You are on page 1of 327

XuPing Li

Numeral Classifiers in Chinese

Trends in Linguistics
Studies and Monographs 250

Editor

Volker Gast
Editorial Board

Walter Bisang
Jan Terje Faarlund
Hans Henrich Hock
Natalia Levshina
Heiko Narrog
Matthias Schlesewsky
Amir Zeldes
Niina Ning Zhang
Editors responsible for this volume

Walter Bisang
Niina Ning Zhang

De Gruyter Mouton

Numeral Classifiers
in Chinese
The Syntax-Semantics Interface

by

XuPing Li

De Gruyter Mouton

ISBN 978-3-11-028763-9
e-ISBN 978-3-11-028933-6
ISSN 1861-4302
Library of Congress Cataloging-in-Publication Data
A CIP catalog record for this book has been applied for at the Library of Congress.
Bibliographic information published by the Deutsche Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data are available in the Internet at http://dnb.dnb.de.
2013 Walter de Gruyter GmbH, Berlin/Boston
Printing: Hubert & Co. GmbH & Co. KG, Gttingen
Printed on acid-free paper

Printed in Germany
www.degruyter.com

This book is dedicated solely to my beloved late mother Xu Yinfeng


(06/09/195927/10/2012), for her deep love in the past thirty years.

30

Preface

This book has grown out of my 2011 dissertation, entitled On the semantics of classifiers in Chinese (Bar Ilan University, 2011). The book is a
substantial revision of the dissertation, incorporates both the research for
the dissertation and the results of my work on classifiers in the years 20112013. Three new chapters, namely, Chapter 2, Chapter 4 and Chapter 10,
have been added to the original dissertation. Specifically, Chapter 4 on
natural atomicity was presented at the workshop on Mass/count in linguistics, philosophy and cognitive science at ENS, Paris in Dec 2012. Chapter
10 is based on my talk at the workshop on Semantic and typological perspectives on definite in Dsseldorf in June 2012. In the meantime, a different version of Chapter 8 is published as Li and Rothstein (2012) in Language and Linguistics. Chapter 9 is a development of Li and Bisang (2012)
in Lingua.
In this book, I discuss both syntax and semantics of classifiers in Chinese. My knowledge of semantics comes directly from Susan Rothstein,
who has been guiding and supporting me since we met in 2006. I am grateful to her for what she has done for me at every stage of my career. Both
Greg Carlson and Fred Landman deserve special thanks. Greg, as one of
the referees of my dissertation in 2010, gave me many constructive comments, which lead to great improvement of my dissertation. Fred read the
whole manuscript and edited and commented on it in the spring of 2011,
resulting in a much improved final version. I also appreciated Freds cutting-ins during my appointments with Susan at their home. I thank them
both very much for their contribution to my work on classifiers. I should
also like to thank Victor Pan, the conversations with whom always make
me think carefully about the question of how much syntax is needed for my
semantics for Chinese classifiers.
Ive worked in Israel, German and France since 2006. I thank everybody
at the institutions I worked in, in particular, my Israeli teachers: Gabi Danon, Edit Doron, Yael Greenberg, Nirit Kadmon, Fred Landman, Susan
Rothstein, and my German linguistic circle: Walter Bisang, Franziska
Kretzschmar, Yuchen Hung, Matthias Schlesewsky, Luming Wang, and the
linguists in Paris: Hilary Chappell (and members in her Sinotype Project),
Katia Chirkova, Redouane Djamouri, Carmen Dobrovie-Sorin, David Ni-

viii

Preface

colas, Victor Pan, Alain Peyraube, Paul Waltraud. Thanks also goes to Lisa
Cheng, James Huang, Rint Sybesma, Dylan Tsai, Niina Zhang.
I would also take this opportunity to acknowledge the financial support
from various sources. Thanks to Bar-Ilan for the Presidential Fellowship
(2006-2010), which allowed me to study at BIU. I also wish to thank Walter Bisang for the financial support which made my visit possible in the
academic year of 2008-2009 in Mainz, and for his guidance which allowed
me to learn so much about the typology of classifiers. The revision of the
book was conducted when I was a postdoc fellow in Hilary Chappells
ERC-Sinotype Project in Paris. I thank her unconditional support during
my stay in Paris.
Finally, very special thanks to the referee and to the editors of the series,
Walter Bisang and Niina Zhang, for their helpful comments. I also thank
my project/book editors Julie Miess and Wolfgang Konwitschny for their
professional editorial guidance.
XuPing Li
Paris, June 2013

Contents

Preface .......................................................................................................... v
Abbreviations ............................................................................................. xv
Chapter 1
Introduction ................................................................................................ 1
1.
Issues .................................................................................................. 1
1.1. Issue 1: the debate on a count/mass distinction in Mandarin ............ 1
1.2. Issue 2: counting and measuring functions of classifiers .................. 2
1.3. Issue 3: definiteness in classifier languages ...................................... 3
2.
Data and source.................................................................................. 4
3.
Structure of the book ......................................................................... 6
Part I: The debate on a count/mass distinction in Chinese
Chapter 2
Defing classifiers ...................................................................................... 13
1.
Chinese classifiers: an illustration ................................................... 13
1.1. Identifying classifiers syntactically ................................................. 13
1.2. Chinese classifiers: a heuristic classification .................................. 16
2.
Classifiers as a closed class ............................................................. 21
3.
Classifiers without descriptive content ........................................ 23
3.1. Classifiers are not nominal .............................................................. 23
3.2. Classifiers have no descriptive content ......................................... 28
4.
Classifiers are complement-taking................................................... 30
5.
Classifiers as stressless .................................................................... 32
6.
Classifiers in English: a contrastive look ........................................ 33
7.
Conclusions...................................................................................... 41
Chapter 3
The count/mass distinction in Chinese revisited ................................... 42
1.
Introduction ...................................................................................... 42
2.
Syntactic distinction between count and mass classifiers ............... 44
2.1. Introduction to count/mass classifiers ............................................. 44
2.2. Lexical/functional distinction of classifiers .................................... 47
2.3. Two syntactic diagnostics ................................................................ 49

Contents

3.
3.1.
3.2.
4.
5.

Possibility of pre-classifier adjectival modification ........................ 55


Adjectives before count and mass classifiers .................................. 55
Two constraints on pre-classifier adjectives.................................... 59
Optionality of post-classifier de....................................................... 62
Conclusions...................................................................................... 67

Chapter 4
Natural atomicity ..................................................................................... 68
1.
Introduction ...................................................................................... 68
2.
Countability and individuation ........................................................ 69
3.
A lexical distinction between mass and count nouns ...................... 73
4.
Individual, stuff and partial-object readings .................................... 77
5.
Natural atomicity as a grammatically relevant phenomenon ........... 81
6.
Conclusions...................................................................................... 85
Chapter 5
Chinese bare nouns .................................................................................. 86
1.
Introduction ...................................................................................... 86
2.
The Krifka-Chierchia hypothesis ..................................................... 87
3.
Bare nouns as kinds ......................................................................... 89
3.1. Strong kind-inducing contexts ......................................................... 89
3.2. Appositives ...................................................................................... 92
3.3. Scope with respect to opacity .......................................................... 93
3.4. Scope with respect to quantifiers ..................................................... 97
4.
Semantics of bare nouns ................................................................ 101
5.
Bare predication ............................................................................. 103
5.1. Some analyses of copula clauses ................................................... 104
5.1.1. Ambiguous BEs ............................................................................. 104
5.1.2. Unambiguous BE ........................................................................... 106
5.2. Post-copula bare nouns as predicates of individuals ..................... 108
5.3. Post-copula bare nouns as predicates of subkinds ......................... 112
5.4. Post-copula bare nouns as predicates of kinds .............................. 114
6.
Definite bare nouns ........................................................................ 116
6.1. Topic-hood and definiteness .......................................................... 116
6.2. Definite bare nouns in object positions ......................................... 120
6.3. Semantics of definite bare nouns ................................................... 121

Contents

xi

Part II: Functions of classifiers: counting and measuring


Chapter 6
Counting and measure functions of classifiers .................................... 127
1.
Introduction .................................................................................... 127
2.
Counting and measuring readings: a crosslinguistic perspective .. 129
2.1. Introducing counting and measuring readings ............................... 129
2.2. Structures for counting and measuring readings ........................... 133
3.
Ambiguity of container classifiers in Chinese ............................... 135
3.1. Counting and measuring readings for Chinese container classifiers
135
3.2. The syntax of counting and measuring readings ........................... 140
4.
A feature analysis of classifiers: [Counting, Measuring] .......... 143
4.1. Four types of classifiers ................................................................. 144
4.1.1. Type 1: [+C, -M] classifiers........................................................... 144
4.1.2. Type 2: [-C, +M] classifiers........................................................... 147
4.1.3. Type 3: [+C, +M] classifiers.......................................................... 148
4.1.4. Type 4: [-C, -M] classifiers............................................................ 150
4.1.5. Concluding remarks ....................................................................... 152
4.2. Syntactic support for the counting and measuring readings .......... 153
4.3. Semantic shifting between counting and measuring readings ....... 158
5.
Semantics of classifiers: counting and measuring ......................... 161
5.1. Krifkas (1995) semantics for Chinese classifiers ......................... 162
5.2. Rothsteins (2010) semantics for English classifiers..................... 164
5.3. Semantics of Chinese classifiers.................................................... 167
5.3.1. Semantics of classifiers in counting readings ................................ 167
5.3.2. Semantics for classifiers on the measure reading .......................... 171
6.
Conclusions.................................................................................... 172
Chapter 7
Adjectival modification in classifier phrases: pre-classifier adjectives
.................................................................................................................. 174
1.
Introduction .................................................................................... 174
2.
Licensing pre-classifier adjectives................................................. 176
2.1. Pre-classifier adjectives before counting classifiers ...................... 176
2.2. Concrete portion reading (Partee and Borschev 2012) .............. 179
3.
The modification relation of pre-classifier adjectives ................... 181
3.1. Adjectival modification in pseudo-partitives................................. 181
3.2. Pre-classifier adjectives in Mandarin ............................................ 183
3.2.1. Rejecting pre-classifier adjectives modifying mass classifiers ..... 184

xii

Contents

3.2.2.
4.
4.1.
4.1.1.
4.1.2.
4.1.3.
4.2.
5.
5.1.
5.2.
5.3.
6.

Rejecting pre-classifier adjectives modifying nouns ..................... 186


Pre-classifier adjectives modifying Cl+N .................................. 189
Contexts of using pre-classifier adjectives .................................... 189
Consumption contexts.................................................................... 189
Contexts of significance ................................................................ 191
Contrastive contexts....................................................................... 192
Syntactic structure ......................................................................... 193
Semantics of pre-classifier adjectives............................................ 195
Introduction to expressives ......................................................... 195
Pre-classifier adjectives as expressives ......................................... 197
Pre-classifier adjectives and plurality ............................................ 200
Conclusions.................................................................................... 201

Chapter 8
Modification marker de in classifier phrases ...................................... 202
1.
Introduction .................................................................................... 202
2.
Licensing the post-classifier de ..................................................... 204
2.1. De with mass classifiers (Cheng and Sybesma 1998) ................... 206
2.2. Information weight (Tang 2005) ................................................ 209
2.3. The indeterminacy account (Hsieh 2008) ................................... 211
3.
Unsolved problems (Li and Rothstein 2012) ................................. 214
4.
Semantics of Num-measure Cl-de-N: as much as.......................... 216
4.1. Measure classifiers and the particle de .......................................... 216
4.2. Semantics of Num-Clmeasure-de-N ................................................... 218
5.
Num-count Cl-de-N: as many as ................................................... 220
5.1. Counting classifiers in measure phrases ........................................ 221
5.2. Semantics of Num-Clcount-de-N ...................................................... 223
6.
Conclusions: two puzzles about [+counting] classifiers................ 226
Part III: Definiteness in classifier languages
Chapter 9
Definite classifiers in southern Chinese languages ............................. 233
1.
Introduction .................................................................................... 233
2.
Cl+N in three Chinese languages ............................................... 236
2.1. Cl+N in Mandarin ...................................................................... 237
2.2. Cl+N in Wu ................................................................................ 240
2.2.1. Preverbal Cl+N........................................................................... 240
2.2.2. Postverbal Cl+N............................................................................. 242
2.3. Cl+N in Cantonese ..................................................................... 243

Contents

xiii

3.
Information structure and (in)definite Cl+N .............................. 245
4.
Syntax of indefinite Cl+N .......................................................... 248
4.1. Indefinite Cl+N as a reduced form of one+Cl+N (L 1944) .. 248
4.2. Indefinite Cl+N as NumPs ......................................................... 249
4.3. Indefinite Cl+N as ClP ............................................................... 251
5.
Syntax of definite Cl+N ............................................................. 254
5.1. From Dem+Cl+N to definite Cl+N ........................................ 255
5.2. Definite Cl+N as ClP ................................................................. 257
5.3. Definite Cl+N as DP .................................................................. 259
6.
Semantic interpretation of Cl+N ................................................ 262
6.1. Cl+N with a counting reading .................................................... 262
6.2. Semantics of indefinite Cl+N ..................................................... 264
6.3. Semantics of definite Cl+N: from counting to definitenessmarking .................................................................................................... 266
6.3.1. A uniqueness-based approach of definiteness ............................... 266
6.3.2. A familiarity-based approach of definiteness ................................ 269
6.3.3. Semantics of definite Cl+N ........................................................ 273
7.
Summary ........................................................................................ 274
Chapter 10
Definite classifiers and their modifiers ................................................ 275
1.
Introduction to modified Cl+N................................................... 275
2.
Syntax of modified Cl+N in Wu ................................................ 277
2.1. Modified Cl+N as a definite expression .................................... 278
2.2. Modified Cl+N as DP................................................................. 280
2.2.1. Definite classifiers as D head ........................................................ 280
2.2.2. Dems as [Spec DP] ........................................................................ 282
2.2.3. Adjs/RCs as [Spec DP] .................................................................. 284
3.
Semantics of non-bare Cl+N ...................................................... 285
3.1. Definite classifiers characterized with familiarity ..................... 285
3.2. Interpret modified Cl-N compositionally ................................... 287
4.
Concluding remarks ....................................................................... 290
References ................................................................................................ 291
Index ......................................................................................................... 307

Abbreviations

ACC

accusative case

CL

classifier

EXP

experiential aspect

DUR

durative aspect

FOC

focus marker

GEN

genitive case

MOD

modification marker

NMLZ

nominalizer

OM

object marker

PASS

passive

PFV

perfective

PL

plural

PRF

perfect

PROG

progressive

PRT

particle

SG

singular

Chapter 1
Introduction

1. Issues
This book is a study on numeral classifiers in Chinese. It explores the
grammatical properties of Chinese classifiers at the syntax-semantics interface. The core task of this study is to look into the question of how classifiers are semantically interpreted in different syntactic contexts or how
different semantic functions of classifiers are realized at the syntactic level.
Its primary goal is to provide the missing semantic component in previous
syntactically oriented works.
The following three issues will be explored in this research: (i) the debate on a count/mass distinction in Mandarin, (ii) counting and measuring
functions of classifiers, and (iii) definiteness-marking in classifier languages.

1.1. Issue 1: the debate on a count/mass distinction in Mandarin


The first issue is concerned with the question of whether there is a lexical
count/mass distinction in the nominal domain in Mandarin. There are three
specific questions to be addressed:
(i)

(ii)

(iii)

Can classifiers encode a count/mass distinction in Mandarin? In


other words, is it plausible to posit a distinction between count and
mass classifiers, as proposed in Cheng and Sybesma (1998)?
How to characterize the contrast between shui water and nanhai
boy in Mandarin? Does it reflect a grammatical distinction between mass and count nouns or simply an ontological distinction between homogenity?
Are there mass-to-count (universal grinding) or count-to-mass
(universal package) shiftings in Mandarin?

Introduction

To start with, I make a distinction between individuation and countability (cf. Joosten 2003). Individuation is a cognitive and/or ontological
notion. It refers to whether the referents denoted by nouns are (discrete)
individuals or not. In contrast, countability is a grammatical notion. It refers to whether nouns can be directly counted by numerals or not. I take the
position that the count/mass distinction is a grammatical phenomenon and
it is thus related to the notion countability only. In view of that Chinese
nouns cannot be directly combined with a numeral, they are all mass nouns
(Krifka 1995, Chierchia 1998a, b).
As for the first question, I argue in chapter 3 that the distinction between count and mass classifiers proposed by Cheng and Sybesma
(1998) cannot be established in Mandarin. The syntactic diagnostics they
suggest cannot be justified in making a dichotomy between count and mass
classifiers and therefore, a contrast between count and mass nouns cannot
be drawn. I argue in chapter 5 that all the Mandarin bare nouns are mass
nouns and they denote kinds (see Chierchia 1998b, Yang 2001, Jiang 2012).
Concerning the second question, I find no morphosyntactic evidence
available for a grammatical distinction between mass and count nouns in
Mandarin. Contra Doetjes (1997), I claim that the contrast between shui
water and nanhai boy simply reflects an ontological distinction between
homogeneity and discreteness but not a lexical mass/count distinction.
Following up our answers to the first and second questions, it is expected that there are no grammatical operations such as count-to-mass or
mass-to-count shifting. It will be argued in chapter 4 that Chinese nouns
may refer to ontologically different entities, they have a genuine ambiguity
between object reading and stuff reading, or even a partial object reading
in some occasions (see Huang and Lee 2009).

1.2. Issue 2: counting and measuring functions of classifiers


I claim that Chinese classifiers have two basic functions: a counting function and a measure function, and that the semantic distinction between
these two readings is reflected at the syntactic level. Based on the feature
[Counting] and [Measure], a four-way classification of Chinese classifiers is proposed. With this, I claim that in Chinese, classifiers cannot be
divided into two lexical groups, like sortal and mensural classifiers (Lyons
1977, Crofts 1994) or count and mass classifiers (Cheng and Sybesma
1998). Instead, the important distinction lies in the different uses of clas-

Issues

sifiers, as expressions introducing counting or expressions introducing


measuring.
The Mandarin example san ping jiu three bottles of wine can either be
interpreted with a counting reading or a measuring reading, as in (1a) and
(1b) respectively.
(1) a. fuwusheng
kai
le
[san ping jiu].
waiter
open PFV three CLbottle wine
The waiter opened three bottles of wine.
b. fuwusheng
he
le
[san ping jiu].
waiter
drink PFV three CLbottle wine
The waiter drank three bottles of wine.

[Counting]

[Measure]

On the counting function, the classifier applies to the denotation of the


bare noun, a kind, and returns a set of atomic entities, which count as one
in a particular context (see Rothstein 2010). With this reading, the Numeral-Classifier-Noun is assigned with a right-branching structure:
[Num [Cl-N]].
On the measure function, the classifier first combines with the numeral
to form a complex modifier, which denotes the set of entities of the head
noun type whose measure value is the quantity denoted by the numera classifier (see Krifka 1995, Chierchia 1998a, Landman 2004, Rothstein 2009).
Therefore, with the measure reading, it is assigned with a left-branching
structure: [Num-Cl [N]].
This approach is significant for several reasons. First, it suggests that
both left-branching and right-branching structures are needed for Num-ClN, which capture two different semantic funcitions of classifiers in Chinese.
Second, the distinction between counting and measuring classifiers also
provide a good explanation of the licensing conditions and functions of
pre-classifier adjectives and postverbal de. These two phenomena will be
discussed in chapter 7 and 8 respectively.

1.3. Issue 3: definiteness in classifier languages


The third issue deals with whether it is possible to have a DP structure in
classifier languages such as Chinese, in which there is no grammaticalized
definite article. Contra Bokovi (2010:13) that if a language has an obligatory classifier system, it does not have DP, I argue that classifier languages, like Wu Chinese and Cantonese, can have a refined DP structure. I

Introduction

also claim that the definiteness encoded by D is characterized with the


pragmatic notion familiarity but not the semantic notion of uniqueness.
In southern Chinese languages, classifiers can mark definiteness in the
construction Cl+N. Cl+N can be used alone, as in (2a) or be preceded
by elements like adjective, relative clause, possessor, demonstrative etc., as
in (2b).
(2) a. [ts
kiu]
CL
dog
The dog died.
ts
b. [
I
CL
My dog died.

si-i die.
die
PRT
kiu]
dog

[Wu: Fuyang]

si-i die.
die
PRT

I propose that definite classifiers are quasi-definite articles that are able
to instantiate a determiner head D0 via Cl-to-D raising. The elements occurring before the definite classifier fall into two groups in terms of their
syntactic position: demonstratives are analyzed [Spec DP] and adjective/relative clauses are DP modifiers.
I argue that definiteness encoded by definite classifiers is characterized
with familiarity but not with uniqueness. To put it more specifically, I
propose that definiteness in the Chinese languages be identified with Robertss (2003) notion of weak familiarity. Definite Cl+N refer to entities
that are directly involved in the situation or are presupposed to be familiar
or identifiable by interlocutors, as part of the background information. The
modifiers preceding Cl+N express the contextual information on familiarity in an overt way.

2. Data and source


I will base the discussion of classifiers largely on the data of Mandarin
Chinese, though I will also draw data from other Chinese languages, such
as Wu and Cantonese. In chapter 9 and 10, I will explicitly discuss differences between classifiers in Mandarin, Wu and Cantonese. Therefore,
when using the expression Chinese, I mean the Chinese languages or the
Sinitic languages, and not just Mandarin.
I now provide some general background information about the three
Chinese languages that I discuss in this book: Mandarin Chinese, Wu Chinese and Cantonese (Yue Chinese).

Data and source

Mandarin Chinese was originally spoken across most of Northern and


South-western areas of China. It has now become the national language of
Peoples Republic of China. The variant of Mandarin studied in this book
is the Putong-hua common language, i.e. Standard Mandarin.
Cantonese, a southern Chinese language, is spoken in the southern provinces of China, including Guangdong Province, Guangxi Province, Hong
Kong Special Administration Region and Macau Special Administration
Region. The variant of Cantonese I study here is Hong Kong Cantonese
(mainly based on Matthews and Yips 1994 grammar).
Wu Chinese is spoken in the Yangtze Delta area including Shanghai
City, Zhejiang Province and southern Jiangshu Province. The Wu data
presented in this dissertation are based on the Fuyang dialect, the mothertongue of the author. The dialect belongs to the Taihu Lake groups of the
Northern Wu dialect. It is spoken in the Fuyang city, in the northwest of
Zhejiang province and to the southwest of Shanghai. The dialect has about
600,000 native speakers.
The language data used in this book follows the following conventions:
Mandarin: Pinyin (People's Republic of China's official Romanization
system)
Cantonese: Jyutping (the Linguistic Society of Hong Kong Cantonese
Romanization Scheme)
Wu (Fuyang): IPA symbols.
Note that the data I use in the book is Mandarin, unless marked otherwise.
Further, tone is not marked in the examples.
The author himself is a bilingual speaker of Mandarin and the Fuyang
dialect of Wu Chinese. The Mandarin examples used in this book come
from various sources, e.g. retrieved from the Peking University Corpus,
googled from the internet, made up by the author or otherwise cited from
others. They are all carefully checked with other Mandarin speakers from
different parts of China. My informants include Chen Yujie (Henan), Hung
Yuchen (Taiwan), Li Luxia (Hunan), Liang Xinliang (Liaoning), Liu Hui
(Beijing), Victor Pan (Hubei), Wang Luming (Zhejiang), Wang Jian
(Jiangsu). The Wu examples are made up by the author and doublechecked with his mother, Xu Yinfeng, and his sister, Li Ping-Er. The Cantonese examples are taken from Matthews and Yips (1994) reference
grammar of Cantonese and Cheng and Sybesmas (1999) paper. In addition,
I also checked some of the Cantonese examples with my colleages, SingSing Ngai and Hilario de Sousa, who are native speakers of HongKong Cantonese.

Introduction

3. Structure of the book


The book is composed of three parts, which deal with the three issues
raised in section I in respective order.
Part I is concerned with the debate on the mass/count distinction in
Mandarin.
Chapter 2 of the book defines classifiers in Chinese languages. I define
classifiers with both syntactic and semantic criteria: a classifier is the mediating element following a numeral or determiner and it expresses the
semantic function of counting or measuring entities. I also discuss the categorical features of classifiers in Mandarin and English: Mandarin classifiers are functional la Abney (1987), while (non-individual) classifiers
in English are relational nouns in nature.
Chapter 3 discusses the question of whether there is a count/mass distinction of nominal phrases in Mandarin. I review Cheng and Sybesmas
(1998) claim that the count/mass nominal distinction is grammatically realized in Mandarin classifiers. I examine the two syntactic tests that they
argue to substantiate this claim: (i) the presence of pre-classifier adjectives
and (ii) the optionality of the particle de after the classifier. I show that the
facts do not support Cheng and Sybesmas distinction between mass classifiers and count classifiers.
In chapter 4, I first argue that the mass/count distinction is understood
as a grammatical phenomenon. It is related to the grammatical notion of
countability but not to cognitive or ontological notion of individuation.
It is then argued that Mandarin only has an ontological distinction between
discreteness and homogeneity but has no grammatical mass/count distinction. Nouns like shui water and nanhai boy merely represent two ontologically different types of nouns, i.e. nouns referring to homogeneous and
discrete entities respectively. Additionally, I argue that there is no masscount shifting or count-mass shifting in Mandarin. Mandarin nouns have a
genuine ambiguity between object reading and stuff reading, and even a
partial object reading in some occasions.
Chapter 5 explores the semantics of bare nouns in Chinese. I, following
Chierchia (1998b), propose that all Chinese nouns are mass nouns. Chierchia (1998b) suggests that Chinese is an argumental language, in which (i)
all the bare nouns occur freely in argument positions in their bare forms,
and (ii) bare nouns make reference to kinds. I show that there is good evidence to accept Chierchias hypothesis. I look into the object-level interpretations of Chinese bare nouns by focusing on the variability of the interpretation of bare nouns in predicative positions (post-copula positions).

Structure of the book

Following Carlson (1977) and Chierchia (1998a, 1998b), I argue that the
kind interpretation is the default reading for Chinese bare nouns and that
object-level readings, including both indefinite and definite readings, are
derived from the kind reading.
Part II deals with the counting and measuring functions of classifiers in
the sequence of Numeral-Classifier-Noun.
In chapter 6, I argue that the counting and measure readings of classifiers are two basic functions of classifiers and that they are distinguished
syntactically in Chinese. Following Rothsteins (2010) semantics for count
nouns, I propose that on the counting function, the classifier applies to the
denotation of bare nouns, i.e. kind terms, and returns a set of atomic entities, which are counted as one in a certain context. On the measure function, the classifier first combines with the numeral to form a complex modifier, which denotes the set of entities of the head noun type whose
measure value is the quantity denoted by the numeral. I follow basically
the semantics of measuring in Krifka (1995) (see also Landman 2004,
Rothstein 2009).
Based on the feature [Counting] and [Measure], I develop a four-way
classification of Chinese classifiers. With this, I claim that in Chinese,
classifiers cannot be divided into two lexical groups, like sortal and mensural classifiers (as in Lyons 1977, Crofts 1994) or count and mass classifiers (Cheng and Sybesma 1998). Instead, the important distinction lies in
the different uses of classifiers, as expressions introducing counting or
expressions introducing measuring.
Chapter 7 and 8 address the two questions left open in chapter 3 respectively: (i) what are the distributional patterns and the semantic function of
pre-classifier adjectives? (ii) what is the licensing condition and semantics
of the post-classifier de? I propose an account which crucially uses the
distinction between counting and measure functions of classifiers.
In chapter 7, we argue that pre-classifier adjectives can appear before
classifiers in the counting context. That is, they can only appear before [+C,
-M] and [+C, +M] classifiers (the latter only on their counting interpretation), but not before [-C, +M] or [-C, -M] classifiers. I propose that preclassifier adjectives modify the constituent of Cl+N but not the classifier
or the noun. As for the semantic function of pre-classifier adjectives such
as da/xiao big/small, I argue that these phrases have expressive meanings in that they express the speakers evaluation of the atomic entity in the
denotation of Cl+N to be big or small from a particular perspective chosen by the speaker.

Introduction

Chapter 8 discusses the licensing condition and the semantics of the


post-classifier de. I argue that Num-Cl-de-N always has a measure reading
in that Num-Cl denotes the quantity of entities represented by N and it has
the syntactic structure [[[Num-Cl](-de-)]N]. Both measuring and counting
classifiers can be incorporated in the measure phrase marked by de, but
they express two types of measure readings: the as much as Num+Cl type
and the as many as Num+Cl type. Measuring classifiers have the as much
as interpretation, and counting classifiers have the as many as reading,
in which the counting classifier is shifted with a measuring reading. I argue
that the particle de subordinates the constituent of Num-Cl to the head
noun as a modifier. It shifts Num-Cl of type <d, t> to a modifier of type
<<d, t> <d, t>>.
Part III discusses the realization of DP structure and the semantics of
definiteness in Chinese, in particular, in southern Chinese languages like
Wu and Cantonese.
Chapter 9 discusses bare classifier phrase Cl+N in Mandarin, Wu and
Cantonese. In Mandarin Chinese, the Cl+N construction is only found in
postverbal positions and has an indefinite reading. In Cantonese and Wu,
this construction is available in both preverbal and postverbal positions. In
Wu Cl+N has a definite reading when appearing preverbally, and indefinite when appearing postverbally (Li and Bisang 2012). In Cantonese, preverbal Cl+N has a definite reading and postverbal Cl+N is either definite or indefinite (see Cheung 1972, Cheng and Sybesma 1999, 2004,
Simposon et al 2011). The following questions will be addressed concerning the distribution and (in)definiteness of Cl+N: (i) What are the factors
that constrain the distribution of indefinite and definite Cl+N? (ii) What
is the syntax of indefinite and definite Cl+N phrases? Specifically, is indefinite Cl+N a classifier phrase or a numeral phrase? Is definite Cl+N a
definite phrase? (iii) Is the Cl+N construction derived from the counting
reading or the measure reading of the classifier? (iv) What is the semantic
function of the classifier in indefinite and definite Cl+N constructions?
Chapter 10 focuses on modified Cl+N, namely, a complex form of
Cl+N preceded by adjectives, relative clauses, demonstratives, possessors etc. In section 2, I claim that modified Cl+N are unambiguously
definite. I propose a unified DP structure for bare and modified Cl+N, in
which the classifier heads DP. The preceding elements fall into two groups
in terms of their syntactic positions: demonstratives are specifiers of DP,
and adjectives/relative clauses and possessors are DP modifiers. In section
3, a compositional semantics is proposed la Bach and Cooper (1978),
namely, DPs can optionally take an extra property argument, which is satu-

Structure of the book

rated by the denotation of a high-adjoined modifier and intersected with


the property contributed by the content of the noun phrase.

Part I:
The debate on a count/mass distinction in Chinese

Chapter 2
Defining classifiers

1. Chinese classifiers: an illustration


1.1. Identifying classifiers syntactically
This chapter defines classifiers. In a broad sense, classifier is used as a
general cover term for noun classification devices, which include noun
class markers, numeral classifiers, possessive classifiers, locative classifiers etc., as discussed in Allan (1977), Craig (1992) and Aikhenvald (2000).
However, in this study, the term classifier is used to refer to numeral classifiers in a narrow sense. Numeral classifiers are an essential part of the
grammar of East and Southeast Asian languages, such as Chinese languages, Korean, Japanese, Lao, Hmong, Vietnamese etc. According to Greenberg (1974) and Bisang (1999), the existence of numeral classifiers is an
areal feature of East and Southeast Asian languages.
To illustrate, in languages like English, nouns with high countability
like woman, table, cat etc. can be directly combined with a numeral, such
as two women, three tables, one cat respectively. However, there are also
languages whose nouns in construction with numerals may occur with an
additional grammatical element even when such nouns are of high countability (Gil 2005). For instance, in Mandarin Chinese, numerals cannot
modify nouns directly and it is obligatory to have a mediating element inbetween, such as ge in (1b) and ke in (2b). They are the so-called numeral
classifiers.
(1)

a.* liang xuesheng


two student
b. liang ge
xuesheng
two CL
student
two students

(2)

a.*san

shu

[Mandarin]

14

Defining classifiers

three tree
b. san ke
three CL
three tree

shu
tree

In Chinese, the term classifier is called ling-c quantity word or


dnwi-c unit word (L 1942). This suggests that its function is to help
to express quantity or to give a unit for entities. The morpheme ge is a general classifier for individuals. In the example of liang ge xuesheng in (1b),
the numeral liang two requires the presence of ge, which provides a
counting unit for the animate noun xuesheng student. It means two students. In (2b), ke is a specific classifier for plants. It expresses the unit of
tree or flower. The numeral three has to combine with ke to modify the
noun shu tree. The phrase means three individual trees.
As the term itself suggests, a numeral classifier typically goes with
numerals. In languages like Mandarin, it is an obligatory syntactic requirement that numerals take a classifier when modifying nouns. The examples in (1) and (2) characterize one of the most prototypical syntactic
contexts where numeral classifiers may occur, i.e. Numeral + Classifier +
Noun (Num-Cl-N for short).
In other classifier languages, nouns are not preceded but followed by
Num+Cl. For instance, Thai has the word order N+Num+Cl, in which
the classifier is obligatory too. Look at (3) for an illustration.
(3)

a.*thrian
sam
durian
three
sam
b. thrian
durian
three
three durians

[Thai]
luu k
CL
(Jenks 2011: 77)

Nonetheless, numeral classifiers are not merely restricted to the cooccurrence with numeral. They also occur with determiners in these classifier languages. Det+Cl+N (or N+Det+Cl) is the second possible syntactic context for numeral classifiers. I use the example from Mandarin and
Lao to illustrate.
In Mandarin, it is obligatory for some determiners to take a classifier
when modifying nouns, such as the universal quantifier mei every. Look
at (4) as an illustration.
(4)

a.*mei

shu

[Mandarin]

Chinese classifiers: an illustration

15

every book
b. mei ben
shu
every CL
book
every book
According to Enfield (2007: 122), a number of quantifiers in Lao also
require numeral classifiers, appearing in the same constructional pattern as
numeral classifier expressions:
(5)

a. kuu3 s4 (paa3) baang3 too3. [Laos]


I
buy
fish
some CL
I bought some (of the fish).
b. kuu3 s4
(paa3) thuk1 too3.
I
buy
fish every CL
I bought every one (of the fish).

Examples from (1) to (5) characterize two prototypical syntactic environments for numeral classifiers:
(i) Numeral-Classifier-Noun (or Noun-Numeral-Classifier)
(ii) Determiner-Classifier-Noun (or Noun-Determiner-Classifier).
Therefore, from the syntactic perspective, the term numeral classifier
can be understood as the mediating element occurring contiguous to numeral or determiner when modifying nouns. Semantically, it has the function
of counting or measuring entities, namely, providing counting or measuring
units.
There are two issues arising immediately. First, the examples I showed
so far are all concerned with classifiers modifying nouns with high countability in Gils term (2005). They are often called sortal or individual
classifiers, such as ben volume (for book) in Mandarin, luuk in Thai and
too3 in Lao. However, there are also elements modifying nouns with low
countability, such as cup in one cup of coffee, pound in two pounds of
sugar, plume in three plumes of smoke. They are called mensural or
non-individual classifiers. Should I consider them to be classifiers too?
My answer to this question is positive. I take the position that as long as
a morpheme is able to satisfy the syntactic and semantic requirements for
classifiers as stated above, they can be categorized as classifiers, including
both individual classifiers and non-individual classifiers.
Second, non-individual classifiers are found in almost every language
(universally available Croft 1994, Bisang 1999), whereas individual

16

Defining classifiers

classifiers are exclusively found in languages like Mandarin, Thai, Lao,


Vietnamese. Languages in the world can be classified into typologically
different types, depending on whether a language has sortal classifiers or
not. The term classifier language is used here to refer to languages with
sortal classifiers. Therefore, Chinese languages like Mandarin are representative classifier languages, while Indo-European languages like English
are non-classifier languages.
According to WALS (the World Atlas of Language Structures Online
2005), among the 400 languages investigated, 260 languages have no sortal
classifier, 62 languages allow an optional use of sortal classifiers, and 78
languages uses sortal classifiers in an obligatory manner. Therefore, in
addition to classifier and non-classifier languages, there are also optional
classifier languages.
In optional classifier languages, there are individual classifiers, but their
occurence in Numeral-Classifier-Noun is optional. One example of such
a language is Minangkabau, in which classifiers such as ikue in (6) are
sometimes present but in other cases absent.
(6)

duo
(ikue) anjiang
two
CL
dog
two dogs

[Minangkabau]
(Gil 2005)

In brief, here, I investigate numeral classifiers in Chinese, that is, the


grammatical morpheme occurring between numeral/determiner and noun.

1.2. Chinese classifiers: a heuristic classification


According to the syntactic criterion that a classifier refers to the element
occurring between numeral/determiner and noun in Chinese, elements from
different sources may fall into this category. In what follows, I will show
how different lexical types of classifiers can be distinguished in a heuristic
way la Chao (1968).
Chao (1968) makes a classification of six lexically different types of
classifiers (measure in his terms)1: individual classifiers, group classifiers,
1

In earlier descriptive works, the element between numeral/determiner and noun is


called classifier by some and measure by others. These two terms are not clearly
distinguished and one is often subsumed under another. For example, Chao (1968:
584) treats all Chinese classifiers as measures in the sense that a measure is a

Chinese classifiers: an illustration

17

partition classifiers, container classifiers, temporary classifiers and standard measures.2 The relevant Mandarin examples are given from (7) to (12):
Individual classifiers: modify nouns according to the entitys shape,
or other properties.
(7)

a. liang ke
two CL
two trees
b. yi
tiao
one CL
a river

shu
tree
he
river

Individual classifiers individualize a given count noun by designating


its semantic boundaries (Bisang 1999: 118-119) or by designating its
natural unit (Croft 1994: 163). The classifier ke in (7a) is a specific individual classifier for plants, such as shu tree and cao grass. The classifier
tiao in (7b) modifies nouns denoting long-shaped entities, including he
river, bahen scar, shengzi rope etc. The lexical meaning expressed by
individual classifiers usually has to be topologically compatible with the
entities denoted by noun, e.g. their shape, size etc.
Group classifiers: used for a group or collection of individuals. According to Lehere (1986: 117), it expresses numerosity.
(8)

a. san qun
xuesheng
three CLgroup student
three groups of students
b. yi
kun
daocao
one
CLbundle straw
a bundle of straws
The classifier qun group in (8a) modifies [+animate] nouns, such as
ren people, xuesheng students or zhu pig etc. It represents an aggre-

bound morpheme which forms a Determinative-Measure compound, where the


Determinative includes demonstratives, numerals, or quantifiers. In contrast, Li and
Thompson (1981: 106) subsume measures under classifiers and say that any measure word can be a classifier. In the current research, I use the term classifier.
2
Chao (1968) listed out nine types of classifiers: seven types of nominal classifiers
and two subtypes of verbal classifiers. I ignore the seventh class of nominal classifiers, i.e. quasi-measure expressions, because they behave more like nouns than
like classifiers.

