
Epaneksetazo [to pote tha fiyi].
re-examine.1sg the when will leave.3sg
‘I am re-examining when s/he will leave.’
Epaneksetazo [to pion tha kalesis].
re-examine.1sg the who will invite.2sg
‘I am re-examining who you will invite.’

In (51) we have nominalized subjunctive CPs – thus the purported empty noun
would have to be interpreted not as ‘fact’ but as something else, although it is
unclear as to what exactly. In (52) the demonstrative tym in Polish and the
article to in Greek introduce nominalized subordinate yes/no questions and in
(53) an embedded wh-question with when and – for Greek – with who.
It becomes obvious from the above that no noun can mediate between the
D-element and the clause in nominalized CPs, even an abstract one meaning
‘fact’ and the like. At the same time, a Switch, a functional nominalizer,
inserted very high so as to take a complete CP as its complement, is compatible
with all the empirical facts presented and discussed in this section.

6.7.2 Phases and Switches

The follow-up question is, if the complements of a Switch can be of any size,
as Ackema and Neeleman (2004, 173) argue, whether a Switch head can take
any verbal/clausal constituent as its complement. As already announced, here
I will examine Lapointe’s (1999) intuition that the subtrees participating in a
mixed projection, the possible complements of a Switch head in our analysis,
have to be of particular sizes. The strong version of its claim is crystallized in
the statement below:
(54) Switches can be merged with phase heads, themselves inducing a phase.

Pausing for a moment, we must ask why we have to argue that complements of
Switch must be the size of phases. The reasons will have to be of a theory-
internal nature but, hopefully, should reflect some more generally received
160 Mixed projections and functional categorizers

ideas about how syntactic structures are dealt with at the interfaces. Consider
(21) and suppose that Ger (our Switch) indeed takes an AspP complement,
as illustrated in the diagram. Is an Aspectual Phrase complete in any sense?
Does it constitute a complete interpretive unit? The answer is most likely
negative: Aspectual Phrases apparently denote time intervals in which events
unfold, but these time intervals are neither anchored in time nor suitably
ordered with respect to an instance and/or another time interval.
Why would this matter? It would if we expect that the two uniform subtrees
in a mixed projection would be able to stand as complete interpretive units. In
previous work, I have precisely argued for this – namely, that the complements
of Switch must be complete interpretive units by themselves. In Panagiotidis
and Grohmann (2009) a central claim is that complements of Switch must be
the size of Prolific Domains (Grohmann 2003), whereas in unpublished work
I take complements of Switch to be the size of a phase, claiming that
‘Switches will be merged with phases and induce themselves a phase’,
exactly as in (54) above.
This last claim brings us to an interesting dilemma. If Switch heads bear
interpretable categorial features, interpretive perspective-setting features, then
we expect them to be phasal heads, like ‘ordinary’ categorizers n and v
(see Chapter 4): in other words, we expect them to induce a phase themselves.
So, the claim that ‘Switches . . . induce themselves a phase’ is consistent
with what we have seen so far on categorial features and phasehood. However,
if a Switch head is phasal, its complement certainly need not be: for instance,
the complement of a phasal head like a Complementizer is certainly not a
phase. Being more precise about the phasehood of categorizers, both n and v
and functional Switch, categorial features on them are sufficient to impose
an interpretive perspective on material that cannot be otherwise interpreted –
for example, root material – and to complete a phase with it. Thus, categorial
features on both a categorizer (n and v) and a Switch could surely also make
a phase out of any other material consisting of UG features, including func-
tional structures. It therefore is not necessary for the complement of Switch,
a phasal head, also to be a phase. A fortiori, if Richards (2007) is correct, it
cannot be a phase: phase and non-phase heads must actually alternate.18

18
Furthermore, if Richards (2007) is indeed correct, then denominal structures (e.g., the verb tape
in Chapters 4 and 5) cannot have a [v [nP]] structure and, respectively, nominalizations cannot
have an [n [vP]] structure: both n and v are phase heads. This would suggest that a projection
must intervene between them, but this is an empirical matter that will not be explored here.
 6.8 Are all mixed projections externally nominal? 161

Given that there is no theoretical need for the complements of Switches to

come in particular sizes – that is, the size of a phase – I will restrict myself to
closer scrutiny of what type of functional material these complements can
contain and how this affects their syntactic behaviour. But before embarking
on this, there is one matter that needs be addressed.

6.8 Are all mixed projections externally nominal?

Let us return to the generalization in (8), the ‘second’ generalization in the
literature regarding mixed projections – the first being of course Phrasal
Coherence in (5). According to (8), mixed projections display ‘nominal exter-
nal behaviour’: they externally behave as nominal constituents. We saw that
the nature and function of Switches, functional categorizers, captures
Phrasal Coherence. What about the curiosity that Nominal External Behaviour
seems to be?
So far, all the mixed projections examined – for example, Poss–ing gerunds,
nominalized infinitives and DþCP clauses – had the following general struc-
ture in terms of their categorial features (see the diagram in (25) as well):
(55) Mixed projections: nominal external behaviour

The higher part of (such) mixed projections – hence their ‘external’

behaviour – is nominal. According to the account developed here, this external
nominal behaviour is a direct consequence of the Switch’s feature specifica-
tion, [uV][N], which effectively nominalizes its complement and shifts the
categorial identity of the whole constituent to ‘nominal’ through its interpret-
able [N] feature. Apparently, a hypothetical Switch head with a [uN][V]
specification would take ‘nominal’ functional phrases and verbalize them,
rendering the mixed projection externally verbal. Such a mixed projection
would behave as a verbal and/or clausal constituent, despite having a noun
at its heart, as its lexical head. However, as Borsley and Kornfilt (2000) admit,

Some suggestions about what these intermediate projections could be may be extracted from
Alexiadou’s (2001) account on different types of nominalizations.
162 Mixed projections and functional categorizers

there are no unambiguous exceptions to (8) and the state of affairs in (55).
Nevertheless, weaker challenges to Nominal External Behaviour exist and
apparent exceptions to Nominal External Behaviour are actually attested, even
if they are not examples of verbal/clausal external behaviour. Following the
discussion in Panagiotidis (2010), I will present and discuss an example of a
mixed projection that externally seems to behave as a non-nominal constituent:
Japanese and Korean verbal nouns.

6.8.1 Verbal nouns

Verbal nouns (VNs) in Japanese and in Korean share some very remarkable
properties: morphologically they are nouns (Yoon and Park 2004) and no
special nominalizing morphology is attached to them, contrasting them, in
the case of Korean at least, with nominalizations of complete TPs, suffixed
by –um (Yoon and Park 2004). Furthermore, unlike what happens with the –um
nominalizations in Korean, no adverbs are possible with verbal nouns, although
adjectives are generally acceptable.19
Verbal nouns, however, display two prototypically verbal/clausal character-
istics: (a) they assign verbal Case in both Japanese (Iida 1987) and Korean
(Yoon and Park 2004), including –ga Nominative in Japanese; (b) they project
full argument structures; see Tsujimura (1992, 477–9), Manning (1993) and
Yoon and Park (2004), from where the following Korean example is adapted:
(56) [Kim-paksa-ka woncahayk-ul yenkwu]-cwung-ey cencayng-i
Kim-dr-nom atom.nucleus-acc research-midst-loc war-nom
ilena-ss-ta.
broke.out-pst-decl
‘The war broke out while Dr Kim was researching the atom nucleus.’
In the example above, (functional material associated with) the VN yenkwu
(‘research’) assigns not just accusative, like English Poss–ing gerunds, but
also nominative, to a Theme woncahayk-ul (‘atom nucleus’) and an Agent
Kim-paksa-ka (‘Dr Kim’) respectively.

19
Yoon (1996b) discusses –um and points out that it can also be used as a lexical nominalizer,
presumably an n, directly attached on verbs: for example, cwuk (‘die’) is nominalized as cwuk-
um (‘death’ or ‘dying’). Now, in (59) I argue that verbal nouns in Korean (and Japanese) also
involve an n head, which is (crucially for the analysis here) phonologically null. If this is the
case, then Korean –um can be the morphological exponence
(a) either of a functional categorizer, a Switch, when it nominalizes TPs,
(b) or of an n, when it nominalizes vPs in the absence of a Switch further up the tree.
In the structure in (62) there exist both a (low) nominalizer n and a (high) functional
nominalizer, a Switch, so n surfaces as a null morpheme.
 6.8 Are all mixed projections externally nominal? 163

Now, unlike the other mixed projections reviewed in this chapter, VNs
cannot be arguments but are typically embedded within modifying expressions
with a temporal interpretation, as this Japanese example from Shibatani
(1990, 247) illustrates:
(57) [Sensei-ga kaigai-o ryokoo]-no sai . . .
teacher-Nom abroad-Acc travel.vn-gen occasion
‘On the occasion of the teacher’s travelling abroad . . .’
Alternatively, VNs can combine with a copula/light verb (the equivalent of do) to
yield the Light Verb Construction (Yoon and Park 2004); in these cases they
contribute the predicative content to the complex verbal predicate. Summarizing:
(58) Verbal nouns
a. are morphologically simplex nouns that may be modified by adjectives;
b. also contain a verbal/clausal layer that may license an Agent and assign
accusative and nominative Case, but not adverbs;
c. must contain a high nominal layer, which enables them to be complements
of temporal adpositions and which makes them possible (incorporated?)
arguments for light verbs.20

Let us put all those ingredients together and see what kind of structure
emerges. Our rationale here will be that the diverse characteristics of verbal
nouns must be the results of features structurally interacting with each other, in
the spirit of the analysis in Ahn (1991), albeit with different results. Thus, we
start with the fact that VNs display nominal morphology, precisely the way
that English gerunds always contain a verbal morphological chunk. Under the
fairly innocuous assumption that adjectives adjoin to noun phrases, we would
get the following structure as the bottom of a VN tree:
(59) A nominal subtree

Apparently, the nominal chunk in (59) is verbalized immediately above the nP.
Here a first crucial dilemma emerges, especially for our theory of categorial
features, which enables grammars to possess both categorizers, n and v, and
functional categorizers – that is, Switches: which of the two links an nP like
the one in (59) with the verbal/clausal functional material directly dominating it?
In principle, both a v and a Switch specified as [V][uN] would be possible here

20
This last point makes sense if one considers that in complex verbal predicates the non-light verb
element is necessarily non-verbal – recall the discussions about Farsi and Jingulu in Chapter 2.
164 Mixed projections and functional categorizers

and both would do the same job.21 If there is some economy metric according to
which a v head, consisting of a single [V] feature, is more economical than a
[uN][V] head, then we can perhaps argue that an ordinary categorizer, a v, will
do here. The intuitive idea behind this assumed economy metric here is as
simple as not inserting uninterpretable and unvalued features unless we have
to. Already suggested in the discussion below (17) is that a lexical categorizer
can categorize lexical material but not functional constituents and that a func-
tional categorizer, a Switch, will not be used to convert purely lexical material
(see also Section 6.10). Incidentally, a VN, which keeps its ability to take
adjectives, is in this analysis nothing more than a transparent denominal verb,
like tape. One of the ways it differs from tape, however, is that there is no
morphological fusion of the material under n with the material under v, and the
two heads remain separate with v being silent:
(60) A verbalized nP

What immediately follows is a fully blown verbal clausal layer, possibly a full
TP, if Nominative in Japanese and Korean is assigned by T, and given that
nominative arguments are possible with VNs. A Voice head, assigning
Accusative, is also required, in order to capture examples like (56) and (57)
above, where accusative Case is assigned.
(61) A verbalized nP with its verbal/clausal projections

Finally, to complete the picture adumbrated in (58), we need to add a

nominal layer on top of the extended verbal/clausal subtree. A Switch will

21
Assuming of course that, for the sake of the argument, the unattested functional verbalizers, [V]
[uN] Switch heads, exist. Borsley and Kornfilt (2000, 120) claim that ‘there are no nominal
properties that reflect a nominal functional category located below a verbal functional category’,
having stipulated that ‘[c]lausal constructions with nominal properties are a consequence of the
association of a verb with one or more nominal functional categories instead of or in addition to
the normal verbal functional categories, appearing above any verbal functional categories’
(102). This is equivalent to saying that no functional verbalizers, [V][uN] Switch heads, exist.
 6.9 The properties of mixed projections 165

be used here, as the transition will now be from (verbal/clausal) functional to

(nominal) functional material.22
(62) VN: a mixed projection containing a transparently denominal verb

The provisional conclusion is that there is still no solid evidence for a

[V][uN] Switch. So, it seems that either there is a mysterious asymmetry
here or that we need to look harder. I am not going to pursue this apparent
asymmetry any further here, with the exception of footnote 3 in the Appendix.

6.9 The properties of mixed projections

We have seen that mixed projections are possible because functional categor-
izers, Switch heads, exist. From this, we have also easily derived Categorial
Uniformity and the dual categorial properties of mixed projections. Functional
categorizers, bearing an interpretable categorial feature, are phasal heads,
meaning that they induce a new interpretive unit. Empirically speaking, we
saw that these verbal/clausal subconstituents that a Switch can turn into an
interpretive unit can be
• Voice Phrases, as in Spanish nominalized infinitives in (44);
• Aspect Phrases, as in English Poss–ing gerunds in (21) and in Dutch
‘plain’ nominalized infinitives in (36);
• Tense Phrases, as in Spanish nominalized infinitives in (46), in Dutch
‘expressive’ nominalized infinitives in (37) and in Korean and
Japanese verbal nouns in (62);
• Complementizer Phrases, as in Greek DþCP in (48) and Turkish
nominalized clauses in (9).
What we now need to turn to, as promised earlier in the chapter, is a closer
examination of how the kind of functional material contained in mixed projec-
tions affects their syntactic behaviour. The nominal subconstituent of mixed
projections seems to be quite restricted: with exceptions like that of

22
More discussion on VNs and mixed projections used exclusively as modifiers can be found in
Panagiotidis (2010) and in references therein.
166 Mixed projections and functional categorizers

adjectivally modified Dutch infinitives in (36), it consists of just the Switch

and a Determiner, possibly needed for the purposes of argumenthood. We will
hence examine the functional structure of the verbal/clausal subtree of mixed
projections and discuss how such scrutiny can explain some of their properties.