18

Defining classifiers

gate or a gathering of animated entities. The group classifier kun in (8b) is


similar to the meaning of bundle in English, referring to the collection of
entities tied together with strings. Classifiers in this group include dui
pile in yi dui pingguo a pile of apples, cuan string in yi cuan lajiao a
string of chili peppers, ba handful in yi ba douzi a handful of beans.
Partition classifiers: represent portions of things. What is speical
about partition classifiers is that they represent some particular part of entity taken from a large part or a whole individual entity. In other words, a
part-whole relation is always implied.
(9)

a. yi
jie
shengzi
one CLsection rope
a section of rope
b. yi
pian dangao
one
CLpiece cake
a piece of cake

The noun shengzi rope can be modified by the individual classifier


tiao, as in yi tiao shengzi a rope, which represent a length of rope. Contrastively, in (9a), yi jie shengzi simply means a small section of rope,
which is cut off from a long one. The partitive classifier phrase in (9a)
presupposes the existence of a long rope. The example (9b) has the similar
presupposition that a piece of cake is partitioned out of a large piece.
Container classifiers: container nouns used as measures.
(10) a. yi
ping
jiu
one
CLbottle wine
a bottle of wine
b. san che
chengke
three CLbus passenger
three busloads of passengers
There are two points worth noting about container classifiers. First, the
container referred to by the container classifier does not have to exist physically in every case. Yi ping shui in (10a) can either refer to the scenario
that there is a physical bottle which contains wine or that there is as much
wine as one bottle, say, 750ml. In the latter case, the container is used as a
standard measure. Second, container nouns are an open class, but container
classifiers are not. Not every container noun can be used as container clas-

Chinese classifiers: an illustration

19

sifiers. There are some general constraints when deriving container nouns
into container classifiers. This point will be elaborated in section 3.
Temporary classifiers: use the outside extent of objects to measure
quantity.
(11) a. yi
shen xue
one
CLbody snow
a bodyful of snow
b. yi
bizi
hui
one
CLnose dust
Lit:a noseful of dust An idiom meaning: being refused
In Lehrer (1986), temporary classifiers are treated as a subtype of container classifiers. The concept container must be construed very generally to include not only objects for putting other objects and substances in,
but also on and even under (Lehrer 1986: 120). Examples include a ceiling of posters, a wall of pictures, and a shelf of books in English.
However, I think that there are some crucial semantic differences between
container classifiers and temporary classifiers. Container classifiers can be
easily used as standard measures (by convention), while generally speaking,
temporary classifiers can only be inexact measures.
Standard measures: they are measures proper. The set of standard
measures is small and closed in Chinese. It is also true in other languages
(see Allan 1977). Standard measures range from measure units for weight,
length, volume and so on.
(12) a. wu
mi
bu
five CLmeter cloth
five meters of cloth
b. liang gongjin pingguo
two CLkilo apple
two kilos of apples
These are the six types of classifiers discussed in Chao (1968). One particular type of classifier that is not discussed by Chao (1968) is kind classifier. They are zhong kind/species and lei type in Chinese. They are
used in the same syntactic contexts as the classifiers shown above, namely,
they can be used in Num/Determiner+Cl+N, as shown in (13).
(13) a. zhe

zhong dongwu

20

Defining classifiers

this CLkind animal


this kind of animal
b. liang lei
butong-de
ren
people
two CLtype different
two different types of people
In this research, I will also take into account kind classifiers.
The classification of these six subtypes of classifiers made by Chao
(1968) is heuristic. It is based on the different origins and lexical meanings
of classifiers. From examples (7) to (13), it appears that Chinese classifiers
do not constitute a semantically homogeneous class, although they can be
identified consistently according to the syntactic criterion, i.e. the syntactic
contexts in which they are allowed. Different subtypes of classifiers emphasize different ways of quantifying entities, such as the natural unit of
entities, the quantity of aggregated entities, or standard measurement of
entities.
In the rest of the chapter, I will examine whether the different subtypes
of classifiers in Chinese constitute a uniform category in terms of their
syntactic properties. Specifically, I am interested in finding out whether the
classifiers have the same categorical feature, namely, having the same degree of lexical/functional make-up. I argue that while Chinese classifiers
come from different sources, they constitute an independent category, to be
distinguished from nouns or numbers. They are functional and belong to a
closed class.
The remainder of the chapter is structured as follows. From section 2
through 5, I will spell out the argument that Chinese classifiers constitute a
functional category la Abeny (1987). Section 6 offers a contrastive
look at classifiers in English. The chapter is concluded in section 7.
Before proceeding to justify our arguments, I summarize Abenys (1987:
64-65) five criteria for functional elements (not in the original order):
a. Functional elements constitute closed classes.
b. Functional elements lack descriptive content. Their semantic
contribution is second-order, regulating or contributing to the interpretation of their complement. They mark grammatical or relational
features, rather than picking out a class of objects.
c. Functional elements permit only one complement, which is in general not an argument. The arguments are CP, PP and DP. Functional
elements select IP, VP, NP.
d. Functional elements are usually inseparable from their complement.

Classifiers as a closed class

21

e. Functional elements are generally phonologically or morphologically


dependent. They are generally stressless, often affixes or clitics, and
sometimes even phonologically null.

2. Classifiers as a closed class


The first issue I am interested in is whether classifiers belong to an open or
a closed word class. I will show that Mandarin classifiers belong to a
closed class.
Greenberg (1972), Allan (1977), Denny (1976, 1979), and others, all
point out that classifier languages have a small, closed, paradigmatic contrasting set of morphemes (cf. Lehrer 1986: 110). The number of classifier
varies in each classifier language. In Mandarin Chinese, one classifier
dictionary lists 199 classifiers, 76 of them sortal (38%), while a teaching
grammar of the 143 most commonly used classifiers lists 66 sortal
(46%) (cf. Erbaugh 2002:39).
I counted the numbers of each type of classifiers listed in Chao (1968),
as shown in the following Table 1. According to this table, there are
around 50 individual classifiers in Mandarin. Roughly speaking, the number of the most frequently used individual classifiers in Mandarin ranges
from fifty to seventy, depending on factors like register and regional differences etc.
Table 1. Numbers of classifiers in each subtype

Classifiers
Individual Cls

Numbers
51

B
C
D
E
F

Group Cls
Partition Cls
Container Cls
Temporary Cls
Standard measures

46
39
36
14
46

Examples
ge (the general classifier), ben volume,
tou head (classifier for animials)
qun group, lie series, kun bundle...
pian piece, jie section, si slice
he box, hu kettle, wan bowl
shen body, lian face, di floor
mi meter, shen liter

Table 1 does not represent an exhaustive list of the number of classifiers, since what Chao lists only represents the most frequently used classifiers in each type. Among the six types of classifiers in Chinese discussed in
Chao (1968), members of individual classifiers, group classifiers, partition
classifiers and standard measures are relatively stable. It is rare to see new

22

Defining classifiers

members join in, though this possibility cannot be ruled out absolutely.
However, those of container and temporary classifiers seem to be open to a
great extent. Intuitively, the number of container and temporary classifiers
should be much larger than what is listed in Table 1. This is because nouns
referring to containers can be easily used as container classifiers, such as yi
dai laji a bag of rubbish, and nouns referring to objects whose surface or
outer area is able to measure quantities of entities can be used as temporary
classifiers in many cases, such as yi lian hanshui a faceful of sweat.
These two types of nouns belong to an open class, so it is very easy for
many to make the analogy that these two types of classifiers, i.e. container
and temporary classifiers, are also open. However, I do not think that this
kind of analogy withstands scrutiny. I use container classifiers to illustrate
some of the problems with this kind of analogy.
In many cases, nouns expressing typical containers cannot be used as
container classifiers. For example, jiaoshi classroom can be seen as a
perfect container for xuesheng student in a metaphorical sense, as in
the example many students sat in the classroom. However, the expression of #yi jiaoshi xuesheng #a classroom of students is not felicitous in
Mandarin. At least, it is not as natural as yi ban xuesheng a class of students or yi wu xuesheng a room of students. A similar example is about
the use of qianbao wallet. We usually put money in a wallet, so it is reasonable to consider wallet to be a prototypical container for money. However, it is not common to use qianbao as a container classifier for money
and the expression #yi qianbao qian #a wallet of money is unacceptable,
though the meaning of the expression is understandable to native speakers.
Many would use yi bao qian a bag of money instead. This shows that
container nouns can express a relation of containment, but they cannot be
used as container classifiers freely. There are some underlying constraints
for such a category shift.
It is a conventionalized practice to choose a certain classifier to modify
a certain group of nouns. There are some selectional constraints between
the classifier and the noun, which may be semantic or conventional. One of
the constraints for the container noun-to-container classifier-shifting is
conventionalization. For example, to talk about cherries, it is common to
say a basket of cherries or a plate of cherries but it is not as common to
say a mug of cherries. This expression is understandable to the addressees,
but mug is not a conventional container to cherries or it is not a conventional practice to use mug to carry cherries. Certainly, in this special context, mug can be (non-conventionally) analyzed as a special container clas-

Classifiers without descriptive content

23

sifier for cherries, but this use will not be conventionalized or accepted in a
general way in the linguistic community.
To sum up, there are conventional and non-conventional uses of classifiers. The members of conventional classifiers are stable to a great extent,
while the innovative use of classifiers may introduce new classifiers into
the system, but innovative classifiers do not change the membership of
classifiers in nature. Historically speaking, some of the core members of
conventional classifiers developed from innovative classifiers, but this kind
of development or conventionalization takes a fairly long process.
I conclude that Mandarin classifiers belong to a closed class. Each subtype of classifier has stable and conventionalized members.

3. Classifiers without descriptive content


This section teases out the relation between lexical meaning and descriptive content of classifiers. The important questions to be pursued are: (i)
are classifiers or some of the classifiers (e.g. mensural classifiers) nominal
in nature? (ii) does the fact that a classifier expresses some lexical meaning
mean that they belong to a lexical category? Our answers to these two
questions are in the negative.

3.1. Classifiers are not nominal


Many hold the opinion that Mandarin classifiers are lexical, or more precisely, being nominal. For example, Zhu ([1982] 1990: 50) considers classifiers to be content words, on a par with nouns. A similar idea dates back
to Wang (1943), who defines classifiers as a special type of noun. They are
called unit noun (danwei mingci in Chinese). A recent account in support
of this view is found in Wu and Bodomo (2009: 488), who claim that
Mandarin classifiers are contentful morphemes which are used to indicate the semantic classes of nouns; hence, they often carry information
beyond that carried by their associated noun (cf. Allan 1977).
Another viewpoint is that different types of classifiers have different
lexical/functional make-ups. According to Cheng and Sybesma (1998: 1417), there are count and mass classifiers in Chinese, which have different categorical features. Assuming that all classifiers have nominal orgins,
they claim that count classifiers are fully grammaticalized classifiers and

24

Defining classifiers

cannot occur as independent nouns, while massifiers are really nouns , as


such are an open class.
One group of classifiers consists of elements which are completely grammaticalized as classifiers. They form a closed class, and they cannot occur
as independent nouns.
The other group of classifiers do not constitute a closed class in the sense
that any noun which can be seen to create a unit for measuring mass nouns
can be used (notably all sorts of container words or words that can be interpreted as containers; e.g. wan bowl, bei cup). So the latter group consists
of elements which also occur as independent nouns.

I first argue against the view that Mandarin classifiers have a nominal
origin. Classifiers in Mandarin grammaticalize from different categories,
which can be nouns, verbs or adjectives. A majority of classifiers have a
nominal origin, but some of them are evolved from verbs and adjectives.
See the illustration in (14) and (15).
The classifiers in (14) are derived from verbs historically. For example,
on its verb use, gua in (14a) originally means hang. It has an extended
use as classifier, meaning string. The morpheme peng in (14b) has the
lexical meaning scoop with hand; in classifier use, it means handful. In
(14c), the morpheme cuo, as a verb, means take up with fingers literally.
Used as a classifier, it means pinch.
(14) a. yi
gua
bianpao
[classifiers derived from verbs]
one CLstring firecrackers
a string of firecrackers
b. yi
peng
xiangtu
one CLhandful soil:from:homeland
a handful of soil from homeland
c. yi
cuo
yan
one CLpinch salt
a pinch of salt
In contrast, those in (15) are adjectives: wan means curved and fang
means square. When they are used as classifiers, they refer to entities
with the shape of being crescent or square respectively.

Classifiers without descriptive content

25

(15) a. yi
wan
mingyue [classifiers derived from adjectives]
one
CLcurved moon
a cresent of moon
b. yi
fang yantai
one CLsquare ink stone
a piece of ink stone
In the two groups of examples in (14) and (15), the element between
numeral and noun cannot be treated as verbs or adjectives any more. They
are reanalyzed as classifiers, which provide units to measure or count entities referred to by noun. Specifically, those in (14) are non-individual classifiers and those in (15) are individual classifiers. If the examples in (14)
and (15) are taken into account, then, surely, it is incorrect to say that all
classifiers are originally nouns.
Second, I show that it is not correct either to treat a subgroup of classifiers as nouns, such as mass classifier or non-individual classifiers.
Among the different types of non-individual classifiers, container classifiers have the highest degree of nominal properties and they have noun uses,
but group classifiers and most partition classifiers have a rather low degree
of nominal properties and they cannot be used as nouns.
The first column of (16) illustrates the classifier use of the morphemes.
In the second column, the corresponding morphemes are modified by the
general classifier ge, as in the construction Num-ge-N. This suggests that
container classifiers have a high degree of nominal properties and they
have noun counterparts.
(16)

container Cl
a. yi
ping
shui
one CLbottle water
one bottle of water
b. yi
xiang pingguo
one
CLbox apple
one box of apples
c. yi
dai
binggan
one
CLbag biscuit
one bag of biscuits

a. yi *(ge)
one CL
one bottle
b. yi *(ge)
one CL
one box
c. yi *(ge)
one CL
one bag

N
ping
bottle
xiang
box
dai
bag

In contrast, group classifiers show a low degree of nominal properties in


that they cannot function as nouns independently. As shown in the second
column of (17), they cannot be modified by the general classifier ge and

26

Defining classifiers

they are not allowed in Num-ge-N. However, they can be a part of a noun
compound, which can be modified by the general classifier ge, as in the
third column.
(17)

group Cl

N
N-compound
a. yi kun daocao a.*yi ge kun
a.yi ge cao-kun
one CLbundle straw
one CL bundle
one CL straw-bundle
a bundle of straws
a bundle
a straw-bundle
b. yi huo liumang b.*yi ge huo
b. yi ge tuan-huo
one CL group
one CL gang-group
one CLgroup hooligan
a group of hooligans
a group
a gang
c. yi
dui huo
c.*yi ge dui
c. yi ge huo-dui
one CL pile
one CL fire-pile
one CLpile fire
a pile of fire
a pile
a bonfire

Most partition classifiers have an even lower degree of nominal properties. They can neither be used as nouns nor can they form noun compounds,
though originally they are nouns. Partition classifiers such as luo pile or
duan section in (18) cannot be formed into noun compounds.
(18)

partition Cl

a. yi
luo wenjian
one CLpile document
one pile of documents
b. yi duan ganzhe
one CLsection sugarcane
one section of sugarcane
c. yi zhang zhi
one CLpiece paper
one piece of paper

N
luo
pile

a.* yi
ge
one CL
one pile
b.* yi ge
duan
one CL section
one section
c* yi
ge zhang
one CL piece
one piece

Therefore, it is not entirely true to say that non-individual classifiers are


nouns. They exhibit different degrees of nominal properties: some have
noun counterparts, some can only be part of noun compounds and others
can simply be classifiers.
Importantly, individual classifiers do not form a unified class either in
terms of their nominal properties. There are indeed a large number of individual classifiers that can be used as independent nouns. For example, in
(19), the count classifiers are as lexical as container classifiers are. As

Classifiers without descriptive content

27

shown in the second column in (19), they, as nouns, must themselves be


preceded by classifiers in order to be modified by numerals.
(19)

count CL
a. yi
tou
niu
one CLhead bull
a bull
b. wu shan men
five CLfan door
five doors
c. san ben
shu
three CLvolume book
three books

N
a.* yi
ge tou
one
CL head
one head
b.* wu ba shan
five CL fan
five fans
c.* san ge ben
three CL exercise-book
three exercise books

In contrast, some individual classifiers have a low degree of nominal properties. They cannot be used as nouns independently, but they can be used in
noun compounds, as in (20).
count CL
N
N-compound
a. yi ba yizi
a.* yi ge ba
a. yi ge che-ba
one CL chair
one CL grip
one CL bicycle-grip
a chair
Intended: a grip
a bicycle-grip
b. yi duo hua
b.* yi ge duo
b. yi duo hua-duo
one CL flower
one CL bud
one CL flower-bud
a flower
Intended: a bud
a bud
c. yi zhi hua
c.*yi ge zhi
c. yi gen shu-zhi

one CL flower
one CL twig
one CL tree-twig
a branch of flower Intended: a twig
a twig

(20)

There are also individual classifiers that have an even lower degree of
nominal properties in that they can only be used as classifiers, as given in
3
(21) .

According to Hanyu Dazidian [Comprehensive Chinese Dictionary], the classifiers in (21) were all nouns in origin. For example, the general classifier ge means
bamboo branch, zhi means bird and mei trunk in classical Chinese. However,
in modern Chinese, the lexical meanings of this group of classifiers are bleached to
such a degree that they cannot be used independently anymore.

28
(21)

Defining classifiers

count CL

a. yi
ge
ren
one
CL
man
a man
b. yi
zhi
dou
one CL
dog
a dog
c. yi
mei
tongqian
one
CL
copper coin
a copper coin

a.* yi ge
one CL

N
ge
CL

b. *yi ge
one CL

zhi
CL

c.* yi ge
one CL

mei
CL

Until now, I have answered the first question raised at the beginning of
this section. I make two claims concerning the question whether classifiers
are nominal. First, Mandarin classifiers are derived from different categories, such as nouns, verbs and adjectives, but most of them have a nominal
origin. Second, among those classifiers with a nominal origin, they exhibit
different degrees of nominal features. This holds true for both individual
and non-individual classifiers. Among the individual classifiers, there are
classifiers with a high degree of nominal properties, and also classifiers
with a low degree of nominal properties, and the same is true for nonindividual classifiers. Therefore, it is inappropriate to claim that all Mandarin classifiers are nominal or that non-individual classifiers are nominal.

3.2. Classifiers have no descriptive content


As for the second question, I argue that classifiers have lexical content, but
they are not content words. It is important to distinguish between lexical
meaning and descriptive content of classifiers.
Wu and Bodomo (2009) argue that Chinese classifiers are contentful
morphemes. They add extra semantic information about the entities expressed by the noun. Specifically, they are used to indicate the semantic
classes of nouns; hence, they often carry information beyond that carried
by their associated noun.
According to them, due to their content, different classifiers can be used
with the same noun to create distinct cognitive effects. The examples in
(22) are taken from them, where the noun shu book can be modified by
different classifiers, such as ben volume, bao bag, luo pile and xiang
box.

Classifiers without descriptive content

29

(22) a. yi
ben
shu
one
CLvolume book
one book
b. yi
bao
shu
one
CLbag
book
one bag of books
c. yi
luo
shu
one
CLpile book
one pile of books
d. yi
xiang shu
one
CLbox book
one box of books
That some element expresses lexical meaning does not imply that it is a
lexical item or a content word. Lets take auxiliary verbs for example. Both
in English and Chinese, auxiliary verbs give further semantic or syntactic
information about a main verb, such as modality, attitude or temporal references. However, it is generally agreed that they are functional words in
both languages. Another similar case regards prepositions. Prepositions are
functional words but express semantic content, such as spatial (or temporal)
meaning. Therefore, the fact that classifiers are able to add extra information about the denotation of nouns does not necessarily lead to the conclusion that they are content words.
As said earlier, a single lexical item may be used as a noun or as a classifier, such as the dual use of bao bag. However, the meaning expressed
by its classifier use is distinctive from what is expressed by its noun use.
Each use has a different meaning. When the lexical item bao bag is used
as a noun, it refers to a certain bag, say, a school bag or a plastic bag.
However, when it is used as a classifier, as in (22b), it is used in a relational way as a container to hold books. In this case, the main function of bao
bag is to express a containment relation. It is even possible that there is
no physical bag in the actual scene. The phrase yi bao shu may simply
mean a package of books, which are wrapped up with craft paper. Therefore, it seems that a single lexical item can be used in two different syntactic contexts, in which they express different semantic meanings. It is possible that there are two different lexical entries for a single morpheme in this
case: a noun entry and a classifier entry (a sortal noun use and a relational noun in Lbner 1985 and Partee and Borschev 2012).

30

Defining classifiers

This provides further evidence in support of the view that classifiers


should be categorically distinguished from the word class, from which they
are derived, which are content words and can be nouns, verbs or adjectives.
What is meant by saying that the classifiers are functional is that they
do not have descriptive content (Abeny 1987: 65), and they express the
grammatical function of counting or measuring entities, as will be defined
later in chapter 6.

4. Classifiers are complement-taking


I now look at Abenys (1987) criterion (c) and (d). Criterion (c) states that
functional elements permit only one complement, which is in general not
an argument. Criterion (d) says that functional elements are usually inseparable from their complement.
It is impossible to use classifiers alone in Mandarin Chinese. They must
always be adjacent to some other element, either following determiners/numerals or preceding nouns. The examples from (23) to (28) illustrate
the possible syntactic contexts for classifiers in Mandarin.4
4

Tang (1990) examines the possible contexts of classifiers in Mandarin as well.


According to her, the demonstrative or the numeral must co-occur with the classifier. According to Tang, the examples in (i) are ungrammatical.
(i) a.*na
shu
(Tang 1990:398) b.*ben shu
(Tang 1990: 400)
that book
CL book
However, Tangs observation is not complete. Both Dem+N and Cl+N are
possible in Mandarin, but they are constrained by some restrictions. Generally
speaking, Dem+N is usually found in subject positions and not in object position
in Mandarin, as exemplified in (ii).
(ii) a. zhe shu shi wo- de.
b. ???wo mai le
zhe shu.
this book be mine
I buy PFV this book
This book is mine.
I bought this book.
As for Cl+N, Zhu (1982: 52) points out that in Mandarin, Cl+N can appear
in object positions and not in subject positions. Cheng and Sybesma (1999) argue
that indefinite Cl+N are restricted in lexically governed positions, such as postverbal positions. Nevertheless, Lin (1997) said that Cl+N can be used after verb
as object, but it cannot appear as the complement of a preposition, as in (iii).
(iii) Ta gen
*(yi)
jia
chubanshe you heyue.
he with
one
CL
publisher have contract
Intended: He has a contract with a publisher.

Classifiers are complement-taking

31

Firstly, classifiers are complement-taking, and noun phrases are the only possible constituent that can be taken as complement to classifiers.
(23) [Cl+N]
wo
gei
ni
pao
1SG
give
you
make
I made you a cup of tea.

le
PFV

[bei
CL

cha].
tea

Cl+N can only have a singular reading, so bei cha in (23) only means
a cup of tea, but not cups of tea. Some believe that in the case of (23), there
is a covert numeral yi one before Cl+N. I will discuss this issue in
chapter 9.
A second context is that the constituent of Cl+N can be modified by
numerals, as the sequence of Num+Cl+N in (24).
(24) [Num+ Cl+N]
ta
qing le
[liang ge
he
invite PFV two
CL
He invited two students.

xuesheng].
student

Thirdly, Cl+N can be directly modified by determiners and Dem+


Cl+N has a singular interpretation, as in (25). In order to have plural readings, numerals larger than one can be inserted between Dem and Cl, as in
the sequence Dem+Num+Cl+N, as in (26).
(25) [Det+Cl+N]
wo
yao
[na
I want that
CL
I want that book.

ben
book

shu].

(26) [Det+Num+Cl+N]
[na
san
ben
shu] shi
wo-de.
that
three CL
book be
mine
Those three books were bought by me.
Fourthly, when the classifier is modified by numerals and/or determiners, the NP can be elided. Num-Cl and Dem-(Num)-Cl are both legitimate constructions. The relevant examples are given below, where the NP
is either elided, which can be recovred from the context, as in (27) or topicalized, as in (28).

32

Defining classifiers

(27) [Det+Cl]
wo
yao
[na
ben shu].
I
want that
CL book
I want that volume (of book).
(28) [Num+CL]
pingguo, wo
zhi
chi
apple
I
only eat
Apples, I only ate five.

le
PFV

[wu
five

ge].
CL

Nevertheless, as I mentioned earlier, a classifier cannot be used by itself.


For example, the example in (29) is ungrammatical, since there is no numeral preceding it or noun following it.5
(29) *[Cl]
* cha, wo
pao
le
bei.
tea
I
make PFV CLcup
Intended: As for tea, I made a cup.
It is clear from examples (23) to (29) that classifiers are complementtaking and they take an NP as complement.

5. Classifiers as stressless
Classifiers are usually stressless in Mandarin. As Tao (2006) said that in
Beijing Mandarin, there is often one accent on the phrase; it falls on the
numeral while leaving the classifier unstressed. Chen (2000:321) also
5

Classifiers can be reduplicated and express distributive meaning. The reduplicated


classifiers behave more like floating quantifiers and they usually requires a plural
nominal phrase as antecedent, as in (i). However, reduplicated classifeirs cannot be
used in argument positions alone, such as subject or object, as in (ii). This
distinguishes reduplicated classifiers from other classifier phrases, which are
argumental, such as Num-Cl-N or Dem-Num-Cl-N.
(i) ta
de
xuesheng ge-ge dou hen congming.
He
Mod
student Cl-CL all very clever
His students all are clever.
(ii) # ge-ge dou hen
congming.
CL-CL all very
clever
Intended: Each is clever.

Classifiers in English: a contrastive look

33

points out that classifiers in Beijing Mandarin and many other Chinese
dialects usually bear a neutral tone (toneless). The example Chen (ibid)
gives is that in the phrase liang ge ren two Cl person, the classifier ge is
stressless, as indicated by the underline. This generalization applies to
standard Mandarin as well (see Chao 1968).
The exception is that classifiers can be contrastively focused by stress
placement. As in (30), the classifier wan bowl and guo pot are stressed,
as indicated by letters in bold. It expresses the contrastive meaning that two
individuals drank two different quantities of soup, say a large quantity, as
expressed by wan, versus a small quantity, as expressed by guo.
(30) wo he le
yi wan tang, ta he
le
yi guo.
I drink PFV one CLbowl soup he drink PFV one CLpot
I drank a bowl of soup, and he drank a pot.
Intermediate summary:
To summarize section 2 to 5, I showed that Mandarin classifiers pass
Abenys (1987) tests: (i) belonging to a closed class, (ii) lacking descriptive content, (iii) being syntactically dependent, (iv) being phonologically
weak. Accordingly, classifiers in Mandarin are considered to be functional
and not lexical (also see Muromatsu 1998 for a similar view on classifiers
in Japanese).
It is a standard assumption that Chinese classifiers head an independent
functional projection of classifier phrase (ClP for short) (Tang 1990 a, b; A.
Li 1999; Cheng and Sybesma 1999, among others). Applying Abenys
(1987) DP hypothesis to Chinese, Tang (1990a, b) argues that in Chinese,
there is an extra projection between the determiner phrase, i.e. DP and the
noun phrase, i.e. NP and this projection is realized by a classifier phrase,
i.e. ClP. Tang (1990) thus proposes a DP structure like
[DP D [NumP Num [ClP CL [NP N]]]] for nominal phrases in Mandarin. I will
discuss in chapter 6 what are the possible syntactic structures for the sequence of Num-Cl-N.

6. Classifiers in English: a contrastive look


This section makes a comparison of classifiers between Mandarin Chinese
and English by focusing on their categorical differences. I said earlier that
Mandarin is a classifier language, whereas English is a non-classifier lan-

34

Defining classifiers

guage. When talking about classifiers in English, I refer to non-individual


classifiers but not the individual classifiers. The construction which allows
these classifiers are usually called pseudo-partitives (Selkirk 1977). I
propose that classifiers in non-classifier languages are borrowed from
nouns and do not have an independent categorical status as classifiers (see
Jackendoff 1977 for English, Riemsdijk 1998 and Corver 1998 for Dutch,
Rothstein 2009 for the analysis of classifiers in English and Modern Hebrew), and that classifiers in Chinese constitute an independent category,
and they are distinguished from nouns.
According to Allan (1977), the following five types of classifiers are
distinguished in English (cf. Lehrer 1986: 111).6 Note that measure classifiers in Allans terminology actually corresponds to three different types
of classifiers in Chaos taxonomy, i.e. standard measures, container classifiers and temporary classifiers.
(31) Unit counters
a. a piece of equipment
b. a head of cabbage
c. an ear of corn
(32) Collective classifiers
a. two clumps of grass
b. a herd of animals.
(33) Varietal classifiers
a. two species of wheat
b. all kinds of flowers
(34) Measure classifiers
a. two pounds of cabbage
b. one liter of wine
c. a box of candy
d. a bowl of sugar.

Allan (1977) considers fractional classifiers, such as three quarters of the cake
and number set of classifiers, such as many hundreds of people and dozens of bird,
to be classifiers as well. In many languages, these expressions are expressed by
quantifiers. For example, Lehrer (1986: 112) did not consider them as classifiers
and said that the syntax and semantics for quantifiers will handle most of them.

Classifiers in English: a contrastive look

35

(35) Arrangement classifiers


a. two rows of beans
b. three stacks of books.
There are striking differences between English and Chinese classifiers
with respect to their grammatical properties and categorical features.
To start with, it is important to make a distinction between unit counters
in English and individual classifiers in Chinese. At first glance, unit counters in (31) seem to be very much like individual classifiers in Chinese, on
the ground that they both provide a counting unit for discrete entities, such
as a head of cabbage in (31b) and an ear of corn in (31c). However, unlike
Chinese individual classifiers, these unit counters in English are not required between numeral and noun in an obligatory manner. This is one of
the crucial differences between them. Moreover, most of these expressions
were codified in the fifteenth century (Lipton 1968), but they are not
productively used in modern English. English employs Numeral + Noun
as a dominant syntactic structure to quantify count nouns. I thus generalize
that in English, unit counters do not constitute an independent class, nor
are they used productively.
Next I am going to show how non-individual classifiers in English are
different from their counterparts in Chinese. Non-individual classifiers are
found in both languages, but those in English are nouns in nature.
Firstly, English classifiers, like other nominals, are marked for number
(i.e. singular/plural); whereas Chinese classifiers are always used bare and
no plural marking is available.
English makes a distinction between mass and count nouns. Count
nouns have a singular/plural distinction, and plural nouns can be marked
overtly by the plural marker -s. The singular/plural distinction is also
reflected on classifiers when they modify nouns, such as a bottle of water
versus three bottles of water in (36).
(36) a. I bought a bottle of water.
b. There stand three bottles of water on the table.
When the noun to be modified is also a count noun, both the classifier
and the noun can be marked by the plural marker. The example of (37b)
not only emphasizes the plurality of bags but also that of oranges.
(37) a. a bag of oranges
b. two bags of oranges

36

Defining classifiers

The number-marking exhibited in (36) and (37) suggest strongly that


English classifiers are nouns (e.g. borrowed from container nouns and other types of nouns), assuming that the plural marker s only attaches to
nouns.
Chinese has no plural marker like English. The classifier ping bottle
does not show the alternation between bottle and bottles, as in (38). It is
always used in bare form without any suffix. Therefore, this is not a good
test itself to to be employed to argue for or against the thesis that Mandarin
classifiers are nominals or not.
(38) a. yi
one
b. san
three

ping
CLbottle
ping
CLbottle

shui
water
shui
water

Nouns in Mandarin can be monosyllabic, disyllabic or polysyllabic.


One of the devices to derive a disyllabic noun from a monosyllabic noun is
to use the nominalization marker zi. As in (39), the morpheme ping bottle
can either be used alone as a noun (39a) or be suffixed with a nominalizer
marker zi, i.e. ping-zi (39a). The nominal nature of ping and ping-zi is
demonstrated by the fact that they can be modified by the general classifier
ge.
(39) a. yi
ge
ping
one CL
bottle
a. yi
ge
ping-zi
one CL
bottle-NMLZ
BOTH: a bottle
b. liang ge
lan
two CL
basket
b. liang ge
lan-zi
two CL
basket-NMLZ
BOTH: two baskets
Although both ping and ping-zi are container nouns, only the monosyllabic ping can be used as a container classifier felicitously, but not the
disyllabic ping-zi, as shown in (40a). Some may accept that both ping and

Classifiers in English: a contrastive look

37

pingzi can be used as classifiers, but the monosyllabic one shows a high
degree of acceptance than the dysyllabic one.7
(40) a. yi
ping(#-zi)
one CLbottle
one bottle of water
b. yi
xiang(#-zi)
one CLbox
one box of books

shui
water
shu
book

If container classifiers in Chinese are borrowed from container nouns,


as in English, then there is no explanation for the asymmetry between monosyllabic and disyllabic container nouns in (40). The majority of classifiers in Chinese are monosyllabic. One of the possible explanations is that
(container) classifiers in Chinese have a higher degree of grammaticalization than those in English and they are required to be more functional than
lexical, e.g. being phonologically light (i.e. monosyllabic) and semantically
bleached (i.e. no nominalizer).
Secondly, a wide range of adjectives can appear before English classifiers, which exhibit complex modification relations, as in (41), while it is
generally impossible to use adjectives before classifiers in Mandarin (except for dimensional adjectives such as da/xiao big/small), as in (42).
(41) a. three blue bottles of water
b. a tasteless cup of coffee
c. an expensive cup of coffee
(adapted from Jackendoff 1977)
In English, pre-classifier adjectives can modify classifiers, as in (41a),
where what is expressed is that the bottle itself is blue, not that the water is
blue. In a second case, pre-classifier adjectives cross the measure words to
modify nouns, as in (41b), where the coffee but not the cup is tasteless.
7

According to Shen (2013), when ping-zi and xiang-zi are used as classifiers, as in
our examples (40), the phrases are interpreted as a full bottle of water or a full
box of books. The numeral yi one is preferred over others, like er two or san
three. In this case, it seems that these monosyllabic classifiers with the
nominalizer -zi provide as a measure unit, like kilo, but not a counting unit. In any
case, even if these examples in (40) are acceptable for some, they have a different
reading from the counterparts in (38) with monosyllabic classifiers.

38

Defining classifiers

This is called a transparent phenomenon by Jackendoff (1977). It is also


possible that the adjective modifies the entity denoted by Cl+N. For example, in (41c), both the cup and the coffee themselves can be cheap, but
when the coffee is served in that cup (say, in a Beach Hotel), the coffee-inthe-cup can be expensive.
The Chinese counterparts of (41) are not grammatical, as shown in (42).
Chinese classifiers reject those attributive adjectives such as lan blue,
anggui expensive or wuwei tasteless, even though they are semantically
compatible with the classifier or the noun.
(42) a.* yi
one
b.* yi
one
c.* yi
one

lan
ping shui
blue CLbottle water
wuwei
bei kaifei
tasteless CLcup coffee
anggu
bei kafei
expensive CLcup coffee

The only exception is that Chinese classifiers allow a few adjectives,


mainly dimensional adjectives, to appear before them, including da/xiao
big/small, chang long or zheng whole (see Chao 1968, Tsou 1976,
Paris 1981). The examples are given in (43).
(43) a. yi
da
ping shui
one big
CLbottle water
a big bottle of water
b. yi
xiao bei
kafei
one small CLcup coffee
a small cup of coffee
c. yi
chang tiao
xianglian
one
long CL
necklace
a long necklace
d. yi
zheng bei
shui
one
whole CLcup water
a whole bottle of water
Even though a small set of adjectives can appear before classifiers in
Mandarin (43), I cannot get the possible readings that are observed in English, illustrated in (41). In Mandarin, it is neither possible to interpret (43)
with a transparent reading, where the pre-classifier adjectives modify the
noun, nor a reading where pre-classifier adjectives modify the classifier.

Classifiers in English: a contrastive look

39

Pre-classifier adjectives cannot cross the classifier to modify the noun.


For example, if yi da ping shui in (43a) had a transparent reading, it would
be expected to get something like #big water. But expressions like big
water are infelicitous, since shui water denotes homogeneous entities
and which cannot have the property of being big or small.
It is also impossible for pre-classifier adjectives to modify the classifier.
Assume that in a restaurant, a basketball player and a five year-old boy
each were served a bowl of rice (the bowls are the same size and the quantity of rice in each bowl is the same). After serving the rice, I get two different responses from them, as in (44).
(44) a. The basketball player says:
zheme yi xiao wan fan zenme gou wo chi?
such one small CLbowl rice how enough I eat
How can such a SMALL bowl of rice be enough for me?
b. The five-year old boy replies:
zheme yi
da wan fan zenme chi-de-xia?
such one big CLbowl rice how eat-up
How can I eat up such a big bowl of rice?
If pre-classifier adjectives modify classifiers giving a property of the
unit denotes by the classifier, in this case its size, then it would seem that I
had to interpret the adjective as a qualifying either the size of the bowl or
the size of the measure in each case in (44). On the strict container reading of the classifier, (44a) asserts that the bowl is big and (44b) asserts that
the bowl is small. If the classifier is taken as expressing a measure, then
(44a) asserts that the quantity of rice is big, while (44b) expresses that the
quantity of rice is small quantity. However, such intepretations are contradictory, given the fact that the size of the bowls is the same and the quantity of rice is the same. So I conclude that it is impossible that pre-classifier
adjectives do not directly modify the classifier. I will not pursue further the
exact modification relation of pre-classifier adjectives in Mandarin here, I
will return to this issue in chapter 7.
This contrast between pre-classifier adjectival modification in English
(41) and Mandarin (42) suggests that English classifiers are nouns which
can be directly modified by adjectives, whereas Mandarin classifiers are
not nouns and they reject direct modification by adjectives. Even if Mandarin classifiers allow pre-classifier adjectives (42), their interpretations are
different from those of the English counterparts.

40

Defining classifiers

A third difference between English and Mandarin is that in NP-deletion


in coordinated clauses, I can either delete CL-N or N in English, as in (45),
while in Mandarin, I can only delete N, as in (46). Note that according to
some native English speakers, (45a) sounds better than (45b).
(45) a. I bought three bottles of water, and he bought two.
b. I bought three bottles of water, and he bought two bottles.
(46) a. wo mai le san ben
shu, ta ye mai le san *(ben).
I buy PFV three CLvolume book, he also buy PFV three CLvolume
I bought three books; he also bought three.
b. wo you wu ping shui, ta zhi you liang *(ping).
I have five CLbottle water he only have two CLbottle
I have five bottles of water; he only has two (bottles).
The contrast between (45) and (46) tells us that in counting contexts,
the numerals in English can be used independently without the classifiers
or classifier + N, but in Mandarin, numeral goes hand-in-hand with classifier. Assuming that only the maximal projection of NP, i.e. phrasal category, can be elided in NP ellipsis (Lobeck 1995), then Cl-N in English can
be considered as a complex nominal constituent, but Cl-N in Chinese cannot be treated a complex noun.
It is a standard assumption by now in generative syntax on Chinese
noun phrases that a classifier heads its own projection and takes an NP
complement (Tang 1990, Cheng and Sybesma 1999, A. Li 1999, among
others). But there are two possibilities for the position of numerals, either
in the specifier position of the classifier phrase (Tang 1990), or heading a
Numeral Phrase, which takes ClP as complement (Cheng and Sybesma
1999, A. Li 1999). However, under either analysis, numeral and classifier
do not form a constituent (Saito et al 2008: 260). According to this view,
the impossibility of eliding Chinese classifiers in (29) is understandable,
since Chinese classifiers are heads and they are not phrasal.
The classifiers in English are always treated as relational nouns, i.e. a
noun which presupposes a noun that can be quantified, they project into
NP (Selkirk 1977, Abeny 1987; cf. Corver 1997: 219). For example, Abeny
(1987: 296) postulates the following structure for pseudo-partitives:
[DP a [NP bottle [PP of water]]], in which the classifier bottles has the maximal projection NP.
To sum up, the three differences I have demonstrated here are sufficient
to show that classifiers in Mandarin and English are different in terms of

Conclusions

41

their categorical features. Specifically, English classifiers share a lot of


properties with nouns: (i) being number-marked, (ii) being modified by
adjectives, and (iii) being elided like nouns. On the contrary, Mandarin
classifiers, though most of them have nominal origins, share none of these
features of English classifiers. They are better treated as a distinctive category from nouns.

7. Conclusions
In this chapter, I defined the syntactic and semantic criterion to identify
classifiers as a separate word class. In prototypical cases, the classifier
occurs adjacent to numeral or determiner, as in Numeral-Classifier-Noun
or Determiner-Classifier-Noun in Mandarin. Classifiers have the semantic function of counting or measuring entities.
I also discussed the categorical properties of classifiers in Mandarin.
They belong to a functional category: forming a closed word class, lacking
syntactic independence, lacking descriptive content, being phonologically
weak. They are distinctive from classifiers in English. Classifiers in nonclassifier languages like English are intrinsic nouns. They are derived from
sortal nouns into relational nouns.

Chapter 3
The count/mass distinction in Chinese revisited

1. Introduction
In English, table and water represent two different types of nouns, count
nouns and mass nouns respectively. The noun table can be preceded by
numeral directly, such as three tables, and it is a count noun, while water
1
cannot, such as *three water , it is a mass noun. The first question to ask
when looking at the mass/count distinction is what we mean by the notions, mass and count. Is the mass/count distinction a linguistic (stemming from grammar) or an extra-linguistic issue (e.g. a cognitive or an
ontological issue) (see Joosten 2003 for a comprehensive review, also see
Chierchia 2010:143)?
I take the position that the mass/count distinction is a grammatical distinction and it is independent of cognitive systems and reality, though they
may be correlated with each other to some extent. From a cross-linguistic
perspective, languages may differ from each other in having this distinction or not. In languages with this distinction, different grammatical mechanisms may be employed to realize this grammatical notion. According
to Chierchia (2010), there is a signature property, with which we are
able to tell whether a certain language has mass nouns or not. That is, in
general, mass nouns do not allow direct modification of numerals,
*[Num +NMASS].2According to this criterion, all nouns in Mandarin are
1

Nouns like water and beer are mass nouns, but they can sometimes be directly
modified by a numeral (with a noun that agrees in plural with the number), such as
three waters and two beers. These expressions are often used in restaurant to
represent a certain portion of water or beer which is contained in some container.
These examples will not be discussed.
2
Perhaps the most steady grammatical property associated with mass nouns in
general is the marked status of their direct combination with a numeral expression:
regardless of word order, constituents of the form] are either outright ungrammatical or are felt as requiring a reinterpretation of sorts (coercion or type-shifting,
on which more below, under the rubric elasticity).

Introduction

43

mass nouns, since, as already seen in chapter 2, Mandarin nouns cannot be


modified by numerals without a classifier (Krifka 1995, Doetjes 1997:
syntactically mass, Chierchia 1998a, Borer 2005: mass reading as default cross-linguistically).
In contrast, Cheng and Sybesma (1998) claim that there is a mass/count
distinction in Mandarin nominal phrases, but it is reflected at the level of
classifiers but not of nouns. They make a distinction between two types of
classifiers in Mandarin, count classifiers and mass classifiers (or
massifiers), which modify count nouns and mass nouns respectively.
This argument requires Cheng and Sybesma to commit themselves to the
position that there are two lexically different types of nouns in Chinese,
count and mass nouns.
Notice that researchers like Tai and Wang (1990), Tang (2005) also
make binary distinctions of classifiers (e.g. classifiers versus measures for
Tai and Wang, sortal versus mensural classifiers for Tang), but they are
silent on the question whether Chinese nouns are count or mass.
In this chapter, I examine Cheng and Sybesmas syntactic analysis of
the count/mass distinction in Mandarin.
Cheng and Sybesma (1998) propose that count and mass classifiers are
structurally different. According to them, count classifiers belong to a
functional class and mass classifiers belong to a lexical class (derived from
nouns). Count classifiers are base-generated as the head of classifier phrases, while mass classifiers start out in N and then undergo N-to-Cl movement. They suggest that the distinction between mass and count classifiers
can be diagnosed by two syntactic tests: (i) the presence of adjectives before classifiers and (ii) the optionality of the particle de after the classifier.
However, a further empirical examination of the data (examples taken
from early descriptive works done by Lu 1987 and Luo 1988) shows that
there are many counterexamples against the two syntactic diagnostics suggested by Cheng and Sybesma (1998) (see Tang 2005, X.P. Li 2007, Hsieh
2007, 2008, Li and Rothstein 2012 for relevant criticisms). According to
our observation, pre-classifier adjectives are possible both before count
classifiers and before mass classifiers, i.e. Num-Adj-Cl count/mass-N, and the
particle de is found both after count classifiers and after mass classifiers,
i.e. Num-Clcount/mass-de-N. Therefore, these two tests are unable to distinguish mass classifiers from count classifiers in a clear-cut way, as Cheng
and Sybesma suggest. I thus claim that these two diagnostics are insufficient to establish the presence of a mass/count distinction in Mandarin.
Since the distinction between mass and count nouns rests on there being a

44

The count/mass distinction in Chinese revisited

distinction between mass and count classifiers, I conclude that there is no


basis for a mass-count nominal distinction.
This rest of the chapter is structured as follows. In section 2, I will lay
out Cheng and Sybesmas (1998) syntactic proposal about the distinction
between mass and count classifiers. Section 3 and 4 re-examine the test of
the presence of pre-classifier adjectives and the test of the possibility of
post-classifier de respectively. A sample of counterexamples will be presented in each section to show that the two syntactic diagnostics are not
legitimate criteria for distinguishing between mass and count classifiers.
Section 5 briefly summarizes the arguments made in this chapter.

2. Syntactic distinction between count and mass classifiers


This section examines Cheng and Sybesmas arguments about count/mass
classifiers in Mandarin. In section 2.1, I will introduce the two lexically
different types of classifier, namely, mass classifiers and count classifiers.
Section 2.2 introduces the lexical/functional distinction of classifiers. Section 2.3 gives an overview of the two syntactic diagnostics for the distinction between count and mass classifiers.

2.1. Introduction to count/mass classifiers


At first glance, Mandarin bare nouns behave grammatically like mass
nouns in English in many respects.
First, Mandarin bare nouns and English mass nouns cannot be modified
by numerals without the assistance of classifiers. As shown in (1), the
Mandarin noun shu book occurs in the same syntactic context as that of
the English mass noun ink in that a classifier is needed between Num and
N (Note: ignore the presence of of in English, since it is required for some
independent syntactic reason, see Landman 2004).
(1)

a. san
*(ben)
shu
three
CLvolume book
three books
b. two *(drops) of ink

Second, Mandarin bare nouns are on a par with English mass nouns in
being able to act as bare arguments (Chierchia 1998b). As shown in (2a),

Syntactic distinction between count and mass classifiers

45

Mandarin bare nouns xuesheng and shu fill in the argument positions, being subject and object respectively.
(2)

a. xuesheng dou mai


shu
le.
student
all buy book PRF
Students all bought some book(s).

In contrast, in English, only mass noun but not count noun can function as
bare argument, as contrasted between water and book in (2b-c).
b. I finished drinking water.
c. * I finished reading book.
Thirdly, neither Mandarin nouns nor English mass nouns can be inflected with number morphology. Mandarin nouns have a general
number, i.e. unspecified with respect to the feature [+/-plurality] (Rullmann and You 2006). As in (3a), hua flower can either have a singular
instantiation, i.e. one single flower, or a plural instantiation of three
flowers. The English expression in (3b) is judged to be ungrammatical
(the form of waters is acceptable only with an abundant plurality (cf.
Corbett 2000)) unless it is interpreted with the universal packaging reading in the context of restaurant (see Bach 1986).
(3)

a. yi / san
duo
one/three
CL
one/three flower(s)
b. *three waters

hua
flower

In view of these similarities between Mandarin bare nouns and English


mass nouns, some linguists, notably Krifka (1995) and Chierchia (1998b),
make the explicit claim that all the Mandarin bare nouns have the semantics of mass nouns. For example, Chierchia (1998b) suggests that Mandarin
is characterized as a [+argument, -predicate] language, whose nouns are
mapped onto arguments and denote mass entities, referring to kinds.
Contra Chierchia (1998b), Cheng and Sybesma (1998) argue that there
is a count/mass distinction in Mandarin nominal phrases, but that this distinction is reflected at the level of classifiers, not of nouns. They suggest
that there are two types of classifiers in Chinese, count classifiers and
mass classifiers, which select count nouns and mass nouns respectively. Semantically, count classifiers simply name the unit in which the

46

The count/mass distinction in Chinese revisited

entity denoted by the noun naturally occurs, and mass classifiers create a
unit of measure (also see Lyons 1976: 463, Tai and Wang 1990, Croft
1994). This implies a semantic distinction between two types of nouns.
Count nouns come with a natural unit of counting, while mass nouns do not
have an intrinsic unit of counting associated with them, and they are still
mass nouns when they are modified by mass classifiers.
Nouns like child, flower and fish all denote discrete entities, which exist
in our world as naturally discrete individuals. Classifiers modifying those
nouns simply spell out the units intrinsic to them, but do not create any
external units. For example, duo bud is the natural unit for hua flower
and wei tail for yu fish. These classifiers in (4) are individual classifiers,
or count classifiers in Cheng and Sybesmas (1998) terminology.
(4)

a. san
duo
three CLbud
three flowers
b. yi
wei
one
CLtail
one fish
c. liang ge
two CL
two apples

hua
flower

[individual classifier]

yu
fish
pingguo
apple

In contrast, some classifiers express external units of entities referred to


by nouns. Consider the examples in (5).
(5)

a. yi
shu
hua
one
CLbouquet flower
one bouquet of flowers
b. liang xiang pingguo
two CLbox apple
two boxes of apples
c. yi
di
shui
one CLdrop water
one drop of water

[group classifier]

[container classifier]

[partition classifier]

The natural unit for hua flower is duo blossom, but hua can also
have an external unit such as shu bouquet in (5a), which indicates how
the discrete flowers are formed into a plural entity such as a bouquet of
flowers. The use of the container classifier xiang box in (5b) is similar to

Syntactic distinction between count and mass classifiers

47

shu bouquet in (5a). The noun shui water in (5c) denotes homogeneous
entities, which itself has no atomic structure, so the classifier di drop
imposes an external unit on it. The classifiers in (5) are non-individual
classifiers or mass classifiers, which impose external units on the stuff in
the denotation of the noun they modify.
To sum up, according to Cheng and Sybesma (1998), individual classifiers, such as duo bud, wei tail and the general classifier ge (as exemplified in (4)), are count classifiers, while container classifiers, group classifiers, partition classifiers etc. (as exemplified in (5)) are mass classifiers or
massifiers.
However, the examples in (5) are actually counterexamples against
Cheng and Sybesmas argument that count classifiers select count nouns
and mass classifiers select mass nouns, since the complement of mass classifiers like hua flower in (5a) and pingguo apple in (5b) are lexically
listed as count but not mass. This goes against the selectional constraint of
these two types of classifiers imposed by Cheng and Sybesma.
Cheng and Sybesmas mass/count distinction implies that there is a lexical distinction between two types of nouns. In other words, in Mandarin,
there are lexically marked count and mass nouns, which go with count and
mass classifiers respectively. However, Borer (2005: 99) argues that if a
noun is lexically determined to be mass or count in Mandarin, then the
distinction is rather arbitrary rather than lexico-semantically determined.
Really and truly, they are ontological count nouns and ontological mass
nouns (Borer 2005: 98). The former refers to discrete entities, i.e. the boy
type of nouns and the latter homogeneous entities, i.e. the water type of
nouns.
I argue in chapter 4 that it is more appropriate to characterize boy and
water types of nouns in Mandarin as from an ontological perspective, and
that this ontological distinction between discrete and homogeneous entities
is not directly encoded as a grammatical distinction between mass and
count nouns.

2.2. Lexical/functional distinction of classifiers


Not all Chinese classifiers are available in other languages. In nonclassifier languages like English and Dutch, it is only possible to find the
so-called mass classifiers but not the count classifiers (see Lehrer 1986 for
English data, Riemsdijk 1998 for Dutch data, cf. chapter 2: section 1.1).
According to Croft (1994: 151-152), container classifiers, partitive clas-

48

The count/mass distinction in Chinese revisited

sifiers, group classifiers and standard measures are universally available


(cf. Zhang 2012: 24). It is a controversial issue whether these classifiers are
functional or lexical. Riemsdijk (1998: 13) says that most of these classifiers in Dutch are semi-lexical, except that standard measures are functional
(also see Stavrou 2003 for a similar conclusion reached for the Greek data).
However, the classifiers in Chinese may have different lexical-functional
make-ups from the counterparts in non-classifier languages like Dutch and
Greek.
Cheng and Sybesma (1998) suggest that count classifiers and mass classifiers belong to different categories: count classifiers form a closed class
and mass classifiers belong to an open class. According to them, count
classifiers, such as the general classifier ge and specific classifiers of wei
tail, duo blossom in (4), are grammaticalized classifiers and they cannot
be used independently as nouns, whereas mass classifiers are full nominal
elements and any noun which can be seen to create a unit for measuring
can be used as a mass classifier. For example, this applies, they assume, to
the container classifiers of xiang box in (5).
Relying on the above assumption, Cheng and Sybesma (1998) argue
that count classifier phrases and mass classifier phrases have different syntactic structures, though both have the same sequence Num-Cl-N at the
surface level. In the case of count classifier phrases, the classifier is basegenerated in the head of the ClP which takes NP as its complement, as in
(6a). For mass classifier phrases, the classifier starts in N and then undergoes N-to-Cl movement, as in (6b).
(6)

a.

ClP

liang two
Cl0
duo bud

b.
Cl
NP
hua flower

ClP

san three

Cl

Cl0
N
wan bowl

NP
NP
tang soup

They also mention that the motivation for the N-to-Cl movement of
nouns lies in the fact that mass classifiers are of a different type from count
classifiers. They are measure expressions that can select another noun as
complement, and in order to realize this property, they must move from N
to Cl position. As Cheng and Sybesma (1998: 17) write: although they
(mass classifiers) are nouns, they are nouns of a certain type: they can be

Syntactic distinction between count and mass classifiers

49

used as a measure and in that capacity they select another noun. One way
to formalize this is by appealing to an idea proposed by Hoeskstra (1988),
who argues that nouns of this type have a feature, lets call it CONT of
content, and this feature enables the noun to thematically select (and
Case-mark) another noun: the content. At some stage during the derivation,
the measure noun moves to fill the head of ClP.
I have made a careful examination of the lexical/functional make-up of
classifiers in Chinese in chapter 2. I claimed that count and mass classifiers
cannot be characterized with the functional/lexical (nominal) distinction.
Firstly, this is because I showed that Chinese classifiers do not always have
a nominal origin and they come from different categories, which can be
nouns, verbs or adjectives. Secondly, I also showed that for those with a
nominal origin, both count and mass classifiers fall into different subtypes
of classifiers which have high, medium and low amounts of nominal properties. I thus argued that there doesnt seem to be a binary distinction between lexical classifiers and functional classifiers, and that distinction cannot be used to argue for a binary distinction between mass classifiers and
count classifiers.

2.3. Two syntactic diagnostics


Cheng and Sybesma (1998) propose two syntactic criteria to distinguish
mass classifiers from count classifiers: (i) the availability of adjectives
before classifiers and (ii) the possibility of the particle de after classifiers.
The first criterion Cheng and Sybesma suggest is that mass classifiers
can be preceded by adjectives like da/xiao big/small, but count classifiers
cannot (also see Chao 1968, Paris 1981 for the use of the da/xiao test).
Compare (7) with (8):
(7)

a. yi
da
zhang zhi
one
big
CLpiece paper
a big piece of paper
b. na
yi
xiao xiang shu
that one
small CLbox book
that small box of books

[mass classifiers]

50

The count/mass distinction in Chinese revisited

(8)

a.* yi
da
wei
laoshi
one big
CL
teacher
Intended: a big teacher
b.* yi da
zhi
gou
one big
CL
dog
Intended: a big dog

[count classifiers]

(7) shows that mass classifiers like zhang piece, xiang box are felicitous with pre-classifier adjectives; (8) shows that individual classifiers
like wei (an honorific classifier for people) or zhi (a classifier for animals)
cannot have pre-classifier adjectives. Cheng and Sybesma explain this difference in terms of differences in lexical properties of mass classifiers and
count classifiers. Assuming that pre-classifier adjectives have a direct modification relationship with the classifier, they argue that mass classifiers are
nouns in nature and can be modified by pre-classifier adjectives. Count
classifiers, on the other hand, are grammaticalized classifiers, and as such,
they are functional elements that cannot be modified by adjectives.
The second criterion that Cheng and Sybesma suggest is that mass classifiers can be followed by the particle de (as in 9)), but count classifiers
cannot (as in (10)) (also see Tsou 1976 for the use of the de test).
(9)

a. san
wan
de
three CLbowl DE
three bowls of soup
b. liang xiang de
two CLbox DE
two boxes of books

tang
soup

(10) a. ba
tou
(*de)
eight CLhead DE
eight heads of cows
b. jiu
gen
(*de)
nine CL
DE
nine tails

niu
cow

[mass classifiers]

shu
book

[count classifiers]

weiba
tail

The marker de in (9) is not syntactically redundant. Num-Cl-N and


Num-Cl-de-N are two distinctive structures. In order to account for their
structural difference, Cheng and Sybesma (1998) claim that Num-Cl-de-N
is not a classifier phrase but a noun phrase. They argue that Num-Cl-de-N
is a relativized noun, in which the head N is modified by Num-Cl-de,

Syntactic distinction between count and mass classifiers

51

which is treated as a relative clause and denotes the amount of entities


referred to by N. Put explicitly, Cheng and Sybesma suggest that the sequence of Num-Cl-de-N is derived from the deep structure [N[Num-Cl]],
which is argued to be a nominal small clause (NC), in which N is the subject and Num-Cl is a predicate. Note the modification marker de is argued
to be a complementizer, which heads the CP. Hence, they propose deriving
(11b) from (11a). The corresponding syntactic structures for the derivation
are given in (12).
(11) a. wo
he
le
tang san
I
drink PFV soup three
Lit.: I drank soup three bowls.
b. wo
he
le
san
wan
I
drink PFV
three CLbowl
I drank three bowls of soup.
(12) a.

wan.
CLbowl
de tang.
DE soup

NC (nominal clause)

ClP
ClP
tang soup
san three Cl
Cl

NP
N
wan

b.

ClP
CP

ClP
tang soup
C

OPi
NC
ti

C
de
ClP

sanwan proi
three bowls

NP
proi

52

The count/mass distinction in Chinese revisited

Cheng and Sybesma (1998) assume that Mandarin bare noun phrases
are structurally not bare, but are fully-fledged ClPs. So in (12a), the bare
noun tang soup is projected into a ClP and it is the subject of the nominal
clause, and the ClP san-wan three bowls is its predicate. They further
propose that the subject tang soup can undergo subject relativization, i.e.
tang soup is modified by a relative clause, as realized by Num+Cl. This
relative clause is taken as a complement by the CP, headed by de, on the
left side. In addition, there is an operator OP, which is generated in the
Spec position of CP and binds the trace left by the subject tang soup in
the relative clause. 3
However, according to Cheng and Sybesma, this subject relativization
is only possible for mass ClPs and not for count ClPs, even though count
ClPs can be used as predicates in nominal small clauses. Let us consider
the examples in (13).
(13) a. ta
mai
le
shi
zhi
he
buy
PFV ten
CL
He bought ten pens.
b. ta
mai
le
bi
shi
he
buy
PFV pen
ten
He bought pens, numbered ten.
c. * ta mai
le
shi
zhi
he
buy
PFV ten
CL
Intended: He bought ten pens.

bi.
pen
zhi.
CL
de
DE

bi.
pen

(13a) illustrates the use of the count classifier zhi in a standard Numeral
Classifier Phrase. (13b) shows that count classifiers like zhi, along with
Num shi ten, can constitute the predicate of the nominal small clause, i.e.
the count ClP shi zhi ten Cl can be predicated of the subject bi pen.
However, Cheng and Sybesma (1998) argue that the nominal small clause

Tang (1996) argues that in the construction of [N [Num-Cl]], N and the [Num-Cl]
stand in a predication relation. For her, these are not small clauses, however,
because verbs like drink in (11) are not subcategorized to take a small clause as
complement. Instead, she (1996: 471) proposes that [Num-Cl] is better treated as
the lowest argument of the verb and N as the next lowest argument, where
semantically they stand in a predication relation. She also argues that [Num-CL-N]
and [N [Num-Cl]] are two distinctive constructions and that [N [Num-Cl]] should
not be transformationally derived from [Num-CL-N].

Syntactic distinction between count and mass classifiers

53

of (13b) cannot be relativized as in (13c) and that the syntactic derivation


from (14a) to (14b) via subject relativization is impossible.
(14) a.

NC (nominal clause)
ClP
bi pen

ClP
shi ten

Cl

Cl
zhi
ClP

b.
CP

ClP
bi pen
C

OPi
NC
ti

NP
proi

C
de
ClP

shi ten
Cl
zhi

Cl
NP
proi

Cheng and Sybesma (1998: 15) assume that Chinese noun phrases, including regular overt and null pronouns, are really full-fledged ClPs.
Though generated in NP, Chinese nouns undergo N-to-Cl movement. As a
consequence, in the regular case of pronominal, the binding domain of a
pronominal is outside ClP. After relativization, as in (12b) and (14b), pro
is in a configuration that leads to a bound variable interpretation (since pro
cannot have independent reference). It is co-indexed with the subject of the
predicate, which in turn is bound by an operator. Cheng and Sybesma argue that this kind of binding is only possible for mass classifiers, as in
(12b), but not for count classifiers, as in (14b). They argue that mass classifiers undergo N-to-Cl movement and the pro in mass ClPs is a ClP, whereas count classifiers are base-generated as classifiers and they must take
NPs as complement, so pro is not a full-fledged pronoun and it stays in

54

The count/mass distinction in Chinese revisited

the NP rather than moves to Cl (Cheng and Sybesma 1998: 16). Therefore,
the pro in (14b) cannot have a proper binding domain.
It is difficult to see how this argument really works, since the pro in
4
each case is in NP position. The difference between the two structures is
that in (12a) pro is the complement of a nominal head which raises to classifier position, while in (14) it is directly the complement of the functional
head. But it is unclear why this difference has the grammatical consequences that Cheng and Sybesma propose. Presumably, if pro is a real
problem for count ClPs, then the problem also remains for mass ClPs. No
matter how mass classifiers move, e.g. N-to-Cl, the pro always stays in N
and is the complement of Cl, exactly like its counterpart in count ClPs. As
shown by their syntactic tree, the pros in both cases in (12b) and (14b) are
in NP position. There doesnt seem to be any independent evidence for the
syntactic distinction of pro as ClP and pro as NP that they propose. As a
consequence, the analysis stipulates a lot of abstract structure and special
rules for interpreting pro in each of the two structures.
Cheng and Sybesmas (1998) syntactic proposal, as presented here, can
thus be summarized as the conjunction of the following three claims:
(i) there is a nominal mass/count distinction in Chinese, as reflected in
count/mass classifiers;
(ii) count classifiers are fully grammaticalized classifiers while mass
classifiers are nouns in nature; and accordingly they have different syntactic structures;
(iii) the distinction between mass and count classifiers is supported by
the syntactic evidence of (a) the possibility of pre-classifier adjectives and (b) the optionality of the post-classifier de.
Cheng and Sybesmas (1998) mass/count proposal is based on the facts
presented in (7) to (10). These facts have been noticed earlier by other
Chinese linguists such as Chao (1968), Tsou (1976), Paris (1981), Tang
(1990) etc. They note that not all classifiers show the same degree of
grammaticality with respect to the tests of pre-classifier adjectives and of
post-classifier de. However, these earlier scholars do not use these diagnostics to make a dichotomy of classifiers as Cheng and Sybesma (1998) do,
rather they correlate different types of classifiers with the count/mass distinction.

I am grateful to Gabi Danon and Victor Pan for going thorough and discussing
with me about the syntax part of Cheng and Sybesma (1998) on several occasions.

Possibility of pre-classifier adjectival modification

55

There are two questions raised by Cheng and Sybesmas syntactic analysis of mass and count classifiers: (i) does the contrast between two different types of classifiers proposed by them really hold? (ii) does the contrast
between the two types of classifiers or nouns really reflect the mass/count
distinction?

3. Possibility of pre-classifier adjectival modification


In this section, I examine the first syntactic diagnostic used by Cheng and
Sybesma (1998), i.e. the possibility of adjectives before classifiers. Section
3.1 presents data from previous descriptive works and corpora, showing
that this diagnostic has many counterexamples. It is empirically inadequate.
Section 3.2 looks at this diagnostic from the theoretical perspective. I show
that the proposal of attributing the pre-classifier adjectival modification to
the nominal nature of mass classifiers is problematic. This is particularly
true if the conclusions reached in chapter 2 that not all mass classifiers are
derived from nouns, are taken into account that. I suggest that as a consequence, this diagnostic cannot be taken as a legitmate criterion for the distinction between count and mass classifiers.

3.1. Adjectives before count and mass classifiers


Attributive adjectives, acting as nominal modifiers, usually occur adjacent
to nouns. In English, they either directly precede or directly follow nouns,
as exemplified in (15a-b). In other words, Adj-N and N-Adj are the only
possible combinations of modified nouns, although their semantic interpretations are distinctive from each other (Bolinger 1967). It is impossible to
put adjectives before a quantified nominal phrase, as in (15c).
(15) a. an invisible star
b. a star invisible
c.* invisible a star
However, expanded nominal phrases, e.g. pseudo-partitives, illustrates
another possibility for adjectival modification. In pseudo-partitives like
Num-N1-of-N2, adjectives can either occur before N1 or N2. When adjectives appear before N1, they semantically modify N1 or N2. Examples in
(16) illustrate that an adjective preceding N1 modifies N1.

56

The count/mass distinction in Chinese revisited

(16) a. a lovely picture of his father


the picture is lovely, her father need not be
b. an expensive painting of the house
the painting is expensive, the house need not be
[Alexiadou et al 2007:398]
In examples in (17), adjectives preceding N1 can enter into a modification relation with N2 by crossing N1. This is the transparent phenomenon discussed in Jackendoff (1977).
(17) a. a useless couple of days
the days were useless
b. a marvelous glass of wine
the wine is marvelous
c. a nice box of cigars
the cigars are nice
[Jackendoff 1977, cf.Alexiadou et al 2007]
What is relevant for our discussion on classifiers is the adjectival modification in pseudo-partitive constructions, as just shown above. In what
follows, I will discuss two issues: (i) what types of classifiers can be preceded by adjectives in Mandarin? and (ii) what are the possible interpretations of adjectives in Num-Adj-Cl-N?
Recall Cheng and Sybesmas (1998) argument about pre-classifier adjectival modification. They argue that mass classifiers can be preceded by
adjectives, while count classifiers cannot. In their analysis, mass classifiers
are nouns in nature, and they can thus be directly modified by adjectives,
while count classifiers belong to a functional category, and cannot be
modified by adjectives. I am going to show below that these two observations are not accurate.
Cheng and Sybesma (1998) argue that examples like (8) (as repeated in
(18a-b)) are not acceptable. They argue that it is odd to put an adjective
like da big before a count classifier. I agree that (18a) is impossible, but I
do not think (18b) is inappropriate. If it is put in an appropriately constructed context, it becomes completely acceptable, as in (18c).
(18) a.* yi
one
b.* yi
one

da
big
da
big

wei
CL
zhi
CL

laoshi
teacher
gou
dog

Possibility of pre-classifier adjectival modification

c. na
tou
shizi bu-dao ban ge
xiaoshi
that
CL
lion
less than half CL
hour
jiu
ba
yi
da
zhi
gou
gei
chi
FOC OM
one
big
CL
dog
give
eat
That lion ate up a BIG dog within half an hour.

57

le.
PRF

There are many cases where count classifiers are felicitous with preclassifier adjectives. It has been noted in the literature that many count
classifiers go naturally with adjectives (e.g. Lu 1987, Luo 1988, and X.P.Li
2007). Lu (1987) studies the possibilities of adjectival modification of
classifiers in great detail. According to his study, 24 individual classifiers
can be preceded by an adjective. Some of Lus examples are illustrated in
(19) (see Lu 1987: 54-57 for more examples).
(19) a. Xiaoling chi le
yi da/xiao ge mantou.
Xiaoling eat PFV one big/small CL steamed bum
Xiaoling ate a BIG/SMALL steamed bum.
b. zhe yi xiao fu hua
hua le
ta bannian shijian.
this one small CL picture take PFV him half year time
It took him half a year to paint this SMALL painting.
c. ta
jian-shang
kang le
yi da/xiao
jian xingli.
he
shoulder-on
carry PFV one big/small CL baggage
He carried a BIG/SMALL baggage on his shoulder.
Ge in (19a) is a general classifier in Chinese, which modifies nouns
denoting discrete entities in general. The other two classifiers, fu in (19b)
and jian in (19c), are also typical individual classifiers, which modify hua
painting and xingli baggage respectively. As shown in (19), those individual classifiers can all be preceded by da big or xiao small.
Cheng and Sybesma (1998) are aware of some of the examples mentioned by Lu (1987), but they treat them as exceptions, as stated explicitly
in footnote 4 in their paper. However, if 24 individual classifiers out of 54
are compatible with pre-classifier adjectives, it is not appropriate to consider those examples as exceptions. More examples like those in (19) can
be freely generated. Examples in (20) are from X.P. Li (2007).
(20) a. wu-mao-qian mai
le
yi da gen huanggua.
fifty-cent
buy
PFV one big CL cucumber.
Fifty cents bought me a cucumber, which is a big quantity (w.r.t.
the money spent).

58

The count/mass distinction in Chinese revisited

b. ta chi le
yi da tiao huanggua.
he eat PFV one big CL cucumber
He ate a cucumber, which is a lot for him.
c. yi
ge yue ta juran
ba yi da ben
one
CL month he unexpectedly OM one big CLvolume
zidian
quan bei
le.
dictionary all recite PRF
Surprisingly, in only one month, he recited the whole dictionary,
which is a lot (a large vocabulary w.r.t. the short time spent).
As a supplementary study to Lus (1987), Luo (1988) argues that many
of the individual classifiers that are thought to be incompatible with preclassifier adjectives in Lu (1987) are actually acceptable, if appropriate
contexts are provided. Consider the examples in (21). The phrases of yi da
zhang chuang one big Cl bed and yi da tou yeniu one big CLhead wild
cow are listed as unacceptable in Lu (1987). But Luo (1988) illustrates the
following two contexts in which they are felicitous:
(21) a. ta wu-li
jiaju
taiduo, baifang-de ye bu jiangjiu,
he house-in furniture too:much arrange
also not carefully
yi
jinmen youbian jiu-shi yi da zhang jiushi
once enter right FOC-be one big CL
old-style
muchuang, zuobian yi
ge da-ligui,
zhuozi, yizi,
wooden bed left
one CL big:wardrobe table chair
shafa ba wuzi gao-de yongjiubukan.
sofa OM house make crowed
He has too much furniture at home. Once entering, on the right is
a BIG old-style wooden bed, and on the left is a big wardrobe, a
table, a chair and a sofa. They are packed like sardines.
b. feizhou shi
hen
xiong. Liang zhi
shizi jiu
African lion very cruel two
CL
lion
Foc
neng ba
yi
da
tou
yeniu
chiwan.
can
OM
one
big
CL
wild cow eat up
African lions are very cruel. Two lions can eat a BIG wild cow in
a short time.
Another empirical problem pertaining to Cheng and Sybesmas observation is concerned with the correlation between the count/mass distinction
and the lexical/functional distinction. According to them, the underlying
reason why massifiers can be modified by adjectives like da/xiao is be-

Possibility of pre-classifier adjectival modification

59

cause mass classifiers can be used as independent nouns. This also implies
a direct modification relation of the adjectives with the classifier.
However, as shown in chapter 2, some non-individual classifiers are derived from verbs and not from nouns. The relevant examples are repeated
in (22).
(22) a. yi
(chang) gua
bianpao
one
long hang firecrackers
a string of firecrackers
b. yi
(da)
peng
xiangtu
one big
scoop-in-both-hands soil:from:homeland
a handful of soil from homeland
c. yi
(xiao) cuo
yan
one small pinch salt
a pinch of salt
As shown in (22), it is completely possible to insert an adjective before
the classifier, such as chang long, da big, xiao small. If the presence
of adjectives before classifiers is really due to the nominal feature of classifiers, these examples are left unexplained.
Additionally, as shown in (19), some of the individual classifiers that
are subject to adjective modification have low degree nominal properties
and they cannot be used as nouns independently, such as the general classifier ge in (19a), (recall the relevant argument in chapter 2). Cheng and
Sybesmas account cannot explain these examples either.
The examples from (19) to (22) are obvious counterexamples to the generalization made by Cheng and Sybesma (1998) that pre-classifier adjectives are only allowed before mass classifiers. Thus, our empirical examination of the data suggests that adjectives like da/xiao big/small are not
only possible before mass classifiers but also before count classifiers. In
addition, I also showed that the nominal property of classifiers may not be
the decisive factor to license adjectives before classifiers.

3.2. Two constraints on pre-classifier adjectives


Pre-classifier adjectival modification is conditioned by the following two
constraints (Paris 1981, Lu 1987, Tang 1990, Cheng and Sybesma 1998).
The first constraint is that only a small number of adjectives, such as

60

The count/mass distinction in Chinese revisited

da/xiao big/small, can be inserted in pre-classifier position. The second is


that the element de, which typically links the modifier and the modifiee,
cannot follow pre-classifier adjectives.
Let us look at the first constraint. The fact that only a small subset of
adjectives can precede classifiers weakens Cheng and Sybesmas claim
that mass classifiers can be preceded by adjectives because they are nominal. If this claim were correct, then I would expect to get a wider range
of adjectives. However, in Chinese, pre-classifier adjectives are restricted
to dimensional adjectives, such as da big, xiao small, chang long etc.
This is different from English. In English, in psuedopartitive constructions
(as in (17)), N1 can be preceded by different types of attributive adjectives,
such as useless, tasteless, expensive etc (Jackendoff 1977).
With regard to this problem, Cheng and Sybesma suggest that the adjectival modification of the classifier is allowed only in its abstract function
as a container. They claim that, in He carted four wheelbarrows of dung
into the fields (from to Hoeskstra 1988), it is possible to modify the noun
wheelbarrow with adjectives like big but not with modifiers like wooden,
because (they claim) big modifies wheelbarrow in its abstract function as a
container, while wooden would modify the concrete thing.
The examples of mass classifiers that Cheng and Sybesma discuss and
analyze are restricted to container classifiers. However, there are other
types of massifiers, including group classifiers, e.g. qun group, dui pile,
and partition classifiers, e.g. duan section, pian piece etc. These three
types of massifiers show different lexical or grammatical properties and it
is not straightforward to extend Cheng and Sybesmas analysis of container
classifiers to them. However, there are some flaws with this claim.
In Cheng and Sybemsas (1998) paper, there are two examples which
are not container mass classifiers, as illustrated in (23):
(23) a. yi
da
zhang
one
big
CLpiece
a big piece of paper
b. yi
da
qun
one
big
CLgroup
a big group of people

zhi
paper
ren
people

The two mass classifiers in (23), zhang piece and qun group, are not
like container classifiers, such as ping bottle and wan bowl, in several
important aspects. One of the important aspects is that mass classifiers like
qun group and zhang piece do not involve the feature of CONTENT.

Possibility of pre-classifier adjectival modification

61

From the syntactic perspective, if I follow Cheng and Sybesmas reasoning,


they would not undergo N-to-Cl movement and be raised to Cl0 in order to
express the thematic selection of another noun which expresses the content. In contrast, they are expected to be base-generated under the node of
Cl0, and hence they are expected to pattern with count classifiers. With this,
it would be expected that they cannot have pre-classifier adjectives, and
this, of course, is not the case.
Semantically, these classifiers like zhang piece and qun group do not
denote container, as container classifiers do. Cheng and Sybesmas argument that pre-classifier adjectives in Chinese modify abstract function
as a container is too narrow to cover these two subtypes of mass classifier.
Thus, Cheng and Sybesmas syntactic analysis may correctly capture
some properties of container classifiers, but it seems difficult to extend it
to the more general category of mass classifiers, which includes group
classifiers and partition classifiers etc.
The second constraint is that pre-classifier adjectives do not take the
modification marker de. The marker de can optionally appear after the
adnominal modifier piaoliang pretty, e.g. yi ge piaoliang (de) nhai a
beautiful (de) girl, but it is disallowed after a pre-classifier adjective da
big, e.g. yi da (*de) ping jiu a big bottle of wine. Cheng and Sybesma
(1998) argue that N-to-Cl movement is an instance of head movement in
which the combination of the adjective and the N forms a complex head.
As a consequence, this structure does not allow the insertion of the modification marker de. This explanation is due to Tang (1990: lexical integrity). Given that Cheng and Sybesma (1998) argue that mass classifiers are
nouns in nature and pre-classifier adjectives have a direct modification
relation with the classifiers, the complex head Adj+Cl should be seen as
a complex noun head.
However, the proposal of lexical integrity is unsatisfied for two reasons.
First, it cannot explain why this kind of lexicalization happens only with
adjectives like big/small. Secondly, it is unable to account for the possibly
unified meaning of Adj-Cl, such as the evaluative meaning, to be proposed in chapter 9.
In view of the empirical and theoretical problems discussed in sections
3.1 and 3.2, the presence of adjectives before classifiers cannot be considered as a legitimate criterion to distinguish count and mass classifiers.