6.9.1 Similarities: Nominalized Aspect Phrases in English and Dutch

It has been argued that both English Poss–ing gerunds in (21) and Dutch
‘plain’ nominalized infinitives in (36) are nominalized AspPs. Indeed they
share a number of common properties, which can be taken to follow from their
lack of a Tense head and from the constitution of their verbal/clausal functional
structure. One interesting similarity between English gerunds and Dutch
‘plain’ nominalized infinitives is their obligatory subjects, albeit never in the
nominative.23 These properties are easy to capture: Dutch and English are both
non-null subject languages, therefore it is impossible for a null subject other
than PRO to be generated inside a verbal/clausal (sub-)constituent. However,
Poss–ing gerunds and ‘plain’ nominalized infinitives both contain an AspP
subtree, complete with the full argument structure of the verb. Given that the
Tense node, the licenser of PRO, is radically absent, the subjects must be overt.
In Poss–ing gerunds subjects are licensed by the possessive Determiner
c-commanding the AspP, whereas in Dutch they surface as van-phrases,
thanks to the licensing properties of the nominal layer dominating the AspP
(Schoorlemmer 2001, secs. 5 and 6) (see also footnote 13).
Concluding, the two mixed projections share the same properties despite the
different morphology on the verb, infinitival in Dutch and aspectual in English.
The similarity is due to the fact that in both structures, ‘plain’ nominalized
infinitives in Dutch and Poss–ing gerunds in English, the same verbal subtree
is nominalized, an Aspect Phrase.

6.9.2 Differences: two types of Dutch ‘plain’ nominalized infinitives

Zooming in on Dutch ‘plain’ nominalized infinitives, as we saw in Section
6.7.1 and in the previous subsection, Schoorlemmer (2001) identifies them as
mixed projections in which the verbal/clausal subtree is the size of an Aspect
Phrase. However, Schoorlemmer (2002) takes a closer look at ‘plain’ nomin-
alized infinitives and finds out that they actually fall into two distinct classes.
Each of these classes is characterized by a clustering of properties.

23
See Siegel (1998) and Pires (2006, chap. 1) on why Acc–ing and Pro–ing gerunds are not
mixed projections but bare TPs.
 6.9 The properties of mixed projections 167

The first class is ‘plain’ nominalized infinitives with adverbial modification,

which are different from the ‘expressive’ nominalized infinitives (Schoorlem-
mer 2001) reviewed in Section 6.7.1. ‘Plain’ nominalized infinitives in this
first class have propositional readings and become severely degraded if an
event reading is forced upon them by the context:24
(63) ‘Plain’ nominalized infinitives with adverbs: propositions only
??Het gebeurde tijdens [het snel tenten opzetten van Jan] Dutch
It happened during the quickly tents pitch.inf of Jan
‘It happened during John’s quickly pitching tents.’
[Het snel tenten opzetten van Jan] staat buiten kijf
the quickly tents pitch.inf of Jan is beyond dispute
‘The quick pitching of tents by John is beyond dispute.’

Furthermore, adverbially modified ‘plain’ nominalized infinitives can support

object shift and specific or definite direct objects. They can also appear in the
‘perfect tense’:
(64) ‘Plain’ nominalized infinitives with adverbs: perfect aspect
[Het hardnekkig scheidsrechters belaagd hebben] ...
the persistently referees harassed have.inf
‘The persistently having harassed referees . . .’

This battery of properties – that is, adverbial modification, exclusively prop-

ositional readings, the possibility of specific objects, Object Shift and perfect
aspect – is compatible with the analysis in Schoorlemmer (2001) of these
nominalized infinitives as containing an AspP verbal/clausal subtree:
(65) ‘Plain’ nominalized infinitives with an Aspect Phrase

The Aspect node can be correlated with the licensing of adverbs, the
obligatory propositional reading, the providing of a specifier for Object Shift
and, of course, the possibility for perfect aspect.

24
All examples in this subsection are from Schoorlemmer (2002). The phrase markers in (65)
and (68) have been adapted in order to reflect this framework, which incorporates the Switch
head hypothesis.
168 Mixed projections and functional categorizers

The above class of ‘plain’ nominalized infinitives contrasts with one that has
very different properties, although it also constitutes a ‘low’ nominalization,
with the Switch taking a ‘small’ verbal/clausal complement. Such ‘plain’
nominalized infinitives, when modified by an adjective, may have event
readings. Compare the following with (63):
(66) ‘Plain’ nominalized infinitives with adjectives: event reading possible
Het gebeurde tijdens [het snelle tenten opzetten van Jan]
It happened during the quick tents pitch.inf of Jan
‘It happened during John’s quick pitching of tents.’

However, in adjectivally modified structures, no specific readings for the direct

object and no perfect aspect is possible; compare the following with (64):
(67) ‘Plain’ nominalized infinitives with adjectives: no perfect aspect
??[Het hardnekkige scheidsrechters belaagd hebben] ...
the persistent referees harassed have.inf
‘The persistent having harassed referees . . .’

This battery of properties – that is, adjectival modification, the possibility

of event readings, the ban on specific objects and Object Shift and the
impossibility of perfect aspect – clearly defines this class of nominalized
infinitives. Schoorlemmer (2002) claims, correctly I think, that these are all
the result of a structure from which the Aspect head (and its specifier) are
radically absent: there is no host for adverbs, there are no aspectual features
to support ‘perfect tense’ and, of course, there is no proposition-creating
Asp category. Additionally, the specifier that would host the shifted object
and give rise to a specific reading thereof with the object taking scope over the
rest of the verbal/clausal subtree, à la Diesing (1992), is not there, either.
Schoorlemmer (2002) proposes the following structure:
(68) ‘Plain’ nominalized infinitives without an Aspect Phrase

Besides the language-specific interest of the above analysis, I believe it

serves as a case study of how we can follow a promising methodological
 6.9 The properties of mixed projections 169

blueprint: every time two types of mixed projections appear to differ along a
battery of properties, the first attempt at explaining their differences should be
to examine if this battery can be correlated with the presence versus the
absence of a particular functional head inside the verbal/clausal subtree.25
In the case of low nominalizations in Dutch examined here, this was achieved
by correlating a battery of properties (event/proposition reading, adverbs vs
adjectives, Object Shift, specificity and aspect) with the presence or absence of
an Aspect head.
However, not all differences among mixed projections can be explained
along the lines of the presence versus the absence of a category: sometimes
structures containing the same verbal/clausal subconstituents can display very
different behaviours.

6.9.3 Fine-grained differences: different features in nominalized

Tense Phrases
Consider now mixed projections with a TP verbal/clausal constituent: Spanish
nominalized infinitives with nominative postverbal subjects in (46), Dutch
‘expressive’ nominalized infinitives in (37) – see also Schoorlemmer (2001),
and Korean and Japanese verbal nouns in (62). These are structures that
behave in very different ways, as digested in (69).
(69) Comparing mixed projections: nominalized TPs
Overt subjects In nominative Adverbs

Spanish nominalized yes yes yes

infinitives
Dutch nominalized infinitives no no yes
Korean/Japanese verbal nouns yes yes no

Following the methodology of Alexiadou (2001), I am suggesting that the

different properties of the above mixed projections can be reduced to the kind
of functional heads that participate in their verbal/clausal subconstituent.
The most predictable behaviour is that exhibited by the Dutch expressive
infinitives: these being infinitives, their TPs are headed by a defective Tense
head that cannot assign nominative, hence cannot license overt subjects in
SpecTP. The Spanish version of a nominalized Tense Phrase is morphologic-
ally an infinitive; however, its T head looks like it can assign nominative

25
Of course, this is the line of reasoning followed in Section 6.7.1, where we teased apart TP
‘expressive’ nominalized infinitives and AspP ‘plain’ ones in Dutch.
170 Mixed projections and functional categorizers

Case and support non-null subjects, hence the two Tense heads have different
feature content. Similar facts hold for Korean and Japanese verbal nouns: both
overt subjects and nominative assignment are possible. What we have in
Spanish and Korean/Japanese is a type of Tense head that is reminiscent of
that in ‘absolute’ Greek gerunds (Panagiotidis 2010), with ‘quasi-independent
temporal reference’.26 In any case, differences can be accounted for by the
different feature makeup of the Tense head. Finally, the inability of adverbs to
be licensed within the TP of verbal nouns must again be a result of the feature
content or even the absence of the relevant functional head.

6.10 Why functional categorizers?

In Chapter 3 we looked into the category-changing function of v with regard to
the derivation of denominal verbs like tape, with identical facts being true of n,
of course. We saw there that a categorizer can take an already categorized
constituent, an nP or a vP, and recategorize it.27 As surveyed there, structures
such as [v [nP]], which give us denominal verbs like tape and dígest, contrast
with verbs such as hammer and digést, derived via direct root categorization –
that is, [v RootP]. Moreover, I have argued in Chapter 4 that categorizers are
lexical heads, the only lexical heads, as they are the only syntactic nodes
capable of categorizing roots and root material and, therefore, support lexical
content. The assumption so far has been that categorizers n and v can have only
two kinds of complement:
a. root material, to which they assign an interpretive perspective in order
to render it interpretable at LF and ‘matchable’ with a concept;
b. nPs or vPs, which they transparently convert.
The obvious question at this point is why a second type of categorizer must be
posited in this chapter: the functional categorizer which we nicknamed
Switch here. Can we not have an n categorizer mediate between the lower
verbal/clausal and the higher nominal part of all the mixed projections
surveyed, all of them displaying nominal external behaviour? Why can mixed
projections not be simple cases of recategorization via a simple nominalizer n?

26
‘Absolute’ gerunds are the closest Modern Greek has to infinitives: their Tense head is not
morphologically expressed, it can license temporal (not just aspectual) adverbs, sanction quasi-
independent temporal reference, license periphrastic perfect tenses with an auxiliary, license a
pro subject, and assign nominative to an overt subject (Panagiotidis 2010, 173).
27
But recall footnote 18.
 6.10 Why functional categorizers? 171

This is a question to which I have not been able to provide a final, rock-solid
answer, so what follow are comments and intuitions of a more speculative
nature, with some statements verging on the programmatic.
First of all, the difference between recategorizations and mixed projections
can generally be described as follows: when a verb is nominalized or a noun is
verbalized, it is lexical material (nP and vP) that participates as the lower
subtree; it is purely lexical material that undergoes the conversion. On the
other hand, in mixed projections, it is both functional and lexical material that
participates as the lower subtree: in mixed projections we ‘recategorize’ a
chunk of functional structure.
I think that we need to explicitly state a, so far elusive, principle to capture
this situation: categorization of root and lexical material is the job of a lexical
categorizer, whereas (re)categorization of functional material is the job of a
functional head, a functional categorizer. In other words, we need to find a
way, following on from general principles, that bans what looks like a counter-
intuitive state of affairs: a lexical head interrupting a purely functional struc-
ture. Future research must derive the impossibility of the following structure,
where the Switch in (25) has been replaced by a lexical nominalizer n:
(70)

Let us now see if any insight can be offered to help ban the likes of (70).
Generally, if n and v could freely take functional complements, then denominal
verbs and deverbal nouns and mixed projections, on the other hand, should
behave in exactly the same fashion; if this is indeed the case then we can
perhaps do away with functional categorizers. A second point is this: n and v
(or, more acrimoniously, the features [N] and [V]) appear to be necessary and
universal; all languages have n and v. If the lexical categorizers n and v were
responsible for mixed projections in addition to (re)categorizations, then
functional categorizers would be unnecessary. However, I am not confident
that all languages have mixed projections – but this is truly a matter of
empirical enquiry. Finally, arguing that the element which makes mixed
projections is a functional element with a particular feature structure – that
172 Mixed projections and functional categorizers

is, [X][uY] and, possibly, only [N][uV] – enables us to better frame the
problem of the apparent absence of a functional verbalizer, of a [V][uN]
Switch.

6.11 Conclusion
This chapter brings together two approaches to mixed projections, one arguing
them to be categorially dual constituents and one conceiving of them as
projections of a nominalizing head. I claim that they are indeed categorially
dual constituents as a result of their containing a functional categorizer: in the
same way that a lexical categorizer recategorizes lexical constituents, a func-
tional categorizer recategorizes functional ones.
Evidence for the existence of functional categorizers was presented from
head-final languages: Basque and Turkish. The existence of functional cat-
egorizers in natural language grammars was shown to be consistent with the
theory of categorial features presented and advanced here: it results from the
possibility of a categorial feature [N] to co-exist with an uninterpretable
categorial feature [uV] as parts of the same feature bundle, of the same head.
This nominalizing verbal/clausal functional head is understood to be a phasal
head and can take verbal/clausal constituents of a variety of sizes. Moreover,
as expected, verbal/clausal constituents of the same size – say, Aspect
Phrases – may contain elements with different feature content. This derives
the different behaviour of different mixed projections, as expected from a
formal approach: different constituents and different features entail distinct
grammatical behaviours and distinct batteries of grammatical properties.
The puzzle of what looks like the exclusive nominal external behaviour of
mixed projections has been recast as the result of only nominalizing Switch,
an [N][uV] head, existing. This merely rephrases the problem, one that future
research will resolve either by deriving empirical evidence that verbalizing
Switch heads – that is, [V][uN] elements – are possible or by deriving their
impossibility on independent principles.
7 A summary and the bigger picture

7.1 A summary
The theory in this monograph attempts to explain why nouns are different from
verbs and why they are most probably universal. In order to achieve this goal,
it builds upon a series of empirical discoveries, theoretical advances and
methodological principles of more than 50 years of work in generative gram-
mar: empirical discoveries like those regarding the nature of long-distance
dependencies, currently captured as the operations Agree and (internal) Merge,
aka ‘Move’; theoretical advances such as analysing the properties of lexical
items on the basis of their formal features – this being a theory of categorial
features, after all; and methodological principles like the functional–lexical
distinction.
At the same time, research in this monograph has deliberately striven to
synthesize insights, concepts and findings from a variety of frameworks and
approaches to grammatical structure and to language in general. Investigating
mixed projections, I turned to both LFG and HPSG in order to gain an
understanding of how these structures are organized and whether they are
exceptional or the consequence of principles applying everywhere else –
Categorial Deficiency, in our case. Far more importantly, in order to gain an
understanding of what kind of interpretive content each of the word classes
‘noun’ and ‘verb’ could have, I turned to functionalist and typological
approaches, like Baker (2003) did. Crucially, Langacker (1987) and Anderson
(1997) also proved major influences, in their combining notional approaches to
lexical categories with a firm conviction that their interpretation is one of
perspective, conceptualization or ‘grammaticalization’ of concepts – as
opposed to a viewpoint according to which lexical categories form large
pigeonholes into which different concepts are sorted.
Finally, this theory of categorial features attempts to deconstruct word class
categories. Beginning with a criticism of weak lexicalism, I embrace syntactic
decomposition of categories, in the version developed within the Distributed

173
174 A summary and the bigger picture

Morphology framework, so as to capture four desiderata: first, the syntactic

relevance of categorial features, which is well-hidden on occasion and hard
to discern in lexicalist approaches; second, a clear and precise formulation
of the actual interpretation of categorial features at the interface between the
Language Faculty in the Narrow sense (FLN) and the Conceptual–Intentional/
SEM systems; third, the impossibility of uncategorized roots participating in
syntax, the Categorization Assumption (Embick and Marantz 2008, 6), which
can also be understood as the reason why there are no free-standing category-
less words; finally, the role of categorial features in making lexical word
classes, creating functional heads, and sanctioning the existence of mixed
projections. This last desideratum is interweaved with the problem of ‘idio-
maticity’ of words, the fact that morphological structure does not necessarily
entail transparent compositional interpretations – something we explained as
the result of categorial features interacting with semantically defective roots.
The theory proposes two unary categorial features: [N], which sets a sortal
interpretive perspective, and [V], which sets a temporal/sub-eventive interpret-
ive perspective. There is evidence that [N] and [V] are possibly different
values – that is, [sortal] and [temporal] – of a [perspective] feature. The
discussion about whether one of the two features (or values) is unmarked
remained inconclusive in the absence of solid evidence to support such claims.
Categorial features ‘make’ categorizers, which are the only lexical heads in
grammar, in that they can support ‘descriptive content’ by taking root material
as their complements. The process of categorization is both about setting the
perspective of the categorizers’ complement, whether this be root material or
an already categorized constituent, and about completing an interpretive unit,
a (First) Phase. Categorizers, lexical heads, are the only necessary elements
within a projection: roots are actually not, as the existence of semi-lexical
heads – that is, n and v without a complement – demonstrates.
The existence of uninterpretable categorial features endows grammar with
functional heads, each of them flagged by one of these features. Functional
heads are thus members of nominal and verbal/clausal supercategories respect-
ively: therefore they enable the association in local configurations of outlying
nominal and verbal/clausal features with other constituents such as arguments
and modifiers.
The interaction – that is, the Agree relations – between categorial features is
also crucial: uninterpretable categorial features probe for interpretable categor-
ial feature Goals, thus deriving two very important aspects of structure-
building: first, categorial Agree indirectly constrains Merge, moulding syntactic
structures to the effect that, for instance, lexical material only appears at the
 7.2 Loose ends 175

bottom of trees; second, categorial Agree relations play a significant role

in deciding the head of the constituent after each application of Merge.
The theory developed in this monograph goes beyond answering the ques-
tions which it was designed to capture: how categorial features define lexical
categories, what nouns and verbs are and how they differ from each other, the
nature of functional heads, and the need for roots to be categorized. Its reliance
on LF-interpretable categorial features enables this theory to make a precise
hypothesis on the Janus-like element that lies at the heart of a mixed projec-
tion. This functional categorizer, a verbal functional head and a nominalizer
at the same time, as far as attested cases go, gives the mixed projection its
‘mixed’ – that is, its categorially dual – character. It is neither a noun nor a
hybrid head interpreted simultaneously as nominal and verbal, but a category-
changing element that also belongs to the functional entourage of a verb.
This in turn fits in nicely with the categorial uniformity of mixed projections,
their being made up of two categorially consistent subtrees, a nominal one and
a verbal/clausal one: at the point of categorial switch, the functional categorizer
is found, gluing together two distinct constituents.
Summing up, a feature-based analysis of word classes enables us to go
beyond lexical word classes as primitives: lexical categories are neither primi-
tives nor are they necessarily organized in word classes – as Farsi illustrates
pretty uncontroversially. At the same time, the fundamental grammatical and
interpretive role of categorial features (licensing root material, supplying
fundamental interpretive perspectives, making a phase) ensures that they will
be found in every syntactic projection, creating the only indispensable element
in it: the lexical head.

7.2 Loose ends

There is a number of loose ends stemming from the theory presented and
argued for in this monograph. Three of these are reviewed below, and they
concern broader matters as opposed to more empirical ones, or matters of
execution (e.g., the true domain of ‘lexical meaning’, the behaviour of roots
as if they are ‘lexically categorized’ or abiding by particular phonological
restrictions in some languages etc.).
An immediate issue is the interaction between Agree and Merge. Suppose
that, as in Section 5.9.4 of Chapter 5, Agree has a role in deciding the label of a
projection. It follows from the analysis developed therein that every time
Merge assembles an SO (syntactic object) from two [uX] (‘functional’)
LIs (lexical items), it will not be possible to decide the label and the object
176 A summary and the bigger picture

should somehow be discarded before it even reaches the LF-interface to be

evaluated. Agree may block the application of Merge. Is this possible? Does it
reveal something about the relation between Agree, Merge and labelling?
And if indeed it does, what decides the label when XP and YP merge, where
no [uX] (uninterpretable categorial) features are at play?
A second matter concerns the uninterpretability of categorial features that
make functional heads. Is there a way to recast uninterpretable categorial
features as unvalued without losing empirical coverage – for example, with
respect to biuniqueness? Speaking of the correct description of (categorial)
features, one has to wonder why [sortal] and [temporal] are the only (?) values
that the [perspective] attribute can take, giving us [N] and [V]. Are other values
inconceivable? Possibly not. How can we then justify the prominence (to say
the least) of sortality and temporality? Apparently, this will have to be done
with reference to what matters as fundamental perspective for the Conceptual–
Intentional systems, with FLN perhaps responding to some ‘virtual conceptual
necessity’. Having said that, at this point we may wonder whether we are better
off replacing [sortal] with something like ‘perceived as extending uninter-
rupted in some spatial domain’, as in Langacker (1987) and Uriagereka
(1999). The abstractly spatial as nominal versus the abstractly temporal as
verbal definitely looks like an elegant pair, but I can present no systematic
arguments, let alone arguments of an empirical nature, to justify the superiority
of a [spatial] interpretation for [N] over the much–better-understood [sortal]
one that I am vouching for.
A third big question is that of adjectives. I think we have seen adequate
evidence for three generalizations: (i) adjectives are not the unmarked lexical
category and they are possibly not universal; (ii) if they are a lexical category,
they are most likely not of the same ilk as nouns and verbs; and (iii) Degree is
not an ‘adjectival functional head’: adjectives possess no functional category
biuniquely associated with them. What are they, then? I hope to scrutinize the
categorial status of adjectives in future research.

7.3 Extensions and consequences

A familiar pattern emerges when one considers the concepts, assumptions,
hypotheses and proposals of this theory: natural objects, like FLN (the
Language in the Narrow sense), reveal their workings under two conditions:
once they are examined through the lens of a precisely articulated theory and
once they are viewed at the right level and with the necessary amount of
abstraction. To rehearse a rather trivial example from Chapter 2, the noun–verb
 7.3 Extensions and consequences 177

distinction is not discernible if we are not mindful of misleading surface

patterns, if we do not separate lexical nouns and verbs from their functional
entourage and before we have some clearly spelled-out theoretical and meth-
odological principles. Gazing at surface patterns is very often fascinating but
hardly ever revealing.
Moving on, it is firmly hoped within the generative tradition that a theory of
Universal Grammar can be formulated on the basis of empirical evidence
coming from a single language. However, looking, for instance, at verbs being
not word classes but syntactic categories in Farsi, Jingulu and the like, one can
hardly overlook the necessity of cross-linguistic evidence in our successfully
detecting the range, parametric or other, of linguistic phenomena: the limits of
variation, in other words. Similar conclusions can be drawn once we look
at mixed projections beyond Poss-ing gerunds: a limited empirical base was
one of the reasons for the impasse that work on mixed projections reached in
the 1970s and the 1980s – another one being the difficulties of having
categorially dual endocentric projections: a theoretical difficulty. An issue
related to the necessity of cross-linguistic evidence is the constant need to
steer clear of the Scylla of eurocentrism and the Charybdis of exoticization: a
reasonable way to go is to recognize the recurrent patterns, categories and
distinctions in natural language looming just below the surface (or even
deeper), while keeping in mind that such patterns, categories and distinctions
are not necessarily the ones prominent in Germanic, Romance, Japanese and
Semitic. To wit, Verb Second, pronominal clitics, Topic prominence and
triconsonantal roots all reveal something crucial about Universal Grammar,
something as crucial as classifiers, anticausatives, applicatives, incorporation
and so on. And all have to be treated at the right level of abstraction.
Another question is that of the lexical–functional distinction. It is my
conviction that understanding functional heads as satellites, or ‘the entourage’,
of lexical categories is on the right track, as suggested already in work from the
nineties reviewed in Chapter 5. On top of that, Categorial Deficiency captures
the defective character of functional categories, a character that becomes
manifest once one takes a look at the acquisition of first and second language,
language breakdown and language change – remember the overview in
Chapter 5 and, in more detail, in Muysken (2008). Unless there is a solid
and generalized FLN-internal factor that contributes to the ‘vulnerability’ or
the ‘late/no acquisition’ of functional categories, such behaviours will remain
curiosities and the pervasive pattern suggesting that functional elements ‘lack’
something will go unexplained. I think that conceiving functional heads as
categorially deficient, despite their otherwise crucial feature content, provides
178 A summary and the bigger picture

a sound basis for capturing their behaviour in language acquisition, diachrony

and language disorders.
The treatment of mixed projections in this monograph captures, as
mentioned before, fundamental properties such as their categorial uniformity,
while upholding biuniqueness. The key to capturing both their ubiquitous
presence and their categorial duality lies not in positing special conditions
and conventions but in holding fast to the conviction that surface complexity
is reducible to structural simplicity and to having the right theory (one of
categorial features, in our case) in which to describe functional categorizers.
The moral in this case is pretty obvious: sometimes one has to wait.
8 Appendix: notes on Baker (2003)

8.1 Introduction
There is a non-negligible point of criticism that can potentially be raised against
the theory of word class categories presented and discussed in this book, one that
views nouns and verbs as by-products of two fundamental LF-interpretable
features that set interpretive perspectives. The point would be roughly as
follows: Baker (2003) has already developed in detail a theory of lexical
categories based on two unary categorial features; moreover, Baker’s theory is
based on a wealth of empirical data. Why is a theory like the one presented here,
one that builds on assumptions, arguments and discoveries by Baker, necessary
at all? Why do we need yet another generative theory of word class categories?
First of all, Baker’s (2003) book exclusively discusses lexical categories:
nouns, verbs, and adjectives. Here, departing from categorial features as
makers of the word classes ‘noun’ and ‘verb’, we see that their uninterpretable
versions are necessary elements in the creation of functional heads and,
consequently, in the assembling of supercategories (Extended Projections);
we also argue that Agree relations among categorial features both affect
structure-building and play a central role in defining the label of a projection
after the application of Merge. We finally make good on our understanding of
the workings of categorial features so as to derive the (ultimately unexcep-
tional) nature of mixed projections. To be fair, some of the issues mentioned
are insightfully touched upon in Baker (2003, 2008), but not in the systematic
fashion we scrutinize them and account for them here.
A second point is that Baker’s theory, while recognizing the importance of
categories and their unary features being LF-intepretable, does not place
an emphasis on features themselves but on nouns, verbs and adjectives as
virtually primitive lexical categories. This is, of course, to be expected since,
in his system, [N] features exclusively appear on lexical nouns (and, possibly,
Determiners) and [V] features on lexical verbs (and, possibly, a functional
predicator Pred). Categorial features are not particularly active in his theory:

179
180 Notes on Baker (2003)

they merely create the above categories, nouns with their referential indices
and verbs with their specifiers; they then quietly wait to be interpreted at LF.
A third matter, of which this Appendix will be dedicated to providing an
overview, is the fact that a number of arguments and theses in Baker’s (2003)
theory of lexical categories (and his understanding of categorial features) are
the object of extensive controversy. Some of those arguments and theses
are quite central to his theory; others are primarily of methodological concern
or matters of interpreting empirical data. The purpose of this Appendix is
to discuss those controversial points in Baker’s (2003) account that merit a
rethink; it will also remind the reader of how the theory presented here
addresses these points, hopefully in a satisfactory fashion.

8.2 Are nouns referential?

Baker (2003, 95) opens his discussion of nouns with the following claim,
‘the leading idea’ of his account:
(1) a. Semantic version: nouns and only nouns have criteria of identity, whereby
they can serve as standards of sameness.
b. Syntactic version: X is a noun if and only if X is a lexical category and
X bears a referential index, expressed as an ordered pair of integers.

Nouns bearing a referential index has as its consequence a Noun Licensing

Condition (Baker 2003, 96 et seq.): nouns can only be licensed as arguments
or as members of chains.1 Baker (2003, 98) explicitly states that ‘nouns . . .
(and their projections) constitute the canonical argument phrases’ and that
‘nouns are always inherently argumental as a matter of Universal Grammar’
(116) – that is, even in the languages in which Chierchia (1998) argues that
Determiner Phrases are argumental instead.
The functioning of nouns as predicates is a problem for Baker’s analysis.
In examples like the ones below, but not exclusively, hero is not referential:
it does not serve as an argument and it is not a member of a chain:
(2) a. Alex is/became a hero.
b. They made/consider Alex a hero.
c. Alex, a hero, has always been close to us.

Baker (2003, 34–9; 2003, chap. 3 passim) sets out to explain away these and
more complex examples by taking predicate nominals to be always embedded
within a predicate-making functional projection: Bowers’ (1993) Pred.

1
Functional projections like CP and DP may also bear referential indices (Baker 2003, 139).
 8.3 Predication and specifiers 181

This Pred head acts as the predicator while its specifier hosts the subject of
predication. According to this recasting of nominal predicates, these actually
have the following structure:
(3) Baker’s predicative noun configuration

The general practice of making nouns (and adjectives) predicative by

embedding them within a functional PredP is itself not without problems, as
we will see below.
A general issue here is with Baker’s thesis that ‘nouns are always inherently
argumental as a matter of Universal Grammar’ (2003, 116), a thesis which is
both empirically very difficult to justify and generally runs against the consen-
sus in the literature. Even if nouns denote sortal concepts – a hypothesis that
seems to be accurate and that has not been seriously contested – this does not
automatically make nouns referential and/or argumental. Indeed, the majority
view among scholars is that nouns are not referential. On the contrary,
referentiality is understood as something that Determiners ‘do’; DPs are
referential, NPs are (usually) not: Higginbotham (1985), Stowell (1991),
Longobardi (1994), Chierchia (1998), Borer (2005, chaps. 3–6), Acquaviva
(2009a).2 I will not say more on this debate here; suffice it to say that it is at
least problematic to suggest that the two ‘versions’ of (1), semantic and
syntactic, are directly related or, a fortiori, that they are different realizations
of the same principle. Nouns are indeed sortal predicates – but they are
inherently predicative, just like verbs and, probably, adjectives as well.