62

The count/mass distinction in Chinese revisited

4. Optionality of post-classifier de
This section examines Cheng and Sybesmas second criterion for the distinction between count and mass classifiers, namely, the possibility of postclassifier de. I will investigate the question: under what condition are we
able to build up the phrase Num-Cl-de-N? Presumably, this not only has
something to do with the lexical meaning of classifiers but also with the
proper interpretation induced by the insertion of de.
Before starting, I want to clarify that the particle de after a classifier is
not semantically vacuous or syntactically redundant. It is commonly agreed
that the particle de is a modification marker which is able to turn different
syntactic elements into modifiers (Zhu 1961, Paris 1979, Li and Thompson
1981, among others). Namely, [XP [de]] has the function of modifying the
noun it is a sister of.
(24) a. congming de haizi
clever
DE childred
clever children
b. wo mai de shu
I buy DE book
the book(s) that I bought
c. ta
de
shu
he/she DE
book
his/her book(s)

[Adj-DE]

[RC-DE]

[Pron-DE]

In addition to the use of de in the contexts in (24), it can also follow


classifiers, as in (25).
(25) san
xiang de
three
CLbox DE
a. three-box books
b. three boxes of books

shu
book

The example (25) has two readings: an attributive reading and a pseudopartitive reading (Jiang 2008). On the attributive reading, the phrase in (25)
refers to a particular collection of books, e.g. the books that are packed in
three boxes (to be distinguished from other collections, e.g. those packed in
five boxes) and this particular property is expressed by Num-Cl-de. On the
pseudo-partitive reading, the phrase simply has a quantity reading, meaning
three boxes of books and Num-Cl-de is quantity-denoting.

Optionality of post-classifier de

63

I only concentrate on the licensing condition of de after classifiers on


the pseudo-partitive reading and leave aside the attributive reading. It is not
immediately clear whether or not the particle de is still a modification
marker on this reading. I will discuss this issue in Chapter 8.
Cheng and Sybesma (1998) argue that mass classifiers can be followed
by the particle de, but count classifiers cannot. This observation is not accurate.
Firstly, not all mass classifiers goes with de. It is easy to find mass classifiers that cannot go with the marker de, especially partition classifiers (26)
and group classifiers (27).
(26) a. yi
kuai
(*de)
one
CLpiece DE
a piece of meat
b. yi
duan (*de)
one
CLsection DE
a section of distance

rou
meat

(27) a. yi
qun
(*de)
one
Clgroup
DE
a group of students
b. yi
chuan (*de)
one
CLstring DE
a string of chilis
c. yi
dui
(*de)
one
CLpile
DE
a pile of books

xuesheng
student

juli
distance

lajiao
chili
shu
book

Further, researchers such as Tang (2005) and Hsieh (2007, 2008), point
out that it is possible for individual classifiers to be followed with de in
some cases. Examples in (28) are taken from Tang (2005), which are coun5
terexamples against Cheng and Sybesma (1998).
5

Tang (2005) attempted to take the examples in (i) and (ii) as counterexamples
against Cheng and Sybesma (1998). She argued that the particle de not only follows
mass classifiers (i), but also count classifiers in examples (ii):
(i) a. liang
mi
de
bu
[mass classifiers]
two
meter DE
cloth
cloth that is sorted in accordance with two meters

64

The count/mass distinction in Chinese revisited

(28) a. mei chao yue sheng-xia


ershi-si
mei de luan.
every nest about give:birth:down twenty-four CL DE egg
Every nest has about twenty-four eggs.
b. yi
nian yue
zhongzhi le
one
year about plant
PFV
yi-bai-sishi-duo-wan
ke
de shumu.
one-hundred-forty-more-million CL DE tree
(They) planted more than 1,4million trees a year.
In the examples in (28), Num+Cl expresses overall quantities of entities: (28a) means that the total number of eggs (per nest) is 24, while (28b)
means that the number of individual trees planted is more than 1,400,000.
(Note that while Tang accepts (28a), other native Mandarin speakers I consulted do not.) Tang suggests that the insertion of post-classifier de in the
examples of (28) is related to the information weight of the modifiermodifiee relation: she assumes that if the Num-Cl is a heavy and complex
constituent, it is possible for de to follow the classifier.
Following Tangs observation about (28), Hsieh (2007, 2008) presents
some more counterexamples (from the Academia Sinica Corpus). According to Hsieh, de can be licensed in this position if the Num-count Cl expresses non-fixed quantities or if a massifier is used. Note that her exam-

b. liang
bang
de
rou
two
pound DE
meat
meat that is sorted in accordance with two pounds
(ii) a. liang
ben
de
shu
[count classifier]
book
two
CLvolume DE
books that are sorted in accordance with two in number
b. wu
ge
de
pingguo
five
CL
DE
apple
apples that are sorted in accordance with five in number
Although Tangs examples in (i) and (ii) bear great similarity to those in Cheng
and Sybesma (1998), they are characterized with two different phenomena. For
Cheng and Sybesma (1998), Num-Cl-de still expresses quantity information about
the entities denoted by N, while for Tang (2005), Num-Cl-de is like an attributive
modifier, which behaves like a classifying adjective.
They are also structurally different. For Cheng and Sybesma (1998), Num-Clde-N is a classifier phrase, as headed by the classifier, while for Tang (2005), NumCl-de-N is a complex noun phrase andNum-Cl-de is projected as an adjunct to the
NP. I will make a detailed discussion Tangs argument about de in chapter 8.

Optionality of post-classifier de

65

ples of count ClPs have the adverbial jin close to before the classifier
phrase.
(29) a. jin
yi-bai
wei de
qianjiu
close one-hundred CL DE rescue
close to one hundred rescue workers
b. hao
ji
bai
tiao de
quite several hundred CL
DE
quite several hundreds of sea snakes

renyuan
worker
haishe
sea snake

More counterexamples are given by X.P. Li (2007), as illustrated in


(30). He argues that de can only be found in a context of quantification of
aboutness, not exactness, as evidenced by yue about or duo more.
Hence Cl+ duo expresses approximative values.
(30) a. ta
peng-zhe
shi
duo
ben
de
shu.
he
carry-PROG ten
more CL
DE
book
He was carrying more than 10 books.
b. ta yilian
xie le
liang-bai
duo feng de xin.
she continuously write PFV two-hundred more CL DE letter.
She continuously wrote letters, more than 200 letters.
Li and Rothstein (2010, 2012) observe independently that if the numeral
is a (high) round number (31) or a fraction, such as , , etc (32), de
can also follow a count classifier (also see Her and Hsieh 2010 for a similar observation). In these two cases, adverbials like yue about or duo
more are not needed.
(31) a. cunchu-ka
man le, wo pai
le
memory card full PRF, I
take
PFV
you
yi-qian zhang de
zhaopian.
have 1000 CL
DE
photo
The memory card is full. I took one thousand photos.
b. zhe-ci
women qing le yibai wei de tuixiu jiaoshi.
this time we
invite PFV 100 CL DE retired teacher
This time, we invited 100 retired teachers.

66

The count/mass distinction in Chinese revisited

(32) a. yi
ge
yi
sui
de
yiner mei-ci
one
CL
one
year DE
baby every time
zhi
neng chi
1/3
li
de
ganmao-yao.
only
can
eat
1/3
CL
DE
cold pill
A one year old baby can only take 1/3 of a cold pill every time.
b. mei
wan
tang pingjun you 1/4
ge de yancong.
each CLbowl soup equally have 1/4
CL DE onion
There is of an onion in each bowl of soup on average.
The problems exposed above indicate that post-classifier de cross-cuts
the distinction between count and mass classifiers and that the presence of
the marker de after classifiers is not a legitimate criterion to tell mass classifiers apart from count classifiers.
If we investigate precisely, we will find that the two tests do not give
consistent result. Recall the examples between (23) and (26-27). Classifiers
like qun group and kuai piece can be preceded by adjectives but they
cannot be followed by the marker de.
Some may wonder whether it is possible to impose an even stricter condition using these two criteria that these two criteria have to be satisfied at
the same time, namely, co-occurrence of pre-classifier adjective and postclassifier de. However, the combined use of the two diagnostics does not
give us a satisfactory result either. It is possible for container classifiers to
have pre-classifier adjectives and post-classifier de, as in (33), but it is not
plausible for mass Cls like qun group and duan section, as in (34).
(33) a. yi
xiao xiang de
one
small CLbox DE
a small box of books
b. yi
da
ping de
one
big
CLbottle DE
a big bottle of water
(34) a.* yi
one
b.*yi
one

da
big
xiao
small

qun
de
CLgroup DE
duan
de
CLsection DE

shu
book
shui
water

xuesheng
student
shengzi
rope

In summary, both tests of pre-classifier adjectives and post-classifier de


face counterexamples and neither provides a clear-cut distinction between
mass classifiers and count classifiers. The tests, hence, do not establish

Conclusions

67

Cheng and Sybesmas (1998) claim that Chinese shows a mass/count distinction.
The exact semantics of pre-classifier adjectives and post-classifier de
will be discussed in depth in chapter 7 and 8.

5. Conclusions
In this chapter, I have examined Cheng and Sybesmas (1998) arguments
about the distinction between count and mass classifiers. I presented some
new data to test the two syntactic diagnostics for the distinction between
count and mass classifiers. This data shows that pre-classifier adjectives
are available for both count and mass classifiers, and so is post-classifier
de. Therefore, I suggest that these two syntactic diagnostics cannot be used
as legitimate criteria for making distinction between two types of classifiers, as proposed by Cheng and Sybesma (1998).

Chapter 4
Natural atomicity

1. Introduction
The last chapter examined the syntactic approach to the mass/count debate
in Chinese. I showed that no legitimate syntactic criterion is available to
draw a distinction between count and mass classifiers in Chinese. The main
task of this chapter is to examine the lexicalist approach to the mass/count
thesis. It is notably represented by Doetjes (1997) and Cheng et al (2008).
While admiting that Mandarin nouns are syntactically mass, Doetjes (1997)
claims that a semantic distinction between count mass nouns and mass
mass nouns can still be made. Count mass nouns refer to entities with
minimal parts and the referents of mass mass nouns have no minimal
parts. A recent argument in support of this view is that count nouns come
out of lexical as count, and there is no universal grinding in Chinese
(Cheng et al 2008).
The core question to be addressed in this chapter is how we should characterize the contrast between shui water and nanhai boy in Mandarin.
Is it appropriate to correlate it with a mass/count distinction at the lexical
level, as suggested in Doetjes (1997) and Cheng et al (2008)?
(1)

a. san *(ge) nanhai


three
CL
boy
three boys
b. liang *(di)
shui
two
CLdrop water
two drops of water

With respect to this question, I make two main claims. First, it is more
appropriate to recast the contrast between mass mass nouns and count mass
nouns as an ontological distinction between discreteness and homogeneity.
Second, there is no grinding (mass-count shifting) or count-mass shifting in
Chinese. Chinese nouns have a genuine ambiguity between object reading
and stuff reading, or even a partial object reading in some occasions.

Countability and individuation

69

This chapter is planned as follows. In section 2, I define the mass/count


distinction from a grammatical perspective and separate the notion of countability from individuation. Section 3 gives an overview over Doetjess
lexical account about the mass/count distinction in Chinese. In section 4, I
argue that Chinese bare nouns have a genuine ambiguity between individual and stuff readings, which is to be distinguished from the mass/count
distinction. In section 5, I suggest that the ontological distinction between
homogeneity and discreteness is a grammatically relevant phenomenon in
Chinese. Section 6 concludes this chapter.

2. Countability and individuation


It is much easier to judge whether some languages have a mass/count distinction than other languages. For example, it is never an issue with respect
to this question in number-marking languages, such as English and French,
but it is less transparent in bare-argument languages like Mandarin, Halkomelem Salish (Wiltschko 2012), Dne Sulin (Wilhelm 2008), Yudja
(Lima 2012), and Brazilian Portuguese (Mller and Oliveira 2004, de Oli1
veira and Rothstein 2011) etc. This is partially because in bare argument
languages, there is no plural marker available to mark count nouns if they
exist at all. Additionally, the long-standing difficulty of defining the
mass/count distinction in these languages is reflected by the term itself.
The mass-count distinction is a misleading term:
It incautiously takes together a primarily grammatical criterion (the (non-)
countability of nouns) with a non-grammatical, ontological criterion (the
denotation of mass vs. discrete entities). (Joosten 2003: 216)

This boils down to the question what criterion we should adopt to judge
whether a language has a mass/count distinction or not. The term countmass distinction is interwoven with two different notions, countability
and individuation. Countability is a grammatical notion. It refers to
whether nouns can be directly counted by numerals or not. Individuation is
a cognitive or ontological notion and it refers to whether the referents de1

Brazilian Portuguese has a count/mass distinction and its count nouns can be
pluralized with plural markers. However, what is controversial is the grammatical
status of bare singular nouns in this language. According to de Oliveira and Rothstein (2011), bare singular nouns are mass nouns.

70

Natural atomicity

noted by nouns are (discrete) individuals or not. These two notions should
be treated separately and they cannot be applied at the same time to judge
whether a language has a mass/count distinction.
The grammatical criterion of countability is independent of the nongrammatical, ontological criterion for individuation, though they are
somehow correlated with each other (Krifka 2008, Rothstein 2010, also see
Bunt 1985).
As Krifka (2008) points out, in some cases, the semantic criterion is
consistent with the cognitive criterion when judging the mass/count distinction. Consistency is found in the following three scenarios. (a) Liquids
and substances are lacking a defined boundary and they are mass, such as
milk, water and gold. (b) Small objects tend to be mass, because their defined boundary is usually irrelevant, as contrasted between rice and beans,
between sand and pebbles. (c) Entities high on the animacy scale tend to be
count (cf. Smith-Stark 1974). However, in other cases, semantic and cognitive criteria might diverge. There are mass nouns that denote entities one
would expect to be denoted by count nouns: jewelry, silverware, furniture,
drapery. These nouns are called fake mass nouns by Cherchia (2010).
Rothstein (2010) argues explicitly that the mass/count distinction is not
based on an atomic/non-atomic nor on a homogeneous/non-homogeneous
distinction. The mismatch between form and denotation is two-way: as
well as mass terms such as furniture and jewellery that denote sets of inherently individuable objects, there are also count terms that denote sets of
entities that do not have spatial properties constant across time. Examples
are fence, wall and bouquet (ibid: 346).
Wiltschco (2012) argues that there are two sources of mass/count distinction: it can either be seen as a property of grammatical structure (fproperty) on the one hand, or as an ontological/perceptual property on the
other. These two properties are realized at two distinctive syntactic levels.
The ontological property is realized as the root level, while the f-property
is categorical and is suggested to be a nominal inner aspect: the locus of
grammaticalized mass/count distinction.
(2)

D
D

F
F

f-properties: categorical
nominal root R-properties: ontological

Countability and individuation

71

In this research, we take the position that the mass/count distinction is a


grammatical phenomenon. When judging whether a language has count
nouns or not, it should be based on the grammatical notion of countability
but not the ontological notion of individuation. According to Chierchia
(2010), the impossibility of directly combining a numeral with a mass noun
is universal. This is a signature property for mass nouns. In other words, to
define a count noun, it must be able to be directly combined with numerals,
and mass nouns reject number modification.
At least three types of languages can be distinguished according to the
signature property: the mass/count language, the count language and the
mass language.
Type 1: Mass/count languages
(a) Germanic and Romance languages are representative mass/count
languages, whose count nouns are subject to number-marking. As in (3),
English count nouns can be pluralized with the marker -s, while its mass
nouns cannot.
(3)

a. three students
b. * three waters

[Count]
[Mass]

(b) Dne Sulin is a bare argument language. There is no numbermarking strategy available to distinguish different types of nouns, but the
mass/count distinction can still be made. According to Wilhelm (2008), the
mass/count distinction in Dne Sulin is dependent on the natural atomicity of nouns. Nouns like table in (4a) denote discrete entities and they can
be directly modified by numeral. They are count nouns. In contrast, nouns
like milk in (4b) denote homogeneous entities and they cannot be modified
by numerals. They are categorized as mass.
(4)

a. ighe/solghe bekeshchelyi
one / five
table
one table/ five tables
b. #ighe/solghe ejretth
one / five
milk

[Count]

[Mass]

Type 2: Count languages


In Yudja, bare nouns can be used as arguments, and there is neither a
general plural marker nor classifiers. In this language, all the nouns can be
directly modified by numerals. In other words, all the nouns have the count
syntax (Lima 2012). Nouns like am monkey in (5a) and ya water in

72

Natural atomicity

(5b) are used in the same syntactic context. Interestingly, (5b) can either
mean three glasses of water or three drops of water, or three puddles of
water etc. The exact counting unit is context-dependent.
(5)

a. txabu am
three
monkey
b. txabu ya
three
water

Type 3: Mass languages


Mandarin represents a different type of bare argument language, in
which there is no plural marker but there is a general classifier system.
Nouns can never be modified by numerals unless a classifier is used between numeral and noun. For example, the omission of the general classifier ge in (6a) led to the ungrammaticality of the phrase. Therefore, all the
nouns in Mandarin behave like mass nouns. It is a mass language.
(6)

a. san
*( ge) beizi
three
CL cup
three cups
b. yi
*(bei) jiu
one
CLglass wine
a glass of wine

I make three generalizations concerning the possibility of having a


mass/count distinction in a certain language. First, number marking is not a
reliable test to tell whether a language has a mass/count distinction or not.
Relying on the data from English and Dne Sulin, we generalize that in
number-marking languages, there is always a mass/count distinction, but
not vice versa.
Second, bare argument languages are not always deprived of a
mass/count distinction. Bare argument languages can be a count language,
such as Yudja, a mass language, such as Mandarin, or a mass/count language, such as Dne Sulin.
Third, bare-argument languages are not bound to have classifiers, such
as Yudja. If we assume that classifiers are to provide a relevant unit for
counting or measuring, then the difference between Mandarin and Yudja
boils down to whether the unit needs to be expressed overtly or can be
inferred from the contexts only.

A lexical distinction between mass and count nouns

73

3. A lexical distinction between mass and count nouns


In this section, I will review Doetjess (1997) lexicalist approach to the
mass/count distinction in Chinese. Doetjes claims that there is a lexical
distinction between count and mass nouns in Chinese, which hinges on the
denotation of nouns, that is, whether the referents of nouns come with a
count structure (minimal parts) or not. I am going to show that this claim is
not tenable from both the theoretical and the empirical perspectives.
I argued in section 2 that Chinese is a mass language, in which nouns
cannot be modified by numerals in a direct way, unless a classifier is used.
In other words, Chinese nouns have a mass noun syntax. However, Doetjes
(1997) argues that the syntactic mass nature of nouns does not conceal the
semantic distinction between mass and count nouns. She argues that two
semantically different types of nouns can be distinguished in Chinese,
namely, nouns that provide us with minimal parts and those which do not
have minimal parts. The former are called count mass nouns and the latter mass mass nouns. Such a distinction implies that there is a lexical
distinction between mass and count nouns in Chinese.
According to Doetjes (1997:31), nouns like haizi child provide us
with the information about what unit can be considred as a single child.
Therefore, it is a count mass noun, which comes with a count structure
(with minimal parts). In contrast, nouns like shui water have no minimal
parts in the denotation, so that we are unable to directly count this noun
by saying *one water. It is thus treated to be a mass mass noun.
Doetjes suggests that the compatibility of a noun with the general classifier ge be one of the diagnostics for the presence of count structure. As
she (1997: 34) says, if a noun can be combined with ge, it must have a
count structure For example, in (7), a count mass noun like haizi child
is compatible with the classifier ge, as contrasted with the mass mass noun
shui water.
(7)

a. yi
ge
one
CL
one child
b.* yi
ge
one CL

haizi
child
shui
water

Doetjes (1997) assumes that count nouns always have a counting structure and that individual classifier + N in Chinese is equivalent to English
count nouns, both of which imply the existence of minimal parts in the

74

Natural atomicity

denotation of nouns. In English, a count noun can be preceded directly by


an indefinite article. The indefinite article a/an has no concrete lexical
meaning, so the mininal parts must come from the count noun itself. However, in Chinese, nouns are either modified by specific classifiers or the
general classifier ge. According to Doetjes (1997: 33), only those specific
classifiers contain information of how the partition of entities is made. For
example, the classifier ben is a counting unit exclusively for books, as in
the phrase yi ben shu one volume book; it signals that we are talking
about book-volumes. However, the general classifier ge does not contain
any information about what counts as a unit and its use depends on the
presence of minimal parts in the denotation of the noun it combines with.
The general classifier ge behaves like the indefinite article in English in
this regard.
Doetjes (1997) suggests that the compatibility of group classifiers, such
as da dozen and qun crowd, flock, be another piece of evidence to show
that there are count nouns in Chinese. She (1997: 34) says that these classifiers are semantically used for a group or a collection of individuals.
The examples in (8) are taken from Doetjes (with slight modification).
(8)

a. yi
da
baima
one
CLdozen white horse
a dozen of white horse
b. yi
qun
baima
one
CLgroup white horse
a group of white horse

Doetjes says that group classifiers da CLdozen and qun CLflock are
similar to plural morphology in the sense that they indicate that there is a
plurality of individuals. However, neither the plural marker in English nor
the group classifier in Chinese indicates the minimal parts in the denotaiton
of nouns. This information is provided by the noun. In other words, in (8),
the noun ma itself provides us with a criterion for counting. The lexical
meaning of ma decides what counts an individual horse.
Doetjes lexicalist approach to the mass/count distinction in Chinese is
incorrect in several regards. First of all, Doetjes attempted to make an
analogy between group classifiers in Chinese and the English plural marker
s, and between the general classifier ge in Mandarin and the indefinite
article in English, in terms of their semantic function. Unfortunately, the
basic assumption Doetjes (1997) adopts about English count nouns is not
correct: that count nouns are lexically count and the plural marker and the
indefinite article simply mark but do not make the countability.

A lexical distinction between mass and count nouns

75

The plural ending in [the books] indicates that there is more than one book.
It does not give information about what unit could be considered to be a
singular book, and therefore we know that this information must somehow
be present in the denotation of the count noun [book] Similarly, when
we use one N or another N, we know what unit is intended. the information about what counts as a unit must be present in the singular noun.
(Doetjes 1997: 32)

The incorrectness of this assumption is two-fold. First, count nouns are


unable to mark countability. This is because count nouns do not always
refer to entities with minimal parts. Count nouns are not dependent on the
ontological property of the referents in the denotation of nouns (Krifka
2008, Rothstein 2010). Second, if a noun is lexically determined to be
mass or count in Chinese, then the distinction is rather arbitrary than lexico-semantically determined (Borer 2005).2
2

Borer (2005) assumes that it is crosslinguistically true that all noun extensions are
mass. She proposes that both mass and count represent properties of functional
structures (or the absence thereof). Specifically, count is crucially a grammatically
construed notion, corresponding to a piece of structure and mass nouns is absent
of such a structure (Borer 2005: 88).
Like Doetjes (1997), Borer (2005) proposes that number marking is a sort of
classifier, but they are able to create partition or division of stuff, as opposed to
Doetjess view of marking countability.
The exact differences between Borer (2005) and Doetjes are discussed in footnote 7 in Borer (2005: 93-94).
First, for Doetjes, the mass/count distinction is a lexical distinction, namely,
nouns are lexically marked as count or mass (specifically count mass,and mass
mass), and the grammatical marking of divisions, through either classifiers or plural
marking, is a form of agreement with such lexical marking. It does not, in and of
itself, determine the mass-count distinction. In the account proposed by Borer, both
plurals and classifiers create, so to speak, count nouns from unstructured stuff. It is
a syntactic operation to realize mass or count nouns.
Secondly, for Doetjes, the similarity of function between plurals and classifiers
is not structurally reflected. Plural marking remains number marking (rather than
classifier marking), while classifiers project as classifiers. In Borer (2005), it is
specifically argued that plurality is not a number specification, and that plurals are,
morphologically and otherwise, classifiers.
Finally, Doetjes assumes that both classifiers and plural morphemes mark sets
of semantically pre-existing minimal parts, while Borer argues that they create
divisions of stuff, which may be singular or plural. In fact, for Borer, singulars are
created by the counting function, that is, by the Quantity Phrase.

76

Natural atomicity

Empirically speaking, the two diagnostics Doetjes used cannot be justified as legitimate criteria for the mass/count distinction.
Concerning the first diagnostic, it is not accurate that the general classifier ge only goes with nouns with inherent minimal parts. Ge cannot only
modify nouns denoting discrete entities, but also modify abstract nouns,
such as xinqing mood in (9a) or nouns denoting entities without minimal
parts, such as pi fart in (9b). Note that these classifier phrases do not
express individual events, as construed by the contexts in (9).
(9)

a. zhu
ni
you
yi
ge
wish you
have one
CL
Wish you have a good mood.
b. ta yi jie ke
wen
le
he one CL class smell PFV
He smelled three farts during a class.

hao
good

xinqing.
mood

san
three

ge
CL

pi.
fart

Secondly, it is true that group classifier phrases express aggregates of


entities. However, this kind of aggregation is not necessarily related to
count nouns. There are both aggregates named by count nouns and aggregates by mass nouns (Wierzbicka 1988, cf. Wisniewski 2010). As shown in
(10), English nouns like rice and furniture are mass nouns but they can be
modified by group classifiers, such as handful or pile.
(10) a. a handful of rice
b. a pile of old furniture
The problem is shared in Chinese. Nouns like jiuzai-wuzhi relief supplies and qian money are mass-like nouns in Mandarin. They cannot be
modified by any individual classifiers, including the general classifier ge,
as shown in (11a) and (12a). Usually, they are quantified by measure
phrases like dun ton (11b) or quantifiers like yixie some and xuduo
much (12b).
(11) a.# liang ge
jiuzai-wuzhi
two Cl
relief supplies
b. yi
dun
jiuzai-wuzhi
one
ton
relief supplies
a ton of relief supplies

Individual, stuff and partial-object readings

77

(12) a.# san


ge
qian
three CL
money
b. yixie /xuduo
qian
some/a lot
money
some/ much money
As in (13), these two nouns, i.e. qian money or wuzhi supplies, can
be taken as complement to group classifiers, though they are mass.
(13) a. yi
ba
qian
one
CLhandful money
a handful of money (coins, notes)
b. yi
pi
jiuzai-wuzhi
one
CLbatch relief supplies
a batch of relief supplies
To sum up, it is wrong to assume that count nouns have minimal parts
and mass nouns have no minimal parts. As argued in section 2, the
mass/count distinction is not based on the ontological distinction, e.g. with
or without minimal parts in the denotation. Moreover, neither of the diagnostics provided by Doetjes can be justified to show that the noun that the
general classifier ge or the group classifier modifies has minimal parts.
They cannot be used to support the view that there are count nouns in Chinese at the syntactic level.

4. Individual, stuff and partial-object readings


According to Cheng et al (2008), in all languages one can find nouns with
mass denotations and nouns with count denotations at the lexical level.
Chinese is no exception to this generalization. There are lexical count
nouns and lexical mass nouns in Chinese.
In English, a count noun can be shifted into mass by depriving its count
syntax. For example, the noun apple in the sentence there is apple all over
the floor is used as arugment in its bare form. It has a mass noun syntax
and it is interpreted as apple stuff but not as individual apples. The
sentence means that there is apple stuff all over the floor. In other words,
the count noun is grouned into mass by a universal grinder (Pelletier
1975). In English, the count-to-mass shifting is triggered by the morphsyntax mismatch.

78

Natural atomicity

In contrast, the count-to-mass shifting in Chinese is not as easily


achievable as in English. Cheng et al (2008) report that the Mandarin sentence (14) only has a count noun interpretation. The noun gou dog refers
to individual dogs only but not dog stuff. And (14) is only able to describe
a situation in which a wall has been docorated with numerous little dogs.
They call it a wall-paper reading. It is impossible for gou to have a
shifted reading by the mechanism of universal grinding.
(14) qiang
shang dou
shi
gou.
wall
on
all
be
dog
WALL PAPER READING: There are dogs all over the wall.
NOT: There is dog all over the wall.
Example (15) is different from (14) in that it not only has a wall paper
reading (15a) but also a mass noun reading (15b). Among these two readings, the count reading is a basic one, from which the mass reading is derived. In Chinese, the count-to-mass shifting is not due to morpho-syntactic
factors but is context-driven, e.g. due to world knowledge factors.
(15) qiang
shang dou
shi
pingguo.
wall
on
all
be
apple
a. WALL PAPER READING: There are apples all over the wall.
b. MASS READING: There is apple all over the wall.
In view of the examples in (14) and (15), Cheng et al (2008: 54) propose that in Chinese, count nouns come out of the lexicon as count, and a
mass interpretation of a count noun is the result of coercion.
If my argument is correct that Mandarin nouns have a mass noun syntax,
then there is no syntactic-driven count-mass or mass-count shifting in general. The Mandarin bare noun pingguo can be equally interpreted with a
count reading or a mass reading in the appropriately construed contexts. I
wonder why pingguo in Chinese should be treated as a count noun (by
default). Is it possible that the stuff reading is (one of) its basic reading(s)?
Contra Cheng et al (2008), I argue that the so-called wall paper reading and
the mass reading of pingguo apple in (15) is not an issue between count
and (shifted) mass nouns at the grammatical level but a genuine ambiguity
between individual and stuff readings of Chinese bare nouns at the ontological level.

Individual, stuff and partial-object readings

79

I claim that Chinese bare nouns are unspecified with individuation and
that they have an ambiguity between an individual and a stuff reading, or
even a partial-object reading (also see Huang and Lee 2009).
I give an illustration of four different types of nouns, i.e. fruit-vegetable,
animate entities, liquid and artifacts, with respect to the availability of
individual, stuff and partial-object readings.
[Fruit-vegetable]: food nouns, such as those referring to fruit and
vegetables, freely alternate between the individual and the stuff reading.
Additionally, a partial-object reading is also possible. These three readings
for pingguo apple are illustrated in (16).
(16) zhuo shang you
pingguo.
table on
there:be apple
a. There are apples on the table.
OR There is an apple on the table.
b. There is apple on the table.
c. There is part of an apple on the table.

[Individual reading]
[Stuff reading]
[Partial-object reading]

In the individual reading, pingguo can either have a singular or plural


reading, as in (16a). In the stuff reading, pingguo does not have to satisfy
the requirement of being an apple in terms of its shape, as in (16b). In addition, Chinese bare nouns also have a partial object reading. The partialobject reading is different from the stuff reading. In the partial object reading, the noun pingguo apple may refer to some part of an apple, e.g. half
an apple or one third of an apple, as in (16c). In English, it is impossible
for the bare nouns to have the partial object reading, which is expressed by
the (partitive) quantifier some, such as There is some apple on the table.
[Animate nouns]: Most animate nouns in Chinese only have an individual reading. They refer to naturally atomic entities, such as student, dog,
pig, cow etc. This is very much different from the English counterparts.
The English noun dog either refers to individual dogs or dog stuff. Recall
the example (14). In Chinese, the stuff reading is expressed by the lexical
device -rou flesh, such as gou-rou dog-flesh, as in (17).
(17) qiang
shang dou
shi
wall
on
all
be
There is dog all over the wall.

gou-rou.
dog-flesh

However, some animate nouns, such as livestock or seafood, can have


partial-object and stuff readings, in addition to the individual reading. Size

80

Natural atomicity

might be relevant factor and it is much easier for small-sized animals to


have the three-way ambiguity. For example, in (18), the bare noun ji
chicken can either refer to a whole chicken (18a), chicken stuff (chopped
into pieces) (18b), or half a chicken (18c).
(18) guo
li you
ji.
pot
in all
chicken
a. There are chickens in the wok.
There is a chicken in the wok.
b . There is chicken in the wok.
c. There is part of a chicken in the wok.

[Individual reading]
[Stuff reading]
[Partial-object]

[Liquid]: nouns referring to liquid, such as shui water, have a stuff


reading as default.
(19) zhuo shang
you
shui.
table on
there:be water
There is (splashed) water on the table

[Stuff reading]

[Artifact]: nouns referring to artifacts are stubbornly naturally atomic


entities and can only have an individual reading. As shown in (20), the
Chinese noun zixingche bicycle cannot refer to parts of bicycle, so it cannot have a stuff reading.
(20) di-shang dou
shi
zixingche.
floor-on all
be
bicycle
There are bicycles all over the floor.
[Individual reading]
NOT: There is bicycle all over the floor.
I claim that there is no grinding in general in Chinese and that Chinese
nouns are unspecified with individuation. Specifically, Chinese bare nouns
may refer to entities existing in different ontological forms, including naturally atomic entities, partial object or stuff. The interpretational variability
of Mandarin bare nouns cannot be captured by the binary distinction between mass and count nouns.
As shown from (16) to (20), these four types of nouns vary with respect
to the possibility of referring to different ontological forms of entities. Artifacts and animate nouns on the one hand and liquid on the other represent
two ontologically different types of nouns: nouns denoting discrete entities
and nouns denoting homogeneous entities respectively. Food nouns, in-

Natural atomicity as a grammatically relevant phenomenon

81

cluding [fruit-vegetable] and a small set of [animate] nouns (e.g. poultry or


seafood like crab, shrimp), have a three-way ambiguity between individual,
stuff and partial-object readings.
According to our understanding, the wall-paper reading of gou in (14) is
not because dog is a lexical count noun which rejects a mass (grinding)
reading, but rather, it is an animated noun, which is stubbornly naturally
atomic. The two readings of pingguo in (15) are not an ambiguity between
count and mass nouns, but a genuine ambiguity between individual and
stuff reading, which is generally available for Chinese bare nouns.
The underspecification of individuation hypothesis is also supported
by experimental studies.
Huang and Lee (2009) argue that older Chinese children (5.0-6.6 years
old) and Chinese adults are sensitive to distinct linguistic contexts in their
interpretations of partial object situations, such as the contrast between
bare noun and Cl-N. For them, it is possible for bare nouns to have a
partial object reading, in addition to individual and stuff readings, while
Cl-N can only have an individual reading. However, younger children (3.65.0 years old) antithetically accepted partial object situations across linguistic contexts.
They propose that Chinese children are not born with noun individuation based on ontological/semantic properties of nouns; instead, they come
to learn noun individuation through the learning of the individuation function of linguistic forms (for Chinese children, the individual classifiers).
This indicates that noun individuation is not an inherent part of lexical
meaning (even for the count nouns in Cheng et als term or count mass
nouns in Doetjes term). A portion/part of individuals could be an instance
of that kind of individuals as long as the portion/part could be used to identify the kind of individuals it belongs to.

5. Natural atomicity as a grammatically relevant phenomenon


In section 3 and 4, I reviewed two lexicalist approaches to the mass/count
distinction in Chinese. Doetjes (1997) claims that Chinese nouns fall into
two basic types, count mass nouns and mass mass nouns, on the basis
of having or not having minimal parts of the entities in the denotation of
nouns. Cheng et al (2008) argue that nouns come out of the lexicon as
count or mass and this is true of all languages in the world. In contrast with
them, I argue that the mass/count distinction distinguished at the lexical
level is a real world difference about the properties of entities in our world.

82

Natural atomicity

It is different from the mass/count distinction distinguished at the grammatical level, as is found in English.
In what follows, I will recast Doetjess (1997) distinction between
count mass and mass mass nouns, or Cheng et als (2008) distinction between count and mass nouns, into an ontological distinction between discreteness and homogeneity in terms of the property of denotation of nouns.
I will be assuming the position that although Chinese nouns are all mass
nouns, two ontologically different types of nouns can be distinguished:
nouns denoting discrete entities and nouns denoting homogeneous entities.
The characterizing property of the first group of nouns is that the entities denoted are conceived as spatially bound or delimited and are
straightforward identified as being a one. Nouns in this group include
xuesheng student, dianshi-ji TV set, shu tree etc. In contrast, the referents of nouns in the second group do not have well-defined or stable spatial configuration. Nouns in this camp include shui water, yun cloud, lu
road etc.3
I will present evidence below to show that the ontological distinction of
nouns between homogeneity and discreteness is a grammatically relevant
distinction in Mandarin.
The first piece of evidence comes from the use of size adjectives as adnominal modifiers, such as da big and xiao small (Rothstein 2010,
Zhang 2012).
Rothstein (2010) notes that in English, nouns denoting entities with salient individual units can be be modified with adjectives like big or small,
and those denoting homogeneous entities reject adjective modification of
big and small. Such a contrast cross-cuts the distinction of mass and count
nouns.
(21) a. Do not buy big furniture. The stairs are too narrow to carry it up.
b. * a glass of big water
This generalization is also applicable to Chinese. Although nouns like
pingguo apple cannot be directly counted without a classifier in Mandarin, it denotes entities whose unit structure can be modified, as in (22). In

These two types of nouns dont correspond to the distinction between individual
and non-individual classifiers. In other words, it is not completely true that individual classifiers can only modify nouns with discrete referents and non-individual
classifiers modify nouns with homogeneous referents.

Natural atomicity as a grammatically relevant phenomenon

83

contrast, those denoting homogeneous entities like shui and kongqi cannot
be modified by dimensional adjectives, as in (23).
(22) a. yi
ge
hen
one
CL
very
a very big apple
b. yi
ke
hen
one
CL
very
a very big tree
(23) a. * yi
bei
hen
one CLglass very
b. * yi
ge
hen
one CL
very

da
big

de
Mod

pingguo
apple

da
big

de
Mod

shu
tree

da
big
da
big

de
Mod
de
Mod

shui
water
kongqi
air

According to Zhang (2012), in addition to [numerablity], nouns are also parameterized with the feature [dimensionality]. Namely, there are
dimensional and non-dimensional nouns. Dimensional nouns, such as he
river, are subject to modification of dimensional adjectives, and nondimensional nouns, including material nouns like jin gold and you oil
and immaterial nouns, such as minzhu democracy and yinyue music,
cannot be modified by dimensional adjectives. The contrast between these
two types of nouns suggested by Zhang corresponds to the contrast between discrete and homogeneous entities, as contrasted in (22) and (23). I
thus will not provide further examples.
Zhang (2012) also suggests that the contrast between these two types of
nouns with respect to dimensional adjectives also shows up in predicative
positions. Look at her examples in (24) and (25).
(24) a. he
hen
chang.
river
very long
The river is long.
b. qiqiu
hen da.
balloon very big
The balloon is big.

84

Natural atomicity

(25) a.* you


hen
oil
very
b.* zhengqi hen
steam very

chang.
long
da.
big

The modification of adjective as a legitimate test is understandable if


we take into account that discrete entities denoted by nouns have an inherent structure, which can be measured along a certain dimension, while
homogeneous entities have no such structure.
The second piece of evidence is that in Chinese, some classifiers and
quantifiers are sensitive to the natural atomicity of nouns: some modify
nouns denoting naturally atomic entities only and others non-naturally
atomic entities.
Doetjes (1997:34) said that group classifiers, such as da dozen, are
semantically used for a group or a collection of individuals, as in the
example of yi da baima a dozen of white horses. We said earlier that not
all group classifiers take nouns with atomic referents, but it is true for the
specific classifier da dozen, since it expresses the special meaning of a
collection or package of individuals (for sale). Interestingly, when saying yi
da shui, it means a dozen of packs of water only, in which the noun water
are forced to denote atomic entities.
(26) yi da
shui
one dozen water
a dozen of packed water
In contrast, the quantifier yi dian-er a bit takes nouns denoting homogeneous entities, such as shijian time and shengyin sound, as in (27a-b).
(27) a. yi
dian-er shijian
one
bit
time
a bit of time
b. yi
dian-er shengyin
one
bit
sound
a bit of sound
It is inappropriate for yi dian-er to modify naturally atomic nouns like
ren people, as in (28a). This kind of shifting happens only under some
circumstance. For example, the naturally atomic denotation, such as ping-

Conclusions

85

guo apple can be shifted into a non-naturally atomic reading as triggered


by yi dian-er, as in (28b).
(28) a. #yi
dian-er ren
one
bit
people
b. yi
dian-er pingguo
one
bit
apple
a bit of apple(*s).

6. Conclusions
In this chapter, I argued against the view that there is a lexical distinction
between count and mass nouns in Chinese. By separating the grammatical
notion of countability from the ontological/cognitive notion of individuation, we claim that the mass/count distinction is independent whether the
referents in the denotation of nouns are individuals or not. Shui boy and
nanhai boy are both mass nouns, but they represent two ontologically
different types of nouns: nouns referring to discrete entities and nouns
referring to homogeneous entities.

Chapter 5
Chinese bare nouns

1. Introduction
Having made the claim in chapter 3 and 4 that the count/mass distinction of
nominals cannot be established in Chinese, I now explore the KrifkaChierchia proposal that all Chinese nouns are mass nouns (see Krifka 1995,
Chierchia 1998a, b). Krifka assumes that every language which allows for
bare NPs at all uses them as expressions referring to kinds (see GerstnerLink 1988). Chierchia (1998b) makes the explicit claim that Chinese is an
argumental language, in which (i) all the nouns are born as arguments, and
(ii) they denote kinds, as do English mass nouns. Though Chierchias hypothesis has been very influential, it is, as he admits himself rather speculative (Chierchia 1998a: 92). Our first and foremost goal in this chapter
is to examine the plausibility of Krifka-Chierchia hypothesis. We show that
there is good evidence to accept it.
It has been widely observed in the literature that Chinese bare nouns
have, besides kind readings, object-level interpretations, which can be both
definite and indefinite (see Li and Thompson 1981, R. Yang 2001, H. Yang
2005, Rullman and You 2006, among others). For example, in (1a), the
bare noun shu book is a kind term, meaning the kind book, and shu
book in (1b) can either mean the book(s) or some book(s).
(1) a. shu shi renlei de jingshen shiliang.
book be human Mod spirit
food
Books are food for the human soul.
[Generic]
Lit: Books are to our mind, as food to our body
b. wo mai le
shu.
I buy PFV book
I bought some book(s).
[Indefinite]
OR I bought the book/the books.
[Definite]
The second goal of this chapter is to answer the question: what is the relationship between the kind reading and the (in)definite reading, of bare

The Krifka-Chierchia hypothesis

87

nouns? There are two possibilities. The first one is that Chinese bare nouns
are ambiguous between a kind reading and object-level reading (Wilkinson
1991, Gerstner and Krifka 1993, Kratzer 1995, and others) The second
possibility is that the kind interpretation is the default reading for Chinese
bare nouns and that the object-level readings are derived from the kind
reading (see Carlson 1977, Chierchia 1998b, R. Yang 2001, Dayal 2004).
In this chapter, we take the neo-Carlsonian approach and argue for the
second possibility.
The rest of the chapter is organized as follows. Section 2 briefly introduces the Krikfa-Chierchia hypothesis that Chinese bare nouns are kind
terms. In section 3, we justify the Krifka-Chierchia hypothesis with various
pieces of evidence. In section 4, we represent the semantics of bare nouns
in formal terms. Sections 5 and 6 look at the object-level interpretations of
Chinese bare nouns. In section 5, we study the predicative use of bare
nouns, e.g. in post-copula positions (see Higgins 1976, Partee 1986). Section 6 looks at definite readings of bare nouns in argumental positions, e.g.
topic positions. We argue that Chinese nouns are kind-denoting by default,
and that the existential and the definite reading are derived (also see R.
Yang 2001, H. Yang 2005).

2. The Krifka-Chierchia hypothesis


Krifka (1995) is among the first to note that Chinese nouns are ambiguous
between a kind-level reading and an object-level reading. Take as an example the noun xiong bear. It can refer to (2a) the kind Ursus or (2b)
some specimens of this kind. There is also a measure construction which
applies to (2c) a specified number of realizations of Ursus.
(2) a. xiong juezhong
le.
bear
become:extinct PRF
Bears become extinct.
b. wo
kanjian xiong le.
I
see
bear PRF
I saw (some) bears.
c. san
qun
xiong
three
CLherd bear
three herds of bears

88

Chinese bare nouns

Krifka (1995: 399) assumes that the bare noun xiong is basically a
name of the kind Ursus, and that the other uses have to be derived from
that (also see Dolling 1992). One reason to take the kind-referring use in
(2a) as basic is that it seems that every language which allows for bare
NPs at all uses them as expressions referring to kinds (see Gerstner-Link
1988).
Chierchia (1998b: 353-4) develops a typology of noun reference. He
assumes that nouns are either classified as predicates or as arguments, and
he introduces the features [argument] and [predicate] as parameters constraining the way in which the syntactic category N (and its phrasal projection NP) is mapped onto its denotation in different languages. [+argument]
means that N can be mapped onto denotations of the argumental type and
[-argument] cannot. The determiner D denotes a function from predicates
to arguments. In a language where N is a predicate, N must combine with a
D to form an argument. In a language where N is born at the argument type,
this is not necessary and we get so-called bare NP arguments.
French is argued to be an [-argument, +predicate] language, in which
every noun is a predicate, and nouns cannot be made into arguments without projecting D.
English is a language parameterized with the setting of [+argument,
+predicate]. This means that nouns in English can be either predicative or
argumental. In English, mass nouns are [+argument], while count nouns are
[+predicate]. This means that count nouns must combine with a determiner
to denote an argument, and hence count nouns cannot occur in their bare
singular form.
Chierchia proposes that Chinese is an argumental language, where the
nouns have the [+argument, -predicate] setting. This means that nouns can
be mapped onto denotations of the argument type but not of the predicate
type. He assumes that in a language with this setting, nouns and their
phrasal projections are mass nouns that uniformly denote kinds. (We refer
to Chierchia (1998b: 355-357) for the detailed discussion of each type of
language.)
To synthesize, the Krifka-Chierchia hypothesis goes as follows: Chinese bare nouns are mass expressions and denote kinds as default.

Bare nouns as kinds

89

3. Bare nouns as kinds


The Krifka-Chierchia hypothesis has been very influential, but it has not
been carefully justified. Especially, under his nominal mapping hypothesis,
Chierchia makes the following predictions concerning properties of nouns
in [+argument, -predicate] languages like Chinese:
a. There is no plural marker;1
b. Bare nouns cannot be modified by numerals without the assistance
of classifiers;2
c. There is no indefinite article.
These three morphosyntactic predictions are borne out in Chinese.
However, this by itself does not entail Chierchias assumption that Chinese
nouns denote kinds.
In what follows, I use the tests of generic contexts (Section 3.1), appositives (Section 3.2) and scope behaviors (Sections 3.3 and 3.4), to examine
the nature of bare nouns. As we will see, these tests provide evidence for
Chierchias hypothesis that Chinese bare nouns denote kind terms.

3.1. Strong kind-inducing contexts


Kind-reference generic sentences express properties that are true of kinds,
species or classes of objects, but not of individual objects (Krifka et al
1995). In English, bare plural nominals and definite singular nominals have
kind interpretations. They can take kind-level properties denoted by kindlevel predicates, such as being extinct, as exemplified in (3a-b):
1

Note that in Chinese, there is a seeming plural marker men (A. Li 1999). However, this marker is not exactly the same as the English plural morpheme s. Firstly,
it carries definiteness; secondly, it can only be attached to Ns denoting humans.
According to Iljic (1994), it is a collective marker but not a plural maker. Though
A. Li (1999) argues -men in Mandarin Chinese is best analyzed as a plural marker,
she accepts the difference between the Chinese men and the English s. She assumes that -men is realized on an element in D, in contrast to plurality on N, which
is what we find in English. We thus do not think that Chinese men can be treated
as a true plural marker.
2
Not all languages with bare arguments are bound to have classifiers. For example,
in Indonesian, classifiers are optional between numeral and noun, although it is an
argumental language (Chung 2000).

90

Chinese bare nouns

(3) a. Dodos are extinct.


b. The dodo is extinct.
In contrast, indefinite singular nominals are individual-denoting only,
and cannot take kind-level predicates, as in (3c).3
c. * A dodo is extinct.
In Mandarin Chinese, there is neither a plural marker nor are there articles, so we cannot find expressions like bare plurals or (in)definite singulars headed by articles. The closest we can come to English bare plurals is
with Chinese bare nouns, and the closest we can come to the singular definite is with the Chinese Demonstrative+N construction, while the closest
we can come to singular indefinites is with Chinese one+CL+N.4 As (4)
shows, only Chinese bare nouns can be used as subjects of kind-level predicates in kind reference generics (4a), while demonstratives phrases (4b)
and indefinite singular phrases (4c) cannot:
(4) a . jing kuai juezhong le.
whale soon be:extinct PRF
Whales will soon be extinct.
b.#na jing
kuai juezhong le.
that whale soon be:extinct PRF
Intended: That whale will soon be extinct.
c.# you
yi
tiao jin
kuai juezhong
le.
there-be one CLtail whale soon be:extinct PRF
Intended: A whale will soon be extinct.
The predicate of jue-zhong be extinct in (4a) has the literal meaning
vanish-kind in Chinese, so it clearly is a kind-level predicate. It requires a
3

The indefinite singular nominals can have a subkind reading. On the subkind
reading, a dodo means a subkind of dodos. We ignore this reading for the time
being.
4
In Mandarin, demonstratives can either be used in the construction of Dem+Cl+N
or in Dem+N. But, as mentioned in Chapter 3, Dem-N is usually found in preverbal
positions, e.g. the subject position, and not in postverbal positions, e.g. object positions. Note that Dem-N can only have an object-level interpretation and not a kindlevel interpretation. For example, zhe yu this fish means this individual fish and
not this kind of fish.

Bare nouns as kinds

91

kind term to be its argument, so the bare noun in (4a) is interpreted as: the
whale kind. The examples of (4b) and (4c) show that this kind-level predicate cannot take as argument a nominal phrase that refers to individuals,
such as na jing that (individual) whale or yi tiao jing an (individual)
whale.
A second point is that Chinese bare nouns can occur in the post-copula
position as a kind-level predicate. The sentence, in that case, is a generic
sentence, as in (5a).
(5) a. haitun
he lanjing
dou shi jing.
dolphin and blue whales all be whale
Both dolphins and blue whales are whales.
b. haitun he lanjing
shi liang zhong jing.
dolphin and blue whale be two CLkind whale
Dolphins and blue whales are two kinds of whales.
c.# haitun he lanjing
shi liang tiao
jing.
dolphin and blue whale be two CLtail whale
Dolphins and blue whales are two (individual) whales.
(5a) expresses a taxonomic relation between the kind of entities denoted
by the subject, i.e. haitun he lanjing dolphins and blue whales, and the
denotation of the post-copula NP, i.e. jing whales. That is, the kind Dolphin and the kind Blue Whale stand in a subkind relation to the kind Whale.
The subkind relation can be overtly realized by the kind classifier, zhong
kind as shown in (5b); the sentence is infelicitous with the individual
classifier tiao tail, as in (5c).
The use of bare nouns in Chinese to refer to kinds has been neglected in
Chinese linguistics, despite attested examples like the following in (6)
from high school textbooks used in mainland China. These examples illustrate strong kind-inducing contexts, such as discussions of biology, invention, taxonomy. The kind terms are given in boldface.
(6) a. niao shi you konglong jinhua lai
de.
bird be from dinosaur evolve come PRT
Birds evolved from dinosaurs.
b. aidisheng faming le
dianhuaji.
Edison
invent PFV telephone
Edison invented the telephone.

92

Chinese bare nouns

3.2. Appositives
In this section we discuss appositives as a test for the kind referring nature
of Chinese bare nouns. Bare nouns easily take appositive nominals, e.g.
Demonstrative+Cl+N. Note that demonstratives before Cl-N are usually
unstressed in the above cases. They lose their deictic function and end up
working like determinate article, in the sense that they yield definiteness.
This phrase, when it follows an anchor noun, can be interpreted appositively or restrictively (Del Gobbo 1999). We are only interested in the appositive use here.
In kind-reference generics, the appositive phrase can be a kind ClP,
headed by the kind classifier zhong kind, as in (7), but not an individual
ClP headed by individual classifiers such as zhi or ge, individual unit, as
in (8).
(7) a. niao, zhe zhong dongwu, you konglong,
bird this Clkind animal from dinosaur
na zhong dongwu jinhua er lai.
that CLkind animal evolve thus come
Birds, this kind of animal, evolved from dinosaurs, that kind of
animal.
b. aidisheng faming le
dianhuaji zhe zhong shebei.
Edison invent PFV telephone this CLkind equipment
Edison invented the telephone, this kind of equipment.
(8) a. #niao, zhe zhi dongwu, you konglong,
bird
this CL animal
from dinosaur
na
zhi dongwu jinhua er lai.
that CL animal
evolve thus come
The bird, this individual animal, evolved from the dinosaur, this
individual animal.
b.# aidisheng faming le
dianhuaji zhe ge shebei.
Edison
invent PFV telephone this CL equipment
Edison invented the telephone, this equipment.
The verbs of jinhua evolve in (7a) and faming invent in (7b) can only take kind terms as arguments, so the arguments only allow kind classifier phrase as appositive phrases, as in (8) and not individual classifier
phrases, as in (8).

Bare nouns as kinds

93

Moreover, kind ClPs can even function as appositive phrases of bare


nouns in episodic sentences, forcing a kind reading of the bare noun. Consider (9).
(9) a. wo chi le
liulian, zhe zhong / # ge
I eat PFV durian this CLkind / CL
I ate durians, this kind of fruit.
b. wo kan guo
jing, zhe zhong /# zhi
I see EXP whale this CLkind / CL
I have seen whales, this kind of animal.

shuiguo.
fruit
dongwu.
animal

Take (9a) for instance. Without the appositive phrase, the sentence of
wo chi le liulian can either mean that I ate the durian(s) or I ate some
durian(s), where the bare noun durian has a definite or an indefinite reading, but not a kind reading. However, with an appositive like zhe zhong
shuiguo this kind of fruit, the bare noun liulian durian is interpreted as a
kind term, referring to the kind durian. As a result, (9a) means that I had a
durian-eating experience. (Note that the phrase zhe ge shuiguo this Cl
fruit can never be used as an appositive on the bare noun, regardless of
whether the bare noun liulian is kind or individual denoting.)

3.3. Scope with respect to opacity


In this section, I will use Carlsons (1977) scope tests in opacity contexts to
investigate the scope behavior of Chinese bare nouns. I will show that in
the object-level interpretation, Chinese bare nouns are ambiguous between
an opaque reading and a definite reading.
Carlson (1977) shows that in English, in opaque contexts, indefinite
singular NPs like a policeman are ambiguous between an opaque and a
transparent reading (10a). In the transparent reading of (10a), there is some
particular person that Paul is looking for, who happens to be a policeman
(a fact that Paul may be unaware of). In the opaque reading, Paul is not
looking for a particular person, but would be satisfied by anyone who can
show him his Police Badge. Indefinite plural NPs, like some policemen in
(10b), show the same ambiguity. However, as Carlson argues, bare plural
NPs, like policemen in (10c), only have the opaque reading.

94

Chinese bare nouns

(10) a. Paul is looking for a policeman.


(opaque or transparent)
b. Paul is looking for some policemen. (opaque or transparent)
c. Paul is looking for policemen.
(opaque)
The opaque and transparent readings of indefinite phrases, such as a policeman and some policemen, have the following two semantic structures
associated with them (following Zimmermann 1993 for the intensional
one), in (11). In the opaque reading, look for relates Paul to the property of
policeman, meaning that he is trying to find an instantiation of that property in the real world. In the transparent reading, look for relates Paul to an
individual x, meaning that he is trying to find that person x, and x happens
to have the property of being a policeman.
(11) a. look for (paul, policeman )
b. x[ policeman(x) look for (paul, x)

(opaque)
(transparent)

Carlson (1977) argues that if the bare noun is ambiguous between a


kind reading and an indefinite reading, the bare plural in (10c) should have
the opaque and transparent readings, just as the indefinites (10a) and (10b),
but does not.
Carlson (1977) argues that English bare plurals only denote kinds and
that the existential reading of the bare noun is derived from the basic kind
reading by a type-shifting operation on the verb or predicate which introduces an instantiation relation between the kind and its individual instances.
Thus, even if the intensional context allows an existential reading, the existential quantifier is introduced locally by the type shifting operation, i.e. in
the scope of the intensional operator: no transparent readings are generated.
Thus, we can interpret (11a) as (11c), meaning: Paul tries to bring himself
into a world where he has found an instantiation of the kind Policeman
(Note: INST stands for INSTANTIATE).
c.Try (paul, w.x[INST (x,Policeman,w) FIND (paul,x,w)])
(opaque)
As mentioned above, one+Cl+N is the closest we can come in Chinese to English indefinite singulars and Chinese bare nouns are the closest
we can come to English bare plurals. In what follows, we will check
whether Carlsons observation about the contrast between English bare

Bare nouns as kinds

95

plural NPs and indefinite singular NPs in (10) carries over to bare nouns
and one+Cl+N in Mandarin:
(12) a. ta zai
zhao yi
ge
baomu.
he PROG seek one CL
maid
He is looking for a maid.
(opaque or transparent)
b. ta zai
zhao baomu.
he PROG seek maid
He is looking for a maid/maids.
(opaque reading)
OR He is looking for the maid(s).
(definite reading)
In (12a), the singular phrase of yi ge baomu either refers to anyone who
has the property of being a maid, which leads to an opaque reading, or
refers to a particular person, who is a maid, in which case we get a transparent reading. In (12b), the bare noun baomu either refers to anyone who is
a maid/are maids, and the reading we get is the opaque reading, or it refers
to the maid(s) familiar to the subject, in which case the sentence gets a
definite reading.
In short, one+Cl+N pattern with English indefinite singular NPs, and
(12a) is ambiguous between an opaque reading and a transparent reading.
Chinese bare nouns, like English bare plurals, allow an opaque reading, but
not a transparent indefinite reading.
However, Chinese bare nouns have an extra reading, a definite reading,
which English bare plurals do not have. The extra reading was also observed by R. Yang (2001:26-28), who argues that it should be distinguished
from the transparent reading and is a phenomenon unique to Chinese.
Yangs arguments for distinguishing transparent and definite readings will
be discussed in section 2.4.
I here discuss a different test which distinguishes definite from transparent readings: the test of relative clauses.
Chinese relative clauses (RCs) are left-branching. In numeral phrases
like Num+Cl+N, the relative clause can modify the head noun, as in the
form Num+Cl+RC+N, or the whole numeral phrase, as in the form
RC+Num+Cl+N. We use our examples in (13) to illustrate these two
types of relative clauses. (13a) has the form Num-Cl-RC-N and (13b) has
the form RC-Num-Cl-N.

96

Chinese bare nouns

(13) a. ta zai
zhao yi ge [RC ta bu renshi de] baomu.
he PPOG seek one CL he not know Mod maid
There is a maid that he is not familiar with that he is looking for.
[transparent]
OR: He is looking for a maid [and wants her to be one] that he is not familiar with.
[opaque]
b. ta zai
zhao [RC ta bu renshi de] yi ge baomu.
he PROG seek
he not know Mod one CL maid
There is a maid that he is not familiar with that he is looking for.
[transparent]
Zhang (2006) claims that Num-Cl-RC-N is ambiguous in specificity,
whereas RC-Num-Cl-N is unambiguously specific, and that this contrast
results from the difference in scope possibilities. We agree with Zhangs
observation about the correlation between the modification of relative
clauses and specificity of numeral classifier phrases. We suggest that (13a)
either means that there is a certain maid that the subject is not familiar with,
and she is the one that is being looked for, which is a transparent reading,
or that he is looking for a maid and anyone will be fine, as long as he is not
familiar with her, which is an opaque reading. The example (13b) only has
the transparent reading. The relative clausethat he is not familiar with
specifies the property of the individual denoted by the ClP, which helps to
identify the individual and thus makes the transparent reading more salient.
Now, when we replace the numeral-classifier phrase in (13) with a bare
noun, and modify it with the same relative clause as the one used, we only
get an opaque reading, as in (14). (Note that strictly speaking, wide-scope
and specificity are independent phenomena, but the difference is not relevant here, what is important is that only the opaque reading is available.)
(14) ta zai
zhao [RC ta bu renshi de ] baomu. [opaque]
he PROG seek
he not know Mod maid
He is looking for maids [and wants them to be ones] that he is not
familiar with.
It is impossible in (14) to get a transparent reading. The interpretational
variability of the modified N in (14) parallels neither that of (13a) nor of
(13b). This strongly suggests that the indefinite readings of bare nouns
must be distinguished from the scopally active indefinite readings of
Num+Cl+N.

Bare nouns as kinds

97

Now, I would like to know why the second definite reading that is
available in (12b) is not available in (14). According to Lyons (1999), in
English, definiteness presupposes familiarity and implicates uniqueness.
Li and Bisang (2012) argue that definiteness in Chinese is characterized by
familiarity and does not presuppose uniqueness. This means that the
referent of a Chinese definite phrase is an individual that is familiar to the
interlocutors or identifiable in a relevant context, but it may not be a
unique individual (see Li and Bisang 2012 for details). Now, in (15) the
semantic content of the relative clause expresses unfamiliarity, which
causes an obvious clash with the presupposition of familiarity associated
with the bare noun in the definite reading. We assume that the lexical content in this case overrides the presupposition. That is why the definite reading is not available in this case. If we change the relative clause to that I
am familiar with, the definite reading comes back. So (15) is ambiguous:
(15) ta zai
zhao [RC ta renshi de ] baomu. [definite/opaque]
he PROG seek
he know Mod maid
He is looking for the maids that he is familiar with.
The stone of the relative clause kills two birds in one. On the one hand,
it shows that the indefinite reading of bare nouns is not that of true indefinite NPs like one+Cl+N. On the other, it shows that Chinese bare nouns
indeed can get opaque or definite readings, but not transparent readings.
When the definite reading is unavailable for independent reasons, the opaque reading is the only possible interpretation available.
With Carlson (1977) and Chierchia (1998), we assume that this means
that the default reading of Chinese bare nouns is the kind reading, and their
indefinite reading is derived.

3.4. Scope with respect to quantifiers


Carlson (1977:11) shows that English bare plurals and indefinite singulars
also behave differently with respect to the scope of quantified noun phrases
like everyone. While indefinites can take wide and narrow scopes, bare
plurals have narrow scope only:
(16) a. Everybody is reading a book about giraffes. [> books or books>]
b. Everybody is reading books about giraffes. [>books]

98

Chinese bare nouns

The indefinite phrase a book about giraffes in (16a) can have wide
scope over the universal quantifier everybody, meaning that there is a book
about giraffes that everybody is reading. It can also have a narrow scope
with respect to everybody, meaning that everybody is reading a book about
giraffes, possibly different books. In contrast, the bare plural books about
giraffes in (16b) can only have a narrow scope reading: everybody is reading a book about giraffes, possibly different books.
In Chinese, a group of linguists (S.F.Huang 1981, Huang 1982 among
others) believe that Chinese sentences with two quantifiers, which function
as subject and object respectively, shows no scope ambiguity in that the
subject quantifier scopes over the object quantifier. According to them, the
only possible interpretation of (17) is that the quantifier mei ge ren every
scopes over yi ge ren someone. In other words, (17) implies that everyone
likes a different person.
(17) Mei ge ren
dou xihuan yi ge ren.
every CL person all like
one CL person
Everyone likes a person (someone).
(everyone > someone)
However, this observation has proven to be wrong. Some later studies
show that quantifiers in Chinese indeed exhibit scope ambiguity, but different types of verbs may exhibit different levels of prominence in this
kind of scope ambiguity (see Lee 1986, Tsai 1994, Zhou and Gao 2009,
among others). For example, Zhou and Gao (2009) conduct an offline
judgment task, which examines the scope relations of quantifiers with respect to three types of verbs, including action verbs, locative verbs and
psych-verbs. They show that their thematic information does affect quantifier scope interpretation. For the action verb and the locative verb, the
S>O reading is more readily accessible than O>S reading, whereas for the
psych-verb, the S>O interpretation and O>S interpretation are equally accessible, but its S>O interpretation is not as preferably accepted as those of
the former two. This study shows that O>S interpretation is available if
appropriate contexts are provided but it is not as accessible as S>O interpretation. In other words, doubly quantified sentences in Chinese are ambiguous, though the two readings are not equally accessible.
In the current study, I take the viewpoint that both indefinite singulars
and bare nouns show scope ambiguity in Chinese. As in English, the indefinite singular is ambiguous between a narrow scope and a wide scope
reading with respect to the universal quantifier as in (18).

Bare nouns as kinds

99

(18) mei-ge
ren
dou
zai
kan
yi feng
every-CL man
all
PROG read one CL
guanyu jiaxin
de
xin.
about
add-wage Mod letter
Everybody is reading a letter about raising salaries. [>xin or xin>]
(18) either means that everyone has a different letter about salary-raises
to read or that there is a letter about salary-raises that everyone reads. A
natural scenario for the first reading is one where each employee gets a
different raise, and hence they all read different letters. A natural scenario
for the second reading is where they all get a 5% raise and the accounting
department sends a standard letter to all.
The Chinese bare noun is also ambiguous, but between a narrow scope
reading and a definite reading, as in (19).
(19) mei-ge
ren dou zai
kan guanyu jiaxin
de
xin.
every-CL man all PROG read about add-wage Mod letter
Everybody is reading letters about raising salaries.
[>xin]
Everybody is reading the letter about raising salaries. [definite]
(19) has the same narrow scope reading, but the definite reading is different from the wide scope reading in (18): on the definite reading, there is
one and only one letter involved in the situation and everyone reads it. It
can be the case that a letter is pasted next to the entrance to the office, so
everyone reads it when entering the office.
R. Yang (2001) suggests the following scope test to support the argument that the definite reading for Chinese bare nouns is different from the
transparent reading. Look at the examples in (20).
(20) dabufeng xuezhe dou kanguo mei-ge
most
scholar all read
every-CL
a. [jiejue na-ge wenti
de] fenxi
fangfa.
solve that-CL problem Mod analysis method
Most scholars have looked at every analysis that solves that problem.
b. [jiejue mou-ge
wenti
de] fenxi
fangfa.
solve certain-CL problem Mod analysis method
Most scholars have looked at every analysis that solves a certain
problem.

100

Chinese bare nouns

c. [jiejue wenti
de] fenxi
fangfa.
solve problem Mod analysis method
Most scholars have looked at every analysis that solves problems.
OR Most scholars have looked at every analysis that solves the problem.
Following Reinhart (1987), R. Yang (2001) argues that the definite NP,
such as na-ge wenti that problem in (19a), takes highest scope relative to
other scope operators, i.e. it scopes over both the universal quantifier
meige every and the quantifier dabufeng most. (20a) means that there is
a particular problem for most scholars and they have looked at every analysis to solve that problem.
In contrast, the specific indefinite NP, such as mou-ge wenti a certain
question in (20b), not only has a wide scope reading, but also allows for
intermediate scope. On the intermediate reading, mou-ge a certain scopes
over the universal quantifier meige every but remains in the scope of
quantifier dabufen most. So (20b) means that for most scholars, there is a
problem such that he/she has looked at every analysis that solves that problem, possibly different problems for different scholars.
Now, when we look at the bare noun, such as wenti problem in (20c),
I observe that there is no intermediate reading: (20c) has only a narrowest
scope meaning or a widest scope meaning, i.e. anything that is problem
or a particular problem that is known to both the speaker and the listener.
The first reading is derived from the kind interpretation of the bare noun
(which is equivalent to the narrow scope indefinite reading), while the
second is a definite reading, where definiteness is defined in terms of familiarity in Mandarin Chinese. This definite reading gives the semantic effect
of a widest scope reading.
We see that definite readings are to be distinguished from transparent
readings, and the scope behavior of the bare noun with respect to quantifiers is similar to the scope behavior with respect to opaque contexts.
To summarize so far, I have discussed two pieces of evidence to prove
that Chinese bare nouns are kind denoting. I showed that bare nouns appear
naturally in kind-referring sentences. I demonstrated that Chinese bare
nouns do not have the same scope behavior as true indefinites. Chinese
bare nouns allow a definite reading which is independent of the scopal
readings I am concerned with here. Ignoring the definite reading, bare
nouns always have narrow scope with respect to quantifiers and opacity
and cannot get intermediate or wide scope readings. The data discussed

Semantics of bare nouns

101

here strongly supports the Krifka-Chierchia hypothesis that Chinese bare


nouns are kind referring.

4. Semantics of bare nouns


I now specify the interpretation of Chinese bare nouns.
Link (1983) proposes that there is a structural difference between the
denotation of count nouns and of mass nouns in English. Count nouns have
their denotation in an atomic Boolean domain and mass nouns have their
denotation in a non-atomic Boolean domain. So mass nouns denote different kinds of objects from count nouns. In contrast to this, Chierchia (1998a,
b) argues that all the nominals are interpreted with respect to an atomic
Boolean algebra. We follow Chierchia (1998a, b) and assume that all nominals are interpreted in a complete atomic Boolean domain D, generated
by a partially vague set of atoms. We assume a part-of relation YD and a
sum operation WD such that for every XD: WDXD. WD is the operation of
complete join with respect to YD, which means that for every XD: WDX is
the smallest element of D such that for all dX: dD WDX.
For example, a domain with three individuals has the structure in (21).
A singular noun denotes a set of atoms, where atoms are the individuals at
the bottom of the structure, a, b, and c; the other entities are pluralities,
which are derived from the atoms via the sum operation W. A plural noun
denotes the closure of a set of atoms under sum. From singular noun interpretation {a, b, c} we get plural noun interpretation: {a, b, c, aWb, bWc,
aWc, aWbWc}.
(21)

aWbWc
aWb

aWc

bWc

Crucially, Chierchia (1998a) argues that count nouns distinguish lexically between the set of atoms and the set of plural elements in the Boolean
algebra, while mass nouns do not. He suggests that mass nouns are grammatically singular but lexically plural: mass nouns come out of the lexicon
with plurality already built in, and that is the only way they differ from

102

Chinese bare nouns

count nouns (Chierchia 1998a: 53). So for Chierchia, a grammatically


singular count noun denotes a set of atomic individuals, the plural of a
count noun denotes the closure of that set of atoms under the sum operation, minus the set of atoms, while the grammatically singular mass noun
denotes just the closure under sum.
If there are three dogs in our world, including Fido, Barky and Spotty,
the singular noun dog will be true of them, i.e. dog denotes the set {Fido,
Barky, Spotty}. The plural noun dogs will be true of Fido and Barky,
Barky and Spotty, and any other plural combination of these three individual dogs, i.e. dogs denotes the set: {FidoWBarky, BarkyWSpotty, FodoWSpotty, FidoWBarkyWSpotty}. However, a mass noun like furniture will not
differentiate between singular pieces of furniture and plural pieces of furniture. The denotation of furniture is lexically plural and it includes all the
possible instantiations of furniture, both singularities and pluralities.
Chierchia (1998a, b) maintains that mass nouns differ from count nouns
in that the former do not have access to the basic atoms at the bottom of the
Boolean structure, while the latter do. According to Chierchia, the atoms in
the denotation of the mass noun are vague or underdetermined for mass
nouns, but the atoms in the count noun denotation are specified or determined.
Extending this work to genericity, Chierchia (1998b) further argues that
mass nouns denote kinds. Kinds are defined via the maximal entity in the
denotation of mass nouns in each world. Chierchia (1998b:349) assumes
that we can model kinds as individual concepts of a certain sort: functions
from worlds (or situations) into the maximal entity in P, i.e. the sum of all
instances of the kind in each world. Kind terms are thus expressions denoting individual entities of type k.
(22) For any property P associated with the kind and the world of evaluation w, ?P=w.WD(Pw)
Since I will look only at extensional contexts, we can assume that the
denotation of a kind term is: ?P, i.e. WDP.
In chapter 3 and 4, I showed that there is no evidence to show that there
is a mass/count distinction in Chinese, and that all nouns behave morphosyntacially like mass nouns. Earlier in this chapter, I argued that Chinese
bare nouns directly denote kinds in various contexts. I now propose that
Chinese bare nouns are interpreted in the same way as English mass nouns:
both denote kinds. Following Chierchias semantics for English mass
nouns, I propose the following semantics for the bare noun jing whale:

Bare predication

103

(23) a. jing kuai juezhong le.


whale soon be:extinct PRF
Whales will soon be extinct.
b. NjingN=?whale
=WD(whale)
[Kind]
c. NjuezhongN=k.be extinct(k)
[Predicate of kinds]
d. NjuezhongN(NjingN) = k.be extinct(k)(?whale)
= be extinct (?whale)
The bare noun jing whale denotes the kind WHALE, which can (ignoring intensionality) be modeled as sum of all the individual whales. This
kind term can be directly taken as argument of a kind-level predicate like
juezhong be extinct, in (23c), giving the predication in (23d).

5. Bare predication
Besides kind interpretations, Chinese bare nouns allow object-level interpretations, both indefinite and definite. In this section, we will focus on
predicative uses of Chinese bare nouns, e.g. their uses in post-copula positions in the spirit of Higgins (1973) and Partee (1986).
The first goal of this section is to study the interpretational variability of
bare nouns in predicative positions. We will make a distinction among
three types of interpretations of the bare noun jing whale in a post-copula
position:
-the bare noun can denote a set of individuals (24a)
-the bare noun can denote a set of subkinds (24b)
-the bare noun can denote a kind (24c).
(24) a. Moby Dick shi jing.

Moby Dick be whale
Moby Dick is a whale.
b. lanjing
shi jing.
blue whale be whale
Blue whales are whales.
c. jing
shi jing.
whale be whale
Whales are whales.

[Individual-denoting]

[Subkind-denoting]

[Kind denoting]

104

Chinese bare nouns

The second goal of this section is to specify the semantics of predicative


bare nouns on the kind reading and the set reading and specify the derivational relation between them. Since I have argued that the kind interpretation is basic, I will derive the set interpretations from it.
In section 5.1, I discuss some basic properties of copula clauses. In sections 5.2-5.4, I discuss the three types of Chinese copula sentences.

5.1. Some analyses of copula clauses


5.1.1. Ambiguous BEs
While analyzing pseudoclefts like (25), Higgins (1973) argues that there
are at least two copulas in English: predicational BE and specificational
BE.
(25) What John is is unusual.
In one reading, (25) is a predicational pseudocleft: it predicates the
property of being unusual to the referent of what John is. If what John is is
a scuba diver, then being a scuba driver is unusual. On the other reading,
(25) is a specificational pseudocleft. It identifies the property of being
unusual as the referent of what John is. Higgins points to the list-like
quality of specificational sentences and offers the following paraphrase:
John is the following: being unusual.
Higgins (1973: 264) extends the study of pseudoclefts to copula clauses
in general and identifies two more types of copulas. All together he identifies four types of copula sentences, as shown in Table 2:
Table 2. Classification of copula sentences
Sentence types
Predicational
Superscriptional
Identificational
Identity

Subject
Referential
Superscriptional
Referential
Referential

Predicate
Predicational
Specificational
Identificational
Referential

According to Table 2, the copula BE is at least four-way ambiguous:


Predicational sentences: the subject is usually referential, and the
predicate is predicational. In other words, a particular object is picked out

Bare predication

105

by the referential NP and this object is ascribed the property expressed by


the predicate (Higgins 1973:212).
For example, in (26a), the entity I am pointing at has the property of
cathood. In this reading, one can conjoin the predicate with another predicational predicate, e.g. What I am pointing at is a cat and is called Jemima.
(26) a. What I am pointing at is a cat.
b. That is heavy.
c. He is a student.
Superscriptional sentences: neither the subject nor the predicate is
referential. The subject merely delimits a domain and the predicate identifies a particular member of that domain (Higgins 1973:213). Look at (27).
(27) a. What I am pointing at is a cat.
b. What John is is unusual.
For example, on the superscriptional reading of (27a), a cat is one of
the items that I am pointing at. (27a) has the interpretation that what I am
pointing at is the following: a cat. In this case, it is impossible to coordinate the predicate is a cat, with other predicates, such as is called Jemima. But we can add other items to the list, e.g. what I am pointing at is
a cat and a dog.
Identificational sentences: the subject is usually that in its common
gender use (in contrast to its inanimate uses). And that can usually be
replaced with that N. The predicate is usually realized as a proper name.
This construction is typically used for teaching the names of people or of
things. See (28).
(28) a. That (man) is Joe Smith.
b. That (building) is the house that I mentioned.
c. That (metal) is alumina.
Identity sentences: Identity sentences are close to identificational sentences. But the subject and the predicate of identity sentences are both
referential and both make genuine references. The copula BE expresses a
relation of equation between the two referents.

106

Chinese bare nouns

In (29a), the individual that the name Cicero refers to is identified as


the same as what the name Tully refers to.
(29) a. Cicero is Tully.
b. Hesperus is Phosphorus.
In the following section, I will focus on predicational and identity uses
of the copula.