8.3 Syntactic predication, semantic predication and specifiers

There is a far more serious issue regarding the foundations of Baker’s theory:
namely, the interpretation of the [V] feature. Baker (2003, 23) posits that
(4) X is a verb if and only if X is a lexical category and X has a specifier.

2
Again, Chierchia (1998) provides the more or less standard account on what happens with
argumenthood in Determiner-less languages. Also to be consulted: Cheng and Sybesma (1999),
Massam, Gorrie and Kellner (2006), Willim (2000), Bošković (2008).
182 Notes on Baker (2003)

This is far from a purely mechanistic syntax-internal characteristic; on the

contrary the projection of specifiers is viewed as the distinctive characteristic
of verbs, which in turn are understood as the one truly predicative category:
‘only verbs are true predicates, with the power to license a specifier,
which they typically theta-mark’ (Baker 2003, 20). In short, throughout the
second chapter of the book (i) verbs are understood as predicates, to the
exclusion of nouns and adjectives, and (ii) specifiers are understood as subjects
(of verbs). Nouns and adjectives can function as predicates only when they
are embedded inside the complement of Pred, a functional category which is
the equivalent of V; see also (3).
There are three very important problems here. The first is that syntactic
predication mechanisms and semantic predication do not correspond to each
other on a one-to-one basis: in his treatment of verbs-as-lexical-predicates,
one may criticize Baker for blurring ‘syntactic’ predication – that is, the
syntactic mechanisms matching a predicate and a subject – with semantic
predication itself. The whole issue has been extensively discussed in
Rothstein (1983, 1999) and elsewhere, and I will not make any more detailed
comments here.
Second, turning to syntax proper and its mechanisms, it is empirically very
odd to make a claim that nouns and adjectives do not project specifiers. Baker
argues against all nominal and adjectival specifiers and recasts them as speci-
fiers of PredP which embeds nouns and adjectives in their predicative uses.
Again, without getting into the heart of the matter, arguing convincingly
against all specifiers of lexical heads would by itself fill a monograph-sized
piece of research.
A third and even more peculiar claim that Baker makes is that specifiers are
essentially identified with subjects of predication. He also notes (2003, 25) that
functional projections acquire specifiers via movement – that is, they do not
have base-generated specifiers – although he reserves doubts about measure
phrases as the specifiers of Degree Phrases and possessors as the specifiers of
Determiner Phrases. In brief, the hypothesis can be summarized as follows:
verbs and Pred heads are the only categories that project specifiers (by virtue of
their [V] feature), which are identified with subjects of predication. The other
two lexical categories, nouns and adjectives, do not project specifiers, and
functional projections acquire specifiers via movement: moved/internally
merged specifiers (like SpecTP) apparently do not function as subjects.
The claim is too big and complex to evaluate, and one suspects it is flawed
and largely misguided.
 8.4 Adjectives as the unmarked lexical category 183

8.4 Are adjectives the unmarked lexical category? Are they roots?
Baker (2003, 190–2) claims that there is ‘nothing special’ about adjectives:
adjectives are exactly the lexical category which is not marked for any
categorial features. This entails that a non-categorized root, one not bearing
an [N] feature (forcing a ‘referential index’ for Baker) or a [V] feature
(resulting in the ‘projection of a specifier’), must surface as an adjective.
Accordingly, adjectives are the lexical categories that lack any particular
properties – that is, that lack categorial features: they are neither predicates
(like verbs) nor referential (like nouns). A very prominent consequence
of this take on adjectives, should it be true, is that adjectives would be
essentially uncategorized – in stark contradiction to the Categorization
Assumption. On top of that, as remarked in Chapter 2, if category-less roots
surface as adjectives, expressing ‘pure properties’, then all languages should
possess an adjective category. This is of course a claim Baker (2003, 238–63)
dedicates extensive discussion to, despite ample typological evidence to the
contrary: again, see Chapter 2. Furthermore, adjectives, as a default category
virtually indistinguishable from uncategorized roots, should at least some-
times surface as morphologically simple elements, with each root available in
a language having at least one adjective derived from it. However, recall that
adjectives in most languages involve more rather than fewer functional
layers – for example, they display concord with the noun inside a DP, just
like functional elements like quantifiers, demonstratives, possessives and
articles. Second, very few adjectives in Romance, Germanic and Greek are
underived, leading us to the implication that Adjective is definitely not the
unmarked category, let alone one co-extensive with roots. Third, if adjectives
are not marked by any categorial feature whatsoever, how are adjective-
making derivational affixes marked? Fourth, not all roots in a language
derive an adjective, contrary to what should be expected for an unmarked
(uncategorized) word class.
Finally, adjectives are taken not to project full phrase structure; thus they
can take no complements when used as attributive adjectives (Baker 2003,
195–6): ‘attributive adjectives cannot take complements, and cannot appear
with a fully fledged degree system, they can be a little more than just a head . . .
Apparently, it is possible for one head to adjoin to another to make a new head
within an attributive modifier, but that is all’ (196 footnote 5). Under a version
of the generalized X’ schema, the above sounds very odd – namely, banning a
whole lexical category from taking complements on the grounds of its being
unmarked. Effectively, this is a ban on Adjective Phrases, at least as attributive
184 Notes on Baker (2003)

modifiers; cross-linguistically speaking, this constitutes a claim that looks

largely misguided – for example, in the face of circumstantial data from Greek.
(5) Enas [poli perifanos ya tin kori tu] pateras.
A very proud for the daughter his father
‘A father very proud of his daughter.’
More detailed discussion of the above and other points regarding the purported
unmarkedness and the ‘elsewhere’ lexical category status of adjectives may be
found in Chapter 2.

8.5 Pred and other functional categories

One of the merits of Baker’s theory is that he sets out to give interpretive
content to the features [N] and [V], rather than simply view them as convenient
labels that create taxonomies. As repeated several times, this is a desirable
move and – hopefully – a momentous one in our understanding of lexical
categories. Given the pervasive, fundamental and universal character of the
noun–verb distinction, he expectedly chooses fundamental interpretations for
his categorial features [N] and [V]. This is a conceptually desirable, if not
necessary, move: it would be eminently odd if categorial features encoded
peripheral or parochial interpretations, after all.
Taking [N] to stand for ‘has a referential index’ and [V] for ‘projects a
specifier’ has some conceptual side-effects. Beginning with [V] as forcing the
projection of a specifier, the conclusion is that all predicative categories, both
lexical and functional, must be marked for [V]; all referential categories must
similarly bear an [N] feature. This observation is of course already incorpor-
ated in the last pages of Baker (2003, 324–5) and the resulting state of affairs is
tabulated in (6), slightly revised from the table in Baker (2003, 325):
(6) Baker’s system: a typology3
Functional/
Lexical transparent Functional/opaque

specifier Verb Aspect, Tense Pred

referential Noun Det, Num, Case functional
index nominalizers
neither Adjective Degree Adposition

3
The typology in (6) suggests that Pred (Bowers 1993) could be perhaps the ‘verbalizing’
equivalent of ‘category-shifting functional heads’ (‘functional nominalizers’), to which switch
(Chapter 6) belongs. This is an exciting topic for future research.
 8.6 Co-ordination and syntactic categorization 185

There are several things to be said about the above table, which effectively
summarizes some key consequences of Baker’s account. First, adjectives
and degree elements ending up lumped together with adpositions is truly
counterintuitive, although it is a claim consistent with the discussion through-
out Baker’s monograph. Second, and crucially, there is no way to capture the
lexical–functional distinction using categorial features only; the lexical–
functional distinction is tacitly assumed to constitute a kind of an unexplained
primitive. Because of this blind spot, some basic questions remain
unanswered: how do we distinguish a verb from a Pred head? How do we
distinguish a noun from a Determiner? And so on. Third, a further distinction
between opaque (category-shifting) categories and transparent categories that
are members of the supercategory/Extended Projection to which the lexical
head belongs is made. Unfortunately, this difference between opaque and
transparent categories remains unexplained, though one account is offered in
Chapter 6 of this monograph: opaque categories must be Switch heads.
My purpose is of course neither to engage in nit-picking nor to vindicate the
theory presented here over its predecessor and a valuable source of intuition,
discoveries and methodological solutions. On the contrary, I wish to show that,
when proposing a theory of lexical categories or, in our case, of categorial
features, we need to look at the consequences for elements beyond lexical
categories.

8.6 Two details: co-ordination and syntactic categorization

In the final section of this Appendix I will quickly look into two more specific
points in Baker’s analysis. The first is with regard to co-ordination data and
their interpretation as support for the non-predicative nature of bare nouns
and adjectives. The second concerns Baker’s lexicalist commitments blurring
into an account of categories that needs to call upon syntactic categorization.
Baker (2003, 37–9) presents potent empirical evidence in support of predi-
cate nouns and adjectives in reality being nouns and adjectives embedded
within projections of Pred. This evidence comes from co-ordination facts like
the following:
(7) a. *Eating poisoned food made Chris [A sick] and [V die].
b. *A hard blow to the head made Chris [V fall] and [N an invalid].
c. I consider John [A crazy] and [N a fool].

The above sentences exemplify co-ordination options between different cat-

egories. Causatives and secondary predication are deliberately selected as the
186 Notes on Baker (2003)

environments in which to illustrate such options, so as make sure that the

co-ordinated constituents contain as little functional structure as possible.
The idea, of course, is to apply the constituency test already known from
syntax textbooks – that is, that only identical categories may be co-ordinated.
In the first two examples in (7) we have the expected mismatch between the
verb and the other category – adjective in (7)a. and noun in (7)b. – inducing
ungrammaticality. The important one is (7)c., where we have a well-formed
case of co-ordination with an adjectival and a nominal conjunct. Baker
takes this to suggest that both nouns and adjectives are embedded within
Pred projections, thus making the two conjuncts categorially identical in
(7)c. Of course, PredPs are still categorially distinct from VPs; that is why
(7)a. and (7)b. are ungrammatical.
The argument is not trouble-free, however. First of all, both V and Pred bear
a [V] categorial feature, so apparently they are distinct because the former is
lexical and the latter functional. Having said that, Baker (2003, 201, 210)
argues that there are two varieties of Pred: PredA, taking adjectival comple-
ments, and PredN, taking nominal complements. Apparently, these are cat-
egorially identical for co-ordination purposes in (7). Finally, and more
generally, in recent years the growing consensus has been that co-ordination
facts involve identical semantic categories, rather than syntactic projections;
this resolves a host of paradoxical situations where, for instance, co-ordination
of a PP and an adverb is licit.4
(8) The package arrived [safely] and [on time].

Without getting into too much detail here, if VPs are – roughly speaking –
events and predicate nouns and adjectives in isolation are predicates, then the
examples in (7) can be accounted for without appealing to PredPs.
Let us now turn to the final issue, that regarding theoretical commitments.
In Chapter 5 of his book, Baker mainly discusses whether grammatical
category is inherent or syntactically assigned, as supported here. He argues
extensively against the syntactic categorization hypothesis, while he acknow-
ledges that his theory runs in parallel with it (Baker 2003, 267). In the
pages that follow (268–75), he offers a solid, coherent and very convincing
critique against radical categorylessness. He first calls upon evidence from
Incorporation to support the finding that distinct categorial behaviour in syntax
exists even in the absence of ‘any functional superstructure dominating the

4
Ellipsis and gapping are notorious for introducing much noise when co-ordinated, and are
set aside.
 8.6 Co-ordination and syntactic categorization 187

functional head’ (268), suggesting that lexical category is inherent, as opposed

to assigned by a categorizer. Thus, it is not the case that anything can
incorporate into anything, as one might expect from bare acategorial roots;
actually, the reverse seems to hold. So, although Greenlandic, Mayali and
Nahuatl allow APs to partly incorporate into verbal roots, they do not permit
incorporation of bare adjectival ‘roots’; similarly in Quechua, Chichewa and
Japanese, causatives take full AP and NP small clause complements but can
only tolerate bare verbal roots to incorporate into them. Finally, incorporation
into nouns is ‘universally impossible’ (272). Baker then argues that syntactic
categorization has a more limited explanatory power than an account where
every element in syntax bears category. The reason is that if lexical elements
are category-less roots, the impossibility of combining some roots with Num,
and some roots with T – biuniqueness, in other words – becomes a mystery
(see Chapter 5). Finally, Baker anticipates Embick and Marantz’s (2008)
Categorization Assumption: acategorial roots do not appear anywhere in
syntax; they must be embedded within a categorizing structure (Baker 2003,
269). Certain morphological processes are rather permissive to category,
seemingly manipulating category-less roots, in stark contrast with comparable
syntactic processes. So, root compounding can have pretty much anything
category-wise as a first member: draw-bridge (V), dog-house, sky-high (N),
dark-room, red-hot (A), cranberry, huckleberry (X). Syntactic processes like
attributive modification, on the other hand, can only be of the adjective–noun
type (271–2).5
However, his account is more syntactic decompositional than the above
criticism suggests. As almost admitted in a footnote in Baker (2003: 269), it is
virtually impossible to distinguish between acategorial roots and categorially
featureless adjectives or degree elements or ‘heavy’ adpositions like above,
behind and so on; consider also (6). This renders moot the distinction between
a category-less root and an adjective, with things becoming quite serious when
adjectives are derived – for example, heart-y, clue-less, industri-al, meteor-ic
and so on – as already mentioned several times.
Second, Baker (2003, 79–83) proposes that we syntactically decompose
verbs into a V head and an adjective (essentially a category-less root).6 His
discussion of the semantics of V is uncomfortably close to accounts of

5
Interestingly, all these points of criticism have been addressed by recent research on syntactic
decomposition and categorization, including this book. De Belder (2011) and De Belder and Van
Koppen (2012) contain some exciting work on compounding.
6
There is no escape from a complex structure for verbs – see also Chapter 4.
188 Notes on Baker (2003)

standard syntactic decomposition for verbs, such as Folli and Harley (2005) or
even Ramchand (2008). In other words, how is verb ¼ V þ A different from
verb ¼ v þ ROOT, especially if adjectives are not marked for categorial
features? Generalizing somewhat, and bringing nouns into the discussion,
how is Baker’s structure for predicative nouns in (3), repeated below for
convenience as (9), different from the one in (10)?
(9) Baker’s predicative noun configuration

(10) Predicative nouns à la syntactic decomposition

The question is anything but facetious. For Baker’s view to have explana-
tory weight as a hypothesis which is in contrast with syntactic decomposition,
it must first be shown that the Pred and the N nodes in (9) are indeed distinct.
This is very difficult to ascertain within Baker’s framework because according
to it nouns by definition do not project specifiers. Consequently, predicative
N will always be strictly adjacent to Pred and, if Pred is phonologically null
and/or morphologically non-distinct from the N head, then it is extremely
difficult to establish the existence of both. Even if this is successfully done,
it is also very hard to show that both the Pred and the N nodes in (9) are
categorially marked.
So, it turns out that the differences between (9) and (10) are indeed very
narrow.
 References

Abney, Steven P. 1987. The English Noun Phrase in Its Sentential Aspect. PhD thesis,
MIT, Cambridge, Mass.
Ackema, Peter, and Ad Neeleman. 2004. Beyond Morphology: Interface Conditions
on Word Formation. Oxford Studies in Theoretical Linguistics 6. Oxford
University Press.
Acquaviva, Paolo. 2009a. ‘The Roots of Nominality, the Nominality of Roots’. http://
ling.auf.net/lingbuzz/000824
2009b. ‘Roots and Lexicality in Distributed Morphology’. In York Papers in Lin-
guistics, edited by Alexandra Galani, Daniel Redinger and Norman Yeo, 2:1–21.
University of York. http://ling.auf.net/lingbuzz/000654
Acquaviva, Paolo, and Phoevos Panagiotidis. 2012. ‘Lexical Decomposition Meets
Conceptual Atomism’. Lingue E Linguaggio XI (2): 105–20.
Adger, David. 2003. Core Syntax: A Minimalist Approach. Core Linguistics. Oxford
University Press.
2006. ‘Remarks on Minimalist Feature Theory and Move’. Journal of Linguistics 42:
663–73.
2010. ‘A Minimalist Theory of Feature Structures’. In Features, Perspectives on
a Key Notion in Linguistics, edited by Anna Kibort and Greville G. Corbett,
185–218. Oxford University Press.
Ahn, Hee-Don. 1991. Clausal Architecture in Korean and English. PhD thesis,
University of Wisconsin, Madison.
Alexiadou, Artemis. 2001. Functional Structure in Nominals: Nominalization and
Ergativity. Amsterdam: John Benjamins Publishing Company.
Alexiadou, Artemis, and Gereon Müller. 2007. ‘Class Features as Probes’. In Inflec-
tional Identity, edited by Asaf Bachrach and Andrew Nevins, 101–55. Oxford
University Press.
Amritavalli, R., and K. A. Jayaseelan. 2004. ‘The Genesis of Syntactic Categories
and Parametric Variation’. In Generative Grammar in a Broader Perspective:
Proceedings of the 4th Asian GLOW, edited by James Hye Suk Yoon, 19–41.
Seoul: Hankook Publishing Company.
Anagnostopoulou, Elena, and Yota Samioti. 2009. ‘Domains for Idioms’. Paper
presented at the Roots Workshop, June, University of Stuttgart.
Anderson, John M. 1997. A Notional Theory of Syntactic Categories. Cambridge
Studies in Linguistics 82. Cambridge University Press.

189
190 References

Arad, Maya. 2003. ‘Locality Constraints on the Interpretation of Roots: The Case
of Hebrew Denominal Verbs’. Natural Language and Linguistic Theory 21:
737–78.
2005. Roots and Patterns: Hebrew Morpho-Syntax. Dordrecht: Springer Verlag.
Aronoff, Mark. 1994. Morphology by Itself: Stems and Inflectional Classes. Linguistic
Inquiry Monographs 22. Cambridge, Mass: MIT Press.
2007. ‘In the Beginning Was the Word’. Language 83: 803–30.
Aronoff, Mark, and Frank Anshen. 1998. ‘Morphology and the Lexicon: Lexicalization
and Productivity’. In The Handbook of Morphology, edited by Andrew Spencer
and Arnold M. Zwicky, 237–47. Blackwell Handbooks in Linguistics. Oxford:
Blackwell.
Baker, Mark C. 1985. ‘The Mirror Principle and Morphosyntactic Explanation’.
Linguistic Inquiry 16: 373–415.
1988. Incorporation: A Theory of Grammatical Function Changing. University of
Chicago Press.
2003. Lexical Categories: Verbs, Nouns, and Adjectives. Cambridge Studies in
Linguistics 102. Cambridge University Press.
2008. The Syntax of Agreement and Concord. Cambridge Studies in Linguistics 115.
Cambridge University Press.
Barner, David, and Alan Bale. 2002. ‘No Nouns, No Verbs: Psycholinguistic
Arguments in Favor of Lexical Underspecification’. Lingua 112: 771–91.
Basilico, David. 2008. ‘Particle Verbs and Benefactive Double Objects in English: High
and Low Attachments’. Natural Language and Linguistic Theory 26: 731–73.
Beard, Robert. 1995. Lexeme-Morpheme Base Morphology: A General Theory
of Inflection and Word Formation. SUNY Series in Linguistics. Albany: State
University of New York.
Bloom, Paul. 2000. How Children Learn the Meanings of Words. Cambridge, Mass.:
MIT Press.
Bobaljik, Jonathan. 1995. Morphosyntax: The Syntax of Verbal Inflection. PhD thesis,
MIT, Cambridge, Mass.
Bobaljik, Jonathan, and Höskuldur Thráinsson. 1998. ‘Two Heads Aren’t Always
Better than One’. Syntax 1: 37–71.
Borer, Hagit. 2003. ‘Exo-Skeletal vs. Endo-Skeletal Explanations: Syntactic Projec-
tions and the Lexicon’. In The Nature of Explanation in Linguistic Theory, edited
by John Moore and Maria Polinsky, 31–67. Stanford, Calif.: CSLI Publications.
2005. In Name Only. Oxford Linguistics 1. Oxford University Press.
2009. ‘Roots and Categories’. Paper presented at the 19th Colloquium on Generative
Grammar, April, University of the Basque Country, Vitoria-Gasteiz, Spain.
Borsley, Robert, and Jaklin Kornfilt. 2000. ‘Mixed Extended Projections’. In
The Nature and Function of Syntactic Categories, edited by Robert Borsley,
101–31. New York: Academic Press.
Bošković, Željko. 2008. ‘What Will You Have, DP or NP?’. In Proceedings of NELS
37, 101–14.
Botwinik-Rotem, Irina, and Arhonto Terzi. 2008. ‘Greek and Hebrew Locative
Prepositional Phrases: A Unified Case-Driven Account’. Lingua 118: 399–424.
Bowers, John. 1993. ‘The Syntax of Predication’. Linguistic Inquiry 24: 591–656.
 References 191

Bresnan, Joan. 1997. ‘Mixed Categories as Head Sharing Constructions’. In Proceed-

ings of the LFG97 Conference, edited by Miriam Butt and Tracy Holloway-King,
1–17. Stanford, Calif.: CSLI Publications.
Brody, Michael. 1995. Lexico-Logical Form: A Radically Minimalist Theory.
Linguistic Inquiry Monographs 27. Cambridge, Mass.: MIT Press.
Caramazza, Alfonso, and Argye E. Hillis. 1991. ‘Lexical Organization of Nouns and
Verbs in the Brain’. Nature 349 (6312): 788–90. doi:10.1038/349788a0.
Chametzky, Robert. 2000. Phrase Structure: From GB to Minimalism. Generative
Syntax 4. Oxford: Blackwell.
2003. ‘Phrase Structure’. In Minimalist Syntax, edited by Randal Hendrick, 192–225.
Oxford: Blackwell.
Cheng, Lisa Lai-Shen, and Rint Sybesma. 1999. ‘Bare and Not-so-Bare Nouns and the
Structure of NP’. Linguistic Inquiry 30: 509–42.
Chierchia, Gennaro. 1998. ‘Reference to Kinds across Languages.’ Natural Language
Semantics 6: 339–405.
Chomsky, Noam. 1957. Syntactic Structures. ’s-Gravenhage: Mouton & Company.
1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.
1970. ‘Remarks on Nominalization’. In Readings in English Transformational
Grammar, edited by Roderick Jacobs and Peter Rosenbaum, 184–221. Waltham,
Mass.: Ginn & Company.
1986. Knowledge of Language: Its Nature, Origin, and Use. Convergence.
New York: Praeger.
1995. The Minimalist Program. Current Studies in Linguistics 28. Cambridge,
Mass.: MIT Press.
2000. ‘Minimalist Inquiries: The Framework’. In Step by Step: Essays on Minimalist
Syntax in Honor of Howard Lasnik, edited by Roger Martin, David Michaels and
Juan Uriagereka, 89–155. Cambridge, Mass.: MIT Press.
2001. ‘Derivation by Phase’. In Ken Hale: A Life in Language, edited by Michael
Kenstowicz, 1–52. Cambridge, Mass.: MIT Press.
2004. ‘Beyond Explanatory Adequacy’. In Structures and Beyond, edited by Adriana
Belletti, 104–31. Oxford University Press.
2008. ‘On Phases’. In Foundational Issues in Linguistic Theory: Essays in Honor of
Jean-Roger Vergnaud, edited by Robert Freidin, Carlos Peregrin and Maria Luisa
Zubizarreta, 133–66. Cambridge, Mass.: MIT Press.
Cinque, Guglielmo. 1999. Adverbs and Functional Heads. Oxford University Press.
Clahsen, Harald, Susanne Bartke and Sandra Göllner. 1997. ‘Formal Features in
Impaired Grammars: A Comparison of English and German Children’. Journal
of Neurolinguistics 10 (2–3): 151–71.
Clark, Eve V., and Herbert H. Clark. 1977. ‘When Nouns Surface as Verbs’. Language
55: 767–811.
Collins, Chris. 1997. Local Economy. Linguistic Inquiry Monographs 29. Cambridge,
Mass.: MIT Press.
Corbett, Greville G. 1991. Gender. Cambridge Textbooks in Linguistics. Cambridge
University Press.
Cormack, Annabel, and Neil V. Smith. 1996. ‘Checking Theory: Features, Functional
Heads, and Checking-Parameters’. UCL Working Papers in Linguistics 8: 1–40.
192 References

Corver, Norbert. 1997. ‘The Internal Syntax of the Dutch Extended Adjectival
Projection’. Natural Language and Linguistic Theory 15: 289–368.
2005. Copular -Ly. Unpublished ms. University of Utrecht.
Corver, Norbert, and Henk C. van Riemsdijk, eds. 2001. Semi-Lexical Categories:
The Function of Content Words and the Content of Function Words. Studies in
Generative Grammar 59. Berlin: Mouton de Gruyter.
Croft, William. 1991. Syntactic Categories and Grammatical Relations: The Cognitive
Organization of Information. University of Chicago Press.
2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective.
Oxford University Press.
Davies, William D., and Stanley Dubinsky. 1998. ‘Sentential Subjects as Complex NPs:
New Reasons for an Old Account of Subjacency’. In CLS 34–1: Papers from the
Main Session, edited by M. Catherine Gruber, Derrick Higgins, Kenneth S. Olson
and Tamara Wysocki, 34: 83–94. Chicago Linguistic Society.
2001. ‘Functional Architecture and the Distribution of Subject Properties’. In Objects
and Other Subjects: Grammatical Functions, Functional Categories, and
Configurationality, edited by William D. Davies and Stanley Dubinsky, 247–79.
Dordrecht: Kluwer.
Davis, Henry. 1999. On Nouns and Nominalizations in Salish. Unpublished ms.
University of British Columbia, Vancouver.
De Belder, Marijke. 2011. Roots and Affixes: Eliminating Lexical Categories from
Syntax. Utrecht: LOT.
De Belder, Marijke, and Marjo Van Koppen. 2012. ‘One Module, Different Levels of
Merge: ANN Compounds in Dutch’. Paper presented at the Brussels Conference
on Generative Linguistics 7, 17–18 December.
Déchaine, Rose-Marie. 1993. Predicates across Categories. PhD thesis, University
of Massachusetts, Amherst.
Déchaine, Rose-Marie, and Martina Wiltschko. 2002. ‘Decomposing Pronouns.
Linguistic Inquiry’. Linguistic Inquiry 33: 409–22.
Den Dikken, Marcel. 2006. Relators and Linkers: The Syntax of Predication, Predicate
Inversion, and Copulas. Linguistic Inquiry Monographs 47. Cambridge, Mass.:
MIT Press.
Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press.
Dixon, Robert M. W. 1982. Where Have All the Adjectives Gone? And Other Essays in
Semantics and Syntax. Berlin: Walter de Gruyter.
Don, Jan. 2004. ‘Categories in the Lexicon’. Linguistics 42: 931–56.
Donati, Caterina. 2006. Labels and Merge. Paper presented at the Interphases
Conference, May, University of Cyprus, Nicosia.
Doron, Edit. 2003. ‘Agency and Voice: The Semantics of the Semitic Templates’.
Natural Language Semantics 11: 1–67.
Drachman, Gaberell. 2005. ‘A Note on “Shared” Allomorphs’. Journal of Greek
Linguistics 6: 5–37.
Dubinsky, Stanley, and Sylvester Ron Simango. 1996. ‘Passive and Stative in
Chichewa: Evidence for Modular Distinctions in Grammar’. Language 72: 749–81.
Embick, David. 2000. ‘Features, Syntax, and Categories in the Latin Perfect’. Linguistic
Inquiry 31: 185–230.
 References 193