5.1.2. Unambiguous BE
In contrast to Higgins, Partee (1986) argues for an analysis of be, in which
it is not ambiguous. Partee suggests that the copula be is not ambiguous,
and that it is always predicational, meaning that its complement is required
to be a predicate. Other analyses along these lines are those by Williams
(1983), Rothstein (2001) and others. The question for these analyses is
how to treat identificational sentences like (26) and identity sentences like
(27)?
Partees unambiguous analysis of be is built upon her theory of typeshifting principles. Partee (1986) argues that NPs have interpretations as a
family of semantic types rather than as a single type, and that natural typeshifting operations map between these types. (Note that she does not distinguish between DP and NP, and uses NP to cover all nominal projections.)
She proposes that NPs have three basic semantics types: d (referential),
<d,t> (predicative) and <<d, t>,t> (quantificational). Those basic types are
lexically assigned to different NPs, but following the general type-shifting
principles, NPs may be shifted from the basic types to others.
Figure 1 gives a schematic representation of a number of type-shifting
principles and their interrelation. In the figure, the circles represent the
three model-theoretical domains: Dd, D<d,t> and D<<d,t>t> and the arrows
represent mapping relations between these domains.
<<d,t>t>

<d,t>
Figure 1. Type shifting principles

Bare predication

107

I refer readers to Partee ([1986] 2004:193) for details about how these
type-shifting principles work exactly. Here, I give an illustration of the
central type shifting principle used in Partees analysis of unambiguous be:
the principle shifting entities of type d to entities of type <d,t>. The basic
type for proper names is type d. In order to occur in a predicative position,
such as a post-copula position, proper names must be type-shifted from d to
<d,t>. Thus, the copula itself doesnt do anything semantically, it can be
treated as the identity function on predicates: P.P of type <<d,t>,<d,t>>.
The requirement that its complement is of type <d,t> triggers the typeshifting operation which lifts the DP from type d to <d,t>.
Partee proposes the type shifting operation IDENT: IDENT(a)=x.x=a.
Namely, IDENT maps each entity a of type d onto the set of entities identical to a, that is, the singleton set {a} (or the property of being a). Look at
the derivation in (30):
(30) a. Cicero is Tully.
b. NTullyN= tully, NCiceroN= cicero, NbeN= P.P
c. IDENT(tully)=x.(x=tully)
d. Nbe TullyN=NbeN( IDENT(NTullyN)) 
= P.P( x.(x=tully) )
=x.(x=tully)
e. NCicero is TullyN=Nbe TullyN(NCiceroN)
=x.(x=tully) (Cicero)
=(cicero=tully)
(30) illustrates how identity sentences can be treated as predicational
constructions, where definite NPs are assigned predicative readings
through type lifting.
According to Partee (1986), specificational be can be treated in the
same way as the identity be. She assumes that specificational sentences
are some kind of identity sentence in the sense that specificational sentences like (27b) perhaps assert identity of properties: the property that is
what John is is the property of being unusual(Partee 1986/2004: 198).
In short, in Partees account, there is a single be, predicational be,
which takes two arguments, one of type d and one of type <d,t>. The examples of identificational be in (28) and identity be in (29) both involve
the be of predication, but one of the argument is shifted from its basic entity-denoting interpretation to an identity predicate of the form x.(x=a)
(cf. Partee 1998:370).

108

Chinese bare nouns

In sections 5.2 to 5.4, I will discuss three types of copula sentences in


terms of the denotation of the post-copula nominal: sets of individuals, sets
of subkind entities, and kinds.

5.2. Post-copula bare nouns as predicates of individuals


In the first type of copula clauses, the subject must denote an individual of
type e, which can be a proper name, a pronoun or a demonstrative, and the
post-copula bare noun is predicative and denotes a set of individuals, as
shown in (31).
(31) a. Xiaowang shi zhongguo-ren.
Xiaowang be Chinese-man
Xiaowang is a Chinese.
b. tamen shi
xuesheng.
they
be
student
They are students.
c. zhe-xie shi
xiangjiao.
those be
banana
Those are bananas.
In (31a), the bare noun zhongguo-ren Chinese-man does not denote
the kind Chinese but the property of being an individual Chinese person.
(31a) means that the individual Xiaowang has the property of being a Chinese, i.e. an instantiation of the kind Chinese. The bare nouns in (31b) and
(31c) can be interpreted in a similar way.
I can unproblematically coordinate predicates denoting properties of individuals, such as hen nianqing very young or hen reqing very enthusiastic, with these bare noun predicates:
(32) a. Xiaowang shi zhongguo-ren erqie hen nianqing.
Xiaowang be Chinese
and very young
Xiaowang is a Chinese and very young.
b. tamen shi xuesheng erqie hen reqing.
they be student and very enthusiastic
They are students and very enthusiastic.
Chinese bare nouns can be taken as a complement by a classifier, and
the classifier phrase of Num+Cl+N can be used as a predicate in the copula

Bare predication

109

clause. However, in this case, the bare noun can only be the complement of
an individual classifier and not of a kind classifier, as shown by the contrast between (33a) and (33b).
(33) a. Xiaowang shi yi ge zhongguo-ren.
Xiaowang be one CL Chinese
Xiaowang is a Chinese.
b.# Xiaowang shi yi
zhong zhongguo-ren.
Xiaowang be one CLkind Chinese
As (33a) and (33b) show, bare nouns in predicative positions in (31)
denote sets of individuals, not kind terms or sets of subkind entities.
However, the fact that (31a) and (33a) are synonomous, does not mean
that the bare noun has a classifier predicate as part of its structure. I argued
in Section 3 that bare nouns with an indefinite interpretation should be
distinguished from true indefinite phrases such as NumPs, because they
exhibit different scope behavior with respect to intensional operators and
quantifiers.
I now show another difference between bare nouns with an indefinite
interpretation and true NumPs, a difference which shows up in predicative
position. According to Aoun and Li (2003), two bare nouns can be conjoined to become a complex bare predicate expressing a complex property
of individuals, whereas two NumPs cannot be coordinated to form a complex predicate. The examples in (34) are from Aoun and Li (2003).
(34) a. Laowang shi fu-xiaozhang jian mishu.
Laowang be vice-President and secretary
Laowang is vice President and secretary.
b.*Laowang shi yi ge fu-xiaozhang jian mishu.
Laowang be one Cl vice-President and secretary
Intended: Laowang is a vice president and secretary.
c.*Laowang shi yi ge fu-xiaozhang jian yi ge mishu.
Laowang be one Cl vice-President and one Cl secretary
Intended: Laowang is a vice President and a secretary.
In (34a), the subject Laowang has the property of both being a vice
President and being a secretary. In other words, a single individual instantiates two different properties or two different roles at the same time. However, such dual properties or dual-roles cannot be expressed by coordinating two NumPs (34b) or coordinating a NumP with a bare noun (34c).

110

Chinese bare nouns

Aoun and Li (2003:143) propose that the connective jian and can only
coordinate property- denoting nominals.
I show now that the copula shi be in Mandarin is at type
<<d,t>, <d,t>>, which requires a predicative interpretation of bare nouns,
e.g. an interpretation at type <d, t>.
When taking a broad look at Chinese copular constructions, one observes that Chinese copula shi be can not only take nouns as complements, as in (35), but also adjectives, as in (36a). In these cases, the copula
cannot be omitted.5
(35) a. Moby Dick *(shi) jing.
Moby Dick be whale
b. Moby Dick *(shi) yi
tiao
jing.
Moby Dick be one CL
whale
Both: Moby Dick is a whale.
(36) a. tian *(shi) lanse-de.
sky be blue
The sky is blue.
In contrast with (36a), adjectives can appear in predicative positions
without the copula if they are modified by a degree modifier, such as hen
very, as in (36b).
b. tian hen lan.
sky very blue
The sky is blue.
The difference between (36a) and (36b) with respect to the presence of
the copula does not concern us here. The fact of (36a), in connection with
the other facts, suggests that the interpretation of the copula shi be is like
that of be in English: the identity function P.P of type <<d,t>,<d,t>>.
Given that the copula shi is of type <<d,t>, <d,t>> and that Chinese bare
nouns denote kind terms, I suggest that the bare noun, when it occurs in
post-copula position must be type-shifted from its basic kind reading at

Note that the adjective blue can either be expressed by lan or lanse-de in Mandarin. The sentence in (36a) is not a cleft sentence, which is usually expressed by
shi de.

Bare predication

111

type k to the predicative reading, at type <d, t>, to fit into the predicative
position in copula clauses.
According to Chierchia (1998a, b), the kind denotation of nouns has a
corresponding predicative counterpart. Chierchia (1998a, b) argues that
this predicate is recoverable from the kind via a version of the Carlsonian
instantiation relation. He proposes that the predicate formation operation
>
maps each kind onto a denotation of type <d,t>: the set of instantiations
of the kind. Chierchia (1998b) defines the >operator as follows:
(37) > is the function from kinds to sets of individuals, such that for every
kind
> ?
( Pw) = {x: x YDPw}
= Pw
Let k be a kind. Then >k = x.x Y k. This means that >k is the set of
parts of k. Since the kind ?P is identified with the maximal entity in Pw:
>?
Pw is the set of parts of WDPw, x.x YD WDPw, which is just Pw. Hence,
>?
Pw=Pw.
Kinds and predicates can be seen as two modes of packaging the same
information. The working of > and ? is visualized in the picture below. I
assume that the property is interpreted in the actual world w.
(38)

Properties
{a, b, c}

Kinds
?

Pw

{a,b}{b,c} {a,c}
a b c

>

aWbWc

Pw

Following Chierchia (1998b), I propose that the type shifting operation


is used to shift kind denoting bare nouns in post-copula positions from
the kind denotation to the predicative interpretation, as the set of
instantiations of the kind.

>

112

Chinese bare nouns

(39) a. Moby Dick shi jing.


Moby Dick be whale
Moby Dick is a whale.
b. NjingN= ?whale
[Kind interpretation]
NMoby-DickN= moby-dick
[Shifted predicative reading]
c. SHIFT(NjingN)= >?whale
=x. x Y W(WHALE<d,t>)
=WHALE<d,t>
d. NshiN(SHIFT(NjingN)) = P.P(WHALE)
=WHALE
e. NshiN(SHIFT(NjingN)) (NMoby-DickN)=WHALE(moby-dick)

5.3. Post-copula bare nouns as predicates of subkinds


I now discuss the second type of copula clause, in which the post-copula
bare noun denotes a set of subkinds and its subject denotes a kind.
The characteristic feature of this type of copula clause is that the subject and the predicate stand in a taxonomic relation, as exemplified in (40).
(40) a. haitun shi
jing.
dolphin be
whale
Dolphins are whales.
b. jing
shi
dongwu.
whale be
animal
Whales are animals.
In a biological taxonomy, the kind Dolphin is a subkind of the kind
Whale. The kind whale includes different subkinds of whales, like Blue
Whales, Dolphins, Sperm Whales, Porpoises etc. So the kind Dolphin instantiates a subkind of Whales (also see Krifka et al 1995).I propose that
(40a) expresses this kind of instantiation relation, where the predicative
noun jing denotes a set of subkinds of whales and the dolphin kind expressed by the subject haitun is a member of that set. In other words, the
whale kind is instantiated by the haitun , which represents itself as a subkind of whale.

Bare predication

113

The subkind instantiation interpretation of (40) is further supported by


the contrast in (41), where the bare noun can only be preceded by a kind
classifier zhong kind but not an individual classifier, tiao tail.
(41) a. haitun shi
yi zhong jing.
dolphin be
one CLkind whale
Dolphins are whales.
b.# haitun shi yi tiao jing.
dolphin be
one CL whale
Again, I do not assume that the bare noun jing whale in (40a) includes
a null classifier phrase as part of its structure, I am just concerned with
evidence for the subkind instantiation relation, and use the synonymy between (40a) and (41a) and the infelicity of (41b) as evidence for that.
This type of copula clauses is in many respects like the first type, discussed above. Both types can be characterized as predicational sentences
according to Higginss (1973) classification. In other words, in both types,
bare nouns express properties of the entity referred to by the subject. What
is different is that in the first type, the subject denotes an individual and the
post-copula bare noun denotes a set of individuals of type <d, t>, while in
the second type the subject denotes a subkind in a taxonomic classification,
and the post-copula bare nouns denote sets of subkinds of a kind of type
<k, t>.
Above I gave Chierchias predicate formation operation >which maps a
kind ?Pw of type k, onto the set of individual instantiations of the kind of
type <d, t>, i.e. >?Pw, which is Pw. I propose that the subkind predicate interpretations are derived through a similar operation W mapping a kind ?Pw
of type k, onto the set of subkinds of type < k, t>, i.e. W?Pw. To put it explicitly, I propose that the set of subkinds denoted by a bare noun is derived via the W operator from non-overlapping subsets of Pw. (This means
that I am here only dealing with relating a kind to a single level of nonoverlapping subkinds.)
Let (Pw) be a partition of the set of atoms in Pw. In the case of natural
kinds like whales, I assume that is derived from real world knowledge
about biology. Let SK(?Pw)={WDX:X(Pw)}, the set of sums of the
blocks of the partition . These sums are (extensionally) the subkinds of
?
Pw, as illustrated in (42).
(42)

W?

Pw = SK(?Pw)

114

Chinese bare nouns

Thus, W applies to a kind ?P and gives you the set of subkinds of ?P according to partition . Note that by the partition, any two subkinds k and k
in the set W?P are non-overlapping. I can introduce the subkind relation
itself, as given in (43):
(43) k1Yk2 iff k1SK( k2)
(44) illustrates the derivation of the subkind predicative interpretation.
(44) a. haitun shi jing.
dolphin be whale
Dolphins are whales.
[kind interpretation]
b. NjingN=?whale
NhaitunN=?dolphin
c. SHIFT(NjingN)=W ?whale [subkind shifted predicative reading]
=SK(?whale)
= k.kY?whale
d. NshiN(SHIFT(NjingN))= P.P(k.k Y?whale)
= k.kY?whale
e. NshiN(SHIFT(NjingN)) (NhaitunN)=k.kY?whale (?dolphin)
=?dolphinY?whale
5.4. Post-copula bare nouns as predicates of kinds
I now proceed to the third type of copula clauses, where both the subject
and the post-copula bare noun are interpreted as kind terms, as in (45).
(45) a. xihongshi jiu shi fanqie.
love apple FOC be tomato
Love apples are tomatoes.
b. fanqie jiu
shi xihongshi.
tomato FOC be love apples.
Tomates are love apples.
This type of copula sentence is unlike the types of the copula clauses
that were discussed above. The post-copula bare noun denotes neither a set
of individuals, nor a set of subkinds of entities. To see the difference, note

Bare predication

115

that in this case I can neither insert an individual classifier nor a kind classifier before the post-copula bare noun:
(46) a.# xihongshi
love apple
b.# xihongshi
love apple

shi yi
zhong fanqie.
be one CLkind tomato
shi yi
ge
fanqie.
be one CL tomato

I claim that copula clauses like in (46) are identity sentences, in which
the copula shi be takes two kind terms as arguments, i.e. k1= k2. As Wiggins (1965) says, for a sentence to express an identity, is or = must
stand between two noun-phrases which, if they are distinct, are serving
independently of one another to make genuine references (see Higgins
1973:262). By this criterion, the evening star is the morning star and
Hesperus is Phosphorus do express genuine identities. Along this line, I
suggest that kind terms, being names of kind entities, are also able to make
genuine references and to establish an equational relation.
The question is how the kind term at type k can fit into a predicative
position, i.e. the post-copula position. As I saw above, Partee (1987) discusses proper names in copula sentences and proposes lifting with the type
shifting rule: IDENT: d<d,t>, where IDENT[]=x.(x=).
In our case I am dealing with bare nouns of type k. The only assumption I need to make is the plausible assumption that IDENT can lift expressions from type k to type <k, t>. Namely, IDENT: k <k,t>, where
IDENT[] = xk.(xk=).
In this case, I assume that the copula is P.P, of type <<k,t>,<k, t>>.
(47) a. xihongshi jiu shi fanqie.
love apple Foc be tomato.
Love apples are tomatoes.
b. N fanqieN= ?tomato
N xihongsiN= ?love-apple
NshiN= P.P
c. IDENT(?tomato) = xk.(xk =?tomato)
d. Nshi fanquiN= NshiN( IDENT(NfanguiN) )
=P.P( xk.( xk =?tomato) )
=xk.(xk =?tomato)
e. Nshi fanquiN(N xihongshi) = xk.(xk =?tomato) (?love-apple)
= (?love-apple = ?tomato)

116

Chinese bare nouns

(47) expresses that the kind love apple is identical to the kind tomato. It
is an identity sentence, where the copula shi be connects two kind terms.
The two bare nouns are interchangeable with each other: the kind love
apple is identified with the kind tomato in (45a) and the kind tomato with
the kind love apple in (45b). This type of copula clauses is commonly used
to introduce new things or new names of entities to people who do not
know them. For example, people may know what fanqie is but they do not
know xihongshi, so by saying (45a), the two are identified.
To summarize so far, in this section, I have examined the interpretational variability of Chinese bare nouns in post-copula positions. They
have at least three different interpretations: sets of individuals, sets of subkinds or kinds. I argued that the kind reading is the basic reading of Chinese bare nouns, and that in all three cases, predicative interpretations are
derived from the kind interpretation by natural type-shifting operations. I
formulated these three operations. The copula shi is just interpreted as the
identity function. Its only semantic function is to trigger the appropriate
type shifting operation.

6. Definite bare nouns


In sections 4 and 5, I discussed the kind reading and the predicative use of
bare nouns. I now move on to the third reading of bare nouns discussed
above, the definite reading. I will not offer a semantic analysis of definite
readings of bare nouns because this would take us too far beyond the research questions I am dealing with in this book, namely, the syntax and
semantics of classifiers. But to complete my survey of bare noun readings,
I will try to show that definite readings of bare nouns are contextually determined. They occur in topic positions and in canonical object positions,
where, for example, they can have an anaphora use or an immediate situation use. I will suggest that definite readings of Chinese bare nouns are
derived by intersecting a predicate derived from the kind term with a contextually determined predicate C, which expresses familiarity.

6.1. Topic-hood and definiteness


The definite reading of Chinese bare nouns has been discussed extensively
by Chinese linguists in the past decades, including Chao 1968, Li and
Thompson 1976, 1981, Paris 1981, D.X. Zhu 1982, Huang 1982, L. D. Li

Definite bare nouns

117

1985, D.X. Shi 1992, Tsai 1994, Yuan 1996, Xu and Liu 1998, Cheng and
Sybesma 1999, R. Yang 2001, Y.Z. Shi 2002 and many others. It is beyond
the scope of the book to review all the relevant literature. In what follows, I
will synthesize some of the findings made by those linguists to give us
some idea under what condition or in what contexts the definite reading of
bare nouns is available.
Assuming that the grammatical meaning of subject and predicate in a
Chinese sentence is topic and comment, Chao (1968: 76) claims that there
is a very strong tendency for the subject to have a definite reference and
the object to have an indefinite reference. Since the subject sets the topic
of the talk and the predicate gives the information by adding something
new, the subject is likely to represent the known while the predicate introduces something unknown (ibid). The following pattern exhibited in
(48) and (49) is noted by Chao (1968: 76).
(48) a. wo yao qing ke.
I want invite guest
I want to invite guests.
b. ke
lai
le.
guest come PRF
The guest has come.
OR The guests have come.
(49) a. nar you shu?
where there-be book
Where are there books? / where is there a book?
b. shu zai nar?
book at where
Where are the books?/ where is the book?
As we can see from (48a) and (49a), when the bare noun is a postverbal
position, it has an indefinite reading, but when the bare noun is in a preverbal position, it has a definite reading, as in (48b) and (49b).
Li and Thompson (1981: 86) argue that nouns that are unmarked for
definiteness are always interpreted as definite or generic when they are
topics... In (50), the bare noun gou dog is the syntactic object of the verb
kan see, and it is also the topic of the whole sentence. It can either have a
definite reading (50a) or a kind reading (50b), but not an indefinite reading
(50c).

118

Chinese bare nouns

(50) gou wo kan


guo
le.
dog I
see
EXP PRF
a. The dog I have already seen.
b. Dogs (generic) I have already seen.
c. Impossible: A dog I have already seen.
Note that Chaos notions of topic and subject are different from Li and
Thompsons. For Chao (1968), most Chinese sentences consist of a subject
(the first NP) and a predicate, but the meaning or the function of the subject and the predicate is topic and comment respectively. According
to Li and Thompson, subject and topic are two distinctive syntactic elements in a sentence, though subjects are in many cases topics (see Li and
Thompson 1981: 94).
I adopt Li and Thomsons uses of the notions topic and subject, and I
consider them to be two different syntactic positions in a sentence (cf. (50).
Accordingly, in Chaos examples, the topic position happens to be the subject position of the sentence. So the first nominal phrase in (48b) and (49b)
are not only subjects but also topics, where the bare noun has a definite
reading. Therefore, the definite reading of the subject can be attributed to
its topic use. In Li and Thompsons examples in (50), the first NP is the
sentential topic but it is the object, and not the subject of the sentence, and
it also has a definite reading.
The definite reading of bare nouns is not restricted to default sentential
topic position. It is also available in secondary topic positions, as in (51)
(Y.Z. Shi 2002: 27).
(51) a. wo shu
yijing
kanwan
I book
already read-finish
I have finished reading the book.
b. ta yifu
yijin
maidao le.
she clothes already buy
PRF
She has already bought the clothes.
c. women fan yijin
zhunbei hao
we
meal already prepare well
We have prepared the meal.

le.
PRF

le.
PRF

The first nominal phrase in the examples in (51), wo I, ta she and


women we are the subjects of the sentences. If I assume that the sentential
initial element is the topic, then the subjects in (51) are also the default
topics of the sentences. The nominal phrase immediately following the

Definite bare nouns

119

subject, shu book, yifu clothes and fan meal, are objects of the sentence, but they move from their base-generated postverbal object positions
to preverbal positions, such as the position between subject and verb. This
movement is a form of topicalization of the object. Semantically, these
objects are presupposed by the interlocutors to be known and they have a
definite interpretation. For example, the bare nouns shu, yifu and fan in (51)
mean the book, the clothes and the meal respectively. These expressions are called secondary topics, which stand in a certain relation with
the primary topic, like a possessive relation.
The examples in (52) illustrate the ba construction in Mandarin Chinese,
where bare nouns following ba have a definite reading.
(52) a. ta yijin
ba zuoye
zuowan
she already OM homework finish
She has finished her homework.
b. ta you
ba yifu
mai le.
she again OM clothes sell PRF
She sold her clothes again.
c. women yijin
ba
fan zuohao
we
already OM meal make
We have made the meal.

le.
PRF

le.
PRF

The ba construction is a construction found in Mandarin Chinese (similar constructions are found in other Chinese languages). Chinese is a SVO
language, in which objects usually follow the verb. Sometimes, if the sentence expresses a causative meaning and the object is definite, the object
can be preposed to a preverbal position and marked by the object maker ba ,
which originally means take or hold. For example, the objects after ba
in (52), zuoye, yifu and fan mean the homework, the clothes and the meal
respectively. The ba-construction is similar to the construction shown in
(52), in which the preposed object functions like a secondary topic (Li and
Thompson 1981).
I can thus make the following generalization: when a bare noun in Chinese functions as a primary topic (sentential initial position) or secondary
topic (including BA constructions), it always has a definite reading.

120

Chinese bare nouns

6.2. Definite bare nouns in object positions


Even though definite bare nouns are frequently found in the (secondary)
topic positions, it is also possible to find definite bare nouns in canonical
object positions, where they are not topics or secondary topics.
For example, as we saw in (18) (repeated as in (53)), the bare noun xin
is the object of the verb kan read and can either be interpreted with a
narrow scope reading with respect to every, meaning letters, as in (53a)
or with a definite reading, meaning the letter, as in (53b).
(53) mei-ge
ren dou zai
kan guanyu jiaxin
de xin.
every-CL man all PROG read about add-wage Mod letter
a. Everybody is reading letters about raising the salary.
b. Everybody is reading the letter about raising the salary.
Some more examples of definite bare nouns are given in (54).
(54) a. wo yijin
xie
le xin
le.
I already write PFV letter PRF
I wrote the letter already.
b. ta he-wan
yao
le.
he drink-finish medicine PRF
He finished drinking the (Chinese herb) medicine.
According to Li and Thompson (1981:86), nouns that are unmarked
for definiteness are always interpreted as definite or generic when they are
topics... and definite bare nouns in object positions, as those in (53) and
(54), are a marked case. In contrast, I assume that there are two types of
definiteness involved here, or rather the definiteness of bare nouns comes
from two different sources. I propose that the definiteness of bare nouns in
topic positions is due to their topic status (a syntactic position for R. Yang
2001), where they refer to the entity to be talked about or commented on.
As definite objects, the bare nouns usually refer to entities that are (perceptually) visible in an immediate situation or particularly salient or familiar
in the discourse context. For example, the definite interpretation of (53)
would be appropriate if the bare object xin letter is used to refer to some
particular letter, which is particularly salient in the local context, or which
had been the topic of previous discourse. The English equivalent might be
Did you see/write that letter, using the demonstrative. Here that letter
has an almost anaphoric use.

Definite bare nouns

121

It is important to distinguish these two types of definite bare nouns, i.e.


definite bare nouns as topics versus definite bare nouns in canonical object
positions. In the case of topics, the definiteness follows because the topic is
assumed to be part of the background knowledge of the discourse participants, and in the less frequent case of direct objects, the definiteness relies
on the discourse or perceptual salience in the immediate situation. I thus
propose that in both cases, the bare noun in the definite reading refers to a
familiar and salient individual under discussion in the context.

6.3. Semantics of definite bare nouns


As argued earlier, Mandarin bare nouns denote kinds. Their indefinite
readings in object/predicative position are derived from a local Carlsoniantype shift reading, as shown by the scope facts in section 3. I have just
showed that Mandarin bare nouns also have a definite reading, when they
denote a specific, salient and contextually familiar entity. This occurs
mainly when a bare NP is in topic position, as in section 6.1 and occasionally when it occurs in direct object position, as in section 6.2. In this subsection, I will not discuss in detail how these readings are derived, but
merely suggest some possible semantic derivations.
Chapter 9 will discuss definite classifiers in Cl+N. I will argue, following Li and Bisang (2012), that definite interpretations of Cl+NP in Chinese are not entities postulated to be unique, but are familiar entities approximately in the sense of Robertss (2003) of weak familiarity. I will
argue there that the Cl+N in Wu and Cantonese, in the definite reading,
denotes a generalized quantifier with existential force la Landman (2004).
I assume that the definite interpretation of bare nouns is derived in more
or less the same way. I assume here that in their definite reading, bare
nouns shift from the kind denotation to an indefinite reading in the normal
way, but that discourse constraints on NPs in topic positions or NPs used
anaphorically/situationally force a familiarity constraint in the interpretation of the bare noun, which leads to the definite or familiar interpretation.
Thus the predicate interpretation of the N is not merely a set of instantiations of the N-kind, but a set of contextually familiar and salient instantiations of the kind denoted by N.
This is supported by R.Yangs observation about definite and indefinite
interpretations of bare nouns in the subject position. She claims that bare
nouns in subject position usually have a definite interpretation, because
subject position is usually also topic position, and topics are required to be

122

Chinese bare nouns

salient or familiar. However, if subject position is a position which cannot


be topic, the indefinite reading peeks out, that is, it is available when the
definite reading is not induced. She offers the following as examples.
Firstly, indefinite readings become immediately available for preverbal
bare NPs, once they occur in the context of left peripheral locatives, as
shown in (55):
(55) waimian / yuanchu gou zai-jiao.
outside far-away dog PROG-bark
a. Outside/Far away, dogs are barking.
b. Outside/Far away, the dog(s) is/are barking.
Secondly, indefinite readings seem to peek out in the context of leftperipheral temporal phrases, in addition to a definite reading, as in (56):
(56) a. jintian jingcha zhua ren
le.
today cop
arrest man PRF
i. Today cops arrested some people.
ii. Today the cop(s) arrested some people.
b. haoxiang jingcha zhua ren
le.
apparently cop
arrest man PRF
i. Apparently cops arrested some people.
ii. Apparently the cop(s) arrested some people.
Thirdly, as shown in (57), in the context of a universal quantifier in
everyones backyard, the subject bare NP dog unambiguously gets an
indefinite reading:
(57) gou zai meigeren-de houyuan-li
jiao.
dog at everyone-Mod backyard-inside bark
a. Dogs are barking in everyones backyard.
b. # The dog is barking in everyones backyard.
In Yangs account of the distribution of definite and indefinite readings
in subject position, the definite reading is induced by topic position and the
indefinite reading is possible only when the definite reading is not induced.
This is predicted by our tentative suggestion that definiteness is induced
via a contextually introduced salience-and-familiarity constraint on the
interpretation of the N: when context does not introduce this constraint, a
pure indefinite interpretation is possible.

Definite bare nouns

123

Within the scope of this book, I am not going to take the issue any further, and I leave further discussion of definite interpretations of bare nouns
for further research.

Part II:
Functions of classifiers: counting and measuring

Chapter 6
Counting and measure functions of classifiers

1. Introduction
In this chapter, I discuss the semantics of classifiers in Chinese. I am not
concerned with the lexical meanings of different classifiers, but with the
properties that characterize the interpretation of classifiers as a class, or put
differently, the interpretative function of classifiers as an independent category. Concerning the semantics of classifiers, I develop two arguments.
First, I propose that classifiers have two basic functions, counting and measuring. Secondly, I claim that the counting function and the measure function of classifiers are distinguishable at the level of syntax in Chinese.
In chapter 2, I introduced several lexical classifications of classifiers,
e.g. a distinction between classifiers and measure words (Tai and Wang
1990), between sortal and mensural classifiers (Lyons 1977, Tang 2005),
and between count and mass classifiers (Cheng and Sybesma 1998). I argue here that the semantic distinction between the counting use and the
measure use of classifiers is a more crucial and fundamental distinction
within the Chinese classifier system.
Previous lexical analyses of classifiers in Chinese assumed that it is the
semantic role of all classifiers to make explicit a set of relevant units in the
denotation of the noun in terms of which to count. In this, two groups of
classifiers are distinguished, based on two different ways in which those
units are determined: count classifiers or sortal classifiers are assumed to
pick out a set of units based on the inherent properties of the entities in the
noun denotation; mass classifiers or mensural classifiers are assumed to
impose a unit structure on the noun denotation (for discussion, see the literature cited above). Against this, I assume, following Rothstein (2009,
2010), that counting and measuring are two very different semantic operations, associated with two different syntactic structures, and that the crucial
distinction between classifiers is which of the operations they naturally
take part in. This distinction cuts across the lexical subclasses of classifiers,
although it is not unrelated to that classification, since some lexical classes

128

Counting and measure functions of classifiers

of classifiers have primarily the counting use, others have by default the
measure use, some have both, and some have neither.
To my knowledge, Krifka (1995) is the only published paper discussing
the formal semantics of classifiers in Chinese in a detailed way.
Krifka (1995:400) proposes that Chinese classifiers denote functions
that map a kind onto a measure which measures the quantity of specimens
of that kind by counting them. Using the rule of functional application, he
represents the semantics of classifiers as follows.
For Krifka, a classifier is a measure expression M which denotes a
function that applies to a number (the denotation of a Num phrase) to form
the interpretation of a measure phrase MP:
(i) N[MP[Num][M]]N = N[M] N (N[Num]N)
The measure phrase applies to a noun to form a noun phrase:
(ii) N[NP [MP ] [N]]N =N[MP]N (N[N]N)
Hence, on this analysis, the classifier combines first with the number to
yield a Measure Phrase, and then in stage two, the Measure Phrase, Num+
Cl, applies to the Noun. Thus, in san tou daxiang three heads of elephants, the classifier tou head is a measure head which combine first
with the number san three to give the measure phase san tou three head,
which then combines with the nominal head daxiang elephant, to give
three head of elephants. On this analysis, counting is a special kind of
measuring, and classifiers are unambiguously measures.
In this book, I will argue for a semantics on which counting and measuring are different semantic functions of classifiers. Rothstein (2010) proposes that English count nouns are derived from root nouns Nroot via an
operation COUNTk, which maps Nroot onto a set of atomic entities of type
dk, which count as one in the context k. I propose that Chinese classifiers,
on the counting interpretation, are the lexical realization of the operation
COUNTk. For the measure reading, I will use the work of Krifka (1995),
Chierchia (1998a) and Landman (2004) to formulate a semantics of measures in which measures are functions from kinds to sets of instantiations of
the kinds that have a certain measure value.
I take the features [Counting] and [Measure] as features constraining
the default interpretations of classifiers. Four types of classifiers can be
distinguished:
[+C, -M] are classifiers that are by default counting operators;
[-C, +M] are classifiers that are by default measuring operators;
[+C, +M] are classifiers that can naturally be used as either operation;

Counting and measuring readings: a crosslinguistic perspective

129

[-C, -M] are kind classifiers, whose interpretation falls outside the operations of counting and measuring in the domain of individuals.
In this chapter, these features are still purely classificatory. Later in the
book, we will see that the features reflect information in the lexical entry
of the classifier.
In section 2 of this chapter, I introduce the counting and the measure
functions of classifiers with some cross-linguistic data. In sectin 3, discusses the counting and measure functions of container classifiers in Mandarin. A syntactic distinction of these two readings will also be offered. In
section 4, I propose the feature system [Counting, Measure] for classifiers. Section 5 gives the semantics of the counting and measure functions of
Chinese classifiers. Finally, this chapter is summed up in section 6.

2. Counting and measuring readings: a crosslinguistic perspective


This section discusses the two basic funcitions of classifiers, counting and
measuring. I will show that the distinction between counting and measuring readings of (container) classifiers is a grammatically relevant phenomenon cross-linguistically. Section 2.1 presents crosslinguistic facts about
the contrast between counting and measuring readings. Section 2.2 proposes the possible syntactic structures for these two readings.

2.1. Introducing counting and measuring readings


Many scholars (e.g. Selkirk 1977, Jackendoff 1977, Corver 1998, Doetjes
1997, Landman 2004, Rothstein 2009, Partee and Borschev 2012, among
others) have observed that container classifier phrases, such as three bottles of water in English, are ambiguous between a counting reading and a
measure reading (or between a container and a quantity/pseudo-partitive
reading in Selkirks 1977 or Jackendoffs 1977 terms). For example, on the
counting reading, the word bottle in (1) refers to concrete bottles, while on
the measure reading, bottle denotes an abstract measure unit and the concrete container does not have be present:
(1)

English
a. John carried three bottles of water home.
b. I poured three bottles of water into the soup.

[Counting]
[Measure]

130

Counting and measure functions of classifiers

The verb carry in the example of (1a) forces a counting reading: carrying requires the water to be carried in something, a container, hence (1a)
involves three concrete bottles, which are filled with water. On the other
hand, the preferred reading of (1b) is a measure reading, where what is
poured in the soup is not three bottles, but a quantity of water amounting to
three bottles, and it is irrelevant whether the pouring is done from three
concrete bottles of water or whether the same amount is poured from a jug.
Pustejovsky (1993) introduced the notion dotted type to represent
the sort of an expression that simultaneously incorporates two distinct
sorts: in a case like this, one might argue that we can refer simultaneously
to the container and the substance contained in it (cf. Partee and Borchev
2012: 458). In the current research, I do not treat the dotted type as a seprate type, but as a counting reading. The phrase refers to a complex entity,
i.e. a container contained with stuff. It is a counting reading.
The two-way contrast of container classifiers between counting and
measuring readings is found crosslinguistically. The examples in (2) and (3)
give an illustration of the Russian and the Hebrew cases respectively.
According to Partee and Borschev (2012), in Russian, the noun butylku
bottle can either refer to the container itself or an abstract measure unit.
Speficially, accompanying verbs may select for one or the other, as in (2):
(2a), with break, is most likely understood as referring to the container,
and example (2b), with drink, favors reference to the content and the
container bottle does not have to show up physically. The modifier pol
half also forces a measure reading, in that it only refers to half of a designated quanity but not a broken bottle.
(2)

Russian (Partee and Borschev 2012: 458-459)


a. My razbili
butylku
ampanskogo.
we
broke
bottle-ACC.SG
champagne-GEN.SG
We broke a bottle of champagne.
b. My vypili
pol-butylki
ampanskogo.
we
drank
half-bottle-ACC.SG
champagne-GEN.SG
We drank half a bottle of champagne.

Rothstein (2009) argues that the counting and measure readings are expressed by two different constructions in Modern Hebrew. The Free Genitive construction only has a counting reading (as in 3a), while the Construct State construction is ambiguous between a counting reading and a
measure reading (3b). This suggests that in Hebrew, the contrast between
counting and measuring readings is syntactically encoded.

Counting and measuring readings: a crosslinguistic perspective

(3)

a. (alo) kosot
el mayim
(three) cup(f.pl.) of water
b. (alo) kosot
mayim
(three) cup(f.pl.) water
Both: three cup s of water

131

[Free Genitive: Counting]


[Construct Strate: Count or Meas]

As Rothstein argues, if only the the measure reading is appropriate, only


the construct state construction can be used. Thus in recipes, where only
the measure reading is appropriate, the free genitive construction with
el/of is impossible:
(4)

mosifim tey kosot


(#el) kemax ve-alo kapiot
(#el)
add
2-f cup(f.pl.) ( #of ) flour and 3(f ) teaspoon(f.pl) (#of )
sukar la-batzeku-mear-be-vim.
sugar to-DEF-dough-and-mix
Add two cups of flour and three teaspoons of sugar to the dough and
mix.

Back to the English case. Rothstein (2009) argues that, even though
there is no grammatical encoding of the distinction between counting and
measure readings in English, it is a real semantic distinction. There is
enough morphosyntactic evidence to draw these two readings apart. The
following observations are due to Rothstein (2009: 110-111).
Measure suffix: On the measure reading, the suffix -ful can often be
added to the classifier; with the suffix, the classifier cannot express an
individuating reading.
(5)

a. Add two cup(ful)s of wine to the soup.


b. Bring two cup(#ful)s of wine for our guests.
c. We needed three bucket(ful)s of cement to build that wall.
d. Three bucket(#ful)s of mud were standing in a row against the wall.

Pronominalization: Plural individuating classifiers can naturally be antecedents for individuating pronouns; measure classifiers cannot:
(6)

There are two cups of wine on this tray.


a. They are blue.
b. They (each) contain 100 milliliters.
c. They (each) cost 2 Euros.

132
(7)

Counting and measure functions of classifiers

There are two cups of wine in this soup.


a. #They are blue.
b. #They (each) contain 100 milliliters.
c. #They (each) cost 2 Euros.
d. It adds flavour/??They add flavour.

Another relevant fact is concerned with number agreement between


number and noun. Corver (1998) argues that in the construction of NumN1-of-N2, under the counting readings, the verb agrees with N1 and on the
measuring reading, the verb agrees with N2. The example in (8) can only
be interpreted with a counting reading, the copula must be used in its plural
form to agree with the noun cups.
(8)

a. Two cups of wine are emptied.


b.# Two cups of wine is emptied.

Taking into account the fact of number agreement, the infelicity of (7)
may be partly due to that the plural pronoun they used. They in the second
clause is supposed to be anaphoric with the plural noun cups but not with
wine in the first clause, which forces a counting reading. It is thus contradictory with the measure context construed in the first clause.
Distributivity: the distributive operator each can distribute to the individuals in the denotation of individuating classifier expressions, as in (9a),
but is infelicitous with measure phrases, as in (9b-c).
(9)

a. Two packs of flour cost 2 euros each.


b. #Two kilos of flour cost 2 euros each.
c. The two cups of wine (#in this soup) cost 2 euros each.

To sum up, the semantic distinction between counting and measure


readings of classifiers is attested across a wide spectrum of typologically
different languages, which obviously can differ considerably in how the
distinction is reflected in the language.
The relevant facts about the ambiguity of Chinese container classifiers
will be discussed separately in section 3.