Embick, David, and Alec Marantz. 2008. ‘Architecture and Blocking. Linguistic
Inquiry’. Linguistic Inquiry 39: 1–53.
Embick, David, and Rolf Noyer. 2001. ‘Movement Operations after Syntax’. Linguistic
Inquiry 32: 555–95.
Emonds, Joseph E. 1985. A Unified Theory of Syntactic Categories. Studies in
Generative Grammar 19. Dordrecht: Foris Publications.
Evans, Nicholas, and Stephen C. Levinson. 2009. ‘The Myth of Language Universals:
Language Diversity and Its Importance for Cognitive Science’. Brain and
Behavioral Sciences 32: 429–92.
Everett, Daniel. 2005. ‘Cultural Constraints on Grammar and Cognition in Pirahã:
Another Look at the Design Features of Human Language’. Current Anthropology
46: 621–34.
Family, Neiloufar. 2008. ‘Explorations of Semantic Space. The Case of Light Verb
Constructions in Persian’. Paper presented at the Conference on Complex Predi-
cates in Iranian Languages, 5 July, Université Sorbonne Nouvelle-Paris III, France.
http://iranianlinguistics.org/complexpredicates/presentations/family-2008.pdf
Felix, Sascha. 1990. ‘The Structure of Functional Categories’. Linguistische Berichte
125: 46–71.
Folli, Rafaella, and Heidi Harley. 2005. ‘Consuming Results in Italian and English:
Flavours of v’. In Aspectual Inquiries, edited by Paula Kempchinsky and
Roumyana Slabakova, 1–25. Dordrecht: Springer Verlag.
Folli, Rafaella, Heidi Harley and Simin Karimi. 2003. ‘Determinants of Event
Type in Persian Complex Predicates’. In Cambridge Occasional Papers in
Linguistics, edited by Luisa Astruc and Marc Richards, 1:100–20. University of
Cambridge.
Fox, Danny. 2000. Economy and Semantic Interpretation. Cambridge, Mass.: MIT Press.
Francis, Elaine J., and Etsuyo Yuasa. 2008. ‘A Multi-Modular Approach to Gradual
Change in Grammaticalisation’. Journal of Linguistics 44: 45–86.
Friedmann, Naama. 2006. ‘Speech Production in Broca’s Agrammatic Aphasia:
Syntactic Tree Pruning’. In Broca’s Region, edited by Yosef Grodzinsky and
Katrin Amunts, 63–82. Oxford University Press.
Fu, Jingqi, Tom Roeper and Hagit Borer. 2001. ‘The VP within Process Nominals:
Evidence from Adverbs and the VP Anaphor “Do-So”’. Natural Language and
Linguistic Theory 19: 549–82.
Fyndanis, Valantis. 2009. Functional Categories in Greek Agrammatism. PhD thesis,
Aristotle University of Thessaloniki.
Galani, Alexandra. 2005. The Morphosyntax of Verbs in Modern Greek. PhD thesis,
University of York.
Gil, David. n.d. What Is Riau Indonesian?. Unpublished ms. Max Planck Institute for
Evolutionary Anthropology, Leipzig. http://sastra.um.ac.id/wp-content/uploads/
2010/01/PU-David-Gil-Riau-Indonesian-.-.-..pdf
1994. ‘The Structure of Riau Indonesian’. Nordic Journal of Linguistics 17:
179–200.
2000. ‘Syntactic Categories, Cross-Linguistic Variation and Universal Grammar’. In
Approaches to the Typology of Word Classes (Empirical Approaches to Language
194 References

Typology), edited by Petra M. Vogel and Bernard Comrie, 173–216. Berlin:

Mouton de Gruyter.
2005. ‘Word Order without Syntactic Categories: How Riau Indonesian Does It’. In
Verb First: On the Syntax of Verb-Initial Languages, edited by Andrew Carnie,
Heidi Harley and Sheila Dooley, 243–63. Amsterdam: John Benjamins Publishing
Company.
2013. ‘Riau Indonesian: A Language without Nouns and Verbs’. In Flexible Word
Classes: Typological Studies of Underspecified Parts of Speech, edited by Jan
Rijkhoff and Eva van Lier. Oxford University Press.
Givón, Talmy. 1984. Syntax: A Functional-Typological Introduction. Amsterdam: John
Benjamins Publishing Company.
Grimshaw, Jane B. 1990. Argument Structure. Cambridge, Mass.: MIT Press.
1991. Extended Projection. Unpublished ms. Brandeis University.
2003. Words and Structure. CSLI Lecture Notes 151. Stanford, Calif.: CSLI
Publications.
Grohmann, Kleanthes K. 2003. Prolific Domains: On the Anti-Locality of Movement
Dependencies. Linguistik Aktuell ¼ Linguistics Today 66. Amsterdam: John
Benjamins Publishing Company.
Guilfoyle, Ethne, and Máire Noonan. 1992. ‘Functional Categories and Language
Acquisition’. Canadian Journal of Linguistics 37: 241–72.
Haegeman, Liliane M. V. 2006. Thinking Syntactically: A Guide to Argumentation and
Analysis. Blackwell Textbooks in Linguistics 20. Oxford: Blackwell.
Haider, Hubert. 2001. ‘Heads and Selection’. In Semi-Lexical Categories, edited by
Norbert Corver and Henk van Riemsdijk, 67–96. Berlin: Mouton de Gruyter.
Hale, Kenneth L., and Samuel Jay Keyser. 1993. ‘On Argument Structure and the
Lexical Expression of Syntactic Relations’. In The View from Building 20: Essays
in Honor of Sylvain Bromberger, edited by Kenneth L. Hale and Samuel Jay
Keyser, 53–109. Cambridge, Mass.: MIT Press.
2002. Prolegomenon to a Theory of Argument Structure. Linguistic Inquiry
Monographs 39. Cambridge, Mass.: MIT Press.
Halle, Morris, and Alec Marantz. 1993. ‘Distributed Morphology and the Pieces of
Inflection’. In The View from Building 20: Essays in Honor of Sylvain Bromber-
ger, edited by Kenneth L. Hale and Samuel Jay Keyser, 111–76. Cambridge,
Mass.: MIT Press.
Harley, Heidi. 2005a. ‘Bare Phrase Structure, a-Categorial Roots, One-Replacement
and Unaccusativity’. In Harvard Working Papers on Linguistics, edited by Slava
Gorbachov and Andrew Nevins, 9: 1–19. Cambridge, Mass.: Harvard University
Press.
2005b. ‘How Do Verbs Get Their Names? Denominal Verbs, Manner Incorporation
and the Ontology of Verb Roots in English’. In The Syntax of Aspect: Deriving
Thematic and Aspectual Interpretation, edited by Nomi Erteschik-Shir and Tova
Rapoport, 42–64. Oxford University Press.
2007. ‘The Bipartite Structure of Verbs Cross-Linguistically, or Why Mary Can’t
Exhibit John Her Paintings’. Paper presented at the ABRALIN Congres 2007,
March, UFMG, Belo Horizonte, Brasil.
 References 195

2009. ‘The Morphology of Nominalizations and the Syntax of vP’. In Quantification,

Definiteness, and Nominalization, edited by Anastasia Giannakidou and Monika
Rathert, 320–42. Oxford University Press.
2012a. On the Identity of Roots. Unpublished ms. University of Arizona. http://ling.
auf.net/lingBuzz/001527
2012b. External Arguments and the Mirror Principle: On the Distinctness of Voice
and v. Unpublished ms. University of Arizona.
Harley, Heidi, and Rolf Noyer. 1998. ‘Licensing in the Non-Lexicalist Lexicon:
Nominalizations, Vocabulary Items and the Encyclopaedia’. MIT Working Papers
in Linguistics 32: 119–37.
1999. ‘State-of-the-Article: Distributed Morphology’. GLOT International 4 (4): 3–9.
2000. ‘Formal versus Encyclopedic Properties of Vocabulary: Evidence from
Nominalisations’. In The Lexicon-Encyclopedia Interface, 1st edn, edited by Bert
Peeters, 349–74. Current Research in the Semantics/pragmatics Interface 5.
Amsterdam: Elsevier.
Haspelmath, Martin. 2001. ‘Word Classes and Parts of Speech’. International Encyclo-
pedia of the Social & Behavioral Sciences. Amsterdam: Elsevier.
Haugen, Jason. 2009. ‘Hyponymous Objects and Late Insertion’. Lingua 119: 242–62.
Hawkins, Roger, and Cecilia Yuet-hung Chan. 1997. ‘The Partial Availability of UG in
Second Language Acquisition: The “Failed Functional Features Hypothesis”’.
Second Language Research 13 (3): 187–226.
Hawkins, Roger, and Hajime Hattori. 2006. ‘Interpretation of English Multiple
Wh-Questions by Japanese Speakers: A Missing Uninterpretable Feature
Account’. Second Language Research 22 (3): 269–301.
Hazout, Ilan. 1994. ‘Nominalizers in Theta Theory’. The Linguistic Review 11: 5–48.
Hegarty, Michael. 2005. A Feature-Based Syntax of Functional Categories: The Struc-
ture, Acquisition, and Specific Impairment of Functional Systems. Studies in
Generative Grammar 79. Berlin: Mouton de Gruyter.
Hicks, Glyn. 2009. The Derivation of Anaphoric Relations. Linguistik Aktuell/Linguistics
Today 139. Amsterdam: John Benjamins Publishing Company.
Higginbotham, James. 1985. ‘On Semantics’. Linguistic Inquiry 16: 547–94.
Himmelmann, Nikolaus P. 2008. ‘Lexical Categories and Voice in Tagalog’. In Voice
and Grammatical Relations in Austronesian Languages, edited by Peter Austin
and Simon Musgrave, 247–93. Stanford, Calif.: CSLI Publications.
Hinzen, Wolfram. 2006. Mind Design and Minimal Syntax. Oxford University Press.
Hudson, Richard. 2003. ‘Gerunds without Phrase Structure’. Natural Language and
Linguistic Theory 21: 579–615.
Iida, Masayo. 1987. ‘Case-Assignment by Nominals in Japanese’. In Working Papers
in Grammatical Theory and Discourse Structure: Interactions of Morphology,
Syntax, and Discourse, edited by Masayo Iida, Stephen Wechsler and Draga Zec,
93–138. CSLI Lecture Notes 11. Stanford, Calif.: CSLI Publications.
Iwasaki, Yasufumi. 1999. Three Subcategories of Nouns in Japanese. PhD thesis,
University of Illinois, Urbana-Champaign.
Jackendoff, Ray. 1977. X Syntax: A Study of Phrase Structure. Cambridge, Mass.: MIT
Press.
196 References

Jelinek, Eloise. 1995. ‘Quantification in Straits Salish’. In Quantification in Natural

Languages, edited by Emmon Bach, Eloise Jelinek, Angelika Kratzer and Barbara
Partee, 487–540. Dordrecht: Kluwer.
1996. ‘Definiteness and Second Position Clitics in Straits Salish’. In Approaching
Second: Second Position Clitics and Related Phenomena, edited by Aaron
Halpern and Arnold M. Zwicky, 271–97. CSLI Lecture Notes 61. Stanford, Calif.:
CSLI Publications.
Johnson, Kyle. 1988. ‘Clausal Gerunds, the ECP, and Government’. Linguistic Inquiry
19: 583–609.
Jouitteau, Mélanie. 2005. ‘Nominal Properties of vPs in Breton, a Hypothesis for the
Typology of VSO Languages’. In Verb First: On the Syntax of Verb Initial
Languages, edited by Andrew Carnie, Heidi Harley and Sheila Ann Dooley,
265–80. Amsterdam: John Benjamins Publishing Company.
Karimi-Doostan, Gholamhosein. 2008a. ‘Classless Words in Persian’. Paper presented
at the Conference on Complex Predicates in Iranian Languages, 5 July, Université
Sorbonne Nouvelle-Paris III, France. http://www.iranianlinguistics.org/papers/
Abstract_Doostan.pdf
2008b. ‘Separability of Light Verb Constructions in Persian’. Paper presented at the
Conference on Complex Predicates in Iranian Languages, 5 July, Université
Sorbonne Nouvelle-Paris III, France.
Kayne, Richard S. 1994. The Antisymmetry of Syntax. Linguistic Inquiry Monographs 25.
Cambridge, Mass.: MIT Press.
2005. Movement and Silence. Oxford University Press.
2009. ‘Antisymmetry and the Lexicon’. In Linguistic Variation Yearbook 2008,
edited by Jeroen van Craenenbroeck, 1–32. Amsterdam: John Benjamins Publish-
ing Company.
Kiparsky, Paul. 1982. ‘Word Formation and the Lexicon’. In Proceedings of the
Mid-America Linguistics Conference, edited by Fred Ingeman, 3–29. Lawrence:
University of Kansas.
Kornfilt, Jaklin. 1997. Turkish. London: Routledge.
Kratzer, Angelika. 1996. ‘Severing the External Argument from Its Verb’. In Phrase
Structure and the Lexicon, edited by Johan Rooryck and Laurie Zaring, 109–37.
Dordrecht: Kluwer.
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Stanford University
Press.
2000. Grammar and Conceptualization. Cognitive Linguistics Research 14. Berlin:
Mouton de Gruyter.
Lapointe, Steve. 1993. ‘Dual Lexical Categories and the Syntax of Mixed
Category Phrases’. In ESCOL ’93, edited by Andreas Kathol and Michael Bern-
stein, 199–210. Columbus, Ohio: Ohio State University.
1999. ‘Dual Lexical Categories vs. Phrasal Conversion in the Analysis of Gerund
Phrases’. In University of Massachusetts Occasional Papers in Linguistics 24:
Papers from the 25th Anniversary, edited by Paul DeLacy and Anita Nowak,
157–89. Amherst, Mass.: GLSA Publications.
Larson, Richard K. 1998. ‘Events and Modification in Nominals’. In Proceedings
of Semantics and Linguistic Theory (SALT) VIII, edited by D. Strolovitch and
 References 197

A. Lawson, 145–68. Ithaca, N.Y.: CLC Publications, Department of Linguistics,

Cornell University.
Larson, Richard K., and Hiroko Yamakido. 2008. ‘Ezafe and the Deep Position of
Nominal Modifiers’. In Adjectives and Adverbs. Syntax, Semantics, and Dis-
course, edited by Louise McNally and Christopher Kennedy, 43–70. Oxford
University Press.
Lecarme, Jacqueline. 2004. ‘Tense in Nominals’. In The Syntax of Time, edited by
Jacqueline Guéron and Jacqueline Lecarme, 441–76. Cambridge, Mass.: MIT
Press.
Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation.
University of Chicago Press.
Levin, Beth, and Malka Rappaport Hovav. 2005. Argument Realization. Research
Surveys in Linguistics. Cambridge University Press.
Levinson, Lisa. 2007. The Roots of Verbs. PhD thesis, New York University.
Li, Yafei. 1990. Conditions on Xo Movement. MIT Working Papers in Linguistics. Cam-
bridge, Mass.: MIT Press.
Lieber, Rochelle, and Sergio Scalise. 2007. ‘The Lexical Integrity Hypothesis in a New
Theoretical Universe’. In On-Line Proceedings of the Fifth Mediterranean Morph-
ology Meeting (MMM5) Fréjus 15–18 September 2005, edited by Geert Booij,
Luca Ducceschi, Bernard Fradin, Emiliano Guevara, Angela Ralli and Sergio
Scalise, 1–24. Bologna: Università degli Studi di Bologna. http://mmm.lingue.
unibo.it/proc-mmm5.php
Longobardi, Giuseppe. 1994. ‘Reference and Proper Names: A Theory of N-Movement
in Syntax and Logical Form’. Linguistic Inquiry 25: 609–65.
Lowenstamm, Jean. 2008. ‘On N, √, and Types of Nouns’. In Sounds of Silence: Empty
Elements in Syntax and Phonology, edited by Jutta M. Hartmann, Veronika
Hegedüs and Henk van Riemsdijk, 107–44. North Holland Linguistic Series,
Linguistic Variations. Amsterdam: Elsevier.
Maling, Joan. 1983. ‘Transitive Adjectives: A Case of Categorial Reanalysis’. In
Linguistic Categories: Auxiliaries and Related Puzzles, edited by Frank Heny
and Barry Richards, 1:253–89. Dordrecht: Reidel.
Malouf, Robert. 2000. ‘Verbal Gerunds as Mixed Categories in Head-Driven Phrase
Structure Grammar’. In The Nature and Function of Syntactic Categories, edited
by Robert Borsley, 133–66. New York: Academic Press.
Manning, Christopher D. 1993. ‘Analyzing the Verbal Noun: Internal and External
Constraints’. In Japanese/Korean Linguistics 3, edited by Soonja Choi, 236–53.
Stanford, Calif.: Stanford Linguistics Association.
Manzini, M. Rita. 2010. ‘The Structure and Interpretation of (Romance) Complementi-
zers’. In The Complementizer Phase: Subjects and Operators, edited by Phoevos
Panagiotidis, 167–99. Oxford University Press.
Marantz, Alec. 1991. ‘Case and Licensing’. In Proceedings of ESCOL 1991, 234–53.
Columbus, Ohio: Ohio State University.
1997. ‘No Escape from Syntax: Don’t Try Morphological Analysis in the Privacy of
Your Own Lexicon’. U. Penn Working Papers in Linguistics 4: 201–25.
2000. Words. Unpublished ms. New York University.
198 References

2005. Rederived Generalizations. Unpublished ms. MIT.

2006. Phases and Words. Unpublished ms. MIT. https://files.nyu.edu/ma988/public/
Phase_in_Words_Final.pdf
Massam, Diane, Colin Gorrie and Alexandra Kellner. 2006. ‘Determiners in Niuean:
Everywhere and Nowhere’. Proceedings of the 2006 Annual Conference of the
Canadian Linguistic Association: 1–16.
Matthewson, Lisa. 2001. ‘Quantification and the Nature of Crosslinguistic Variation’.
Natural Language Semantics 9: 145–89.
Matushansky, Ora. 2006. ‘Head Movement in Linguistic Theory’. Linguistic Inquiry
37: 69–109.
McGinnis, Martha. 2002. ‘On the Systematic Aspect of Idioms’. Linguistic Inquiry
33 (4): 665–72.
Milsark, Gary. 1988. ‘Singl-Ing’. Linguistic Inquiry 19: 611–34.
Miyagawa, Shigeru. 1987. ‘Lexical Categories in Japanese’. Lingua 73: 29–51.
Moulton, Keir. 2004. ‘External Arguments and Gerunds’. Toronto Working Papers in
Linguistics 22: 121–36.
Muysken, Pieter. 2008. Functional Categories. Cambridge University Press.
Neeleman, Ad, Hans van de Koot and Jenny Doetjes. 2004. ‘Degree Expressions’.
The Linguistic Review 21: 1–66.
Nevins, Andrew, David Pesetsky and Cilene Rodrigues. 2009. ‘Pirahã Exceptionality:
A Reassessment’. Language 85: 355–404.
Newmeyer, Frederick J. 1998. Language Form and Language Function. Language,
Speech, and Communication. Cambridge, Mass.: MIT Press.
2004. ‘On Split-CPs, Uninterpretable Features, and the “Perfectness” of Language’.
ZAS Papers in Linguistics 35: 399–422.
Nordlinger, Rachel, and Louisa Sadler. 2004. ‘Nominal Tense in Cross-Linguistic
Perspective’. Language 80: 776–806.
Nunberg, Geoffrey, Ivan Sag and Thomas Wasow. 1994. ‘Idioms’. Language 70:
491–538.
Oga, Kyoko. 2001. ‘Two Types of “of” and Theta-Role Assignment by Nouns’. In
Newcastle and Durham Working Papers in Linguistics, edited by Mamiko Akita
and Kyoko Oga, 6: 95–108. University of Durham, Department of English and
Linguistics.
Ouhalla, Jamal. 1991. Functional Categories and Parametric Variation. Theoretical
Linguistics. London: Routledge.
Panagiotidis, Phoevos. 2000. ‘The Categorial Features of Functional Heads’. Unpub-
lished ms. University of Essex.
2002. Pronouns, Clitics and Empty Nouns: ‘Pronominality’ and Licensing in Syntax.
Amsterdam: John Benjamins Publishing Company.
2003a. ‘One, Empty Nouns and Theta Assignment’. Linguistic Inquiry 34: 281–92.
2003b. ‘Empty Nouns’. Natural Language and Linguistic Theory 21: 381–432.
2004. ‘Categorial Deficiency, Head Movement and Phrase Structure’. Paper pre-
sented at the Lisbon Workshop on Alternative Views on the Functional Domain,
8–9 July, Universidade Nova de Lisboa.
2005. ‘Against Category-Less Roots in Syntax and Word Learning: Objections to
Barner and Bale (2002)’. Lingua 115: 1181–94.
 References 199

2008. ‘Diachronic Stability and Feature Interpretability’. In The Limits of Syntactic

Variation, edited by Theresa Biberauer, 441–56. Amsterdam: John Benjamins
Publishing Company.
2010. ‘Non-Argumental Mixed Projections’. Syntax 13: 165–82.
2011. ‘Categorial Features and Categorizers’. The Linguistic Review 28: 325–46.
Panagiotidis, Phoevos, and Kleanthes K. Grohmann. 2009. ‘Mixed Projections:
Categorial Switches and Prolific Domains’. Linguistic Analysis 35: 141–61.
Panagiotidis, Phoevos, Anthi Revithiadou and Vassilios Spyropoulos. 2013. ‘Verbal-
izers Leave Marks: Evidence from Greek’. Submitted for publication. University
of Cyprus, Aristotle University of Thessaloniki and University of Athens.
Pensalfini, Robert. 1997. Jingulu Grammar, Dictionary, and Texts. PhD thesis, MIT,
Cambridge, Mass.
Pesetsky, David. 1995. Zero Syntax: Experiencers and Cascades. Cambridge, Mass.:
MIT Press.
Pesetsky, David, and Ester Torrego. 2004. ‘Tense, Case and the Nature of Syntactic
Categories’. In The Syntax of Time, edited by Jacqueline Guéron and Jacqueline
Lecarme, 495–537. Cambridge, Mass.: MIT Press.
2005. ‘Subcategorization Phenomena and Case-Theory Effects: Some Possible
Explanations’. Lecture delivered at the LAGB Annual Meeting, 3 September,
Fitzwilliam College, University of Cambridge.
Pires, Acrisio. 2006. The Minimalist Syntax of Defective Domains: Gerunds and Infini-
tives. Linguistik Aktuell ¼ Linguistics Today 98. Amsterdam: John Benjamins
Publishing Company.
Plann, Susan. 1981. ‘The Two El þ Infinitive Constructions in Spanish’. Linguistic
Analysis 7: 203–40.
Prasada, Sandeep. 2008. Aspects of a Fully Psychological Theory of Sortal Representa-
tion. Unpublished ms. Hunter College, CUNY, New York.
Preminger, Omer. 2010. ‘Nested Interrogatives and the Locus of Wh’. In The
Complementizer Phase: Subjects and Operators, 200–35. Oxford University Press.
Pullum, Geoffrey. 1991. ‘English Nominal Gerund Phrases as Noun Phrases with Verb
Phrase Heads’. Linguistics 29: 763–99.
Pylkkänen, Liina. 2008. Introducing Arguments. Cambridge, Mass.: MIT Press. http://
site.ebrary.com/id/10229589
Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax: The
Nature of Early Child Grammars of English. Oxford: Blackwell.
1996. ‘Towards a Structure-Building Model of Acquisition’. In Generative Perspec-
tives on Language Acquisition, edited by Harald Clahsen, 43–89. Amsterdam:
John Benjamins Publishing Company.
Ramchand, Gillian. 2008. Verb Meaning and the Lexicon: A First-Phase Syntax.
Cambridge Studies in Linguistics 116. Cambridge University Press.
Rappaport Hovav, Malka, and Beth Levin. 1998. ‘Building Verb Meanings’. In
The Projection of Arguments: Lexical and Compositional Factors, edited by Miriam
Butt and Wilhelm Geuder, 97–134. Stanford, Calif.: CSLI Publications.
Rauh, Gisa. 2010. Syntactic Categories: Their Identification and Description in
Linguistic Theories. Oxford Surveys in Syntax & Morphology 7. Oxford
University Press.
200 References

Reuland, Eric. 1983. ‘Governing “–ing”’. Linguistic Inquiry 14: 101–36.

Richards, Marc. 2004. Object Shift and Scrambling in North and West Germanic:
A Case Study in Symmetrical Syntax. PhD thesis, University of Cambridge.
2007. ‘On Feature Inheritance: An Argument from the Phase Impenetrability
Condition’. Linguistic Inquiry 38: 563–72.
Richards, Marc, and Theresa Biberauer. 2005. ‘Explaining Expl’. In The Function of
Function Words and Functional Categories, edited by Marcel Den Dikken and
Christina Tortora, 115–54. Amsterdam: John Benjamins Publishing Company.
Rizzi, Luigi. 1997. ‘The Fine Structure of the Left Periphery’. In Elements of Grammar:
A Handbook of Generative Syntax, edited by Liliane M. V. Haegeman, 281–337.
Dordrecht: Kluwer.
Robins, Robert Henry. 1964. General Linguistics: An Introductory Survey. Indiana
University Studies in the History and Theory of Linguistics. Indiana University
Press.
Rosemeyer, Malte. 2012. ‘On the Interplay between Transitivity, Factivity and Infor-
mativity: Spanish Nominal and Verbal Infinitives’. In Aspectualidad –
Transitividad – Referencialidad, edited by Valeriano Bellosta von Colbe and
Marco García García. Studia Romanica et Linguistica. Frankfurt am Main:
Peter Lang.
Ross, John. 1973. ‘Nouniness’. In Three Dimensions of Linguistic Theory, edited by
Osamu Fujimura, 137–258. Tokyo: TEC Co.
Rothstein, Susan. 1983. The Syntactic Forms of Predication. PhD thesis, MIT,
Cambridge, Mass.
1999. ‘Fine-Grained Structure in the Eventuality Domain: The Semantics of
Predicate Adjective Phrases and “be”’. Natural Language Semantics 7: 347–420.
Roussou, Anna. 1990. Nominalization in the Syntax of Modern Greek. MA dissertation,
University College London.
Schachter, Paul. 1985. ‘Parts-of-Speech Systems’. In Language Typology and
Syntactic Description, edited by Timothy Shopen, 1: 3–61. Cambridge University
Press.
Schoorlemmer, Maaike. 2001. ‘Dutch Nominalised Infinitives as Non-Identical Twins’.
UiL OTS Working Paper uil-ots-01005-CL/TL. Utrecht University.
2002. ‘Adjectives and Adverbs in Dutch Nominalised Infinitives’. Paper presented at
the TIN Dag Workshop, 26 January, Utrecht.
Schütze, Carson. 2001. ‘Semantically Empty Lexical Heads as Last Resorts’. In
Semi-Lexical Categories, edited by Norbert Corver and Henk van Riemsdijk,
127–87. Berlin: Mouton de Gruyter.
Shibatani, Masayoshi. 1990. The Languages of Japan. Cambridge Language Surveys.
Cambridge University Press.
Siddiqi, Daniel. 2006. Minimize Exponence: Economy Effects on a Model of the
Morphosyntactic Component of the Grammar. PhD thesis, University of Arizona.
Siegel, Laura. 1998. ‘Gerundive Nominals and the Role of Aspect’. In Proceedings of
ESCOL ’97, edited by Jennifer Austin and Aaron Lawson. Ithaca, New York: Cornell
Linguistics Club Publications. http://www.ling.upenn.edu/~lsiegel/escolrev.pdf
Simons, Peter. 1987. Parts: A Study in Ontology. Oxford: Clarendon Press.
 References 201

Starke, Michal. 2009. ‘Nanosyntax – A Short Primer to a New Approach to Language’.