Counting and measuring readings: a crosslinguistic perspective

133

2.2. Structures for counting and measuring readings


It is controversial whether the ambiguity of container classifiers between
counting and measure readings is purely a lexical ambiguity or a syntactic
issue. Stavrou (2003) attributes this contrast to the semi-lexical status of
container classifiers. Specfically, on the container reading, it is lexical,
while on the quantity reading, it is (semi-)functional. However, for most
researchers, the counting reading and the measure reading of classifier
phrases in English are two distinctive semantic functions, which are assocaited with different syntactic structures (Rothstein 2009, also see Selkirk
1977, Lbel 1989, Corver 1998, Landman 2004).
Selkirk (1977) is among the first to recognize the two readings of constructions like a bottle of water in English and posits two distinctive structures for these two readings. On the pseudo-partitive reading (i.e. quantity
reading), a bottle is base-generated in the lower specifier position within
the noun phrase, as illustrated in (10a). On the container reading, bottle is
projected into the head of the nominal phrase and takes the preposition
phrase of water as its complement, as shown in (10b).
(10) a. [NP [N [NP a bottle] of [N water]]]
b. [NP a [N [N bottle] [PP of [NP water]]]]]

(quantity reading)
(container reading)

In the post-Abenian syntactic framework, the two structures given by


Selkirk in (10) have to be represented in terms of functional projections,
such as DP structures. Abeny (1987) did not recognize the ambiguity of the
pseudo-partitive construction and proposed a single DP structure for the
phrase a bottle of water: [DP a [NP bottle [PP of water]]]. In fact, Abenys
structure only corresponds to (10b) on the container reading, but not to
(10a) on the quantity reading.
Corver (1998) proposes two structures for the counting and the measuring readings of container classifiers. On both counting and measuring readings, the pseudo-partitive construction is involved with predicate movment. To put it in a simpe way, the phrase a box of apples has the the deep
structure [DP D [SC apples [Pred box]]], in which apples and box instantiate a
subject-prdicate relation in a small clause. The surface order is realized by
predicate movement, such as to raise the predicate box via XP-movment (in
English) or head movement (in Dutch). The crucial difference between
counting and measuring readings lies in the the predicative relation holding
between the subject and the predicate in the small clause: on the measuring
reading, the subject and the predicate is construed as a copula clause, and

134

Counting and measure functions of classifiers

their relation is represented by the copula BE, while on the counting reading, they embody a containment relation HAVE, which can be decomposed
into an incorporation of a preposition and be. That is, P+BE = HAVE. This
syntactic approach nicely captures the interpretational difference between
counting and measuring, but it fails to reflect the constituent relations between numeral and classifier. In Corver (1998), the numeral (or determiner)
always taken to be the head of an independent projection, e.g. D0 and it
does not form a single consitutent in either case. I think that this cannot be
correct.
More recently, Rothstein (2009), following Landman (2004), assume
that (11a) is the correct structure for the counting reading and (11b) is the
correct structure for the measure reading. Note that of insertion is presumed to be a late phenomenon taking place to satisfy surface constraints
and projecting no PP node.
(11) a. Counting reading
DP
D
threei

b. Measure reading
DP

NumP
Num
ti
N
bottles (of)

NP

NP
NP
water

MeasP

Num
NMeas
three bottles (of) water

The constituent relations represented in the structures in (11) are the


same as those encoded in (10). Therefore, they can be seen as a postAbenian version of Selkirks structures. As reflected in the different syntactic structures for counting and measure readings in (11), Rothstein
(2009) explicitly argues that on the counting reading, bottles is a relational
noun which denotes a relation between entities which are bottles and the
substance they contain. So the classifier bottles is the head of the NP and it
takes the noun water as its complement; the whole structure is the complement of the numerical Num, as in (11a). In contrast, on the measure
reading in (11b), bottles is a modifier which combines with the numerical
three to form a complex modifier three bottles, which then modifies the
nominal head water.
According to Partee and Borschev (2012), the counting and measuring
readings of elements like bottle can be seen as a sort of semantic shifts.
Container nouns have a basic sortal meaning, but they shift productively

Ambiguity of container classifiers in Chinese

135

to a relational reading when used in the Genitive of Measure construction


and may shift to a functional reading over time to gain an abstract measure
reading analogous to litr liter (ibid: 446). Along this line, bottle in the
structure (11a) can be seen to be shifted from a sortal noun into a relational
noun, and in (11b), shifted from a relational noun to a functional noun (e.g.
a unit of measure corresponding to the volume of the containee that the
container can hold).
In this research, I take the Selkirk-Rothsteinien structure, as shown in
(10) and (11), to be the correct structure for counting and measuring readings of classifiers.

3. Ambiguity of container classifiers in Chinese


In this section, I discuss container classifiers in Chinese. I present different
syntactic evidence to show that the distinction between counting and measuring readings shows up with Chinese container classifiers. Two different
syntactic structures for Chinese container classifiers on the counting and
measure readings will also be posited la Selkirk (1977) and Rothstein
(2009).

3.1. Counting and measuring readings for Chinese container classifiers


I now discuss container classifiers in classifier languages like Chinese. I
show that container classifiers in Chinese also express both counting and
measure readings in different contexts. Look at (12):  
(12) a. wo ling le
liang ping jiu, zuo shou yi ping,
I lift PFV two CLbottle wine left hand one CLbottle
you shou yi ping.
right hand one CLbottle
I carried two bottles of wine, one in the left hand and the other in
the right hand.
b. ta-de jiuliang
shi liang ping
hongjiu.
his drinking-capacity be two CLbottle red wine
His drinking-capacity is two bottles of red wine.  
The classifier phrase liang ping jiu two bottles of wine in (12a) has a
counting reading which involves two concrete bottles, one in the left hand

136

Counting and measure functions of classifiers

and one in the right. In contrast, in (12b), liang ping jiu can only be interpreted with a measure reading, which means that the maximal amount of
red wine that he can drink is two bottles. It means that he can drink 1500ml
of red wine if I assume each bottle is 750 ml. In this case, the real bottles
are not required to be present and the counting reading is not available.
The two readings of container ClPs are sensitive to different syntactic
contexts in Chinese. In other words, in some contexts some classifiers
strongly prefer one reading over the other and different classifiers in some
syntactic contexts prefer to have one reading over another. I discuss four
kinds of syntactic contexts.
First, the Cl+N construction is the construction where the classifiernoun phrase is used without a numerical. In this construction the classifier
can have a counting interpretation, but not a measure interpretation. Yip
(2008) uses this test to distinguish classifier like ge from true measure
words like mi meter. I extend Yips argument to distinguish between the
counting and the measure reading of [+Counting, +Measure] classifiers.
(13) a. wo ling le
ping jiu.
[Mandarin]
I carry PFV CLbottle wine
I carried a bottle of wine.
b.* ta-de jiuliang
shi ping
hong-jiu.
his drinking-capacity be CLbottle red wine
Intended: His drinking capacity is a bottle of red wine.
As shown in (13), the Cl+N construction is only possible in the
counting context, as forced out by the verb ling carry in (13a), but not in
the measure context of (13b). Moreover, in Mandarin, Cl+N is used in
postverbal position and has an indefinite reading. Ping jiu in (13a) can only
mean a bottle of wine.
However, in some Chinese languages, the Cl+N construction can
have a definite interpretation (see Shi and Liu 1985, Li and Bisang 2012
for Wu; see Cheung 1972, Cheng and Sybesma 1999 for Cantonese). The
data from Wu in (14) show that the definite Cl+N construction is possible only when container classifiers have a counting reading, as in (14a) and
not a measure reading, as in (14b). This holds for the Cl+N construction
in Cantonese (Yip 2008).

Ambiguity of container classifiers in Chinese

137

(14) a. o thi ko phi tiu i-kuo tinkhiu ko. [Wu:Fuyang]


I eat Mod CLbottle wine England import PRT
The bottle of wine that I drank was imported from England.
tiu te-le u-li
ta
sao la.
b.* o thi ko phi
I eat Mod CLbottle wine at stomach-in PROG burn PRT
Intended: The bottle of wine that I drank is burning in my stomach.
The second context concerns the the additive modifier duo more. If
duo occurs between the the numeral and the classifier, as in
Num+duo+Cl+N, the classifier can have both a counting and a measure
reading; if duo occurs between the classifier and the noun, as in Num
+Cl+duo +N, the classifier only has a measure interpretation.
The observation that duo can follow either the numeral or the classifier
was made in L (1980). The construction where duo precedes the classifier
prefers the numeral to express a round number like shi ten, bai hundred,
qian thousand etc., as in (15a). In the construction where duo follows the
classifier, the classifier is usually a container classifier, a standard measure,
and it is preferable for the numeral to be a cardinal below ten, as in (15b)
(L 1980/1999: 184, also see Zhu 1984).
(15) a. shi duo
feng xin
ten more CL letter
more than ten letters
b. liu chi
duo
bu
six CLinch more cloth
more than six inches of cloth
(16) shows that when duo comes between the numeral and the classifier,
the counting and the measure readings are equally available for the classifier phrase, as in (16a) and (16b) respectively. (17) shows that when duo
follows the classifier, the sentence strongly prefers a measure reading.
Compare (16) with (17).
(16) a. ta ling le shi duo ping
hongijiu.
[counting]
he carry PFV ten more CLbottle red wine.
He carried more than ten bottles of red wine.
b. ta zhishao neng hexia shi duo ping
hongijiu. [measure]
he at least can drink ten more CLbottle red wine
He can at least drink more than ten bottles of red wine.

138

Counting and measure functions of classifiers

In (16a), duo expresses that the number of individual bottles (filled with
red wine) is more than ten, i.e. (16a) has a counting reading. (16b) has a
measure reading: it expresses that the overall quantity of wine that his stomach can hold is more than ten bottles: here ping is understood to mean
bottleful, i.e. it has a measure reading.
(17) a. ta zonggong he le you san ping duo hongjiu. [measure]
he altogether drink PFV have three CLbottle more red wine
He drank more than three bottles of red wine.
b. # ta ling le shi ping
duo hongjiu.
he carry PFV ten CLbottle more red wine
Intended: He carried more than ten bottles of red wine.
In (17a), the most natural reading is the measure reading. It means that
the overall quantity of wine is larger than a certain value, e.g. three bottles.
The concrete bottles are irrelevant. Similarly, (17b) is infelicitous unless
you force it into the measure reading: pour the wine in a big container and
then carry it. Thus Cl duo N forces a measure interpretation on the classifier phrase.
The third context concerns the particle de. While Num-Cl-N can be ambiguous between a counting and a measure reading (as in 18a), the particle
de can induce a measure reading for some classifiers, in particular container classifiers, as Cheng and Sybesma (1998) argue (as in 18b). However,
the particle de can not occur between Num and Cl, as (18c) shows.
(18) a. san ping shui
[Counting or Measure: our observation]
three CLbottle water
three bottles of water
b. san
ping de shui [Measure: Cheng and Sybesma 1998]
three CLbottle DE water
three bottles of water
c.* san de ping shui
three DE CLbottle water
(19a) is a typical counting context, in which the verb kai open requires
concrete containers. It is impossible to have de after the classifier in this
case. In contrast, (19b), with the verb zhuang-xia contain/hold, talks
about the volume of the container and it is thus concerned with quantity of
wine only. It is a measure context and it can have an optional de.

Ambiguity of container classifiers in Chinese

139

(19) a. fuwuyuan kai


le
san ping (#de) jiu.
[Counting]
waiter
open PFV three CLbottle DE wine
The waiter opened three bottles of wine.
b. zhe ge peng keyi zhuan san ping (de) jiu.
[Measure]
this CL basin can contain three CLbottle DE wine
This basic can contain three bottles of wine.
The fourth context concerns reduplication of the container classifier. In
Chinese the container classifier can be reduplicated on the counting reading, as in (20), but not on the measure reading, as in (21).
(20) a. wo mai le
liang ping
jiu, ping-ping
dou hen gui.
I buy PFV two CLbottle wine CLbottle-CLbottle all very expensive
I bought two bottles of wine, each of which is expensive.
b. wo he le
liang ping jiu, ping-ping
dou hen haohe.
I drink PFV two CLbottle wine CLbottle-CLbottle all very delicious
I drank two bottles of wine, each of which tastes good.
(21) a.# zhe ge tong
zhuang le
san ping jiu,
this CL bucket contain PFV three CLbottle wine
ping-ping
dou hen gui.
CLbottle-CLbottle all very expensive
This bucket holds three bottles of wine, each of which is expensive.
b.# zhe ge tong
zhuang le
san ping
jiu,
this CL bucket contain PFV three CLbottle wine
ping-ping
dou hen haohe.
CLbottle-CLbottle all very delicious
This bucket contains three bottles of wine, each of which tastes
good.
As often noted in the literature, reduplicated classifiers in Chinese of
the form of Cl-Cl have distributive readings; they are usually translated as
each or every. Reduplicated classifiers obligatorily occur with the
distributive marker dou all (J.-W. Lin 1996, Cheng 2009). The reduplicated classifier, ping-ping bottle-bottle, is co-referential with the classifier phrase in the preceding clause. This means that in (20) the antecedent
ClP ling ping jiu two bottles wine must refer to concrete bottles of wine.
As a result, the distributive predicate, hen gui very expensive in (20a) and
hen hao-he taste good in (20b), applies to each bottle of wine (cf.

140

Counting and measure functions of classifiers

Schwarzschild 2009 stubbornly distributive predicates). In the measure


context of (21), the plural classifier phrase cannot act as an antecedent for
the reduplicated classifier, since the plural classifier phrase on the measure
reading expresses quantities of entities, and does not refer to a sum of
atomic entities.
We see then that the distinction between counting and measure readings
is a grammatically relevant phenomenon in Chinese, which shows up in a
variety of different syntactic environments.

3.2. The syntax of counting and measuring readings


In this subsection, I will argue that the syntactic structures assumed in
Rothsteins (2009) for counting and measure readings of classifiers in English are appropriate for Chinese classifier constructions as well. I will argue that on the counting reading, Chinese container classifier phrases have
the structure [Num [Cl+N]], and on the measure reading, they have the
structure [Num+Cl [N]].
What came out of the discussion of the different contexts in the previous section is that on the counting readings, Num and Cl behave like a
single constituent, while they do not, on the measure reading. Hence, we
propose the following two structures for the counting and the measure
reading for Chinese ClPs:
(22) a. Counting reading
NumP
Num
santhree

ClP

b. Measure reading
NP
ClP

ClCounting NP
Num
pingbottles shuiwater santhree

ClMeasure
pingbottle

shuiwater

There are two differences between our structures in (22) and those in
(11).
First, I do not analyze ping bottle as a noun but as a classifier. In chapter 2, I argued that Chinese classifiers constitute a category separate from
nouns, and that they have an independent projection of ClP. In this, Chinese classifiers differ from English ones which still retain nominal features.
Hence, in Chinese, the classifier always projects into be the head of the

Ambiguity of container classifiers in Chinese

141

functional projection ClPs, both in the counting and in the measuring structure.
Second, I do not posit a DP structure for Numeral Classifier phrases in
Chinese. There are obvious syntactic differences in NumPs between English and Chinese. Chinese NumPs cannot be used in argument positions
like the subject position, while English NumPs occur unproblematically in
subject position (cf. the discussion in Chapter 5):
(23) a. Three bottles of water stand on the table.
b.* san ping
shui
zai zhuo shang.
three Cl-bottle water at table on
Intended: Three bottles of water are on the table.
The ill-formed sentence (23b) can be made grammatical by inserting an
existential quantifier you there be at the beginning of the sentence, as in
(23c):
c. you
san ping shui zai zhuo shang.
there-be three Cl-bottle water at table on
There are three bottles of water on the table.
In view of these differences, I propose that NumPs in Chinese are always predicates, while NumPs in English can be arguments or predicates.
This means that it is possible for English NumPs but not Chinese NumPs to
have a DP level.
I now show how these two different structures in (22) are justified by
the four contexts I presented in Section 3.1.
The first fact presented was that in the Cl+N construction, the classifer
could only have a counting interpretation (as illustrated by the examples in
(13) and (14)). The structure in (22) accounts for this fact, because on the
counting reading the classifier heads the projection of ClP, which can then
be a complement of a higher functional projection NumP.
But the projection NumP does not have to be projected in all cases. (I
will argue in Chapter 9, that the indefinite phrase Cl+N has a maximal
projection ClP and has no NumP above it.) In this case, the cluster Cl+N
is interpreted as semantically atomic, and the singularity is implied by the
semantics of the cluster and not by an explicit number.
In contrast, on the measure reading, the numeral is part of the complex
modifier of Num-Cl; it is impossible to drop any internal element of the
complex modifier, and the whole constituent including the lexically rea-

142

Counting and measure functions of classifiers

lized Num is adjoined to NP. Num +Cl modify the NP, thus a string containing Cl+N as a constituent cannot be a realization of the measure
structure.
The second fact is that Num+duo+Cl+N has both counting and the
measure readings, while Num+Cl+duo+N has measure reading only, as
shown in (15)-(16).
The structures proposed readily provide an account for these facts. The
modifier duo is an additive modifier which means more and can only
modify constituents that express quantity (or degree). Numbers are obviously quantity expressions, but classifiers express units of counting or
measuring and do not themselves express quantity. NPs do not express
quantity either. I assume that this means that duo can be adjoined to Num
in both counting and measuring readings. Thus, when the word order is
Num+duo+Cl+N, with duo as a modifier of the Num, I can treat Num-duo
as the head of a counting structure: [[Num Num-duo] [Cl+N]], which leads to
a counting reading. Alternatively, I can let Num-duo form a modifier with
the classifier: [[[Num Num-duo] +Cl] N], which leads to a measure reading.
When duo occurs after the classifier, that is when the word order is
Num+Cl+duo+N, it can only be interpreted as a modifier of the complex
NumP [NumP Num Cl], since by assumption it cannot modify the classifier,
and anyway nothing can intervene between the classifier and its complement. Hence, only the measure structure allows duo to be realized in this
position: [[NumP[Num+Cl] duo ] N].
The third fact was that Num-Cl-N is ambiguous between a counting and
a measure reading, while Num-Cl-de-N only has a measure reading (as
discussed in (19) and (20)). The explanation is similar to the previous case.
The Mandarin modification marker de can, as the descriptive name suggests, only occur with modifiers, not with other syntactic constituents. If
so, it follows that de can occur after the classifier in the measure structure
(22b), since there the classifier is part of a modifier phrase, but not in (22a),
because there the classifier is not part of a modifier phrase. Hence when de
is inserted, only the measure structure is available.
The fourth fact discussed was that classifiers can be reduplicated on the
counting reading, but not on the measure reading (as shown in (20)-(21)).
I point out here a related fact. The measure classifier cannot be reduplicated and have an anaphoric relation to antecedent Num-Cl-N, as repeated
in (24a). But if the reduplicated classifier is followed by the noun, then the
acceptablity of the sentence is improved, as in (24b). 1
1

This observation is due to Victor Pan (p.c.).

A feature analysis of classifiers: [Counting, Measuring]

143

(24) a.# zhe ge tong


zhuang le
san ping jiu,
this Cl bucket hold
PFV three CL bottle wine
ping ping
dou hen gui.
CLbottle CLbottle all very expensive
This bucket holds three bottles of wine, each of which is expensive.
b. zhe ge tong
zhuang le san
ping
jiu,
this CL bucket hold
PFV three CLbottle wine
ping-ping
jiu dou hen gui.
CLbottle CLbottle wine all very expensive
This bucket holds three bottles of wine, each of which is expensive.
The contrast between (24a) and (24b) suggests that the real generalization is that if the antecedent of the reduplicated classifier has a measure
reading, the sentence is felicitous if the reduplicated classifier is followed
by a NP, while if the antecedent of the reduplicated classifier has a counting reading, adding an NP after the classifier is not necessary.
This difference can be explained in terms of the structures in (22). In
the counting structure of (24a), the NP is the complement of the ClP and
can be easily deleted as an instance of NP ellipsis. In the measure structure
of (24b), the NP is the head of the structure and the ClP is its specifier, and
a similar deletion is impossible.
In sum, the two syntactic structures I propose for counting and measure
classifiers in (22) elegantly account for the differences discussed.

4. A feature analysis of classifiers: [Counting, Measuring]


Container classifiers allow us to observe the two different uses of classifiers, since they move easily between a measure use and a counting use.
Chinese, as a classifier language, has other types of classifiers, though.
According to Chao (1968)s classification, Chinese classifiers at least include the following subclasses: individual classifiers, container classifiers,
group classifiers, partition classifiers, standard measures, and temporary
measures (recall Table 1 in chapter 2). In this section, I analyze Chinese
classifiers in terms of the availability of counting and measure readings. I
show that not all classifiers easily allow both counting and measure readings and I propose a feature analysis of classifiers in terms of the availability of these two functions.

144

Counting and measure functions of classifiers

Assuming that counting and measure are the two basic functions of
classifiers, I propose to take [Counting] and [Measure] as features constraining the way classifiers can be interpreted and syntactically realized.
With these two features, we predict four types of classifiers: [+C, -M] classifiers are by default counting classifiers, [-C, +M] classifiers, which are
by default measure classifiers, [+C, +M] classifiers for which measure and
counting readings are equally available, and [-C, -M] classifiers, which can
neither count nor measure individuals. I show that all the four types of
classifiers are instantiated in the language.

4.1. Four types of classifiers


4.1.1. Type 1: [+C, -M] classifiers
[+C, -M] classifiers have by default only a counting reading, not a measure
reading. They can be called counting classifiers. The core instances of
[+C, -M] classifiers are the so-called individual classifiers as a lexical
class. Generally speaking, individual classifiers are those classifiers whose
complements are nouns denoting naturally discrete entities. The class includes the general classifier ge, the classifier for individual animals zhi,
and duo blossom etc.
I propose that for their core use, [+C, -M] classifiers spell out the inherent counting unit intrinsic to sets of discrete entities (see also Allan 1977,
Tai and Wang 1990, Cheng and Sybesma 1998). For example, as shown in
(25), hua flowers are counted as duo blossoms, mi rice is counted as li
grains, and shu book are counted as ben volumes:
(25) a. san duo
hua
three CLblossom flower
three blossoms of flower
b. yi
li
mi
one CLgrain
rice
a grain of rice
c. liang ben
shu
two CLvolume book
a volume of book

A feature analysis of classifiers: [Counting, Measuring]

145

In these classifier phrases, the classifier and the noun stand in a relation
that is constrained by selectional restrictions. The property expressed by
the classifier, e.g. shape, dimension etc., must be semantically compatible
with the shape of entities in the denotation of N. I suggest that the classifiers make available the atomic structure of the entities in the denotation of
the N and specify a natural counting unit for these entities.
This idea about the semantics of [+C, -M] classifiers is close to the
function of what Lyons (1976) calls sortal classifiers, or classifiers that
go with counting or atomic predicates. It is important to stress, however,
that in the present account the [+C, -M] feature characterizes the semantic
function of the classifier, and not the kind of noun it combines with. Although nouns denoting discrete atomic entities naturally occur as the complements of [+C,-M] classifiers, the classifier does not require the noun
complement to denote a set of discrete entities. Thus, individual classifiers
can also take as complements abstract nouns, or nouns denoting homogeneous entities. In this case, the classifier picks out a set of contextually
relevant minimal entities which instantiate the kind denoted by the noun.
For instance, the classifier duo blossom can not only modify nouns
like hua flower, but also nouns which have some flower-like characteristics like yun cloud or mogu mushroom:
(26) a. yi
duo
yun
one CLblossom cloud
a blossom of cloud
b. yi duo
mogu
one CLblossom mushroom
a blossom of mushroom
The noun cloud is homogeneous in the sense that it does not have an inherent unit of counting associated with it. For speakers of English, the idea
that a noun like cloud is homogeneous is perhaps counterintuitive, since in
English, cloud is a count noun, and we are used to think of singular count
nouns in English as denoting sets of inherently atomic entities. However,
as showed earlier in Chapter 4, nouns like boy or furniture denote discrete
entities, and those like water and fence denote homogeneous entities. I
argued that the grammatical distinction between count and mass nouns is
independent of the ontological distinction between discreteness and homogeneity. Rothstein (2010) argues explicitly that the naturally atomic/nonnaturally atomic distinction is orthogonal to the count/mass distinction.
Although cloud is a count noun in English, it does not imply that the noun

146

Counting and measure functions of classifiers

denotes a set of inherently individuable atomic clouds, which can be directly counted. The boundaries of clouds are often not clearly defined, and
there will be no absolute agreement as to how they are to be counted. In
Rothsteins terms, a context-dependent decision must be made as to what
counts as one cloud before they can be counted one by one. The interpretation of the plural form clouds is dependent on the choice of the denotation
of the singular. The apparent atomicity of nouns like fence (and cloud in
our discussion) is context dependent (Rothstein 2010).
Another example is concerned with the individual classifier tiao. It
usually modifies thin, long-shaped entities. (27a) and (27b) illustrate the
prototypical cases, in which it modifies three-dimensional entities like
chong worm and two-dimensional entity, like he river, while in (27c), it
modifies an abstract noun, shengming life.
(27) a. yi
one
b. yi
one
c. yi
one

tiao chong
CL worm
tiao he
CL river
tiao shengming
CL life

What is important here is that [+C, -M] classifiers combine with nonnaturally atomic predicates like yun cloud or abstract entities like
shengming life, in addition to naturally atomic predicates as a more general case.
I thus suggest that the individuation function of [+C, -M] classifiers are
crucially derived from the semantics of the classifier itself, and not from
the structural properties of the denotation of the complement noun.
Note also that on the individuating or counting function, the classifier
does not imply any information about the quantity, weight, length or volume of the entities it counts. We do not know how heavy a blossom of
cloud is when we say yi pian yun a piece of cloud. Thus [+C, -M] classifiers simply provide a counting unit, and make available a set of atomic
countable entities, but do not give a way of measuring them along any other dimension. Even when they combine with homogeneous expressions,
these classifiers allow us to say how many but not how much.
As we will see later, [+C,-M] classifiers can occur in some measure
contexts. But these cases are highly restricted, mainly recipe contexts. I
regard such uses as highly context-dependent, non-default uses. I discuss
them in section 4.3 and in chapter 8.

A feature analysis of classifiers: [Counting, Measuring]

147

4.1.2. Type 2: [-C, +M] classifiers


[-C, +M] classifiers only have by default a measure reading, not a counting
reading. This category includes only the classifiers that are pure measure
words. This group of classifiers includes standard measures and temporary
measures in Chaos system.
As Chierchia (1998a) points out, standard measure words like kilo,
pound, meter do not impose any atomic structure on entities, and they
simply measure the size of entities along a certain dimension, e.g. weight,
length. Chinese standard measure words, such as gongjin kilo, mi meter,
behave semantically like their counterparts in English. Look at (28):
(28) a. san
gongji pingguo
three
kilo
apple
three kilos of apples
b. wu
mi
dianxian
five
meter cable
five meters of cable
For example, in (28a), san gongjin tang three kilos of sugar does not
imply that there are three distinctive packs of apples, each of which weighs
exactly one kilo. It can be the case that there are six packs of apples, and
each of them weighs half kilo. This is exactly the opposite of what we find
for individual classifiers.
Another type of measure words are temporary classifiers in Chaos
(1968) term, like shen body, lian face, di floor. Consider (29).
(29) a. yi
lian
hanshui
one
CLface sweat
a faceful of sweat
b. yi
di
mi
one
CLfloor rice
a floorful of rice
c. yi
shen xue
one
CLbody snow
a bodyful of snow
There are two differences between standard measure words and temporary measure words. First, unlike container classifiers, temporary measures
are usually realized by body parts or objects that are able to express at-

148

Counting and measure functions of classifiers

tachment or spreading relations. For example, in (29a), the expression yi


lian hanshui a faceful of sweat means that ones face is covered with
sweat, not that ones face contains sweat.
Secondly, unlike standard measure words, temporary measures express
inaccurate measure values. For instance, yi lian hanshui a faceful of
sweat in (29b) doesnt tell you how much sweat there is. This is not only
because the measure unit expressed by the classifier itself is vague, but
also because the whole classifier phrase is a hyperbole. In this example, the
numeral yi one does not behave like numbers in normal classifier phrases.
It cannot be replaced by other numerals, such as liang two, as shown in
(30a). However, it can be replaced by degree modifiers like man full or
zheng whole, as shown in (30b) and (30c) (as mentioned in Chao (1968).
(29a) expresses in a hyperbolic way that ones face is coverd with sweat
(as in the English, I was bathing in sweat). This interpretation makes
degree modifiers natural.
(30) a. #liang lian
hanshui
two CLface sweat
two facefuls of sweat
b. man
lian
hanshui
full
CLface sweat
a full face of sweat
c. zheng lian
hanshui
whole CLface sweat
a whole face of sweat
We see that temporary measure words like lian face provide measure
units to measure the quantity of entities, in particular, that they emphasize
that the measure values involved are big: in (29b), by its covering meaning
the classifier expresses that the quantity of sweat on the face is large.
Temporary measures are by default measuring classifiers. As I will discuss in section 4.3, some measure words can shift their meaning to a counting interpretation, but, as will be shown, such interpretations are highly
context restricted and conventionalized.

4.1.3. Type 3: [+C, +M] classifiers


The third type of classifiers has both the features of [+C] and [+M]. This
means that unlike [+C,-M] or [-C,+M] classifiers, both counting and meas-

A feature analysis of classifiers: [Counting, Measuring]

149

ure readings are equally available for [+C, +M] classifiers, although context may favor one or the other. [+C, +M] classifiers include container
classifiers, group classifiers and partition classifiers.
For example, the container classifier ping bottle in (31) can either
mean a concrete bottle or an abstract measure unit.
(31) wo he le yi
ping hongjiu.
I drink PFV one CLbottle red wine
I drank a bottle of wine.
In the counting reading, (31) means that I drank (some of) wine out of a
particular bottle. Note that perfective markers in Chinese do not express
telicity in the way such markers do in other languages. Here, the perfective
marker le (31) does not imply a telic event (see Soh and Kuo 2005): (31)
does not express that the bottle of wine got finished.
In the measure reading, I drank the quantity of a bottle, i.e. 750 ml of
red wine. In this case, the event is a telic event. It doesnt mean that I drank
some of 750 ml, but the full amount.
Since I have already discussed the ambiguity of container classifiers in
the sections 3, I here discuss partition classifiers (as in 32) and group classifiers (as in 33). I show that they too are ambiguous between a counting
and a measure interpretation.
(32) a. ta di le
san
di
Este Lauder,
she drip PFV three CLdrop Este Lauder,
e-tou,
liang jia
ge
yi di.
forehead, two-cheeks each one CLdrop
She dripped three drops of Este Lauder essense: one on the forehead and two on the cheeks.
b. jiu nongdu
eryan, san di
xin de
Este Lauder
FOC concentration as:for three CLdrop new Mod Este Lauder
xiangdang yu jiu de yi di.
equivalent to old Mod one CLdrop
As for the intensity, three drops of the new Este Lauder essence is
equivalent to one drop of the old Este Lauder.

150

Counting and measure functions of classifiers

(33) a. you
liang pai xuesheng chao
wo zou-lai,
there-be two CL student toward me walk-come
qianmian yi pai, houmian yi dui.
front
one CLrow back
one CLrow
Two rows of students are walking toward me. One in the front and
on in the back.
b. zhe-ge jiaoshi
zhi neng rongxia liang pai xuesheng.
this-CL classroom only can contain two CLrow student
This classroom can only hold two rows of students.
(32a) and (33a) show counting readings of partition classifiers and
group classifiers. The classifier di drop in (32a) partitions the Este
Lauder essense into a plural entity composed of three drops. There must be
three separate drops on the face. In (33a), the expression that one on the
front and one on the back forces a count reading: there must be two distinctive rows of students.
In contrast, (32b) and (33b) express the measure reading. For example,
(32b) requires the amount of Este Lauder essense to be as much as three
drops, but it is not required that essense comes in three individuated drops.
Similarly, (33b) only requires the number of students to be as many as two
rows, though they may be sitting together.
Hence, container Cls, group Cls and partition Cls indeed are equally
open for counting and measure readings in different contexts. On the
counting reading, the classifiers indicate how the plural entities or mass
entities are packed or accumulated into single units. On the measure reading, they express the quantity of plural entities or mass entities along a
certain dimension, e.g. weight or length.

4.1.4. Type 4: [-C, -M] classifiers


At first glance, it looks a bit strange that there could be classifiers that neither count nor measure individuals. But there is such a type of classifiers,
the kind classifiers that denote predicates of subkinds and not of individuals. In Chinese, kind classifiers include zhong kind and lei class:
(34) a. yi zhong yu
one CLkind fish
one kind of fish

A feature analysis of classifiers: [Counting, Measuring]

151

b. san lei
shu
three CLclass book
three classes of books
Individual classifiers (and other types of non-kind classifiers) denote
sets of individuals. However, kind classifier phrases denote sets of entities
of a different type. They denote sets of subkind entities. I showed in Chapter 5 that bare nouns in Mandarin denote kinds. I assume that the kind classifier turns the kind into a set of well-established subkind entities (see
Krifka 1995 for the semantics of kind classifiers in Chinese). This means
that kind classifiers neither count nor measure individuals, which is what
makes them [-C, -M].
The numeral before the kind classifier can only count the number of
subkinds of entities, not the number of individuals constituting the subkind.
For example, (34b) shows that the cardinalty of subkinds of book is three.
It does not matter how many books are in each type or how much book is
in each type.
Subkinds can be seen as plural entities (i.e. sums of the atoms) whose
members of a subkind must share the same natural properties. Crucially,
kind classifiers are sensitive to the (natural) properties of individuals but
not to their quantity, while other types of classifiers, like individual and
group classifiers, are sensitive to quantities. This difference is seen in (35)
and (36):
(35) a. yi ge ershi ke
de pingguo
one CL twenty gram DE apple
an apple of twenty grams
b. yi qun
ershi
ge de xuesheng
one CLgroup twenty CL DE student
a group of twenty students

[+C, -M]

[+C, +M]

(36) a.# yi zhong ershi tiao de yu


one CLkind twenty CL DE fish
Intended: a kind of fish of twenty individuals
b.# san lei
wu ben
de shu
three CLkind five CLvolume DE book
Intended: three classes of books of five volumes
As shown in (35), in ClPs with the features of [+C M] and [+C, +M],
we can insert modifiers expressing number or quantity. For example, in

152

Counting and measure functions of classifiers

(35a), the apple weighs 20 gram and in (35b), the group entity is composed
of twenty students. However, kind ClPs cannot take such modifiers to specify the numbers or the quantity of the instantiations of the relevant kinds,
as shown in (36).
The difference between kind classifiers and other types of classifiers is
also evidenced in (37). For example, yi qun dongwu a group of animals
and yi zhong dongwu a kind of animal are both instantiated by individual
animals, but the former requires enough instances for the animals to form a
group, while the latter, as a subkind, simply requires some instance which
is an animal. Look at the difference between (37a) and (37b).
(37) a.#wo kandao le liang qun dongwu, yi zhi mao he yi zhi gou.
I see
PFV two CLgroup animal one CL cat and one CL dog
I saw two groups of animal: one (individual) cat and one (individual) dog.
b. wo kandao le liang zhong dongwu: yi zhi mao he yi zhi gou.
I see PFV two CLkind animal one CL cat and one CL dog
I saw two kinds of animals: one (individual) cat and one (individ
ual) dog.
Liang qun dongwu two groups of animals in (37a) postulates the existence of a lot of animals, two different groups, each containing a large
enough number of animals. In contrast, (37b) with two kinds of animals
can be two animals, one of each kind.
I thus conclude that kind classifiers are not sensitive to quantity information of the entities. They have a categorization function, i.e. to categorize entities into different sorts or types. They neither count nor measure
individuals. They are [-C, -M] classifiers.

4.1.5. Concluding remarks


I categorized classifiers in Chinese into four types in terms of their semantic functions. The present classification differs from previous accounts in
the following ways. Previous accounts typically make a distinction between the types of lexical information conveyed by different types of classifiers, like sortal and mensural classifiers, etc. My classification of classifiers is based on the basic semantic functions of counting and measuring.
It has the following advantages over the other approaches:

A feature analysis of classifiers: [Counting, Measuring]

153

First, our classification of classifiers is built upon the semantic function


of classifiers themselves, but not on the properties of their complements.
For example, Lyons (1976) defines sortal classifiers in terms of the properties of the noun being sortal or not. I argued in 4.1.1 that while [+C, -M]
classifiers naturally modify nouns denoting discrete entities, they can also
modify nouns denoting homogeneous entities, such as yun cloud or even
abstract entities. My approach can also avoid the difficulty of categorizing
the seeming dual use of certain classifiers, such as ba handful in yi ba
mianfen a handful of flour and yi ba yizi a chair. For us, it is not an
issue whether ba is a group classifier or an individual classifier in these
two cases, but what is important is that they can express counting or measuring functions in both cases. In our classification, nouns are not decisive
in interpreting the function of classifiers or categorizing classifiers.
Second, the feature system expresses the fact that container, group and
partition classifiers are semantically both like [+C, -M] and [-C, +M] classifiers. Instead of either focusing only one of the readings or postulating an
ambiguity for some classifiers, the classification tells us that the fact that a
large group of classifiers falls into both categories is not an exception to be
dealt with, but something to be expected.
Third, in my analysis, both left-braching and right-branching structures
are possible for classifier phrases in Chinese. Unlike previous syntactic
analysis, lexically different subtypes of classifiers do not go randomly for a
certain structure and different syntactic structures are chosen to represent
different semantic functions of classifiers.
I will below give a semantic intepretation for the feature combinations.
What is left open is how to represent the basic lexical meaning of the classifiers in such a way as to explain why some classifiers have both uses and
some have only one. This is not a question I will deal with in this book.

4.2. Syntactic support for the counting and measuring readings


I have argued so far that there are good semantic reasons for making a classification on the basis of the features of [Counting, Measure]. In this
section I discuss some syntactic evidence for the classification.
I gave four syntactic tests for distinguishing the syntactic differences
between the counting and the measuring functions of [+C, +M] classifiers
in section 3. I now show that the syntactic properties of [+C, -M] classifiers are on a par with [+C, +M] classifiers on counting contexts and that of

154

Counting and measure functions of classifiers

[-C, +M] classifiers patterns with [+C, +M] classifiers on the measure contexts.
First, the Cl-N construction only allows counting classifiers. This predicts that [+C, -M] classifiers, i.e. individual classifiers, can be used in
Cl+N construction, as shown in (38a-b), and that [-C, +M] classifiers, i.e.
true measure words, must always be accompanied by a lexically realized
Num, as shown in (38c-d). Note that Cl+N in Mandarin is indefinite, as
in (28a), and the one in Wu is definite, as in (38b).
(38) a. wo xiang mai ba dao.
[Mandarin]
I want buy CL knife
I want to buy a knife.
[Wu: Fuyang]
b. kiu ti
khunk.
CL dog PROG sleep
The dog is sleeping.
c.* wo xiang zou gongli
lu.
[Mandarin]
I want walk kilometer road
Intended: I want to walk one kilometer.
d.* di
lu o u f uan.
[Wu: Fuyang]
mile road I walk not finish
Intended: The mile of road, I cannot walk to the end of it.
Secondly, in the classifier phrase of [+C, -M] classifiers, the modifier
duo more can occur between Num and Cl, but not between Cl and N.
This is shown in (39). In the classifier phrase of [-C, +M] classifiers, the
element duo more occurs either between Num and Cl or between Cl and
N, as shown in (40).
(39) a. shi duo ge pingguo
ten more CL apple
more than ten apples
b.#/???shi ge duo pingguo
ten CL more apples
(40) a. na ge tong
zhuang le
shi duo gongjin pingguo
apple
that CL bucket contain PFV ten more CLkilo
That bucket contains more than ten kilos of apples.
b. na ge dong zhuang le
san gongjin duo pingguo
that CL bucket contain PFV three CLkilo
more apple
That bucket contains more than three kilos of apples.