In Norlyd, edited by Peter Svenonius, Gillian Ramchand, Michal Starke and Knut
Tarald Taraldsen, 36.1:1–6. Tromsø: CASTL. http://ling.auf.net/lingbuzz/001183
2011. Towards Elegant Parameters: Language Variation Reduces to the Size of
Lexically Stored Trees. Unpublished ms. Barcelona. http://ling.auf.net/lingbuzz/
001183.
Stiebels, Barbara. 1999. ‘Noun–Verb Symmetries in Nahuatl Nominlizations’. Natural
Language and Linguistic Theory 17: 783–836.
Stowell, Tim. 1981. Origins of Phrase Structure. PhD thesis, MIT, Cambridge, Mass.
1991. ‘Determiners in NP and DP’. In Views on Phrase Structure, edited by
Katherine Leffel and Denis Bouchard, 37–56. Dordrecht: Kluwer.
2007. ‘The Syntactic Expression of Tense’. Lingua 117: 437–63.
Svenonius, Peter. 1994. Dependent Nexus: Small Clauses in English and the Scandi-
navian Languages. PhD thesis, University of California at Santa Cruz.
2004. ‘On the Edge’. In Peripheries: Syntactic Edges and Their Effects, edited by
David Adger, Cécile De Cat and George Tsoulas, 261–87. Dordrecht: Kluwer.
2005. Idioms and Domain Boundaries. Unpublished ms. CASTL, University of Tromsø.
2007. ‘Adpositions, Particles, and the Arguments They Introduce’. In Argument
Structure, edited by Eric Reuland, Tanmoy Bhattacharya and Giorgos Spathas,
63–103. Amsterdam: John Benjamins Publishing Company.
2008. ‘Projections of P’. In Syntax and Semantics of Spatial P, edited by Anna
Asbury, Jakub Dotlacil, Berit Gehrke and Rick Nouwen, 63–84. Amsterdam: John
Benjamins Publishing Company.
Terzi, Arhonto. 2010. ‘Locative Prepositions and Place’. In Mapping Spatial PPs,
edited by G. Cinque and L. Rizzi, 196–224. The Cartography of Syntactic
Structures. Oxford University Press.
Thráinsson, Höskuldur. 1996. ‘On the (non-) Universality of Functional Categories’.
In Minimal Ideas: Syntactic Studies in the Minimalist Framework, edited by
Werner Abraham, Samuel D. Epstein, Höskuldur Thráinsson and Jan-Wouter
Zwart, 253–82. Amsterdam: John Benjamins Publishing Company.
Tonhauser, Judith. 2005. ‘Towards an Understanding of the Meaning of Nominal
Tense’. In Proceedings of Sinn und Bedeutung 9, edited by Emar Maier, C. Bary
and J. Huitnink, 475–88. Nijmegen Centre of Semantics.
2007. ‘Nominal Tense? The Meaning of Guarani Nominal Temporal Markers’.
Language 83: 831–69.
Tosun, Gülsat. 1999. Fused Functional Heads and Turkish Facts. Unpublished ms.
Harvard University.
Traugott, Elizabeth. 2007. ‘The Concepts of Constructional Mismatch and Type-
Shifting from the Perspective of Grammaticalisation’. Cognitive Linguistics
18 (4): 523–57.
Tremblay, Mireille. 1996. ‘Lexical and Non-Lexical Prepositions in French’. In
Configurations: Essays on Structure and Interpretation, edited by Anna-Maria
Di Sciullo, 79–98. Sommerville, Mass.: Cascadilla Press.
Trommelen, Mieke. 1989. ‘Lettergreepstruktuur en Woordkategorie’. De Nieuwe
Taalgids 82: 64–77.
202 References

Tsimpli, Ianthi Maria. 2003. ‘Clitics and Determiners in L2 Greek’. In Proceedings of

the 6th Generative Approaches to Second Language Acquisition Conference
(GASLA 2002), edited by Juana M. Liceras, 331–9. Sommerville, Mass.:
Cascadilla Press.
Tsimpli, Ianthi Maria, and Maria Dimitrakopoulou. 2007. ‘The Interpretability
Hypothesis: Evidence from Wh-Interrogatives in Second Language Acquisition’.
Second Language Research 23: 215–42.
Tsimpli, Ianthi Maria, and Maria Mastropavlou. 2007. ‘Feature Interpretability in L2
Acquisition and SLI: Greek Clitics and Determiners’. In The Role of Formal
Features in Second Language Acquisition, edited by Juana M. Liceras, Helmut
Zobl and Helen Goodluck, 143–83. London: Routledge.
Tsimpli, Ianthi Maria, Antonella Sorace, Caroline Heycock and Francesca Filiaci. 2004.
‘First Language Attrition and Syntactic Subjects: A Study of Greek and
Italian Near-Native Speakers of English’. International Journal of Bilingualism
8: 257–77.
Tsimpli, Ianthi Maria, and Stavroula Stavrakaki. 1999. ‘The Effects of a Morphosyn-
tactic Deficit in the Determiner System: The Case of a Greek SLI Child’. Lingua
108: 31–85.
Tsujimura, Natsuko. 1992. ‘Licensing Nominal Clauses: The Case of Deverbal Nouns
in Japanese’. Natural Language and Linguistic Theory 10: 477–522.
Uriagereka, Juan. 1999. ‘Warps: Some Thoughts on Categorization’. Theoretical
Linguistics 25: 31–73.
Vainikka, Anne. 1994. ‘Case in the Development of English Syntax’. Language
Acquisition 3: 257–325.
Valgina, N. S., D. E. Rosental and M. I. Fomina. 2002. Современный Русский Язык
Учебник [Modern Russian Language Textbook]. Московский государственный
университет печати [Moscow State University Press].
Van Hout, Angelique, and Tom Roeper. 1998. ‘Events and Aspectual Structure in
Derivational Morphology’. MIT Working Papers in Linguistics 32: 175–220.
Van Riemsdijk, Henk. 1998a. ‘Categorial Feature Magnetism: The Endocentricity and
Distribution of Projections’. Journal of Comparative Germanic Linguistics 2: 1–48.
1998b. ‘Head Movement and Adjacency’. Natural Language and Linguistic Theory
16: 633–79.
Volpe, Mark. 2009. ‘Root and Deverbal Nominalizations: Lexical Flexibility in
Japanese’. Unpublished ms. http://ling.auf.net/lingBuzz/000789
Wasow, Thomas. 1977. ‘Transformations and the Lexicon’. In Formal Syntax, edited
by Peter Culicover, Thomas Wasow and Joan Bresnan, 324–60. New York:
Academic Press.
Willim, Ewa. 2000. ‘On the Grammar of Polish Nominals’. In Step by Step: Essays on
Minimalist Syntax in Honor of Howard Lasnik, edited by Roger Martin, David
Michaels and Juan Uriagereka, 319–46. Cambridge, Mass.: MIT Press.
Yoder, Brendon. 2010. ‘Syntactic Underspecification in Riau Indonesian’. In Work
Papers of the Summer Institute of Linguistics, University of North Dakota Session,
edited by Joan Baart, 50: 1–15. Grand Forks, N. Dak.: SIL International.
Yoon, James Hye Suk. 1996a. ‘Nominal Gerund Phrases in English as Phrasal Zero
Derivations’. Linguistics 34: 329–56.
 References 203

1996b. ‘A Syntactic Account of Category-Changing Phrasal Morphology: Nominal-

izations in English and Korean’. In Morphosyntax in Generative Grammar:
Proceedings of the 1996 Seoul International Conference on Generative Grammar,
edited by Hee-Don Ahn, Myung-Yoon Kang, Young-Suck Kim and Sookhee Lee,
357–68. Seoul: Hankook Publishing Company.
Yoon, James Hye Suk, and Neus Bonet-Farran. 1991. ‘The Ambivalent Nature of
Spanish Infinitives’. In New Analyses in Romance Linguistics: Selected Papers
from the Linguistic Symposium on Romance Languages XVIII, edited by Dieter
Wanner and Douglas A. Kibbee, 353–70. Amsterdam: John Benjamins Publishing
Company.
Yoon, James Hye Suk, and Chongwon Park. 2004. Process Nominals and Morpho-
logical Complexity. Unpublished ms. University of Illinois, Urbana-Champaign.
 Index

action nominalization, 149 interpretable, 15–16, 22, 82

adjective, 41, 137 interpretation of, 84
as the unmarked category, 41, 43, 46, 183 on functional heads, 111
morphological complexity of, 46 phase-inducing, 160
possessive, 45 privative, 14, 106
adposition, 49, 99 uninterpretable, 117
as a relational category, 50 values of, 100
as nominal, 112 Categorial Uniformity, 154
as the unmarked category, 50 categorization, 17, 22, 95
adverb, 48 Categorization Assumption, 81, 89, 122, 141,
affix 183, 187
category-changing, 144–5 categorizer, 17–18, 54, 62, 64, 74, 79, 82,
Agree, 94 91–2, 95
and labelling, 129, 131 as lexical, 99
categorial, 128, 146 as licenser, 96
definition of, 124 as phase head, 95, 152, 160
Amherst system, 12, 21, 49 as the limit of idiosyncrasy, 97
Arabic, 137 indispensable in structure building, 100
argument structure, 7, 37, 51, 59–60 not functional heads, 98
auxiliary verb, 39 obligatory in all syntactic structures,
128
Basque, 149 categorizer projection, 68
biuniqueness, 111, 126, 187 category-specific behaviour, 141, 186
and adjectives, 112 causativizer, 69
and adpositions, 112 Chichewa, 44, 187
definition of, 114 classifier, 88
Classless Word, 103
Cartography, 109 coercion, 114
Categorial Deficiency, 119, 129, 173 Cognitive Grammar, 19
and head movement, 121 Complementizer, 157
consequences of, 118–19 Complete Functional Complex, 119
definition of, 117 Complex Predicate, 37, 39, 103
deriving the Categorization Assumption, compositionality, 58, 61, 93, 97
123 compound, 90, 187
categorial feature, 11 concept, 5, 20, 84
as [perspective] feature, 101, 104, 147 conceptualization, 19–20
as taxonomic marker, 116 conversion, 62, 73, 78
for functional heads, 107–8 co-ordination, 185

204
 Index 205

criterion gender, 80
of application, 85 Ger head, 143, 146, 160
of identity, 85 not a noun, 144
of individuation, 88 German, 137
c-selection, 121 Germanic, 37
gerund, 135, 138–9, 143–4, 161, 165–6
Dagaare, 137 Goal, 127
defective intervention effect, 126, 147 grammatical noun and verb, 99
Degree, 47, 112 Greek, 45, 71, 90, 101, 112, 156–8, 165,
denominal verb, 63, 67–8, 76, 170 170, 184
descriptive content, 99, 174 Greenlandic, 187
Diesing effect, 152, 168
Distributed Morphology, 54, 58, 71 Head Feature Convention, 136
Dutch, 73, 76, 137, 153, 165–6 Hebrew, 93, 102, 137
Hindi, 73
Economy, 57, 164
empty noun, 99, 157–8 idiom, 61
Encyclopedia, 62 lexical, 93, 97
English, 49–50, 72, 139, 144, 166 idiosyncrasy, 58, 60–1, 64, 97
entity, 85, 88 Incorporation, 186
Eurocentrism, 24, 30, 177 indexical, 85
event, 87–8 Indonesian
exocentricity, 135 Riau, 30
exoticization, 28, 31, 177 Standard, 31
Extended Projection, 115, 119 inflected language, 27
deriving of, 126 inner morpheme, 69–70, 96, 98
interpretation of [N]
factive nominalization, 150 as sortality, 86
Farsi, 37, 39, 103 interpretation of [V]
feature, 57 as extending into time, 86, 89
affixal, 117 interpretive perspective, 3, 21, 83, 87, 89, 92,
as interface instruction, 2 96, 123
categorial, 80 provided by categorial features, 95
distinctive, 80 Italian, 90, 137
motivation for, 2
taxonomic, 12, 80 Japanese, 37, 43, 69, 138, 162, 165, 187
First Phase, 62, 68 Jingulu, 38
non-compositional, 97
Full Interpretation, 80, 146 Kabardian, 157
functional categorizer, 143, 145, 170 Kannada, 44
overt, 148 Kase, 149
functional category, 114, 118 Kikuyu, 137
functional head kind, 85
as satellite of lexical elements, Korean, 43, 152, 162, 165
110
cannot categorize, 91 label, 129
functional nominalizer, 148 language acquisition, 119
functional projection, 127 language attrition, 120
Fusion, 152 language disorder, 120
206 Index

Language Faculty, 2, 19, 22, 25, 57, Number, 2, 88, 142

94, 174 biunique relation with nouns, 88
Late Insertion, 58 Numeration, 118
lexical category
cross-categorization of, 13, 15 Object Shift, 167
default, 104
impairment of, 101 part of speech, 3
nouns as the only, 100 part of speech membership
universality of, 25 conceptually significant, 9
lexical idiom, 97 continuum approach, 9
Lexical Item, 128 morphological criteria, 6
lexical–functional continuum, 119 notional criteria, 5
lexical–functional distinction, 99, 110, 114 syntactic criteria, 7
and theta assignment, 115 taxonomic approach, 5, 9
as a matter of descriptive content, 115 particle verb, 37
derived from Categorial Deficiency, 119 Persian, see Farsi
lexicalism, 55, 58 Phase, 153, 159
weak, 55–6 Phase Theory, 65
lexicon, 54–5 Phrasal Coherence, 137, 139, 146,
light verb, 38, 103 155
Light Verb Construction, 37, 103, 163 Pirahã, 24
Lillooet Salish, 32, 35–6 Polish, 157–8
Pred head, 181
Makah, 33 predication, 16
Malayalam, 44 preverb, 38, 103
Mayali, 187 category of, 103
minimality, 147 Principle of Decompositionality, 109
mixed projection, 135 Probe, 125, 127
Morphology (module), 57, 67, 71 projects, 125, 130
productivity, 58, 71
Nahuatl, 187 inflectional, 71
nanosyntax, 109 projection line, 111, 116–17
natural class, 13 lexical heads at the bottom of, 127
nexus constructions, 141 Prolific Domain, 160
Nominal External Behaviour, 139, 161 pronoun, 132
nominalized infinitive, 139, 153, 156, prototype, 9
165, 169
expressive, 154, 165 Quechua, 187
plain, 154, 165–6
nominalizer, 74 radical categorylessness, 140, 155, 186
types of, 83 Reference–Predication Constraint, 148
Nootka, 36 root, 17, 29, 34, 36, 38, 40, 54, 59, 61, 68,
Nootka debate, 32 81, 83
Noun Licensing Condition, 180 acategorial, 90, 95
Noun Phrase as bound morpheme, 90
gradable, 114 as polysemous, 93
noun–verb distinction, 79, 83 as UG-external element, 94
criteria for, 29 as underspecified, 94
in borrowing, 102 categoryless, 72–3
lack of, 28, 30, 33 content of, 92–3
 Index 207

derivation without, 94 Tagalog, 29

licensing of, 95 telicity, 89
syntactically unexceptional, 92 Tense, 51, 135, 150, 169
uncategorized, 96 biunique relation with verbs, 88
root allomorphy, 75 nominal, 7
Russian, 44, 102 tense marker, 34
Tense Phrase, 26
semi-lexical category, 99, 110, 115, 119 time stability, 9, 85–6
as bearing only formal features, 99 Turkish, 34, 137, 139, 144, 149, 165
sortality, 16, 84
Spanish, 90, 139, 156–7, 165 UG feature, 109, 118
specifier, 130, 182, 188 uninterpretable categorial feature
St’át’imcets, see Lillooet Salish motivation for, 123
state, 87, 89 Universal Grammar, 177
stored form, 72 Urdu, 73
strong verb, 75
subevent, 87 verbal noun, 138, 162–3, 165, 169
supercategory, 110, 114, 117, 120 as a denominal verb, 164
switch, 142, 145, 159 verbalizer, 38, 64–5, 91
as a categorially dual head, 148 overt, 91, 102
as a functional head, 145 types of, 83, 99
as phasal head, 153 Vocabulary, 71
encoding tense, 151 vocabulary item, 96
syncretism, 27 Voice, 65
syntactic categorization, 17, 54, 77, 186 biunique relation with verbs, 88
syntactic category, 3 Voice Phrase, 64
syntactic decomposition, 3, 18, 34, 39, 58, 78, as the minimum verbal constituent,
81, 173, 188 102
syntactic deconstruction, 17
Syntactic Object, 129 word class, 3
syntax-all-the-way-down, 58 word-making, 56–7, 70