A feature analysis of classifiers: [Counting, Measuring]

155

Note that both examples of (40) have the measure reading. Since gongjin is a true measure word, the counting reading for Num-duo-Cl-N that I
observed for container classifiers is impossible in (40a).
Thirdly, [+C, -M] classifiers cannot be followed by the marker de,
while [-C, +M] classifiers can be naturally followed by the marker de.
(41) a.# you
san ge de pingguo cong louti shang gun xialai.
there-be three CL DE apple from stair on
roll down
Intended: Three apples rolled down from the stairway.
b. wo mai le
san gongjin de pingguo.
DE apple
I
buy PFV three CLkilo
I bought three kilos of apples.
Fourthly, [+C, -M] classifiers can be reduplicated, as in (42a), but [-C,
+M] do not. This is shown in (42b).
(42) a. ge-ge
pingguo duo hen tian.
CL-CL apple
all very sweet
Each apple is sweet.
b.# gongjin-gongjin pingguo dou hen tian.
kilo-kilo
apple
all very sweet
Each kilo of apples is sweet.
Next I look at group classifiers and partition classifiers, and study
whether they behave in the same way as container classifiers in the relevant syntactic contexts.
First, group and partition classifiers can be used in the form of Cl+N
only in the counting reading, not in the measure reading, as shown by the
contrasts in (43) and (44):
(43) a. you
pai xuesheng chao wo zoulai.
[group Cl]
there-be CLrow student toward me walk.
A row of students are walking toward me.
b.# zhe-ge jiaoshi
zhi neng rongxia pai xuesheng.
this-Cl classroom only can contain CLrow student
Intended: this classroom can only hold a row of students.
(44) a. wo-de bai tixu shang you
di moshui.
my white T-shirt on there-be CLdrop ink
There is a spot of ink on my white T-shirt.

[partiton Cl]

156

Counting and measure functions of classifiers

b.# xie zhe pian wenzhang yong le wo di


moshui.
write this Cl article
use PFV I CL drop ink
It took me one drop of ink to write this article.
Secondly, group and partition classifiers can be reduplicated only when
they are read with a counting reading, but not with a measure reading:
(45) a. ta di le
san
di
Este Lauder, e-tou,
she drip PFV three CLdrop Este Lauder, forehead,
liang jia
ge
yi di.
Di-di
dou hen da.
two-cheeks each one CL-drop Cl-Cl-drop all very big
She dripped three drops of Este Lauder essense: one on the forehead and two on the cheeks. Each drop is big.
b.#jiu
nongdu eryan, san di
xin de
Este Lauder
as for concentration say three CLdrop new Mod Este Lauder
xiangdang yu jiu de yi di.
Di-di
dou hen da.
equivalent to old Mod one CLdrop Cl-Cl-drop all very big
As for the intensity, three drops of the new Este Lauder essence is
equivalent to one drop of the old Este Lauder. Each drop is big.
(46) a. you
liang zu
xuesheng chao wo zoulai.
there-be two CLgroup
student
toward me walk.
zu-zu
dou hen zhengqi.
CL-CLrow all very in order
Two groups of students are walking toward me. Each group is in
good order.
b.#zhe-ge jiaoshi
zhi neng rongna liang zu xuesheng.
this-CL classroom only can contain two CL student
zu-zu dou shi shiwu
ren.
CL-CL all
be fifteen
people
Intended: This classroom can only hold two groups of students.
Each group has fifteen people.
We cannot use the insertion of de/duo test for group and partition classifiers, since native speakers do not accept de/duo for either interpretation:
(47) a.# yi
one
b.# yi
one

qun
CLgroup
kuai
CLpiece

de haizi
DE children
de
xigua
DE watermelon

[group Cl]
[partition Cl]

A feature analysis of classifiers: [Counting, Measuring]

(48) a.# yi
one
b.# yi
one

qun
CLgroup
kuai
CLpiece

duo
more
duo
more

haizi
children
xigua
watermelon

157

[group Cl]
[partition Cl]

There are two different ways to account for these facts. One possibility,
based on the data in (47) and (48) is that we should treat container classifiers on the one hand and group and partition classifiers on the other as two
different types of classifiers. The other possibility is that they are the same
type of classifiers, but that there are extra constraints on the use of de and
duo which mean that they do not occur with group and partition classifiers.
I think that the second possibility is more plausible, because of the similarities between all three types of classifiers as shown in (43) to (45).
However, despite these similarities in behaviour, container classifiers
nonetheless differ semantically from group and partition classifiers. On the
counting function they make use of actual entities in the world to classify
(and thus count), while on the measure reading they assume that the volume of the container can be treated as directly analogous to a standard unit
of measure such as kilo or liter. I will give interpretations for container
classifiers in section 5 below. However, group and partition classifiers are
more complex. Group classifiers require constructing abstract entities (see
e.g. Landman 1989 and also the discussion in Rothstein 2010) and partition
classifiers require doing the inverse, i.e. imposing a part of structure on
instantiations of the kind (whether the instantiations are naturally atomic or
not). Giving a precise interpretation for these classifiers on both the counting and the measuring reading is thus considerably more difficult than giving an interpretation for container classifiers. We will not attempt it here,
but we do assume that the difficulty of using de and duo with group and
partition classifiers is related to the complexity of their interpretation.
I draw two conclusions:
(i) the counting reading and the measure reading can be distinguished
syntactically;
(ii) the tests support the classification which associates [+C,-M] with
the default interpretation of counting, [-C, +M] with the default interpretation of measuring, and [+C, +M] with classifiers that allow both counting
and measuring interpretations.
Since [-C, -M] classifiers are not associated with counting and measuring interpretations, the present tests are not relevant for this class.

158

Counting and measure functions of classifiers

4.3. Semantic shifting between counting and measuring readings


In this section I stress that the features of [Counting, Measure] are characterizations of the default interpretations of the classifiers. As I will show
in section 5, the features [+C] and [+M] associate lexical entries of different types to the classifiers, hence what the parameter setting indicates is
which lexical entries are naturally available. In this, I do not rule out the
possibility that some types of classifiers can have derived readings in certain contexts via type-shifting operations. I will discuss this in detail in
Chapter 7 and 8. In the present section, I show that in appropriate contexts,
[+C, -M] classifiers can have a derived measure interpretation and similarly, that [-C, +M] can have a derived counting reading.
In the default use, [+C, -M] classifiers modify nouns denoting individuable entities and spell out the atomic structure of those entities, i.e. they
indicate the inherent (or imposed) counting units for these entities. However, in the contexts of recipes or menus, the counting units of [+C, -M] classifiers can be coerced into measure units. Look at (49).
(49) zhe ge dangao wo zonggong yong le
yi ge pingguo,
this CL cake I altogether use PFV one CL apple
ban ge zuotian shengxia de, ban-ge shangwu shengxia de.
half CL yesterday left
Mod half-Cl morning left
Mod
To make this cake, I altogether used an apple, half of yesterdays leftover and half of mornings leftover.
(49) provides a context of measurement, where one is told to use as
much as one apple in a cake. A counting reading is not salient in this context, because it is more natural to put chopped apple slices in a cake than a
whole apple. We see then that appropriate measure contexts can be constructed, where the [+C, -M] classifiers gets a measure interpretation.
Fractions are another device that triggers measure readings for [+C, -M]
classifiers, because they, by their nature, express proportions. Consider
(50):
(50) a. ta meitian zuiduo zhi neng chi san-fen-zhi-yi ge xigua.
he every day at most only can eat one-third
CL watermelon
He only eats at most one third of a watermelon every time.
b. zhe zhi fengbi wo zhi yong le ban-zhi.
this CL chalk I only use PFV half-zhi
This chalk, I only used half.

A feature analysis of classifiers: [Counting, Measuring]

159

In (50a), the classifier phrase specifies the quantity of the watermelon to


be 1/3 of a watermelon in the context. It does not matter whether this part
of watermelon is cut into small slices or is made into juice. Similarly, the
natural reading in (50b) is a measure reading. The first zhi in the Dem-Cl
phrase is used as a counting unit which refers to a particular individual
piece of chalk in the context, but the second zhi in half-zhi refers to the
length of the chalk, it tells us that half of the chalk is used, which leads to a
measure reading.
However, not every individual classifier phrase can get a measure reading. The examples in (51) are not felicitous, because it is extremely hard to
find a natural context in which entities like people allow a partition that
allows the measure reading.
(51) a.# ban
half
b.# ban
half

ge
CL
tai
CL

xuesheng
student
diannao
computer

One of the possible contexts in which (51b) can be used as a legitimate


measure phrase is in (51c). But, here, ban tai diannao half a computer
does not denote an object which is half a computer, but measures the quantity of computer that Susan and Fred each owns.
c. Susan he Fred yiqi
mai le
yi tai diannao.
Susan and Fred together buy PFV one CL computer.
Tamen meiren yongyou ban tai diannao.
they each
possess half CL computer
Susan and Fred bought a computer together. Each of them pos
sesses half of a computer.
Note that in (49) and (50), I find two different types of coercion into
measure readings. In (49), the coercion is triggered by the context, while in
(50) the coercion is triggerd by a type mismatch between numeral and classifier. I will return to these two types of coercion in our discussion of the
particle de in Chapter 7 and 8.
Now I look at shifts of measure words, i.e. [-C, +M] classifiers, into
counting classifiers. In their default measure interpretation, measure words
measure quantities but do not impose any atomic structure onto the entities
they measure.

160

Counting and measure functions of classifiers

In some contexts, we can derive a counting reading. For instance, when


the quantity expressed by the measure phrase refers to some individuable
quantities in the context.
For instance, the standard measure word jin pound is a very old and
frequently used measure unit. In contrast, gongjin kilo is a relatively new
measure unit and has been generally accepted in everyday use in some
parts of China only in the past two decades. Due to the frequent use of jin
pound in daily life, it has developed a counting use, in addition to its
original measure use. San jin rou three pounds of meat can have a counting reading if the meat is individually packaged into 500 gram quantities.
In contrast, the new measure unit gongjin kilo does not have this use.
If we consider the measure jin pound to be able to denote individual
packages, then we expect, that on that reading it patterns syntactally with
the [+Counting] classifiers, and in this respect it contrasts with the measure
phrase of Num-gongjin-N, which doesnt ever have the packaging reading.
The examples from (52) to (54) show that this is indeed the case. Note that
it does not matter whether the noun denotes entities that are more discrete,
such as apple, or less discrete, such as meat.
(52) a. wo mai le
jin
pingguo/rou.
I buy PFV CLpound apple/meat
I bought a pound of apples/meat.
b. * wo mai le
gongjin pingguo/rou.
I buy PFV CLkilo apple/meat
Intended: I bought a kilo of apples/meat.

[Mandarin]

(53) a. jin
jin
pingguo/rou dou hen xinxian.
CLpound CLpound apple /meat all very fresh
Each pound of apples/meat is fresh.
b.*gongjin gongjin pingguo/rou dou hen xinxian.
CLkilo CLkilo apple meat all very fresh
Intended: Each kilos of apples/meat is fresh.

[Mandarin]

[Wu: Fuyang]
(54) a. tin
phiku /io man inin.
CLpound apple / meat very fresh
The kilo of apples/meat is very fresh.
b.* kutin phiku/io man inin.
CLkilo apple/ meat very fresh
Intended: The kilo of apples/meat is very sweet.

Semantics of classifiers: counting and measuring

161

In (52a), jin pingguo/rou pound of apples implies singularity, i.e. a


pound of apples/meat. In (53a), tin phiku/io pounds of apples/meat
refers to a particular pound of apples/meat that is assumed to be familiar to
the interlocutors in the context.
I summarize the classification of classifiers in the following table.
Table 3. A four-way distinciton of classifiers: [Counting, Measure]
Classifier types
[+C, -M] classifiers
[-C, +M] classifiers
[+C, +M] classifiers
[-C, -M] classifiers

Default reading
Counting
Measure
Counting or Measure

Possible derived reading


Derived measure reading
Derived counting reading

5. Semantics of classifiers: counting and measuring


In this section I formulate a proposal for the semantics of classifiers on the
counting interpretation and on the measure interpretation. There are only a
few papers in the literature on the semantics of classifiers, namely, Krifka
(1995), Chierchia (1998a), Landman (2004) and Rothstein (2010). Of these
Krifka is the only paper that explicitly discusses the semantics of Chinese
classifiers. Krifka proposes that Chinese classifiers map kinds onto measure functions that measure the number of specimens of that kind. Krifkas
analysis treats all classifiers as measures. In contrast, Chierchia (1998a)
and Rothstein (2010) argue for a counting-based approach to classifiers in
English. They propose that English classifiers have an atomizing function.
Chierchia proposes that classifiers map mass noun denotations onto sets of
atoms. Rothstein (2010) proposes that English classifiers such as unit and
piece explicitly denote a COUNTk function, which applies to root noun
denotations and picks out a context-dependent set of atoms, the set of entities which count as one in the relevant context k.
In what follows, I will review the measure-based approach and the
counting-based approach, and then formulate a proposal for the semantics
of Chinese classifiers on both the counting and measure readings.

162

Counting and measure functions of classifiers

5.1. Krifkas (1995) semantics for Chinese classifiers


Krifka (1995) was one of the first papers to argue that Chinese bare nouns
are kind denoting. Krifka assumes that Chinese bare nouns are basically
names of kinds, and he assumes that other uses of bare nouns, like the indefinite use or the predicative use, are derived from the kind interpretation.
(55) a. xiong jue-zhong le.
bear vanish kind PRF
Bears are extinct.
b. wo kanjian xiong le.
I see
bear PRF
I saw (some) bears.
According to Krifka, the object-level indefinite interpretation of xiong
bear in (55b) is derived from the kind interpretation in (55a), though an
operation R, which is in essence Carlsons (1977) instantiation relation.
The operator R applies to a kind term and returns specimens or individual
sums of subspecies of the kind. Krifkas operator R is similar to Chierchias (1998b) operator >, which shifts kinds into properties. The R relation is given in (56a) and the semantics of the two uses of the bare noun are
given in (56b-c).
(56) a. If x is an individual and k a kind then R(x, k) holds if x is an instantiation of kind k.
b. NxiongN=BEAR
[kind denoting]
c. NxiongN=x.R(x, BEAR)
[object denoting]
For measure phrases (classifier phrases in our terms) like (57), Krifka
(1995) proposes that classifiers are expressions of the lexicalized R operators. The measure words, e.g. qun group and zhi (individual classifier for
animal) take a kind and yield a measure function that measures objects
instantiating that kind.
(57) a. san qun
xiong
three CLgroup bear
three groups of bear

Semantics of classifiers: counting and measuring

163

b. san
zhi
xiong
three CL
bear
three (individual) bears
As we have seen at the beginning of this chapter, for Krifka a classifier
is a measure expression M which combines with a number phrase to form
the interpretation of a measure phrase MP:
(i) N[MP[Num][M]]N = N[M]N(N[Num]N)
And the measure phrase applies to a noun to form a noun phrase:
(ii) N[NP [MP ] [N]]N =N[MP]N (N[N]N).
The semantics for the measure phrases in (57a) and (57b) are given in
(58) and (59) respectively (Note that Krifka gives an intensional semantics,
which I have simplified here into extensional semantics.)
(58) a. NqunN =nkx.R(x, k) herd(x)=n
b. Nsan qunN =kx.R(x, k) herd(x)=3
c. Nsan qun xiongN =x.R(x, BEAR) herd(x)=3
(59) a.Nzhi N=nkx.R(x, k)NATURAL-UNITk(x)=n
b.Nsan zhi N=x. R(x, k) NATURAL-UNITBEAR (x)=3
c.Nsan zhi xiongN =x.R(x, BEAR) NATURAL-UNITBEAR(x)=3
There is a fundamental difference between (58) and (59).
(58) is a straightforward measure reading, where the classifier n herd
applies to a kind and yields a set of instantiations of the kind which measures n herds. In this semantic statement, (58c) is true of an object if it is a
plurality consisting of three herds of bears.
(59), on the other hand, is the equivalent of our counting reading, in
which the classifier counts the number of individual bears. This is encoded
by the natural unit operation which takes a kind and yields a measure
function that measures the number of specimens of that kind. In this semantic statement, (59c), is true of an object if it is a plurality of three individual bears.
Krifka does not discuss the dual functions of classifiers beyond positing
the Natural Unit function, nor does he discuss the fact that a single classifi-

164

Counting and measure functions of classifiers

er can have both a counting and a measure reading. According to the syntax he adopts and the semantics he proposes in (58) and (59), he treats the
counting reading as a particular kind of measure reading.
The structure and interpretation that Krifka gives, where the classifier
combines with the Num, and the whole modifies the noun, fits the structure
that I am proposing for the measure reading. I have argued, however, that
the counting reading of classifiers is built from a different syntactic structure. On the counting reading, the classifier takes the noun as a complement, and this structure is the complement of the NumP.
From this perspective, the analysis in Krifka (1995) cannot account for
the syntactic facts that I have discussed in this chapter.

5.2. Rothsteins (2010) semantics for English classifiers


In contrast with Krifkas (1995) measure-based account, Chierchia (1998a)
and Rothstein (2010) claim that classifiers have an individuating/atomization function (a counting function in our terms). In this section, I will focus on the proposal in Rothstein (2010).
Rothstein (2010) argues that counting is a grammatical operation of
putting entities in one-to-one correspondence with the natural numbers,
which requires a decision as to what counts as one entity. She assumes that
all nouns are interpreted with respect to a domain D, which is a complete
atomic Boolean algebra generated by a (possibly vague) set of atoms. All
lexical nouns N are associated with an abstract root form Nroot, the interpretation of which is a subset of D. As predicates, mass expressions have the
same interpretation as Nroot. As kind denoting expressions, the kind interpretation is derived from the interpretation of Nroot via Chierchias kind
forming operation ?. Thus the denotation of a mass noun, Nmass,
MASS(Nroot)=?Nroot.
For count nouns, Rothstein argues that their interpretation is relativized
to a counting context k, which is a subset of D, and taken to be the set of
entities that in this context count as one. I associate with the model a set K
of such counting contexts on D. In context k, the set of atomic N-entities is
the set Nroot k. Rothstein assumes that in a normal context k, the objects
in Nroot k are non-overlapping: they are the N-entities that count as one in
k.
Technically, the non-overlap requirement for normal contexts can be
made part of a counting function COUNT:

Semantics of classifiers: counting and measuring

165

{<d,k>: d Nroot k} if the objects in


Nroot k do not overlap
(60) a. COUNTk(Nroot)=
undefined otherwise
b. the interpretation of a count noun Ncount in context k is Nk, where
Nk=COUNTk(Nroot).
As can be seen in (60b), Rothstein assumes that a singular count noun
denotes a set Nk, which is a set of ordered pairs, where the first projection
of the ordered pairs in Nk are the entities in Nroot k, and the second projection of the ordered pairs is k. So a singular count noun denotes a set of
indexed atomic N entities, where the index indicates the context with respect to which the entity counts as atomic. The set Nroot k is a set of nonoverlapping N-entities which count as 1 in context k. Such sets Rothstein
calls semantically atomic sets, which are grammatically accessible and
can be counted directly by the numerals (Rothstein 2010).
I define the obvious notions of projection:
(61) 1(Nk) = {d: <d ,k> Nk}
2(Nk) = k
and:
(62) 1(<d,k>) = d
2(<d,k>) = k
Plural count nouns are derived by lifting Links pluralisation operation
from D to the denotations of count noun. Links plural operation * is defined as follows:
*(P)={d D: Y P: d = WY}
Lifting this to the domain D{k} gives:
(63) *(P) = {<d,k> D{k}: d *(1(P) }
Hence, the plural of a singular count noun is the set of ordered pairs,
where the first projection is an entity in *(Nroot k), the closure of Nroot k
under sum, and the second element is k.

166

Counting and measure functions of classifiers

I see that, according to Rothstein (2010), count nouns are of a different


type from mass nouns: mass nouns, as predicates, are of type <d,t>, set of
individuals, and as kind denoting expressions, they are of type k; count
nouns are of type <dk, t>, sets of ordered pairs, objects indexed by k. I
will use x as a variable over individuals of type d and x (in bold italics) as
a variable over objects of type dk.
Rothstein claims that this way of distinguishing mass nouns from count
nouns directly captures the fact that count nouns can be directly modified
by numerals, while mass nouns cannot. Count nouns allow direct grammatical counting, since they have been derived via the implicit lexical COUNT
operator; mass nouns cannot be directly modified by numerals, since they
are derived via COUNT. If we want to count the elements of Nmass, we
introduce the COUNT operation in the syntax, and this requires a classifier.
The most neutral classifiers are unit of and piece of and they can be thought
of as a direct explicit expression of the operation COUNT:
(64) I bought a unit of furniture/ one piece of furniture.

(65) a. Nunit ofNk = Px.

1(x)(Pk)2(x) =k if the objects in


P k do not overlap

undefined otherwise
b. Nunitk of furnitureN=Nunitk of N(NfurnitureN)
= x.1(x)(FURNITURE k)2(x)=k
Hence, the interpretation of unit of in k is like the function COUNTk, it
takes a set like FURNITUREroot and maps it onto the set of pairs of objects
<x, k>, where x is in FURNITURE k, on the assumption that k is a normal context where FURNITUREk is a set of non-overlapping objects.
Some classifiers add more lexical information than unit e.g. strands of
hair, cups of coffee and so on. Rothstein assumes that these add further
properties specified of the semantic atoms (66) (suppressing the definedness condition).
(66) a.NstrandkofN=Px.1(x)(Pk)LONG-AND-THIN(1(x))2(x)=k
b.Nstrandk of hairN = Nstrandk ofN(NhairN)
= x.1(x)(HAIR k)LONG-AND-THIN (1(x))2(x) = k

Semantics of classifiers: counting and measuring

167

As far as classifiers are concerned, Rothstein (2010) explicitly notes


that this account of English count nouns is only relevant for the individuating (counting) function of classifiers: she assumes that measure classifiers
will have to be treated differently.2

5.3. Semantics of Chinese classifiers


In this section, I will work out the semantics of classifiers in Chinese in
their counting reading and in their measure reading. I will extend Rothsteins semantics of English classifiers to the counting function of classifiers in Chinese. Then I will give a semantic account for the measure function of Chinese classifiers, following Krifka (1995), Chierchia (1998a),
Landman (2004) and Rothstein (2009).

5.3.1. Semantics of classifiers in counting readings


Rothsteins (2010) theory is worked out for classifiers in English. In applying it to Chinese classifiers, it is necessary to make some adaptations so
that it reflects the peculiar properties of Chinese bare nouns and classifiers.
First, in Rothsteins analysis, mass nouns are treated as denoting sets,
with the same interpretation as root nouns, rather than as kinds.3 However,
as I argued in Chapter 3, Chinese bare nouns are mass nouns that denote
kinds. I assume that Chinese classifiers apply directly to kind denoting
expressions, and denote functions from kinds into sets of k-indexed atomic
instantiations of the kind, i.e they are functions of type <k,<dk, t>>.
I assume that this difference in interpretation reflects the fundamental
difference between English-type classifiers and Chinese-type classifiers:
English classifers are essentially nominal, while Chinese classifiers are
functional heads which have the function of deriving sets from kinds.
Secondly, for the semantics of classifiers, I want to make explicit how
and where the context k comes in. For Rothstein (2010), the COUNTk
function is either realized by an implicit lexical operator, which is applied
to root nouns to derive count nouns, or by an overt operator, such as piece,
unit etc, which is applied to root nouns to derive Classifier Phrases.
2

Rothstein (2009), discusses the measure function of classifiers in Modern Hebrew.


Rothstein (2010ms) adapts the analysis to an account in which mass nouns denote
kinds.

168

Counting and measure functions of classifiers

If I adapt Rothsteins proposal for English directly as an operation of


the type I want for Chinese, I can assume the following:
(67) NClassifier (NP)N=NClassifierN(NNPN)
=COUNTk(NNPN) = {<d,k>:d>NNPN k}.
The noun denotes a kind; the classifier is interpreted as COUNTk. Applying the classifier to the noun is the result of applying COUNTk to the set
of instances of the kind.
The classifier applies to the set of instantiations of the kind and gives
the set of k-indexed entities which are atomic instances of the kind in context k. However, the classifier system in English is not very productive,
and most of the classifiers like piece or unit, have very general meanings.
Specific classifiers like strand in strands of hair are rare. This makes it
difficult to see how the classifier expresses constraints on which elements
in the context k are relevant, and, since more than one counting context can
be invoked in one background context, which context k is relevant.
Mandarin has a productive classifier system, and most of the classifiers
have very specific lexical meanings, like duo blossom, ben volume, tiao
branch etc. Because of this, Mandarin classifiers are very suited to illustrate what sorts of constraints on contexts classifiers can introduce. Counting context k is a set of entities of D, which will be interpreted as the set of
entities that count as atoms in k. The Chinese classifier applies the
COUNT function in k, but simultaneously adds information which constrains the choice of the atomic entities in k (and possibly the relevant
choice of k).
Let us take the [+Counting] classifier duo blossom as an illustration.
The classifier duo blossom can modify flowers, but also flower-like entities such as yun cloud, as in yi duo yun a blossom of cloud. We know
that the referents of cloud are homogeneous entities, i.e. they do not come
in natural units. On the predicative use, the bare noun yun denotes the set
of instantiations of the kind cloud. In this set, a cloud can have any shape,
flower-like, animal-like, blanket-like etc. When applying the classifier duo
blossom to the noun yun cloud, only flower-like clouds are picked out
to be included in the set of what will be atomic clouds, as denoted by duo
yun blossom cloud. In other words, the classifier duo blossom adds
lexical information, which constrains the set of atomic clouds to be to be
those which are in k and flower-like. Pragmatically, this may constrain the

Semantics of classifiers: counting and measuring

169

choice of relevant context k to be one in which there are enough flowerlike atomic clouds, as Rothstein (2010) implies.
I propose the interpretation schema in (78.a) as the general [+C, -M]
classifier interpretation schema. In this schema is the particular lexical
meaning of the classifier. Thus, the meaning of the classifier duo is the
one given in (78.b):
The semantic template for the counting function of a classifier is given
in (68a). As for the semantics of duo, I propose to compose the classifier
template with the nominal predicate duo blossom, in the lexicon to construct the classifier, which is then applied to the NP meaning. Remember
that k is a variable for kind and k is a variable for context.
(68) a. NClN=kx.1(x)( >k k)
(1(x))2(x)=k
b. NduoN=kx.1(x)( >k k)Blossom-form(1(x))2(x)=k
c. NyunN=?CLOUD
d.Nduo yunN
=x.1(x)(>?CLOUD)k)Blossom-form(1(x))2(x)= k
I propose that the property that the general classifier ge contributes to
the interpretation schema is the trivial property pow(D), which means that
the restricting conjunct is vacuous. This gives the following semantics to
the general classifier:
(69) a. NgeN= k x.1(x) (>k k) 2(x)= k
b. NgeN(NpingguoN)
= NgeN(?APPLE)=x.1(x)(>?APPLEk) 2(x)=k
I illustrate the counting reading of [+C, +M] classifiers with container
classifiers. In fact, their semantics is much easier to specify than that of
partition and group classifiers, since the latter involve second order individuation (see Landman 1989a, b, and discussion in Rothstein 2010). I
leave the semantics for future research.
The semantics of [+C, +M] works a bit differently from that of [+C, -M]
classifiers. I propose that [+C, +M] classifiers impose an external unit on
the instantiations of the kind denoted by N, when these instantiations N do
not come in inherent atomic units or inherent atomic unit are not relevant.
The difference is the following: in a [+C, -M] case like ge pingguo, the
noun phrase denotes an atomic set of apples, derived through the COUNT
function from k: the only thing you need to do in k is pick out the relevant

170

Counting and measure functions of classifiers

k entities that are apples. So, the constraint on 1(x) is that it belongs to
>?
APPLEk. With container classifiers, the relevant k-units are determined
by the semantics of the classifier and not by that of the noun. That is, for
[+C, +M] ClPs, the semantic atom to be counted by the Num is determined
by the lexical property expressed by the classifier. I propose that there is a
relation CONTAIN for container classifiers on the counting reading, such
as ping bottle. I assume here a classifier schema as (70a), in which the
classifier determines what entities will count as atoms, and the kind complement only determines what the container contains:
(70) a. NClN=kx.1(x)(
k)CONTAIN(1(x),k)2(x) = k
b.NpingN=kx.1(x)(BOTTLE k)CONTAIN(1(x),k)2(x) = k
c. Nping shuiN=NpingN(NshuiN) =NpingN(?WATER)
=x.1(x) (BOTTLEk)CONTAIN(1(x),?WATER)2(x)=k
(In this, it needs to be specified what CONTAIN(x, k) means. I will not
be concerned with that here.)
Note that on the counting use, Cl+NP always denotes a set of atomic
entities, a set of singularities. As I will discuss in chapter 9, Cl+NP, like
duo hua CLblossom flower can indeed only refer to singular flowers, but
Cl+NP can be modified by a plural numeral, to express a plurality, as in
wu duo hua five Cl-blossom flowers. Following Rothstein (2010), O assume
that the numeral wu five denotes a function from count noun denotations
into count noun denotations of type <<dk, t>, <dk, t>> which requires a
semantically plural input:
(71) NwuN(Nk) =Px.P(x)|1(x)|k = 5
Wu denotes a function which applies to a count predicate Nk and gives
the subset of ordered pairs in Nk, where the first projection of each ordered
pair has five parts which count as atoms in k. I assume that this semantics
triggers semantic pluralization on the complement of wu. I give the semantic derivation of the plural classifier phrase of wu duo hua five CLblossom
flowers in (72) and (73):
(72) a. NhuaN=?FLOWER
b. NduoN=kx.1(x)(?k k)BLOSSOM(1(x)) 2(x) =k
c. Nduo huaN
= x.1(x)(>?FLOWER k)BLOSSOM(1(x))2(x) =k

Semantics of classifiers: counting and measuring

171

The meaning of wu should apply, but it cannot since it requires a set of


pluralities as input, but (72c) is a set of atoms in k. This is a mismatch
which is resolved by semantic pluralization:
(73) a. *Pluralization applies to the denotation of duo hua
NduohuaN =*(x.1(x)(>?FLOWERk)BLOSSOM(1(x)))2(x)=k
= x.1(x)*(>?FLOWERk)*BLOSSOM(1(x)))2(x)=k
b. Nwu duo huaN
=x.1(x)*(>?FLOWER k)*BLOSSOM(1(x)))|1(x)|k=52(x)=k
That is, it denotes the set of plural entities in the pluralization of the set
FLOWERk which are blossoms and which have five parts that count as
atoms in k.
>?

5.3.2. Semantics for classifiers on the measure reading


I now give a semantics for the measure function of classifiers, including [C, +M] and [+C, +M] classifiers. Let us first look at the semantics of true
measure words, i.e. [-C, +M] classifiers.
Chierchia (1998a) claims that measure words, i.e. [-C, +M] classifiers,
do not individuate or atomize entities. For example, a pound of rice is not
individuated into naturalistic objects like piles or packs to be measured.
Krifka (1989, 1995), Chierchia (1998a), and Landman (2004) all treat
measure expressions as complex modifiers constructed out of a measure
head and a number expression which denotes a number and is of type n.
The measure word, e.g. kilo, is of type <n,<d,t>> and combines first with
the numeral to form a complex modifier which modifies N (or NP). It expresses the quantity of entities or stuff which is a property of the relevant
instantiations of the head noun. Following this approach, Rothstein (2009)
proposes the following semantics for measure classifiers:
(74) a. NbottleN <n,<e,t>> =nx.BOTTLE-FUL(x)=n
b. None bottleN=NbottleN(NoneN)
=nx.BOTTLE-FUL(x)=1
=Qx. Q(x) BOTTLE-FUL(x)=1 (modifier interpretation)
c. None bottleN(NwaterN)= x.WATER(x) BOTTLE-FUL(x)=1
The set of quantities of water which equal one bottleful.

172

Counting and measure functions of classifiers

Again I need to adapt this to Chinese, since I assume that Chinese


measure classifiers are true classifiers and hence take kinds as input. In fact,
what I assume is that measures are functions from numbers to classifier
interpretations: they map a number and a kind onto a set of instantiations
of the kind, the instantiations with a certain measure property. This means
that I treat measure classifiers as expressions of type <n,<k,<d,t>>>, and
measure expressions Num + Cl as expressions of type of <k,<d,t>>, i.e.
functions from kinds into sets of instantiations of the kinds.
For example, in the measure phrase thirty kilos of apples, the measure
word kilo denotes an expression which combines with the thirty to give
thirty kilos. The modifier thirty kilos denotes a function that maps kinds,
like the kind apple, onto instantiations of the kind, which weigh thirty kilos.
I assume that in Chinese, classifiers with the function of [+Measure] are
interpreted as follows. Both [+C, +M] classifiers in the measure reading
and [-C, +M] classifiers in Chinese are measure heads, which combines
with Num to form a modifier, which in its turn applies to the head noun, a
kind-denoting expression.
The semantics of the Chinese measure unit jin pound (i.e. 500 grams)
in the measure phrase of yi jin mi a pound of rice is given as in (85).
(75) a. NjinN =nkx.x>k POUND(x)=n
b. Nyi jinN =kx.x>k POUND(x)=1
c. Nyi jin miN=x.x>?RICE POUND(x)=1
Similarly, [+C, +M] classifiers like container classifiers can denote
measure functions. For example, the container classifier ping bottle can
be used in the measure phrase yi ping jiu a bottle of wine, in which it
denotes a measure unit, e.g. bottleful, as equivalent to 750 ml.
(76) a. NpingN =nkx.x>kBOTTLE-FUL(x)=n
b. Nyi pingN =kx.x >kBOTTLE-FUL(x)=1
c. Nyi ping jiuN =x.x>?WINEBOTTLE-FUL(x)=1

6. Conclusions
In this chapter, I showed that classifiers can have a count function or a
measure function, according to the classification system [Counting,
Measure]. I showed that the measure and the counting functions of clas-

Conclusions

173

sifiers are associated with different syntactic structures. I argued that a


Krifka style semantics is appropriate for measure words and classifiers on
a measure interpretation, while a Rothstein style semantics is appropriate
for counting interpretations. I have modified these analyses to formulate a
semantic interpretations for classifiers of the type s [+C, -M] and [-C, +M]
and for the container classifiers of type [+C, +M] on both counting and
measure readings.
A number of open questions remain:
I have not yet given a semantic template for group and partition classifiers. I assume that the basic principles of counting and measure interpretation apply, but working out the semantics of groups and partitions is
beyond the scope of this dissertation.
I have not shown how non-default interpretations of [+C, -M] and
[-C, +M] classifiers come about, i.e. the derived measure interpretation for
[+C,-M] and the derived counting interpretation for [+M, -C]. It is quite
straightforward to show how the default measure reading can shift into a
count reading for [-C, +M] classifiers (Chierchia 1998a proposes an account), but it is more difficult to see how the counting reading shifts back
into a measure reading for [+C,-M]. I discuss some aspects of this issue in
chapter 8.
I have not discussed how the counting and measure readings of
[+C, +M] container classifiers are related to each other.
All these issues must be left for further research.

Chapter 7
Adjectival modification in classifier phrases:
pre-classifier adjectives

1. Introduction
In Mandarin, at least three types of adjectives can be distinguished in terms
of their syntactic positions in the nominal phrase, namely, adnominal adjectives, pre-classifier adjectives and left-peripheral adjectives (this term
is due to Zhang 2012). Adnominal adjectives are those immediately preceding nouns, as in (1a), and pre-classifier adjectives refer to those occurring between numeral and classifier, as in (1b), and left-peripheral adjectives are those appearing on the left most of the nominal phrase, i.e. before
the (demonstrative-)numeral-classifier-noun cluster, as in (1c).
(1) a. yi
ge [hen da de] xigua
one CL very big Mod watermelon
b. yi [da] ge xigua
one big CL watermelon
c. [hen da de] yi ge xigua
very big Mod one CL watermelon
a very big watermelon

[adnominal adj]
[pre-classifier adj]
[left-peripheral adj]

Among these three types of adjectives, adnominal adjectives are the


most thoroughly explored (see Paul 2005 for a recent review), while the
other two types receive relatively less attention. Intuitively, the meaning
expressed by (1a) and (1c) are very close, expressing that the watermelon
itself is big with respect to a certain standard. In contrast, (1b) with a preclassifier adjective has a different meaning, emphasizing that the quantity
of watermelon is evaluated to be big from a certain perspective adopted by
the speaker.
In this chapter, I will concentrate on the properties of pre-classifier adjectives only. I will address some questions that were raised in Chapter 3.
Specifically, (i) what are the distributional patterns of pre-classifier adjec-

Introduction

175

tives? That is, under what situation can they be used? (ii) What are their
proper interpretations? Can they still be interpreted intersectively?
In chapter 3, I showed that the presence of adjectives before classifiers
cannot be taken to be one of the diagnostics to distinguish two types of
classifiers, i.e.mass and count classifiers, as Cheng and Sybesma (1998)
propose. In chapter 6, I argued for a distinction between the counting and
measure functions of classifiers. In this chapter, following the discussion in
chapter 6, I propose that it is more appropriate to use the distinction between counting and measure functions of classifiers to account for the distributions and the functions of pre-classifier adjectives.
I make three points about the distribution and the semantics of preclassifier adjectives in Chinese:
First, I argue that pre-classifier adjectives appear before classifiers in
counting contexts: before [+C, -M] and [+C, +M] classifiers (on their
counting interpretation), but not before [-C, +M] or [-C, -M] classifiers.
Second, I argue that, while pre-classifier adjectives precede the classifier and the noun at surface structure, they do not stand in a direct modification relation to the classifier or the noun. I propose that pre-classifier adjectives modify the complex constituent Cl+N (also see Corver 1998 for the
discussion on pseudo-partitives in Dutch). Since I argued that there is no
Cl+N constituent in measure phrases, this gives a natural explanation for
why pre-classifier adjectives do not appear in measure expressions.
Third, I argue that pre-classifier adjectives like da/xiao big/small have
expressive meanings la Schlenker (2007): they express that the speaker
regards the atomic entity in the denotation of Cl+N to be big or small
from a particular perspective chosen by him or her.
This chapter makes a systematic examination of the syntactic and semantic properties of pre-classifier adjectives. Section 2 discusses the situation where adjectives can be used before classifiers. In section 3, I examine
the modification relation of pre-classifier adjectives. I rule out the possibility that pre-classifier adjectives modify the classifier or the noun. In section
4, I first present the relevant contexts in which pre-classifier adjectives can
be used and then address the syntactic issue of what constituent preclassifier adjectives modify. In section 5, I discuss the semantics of preclassifier adjectives.

176

Adjectival modification in classifier phrases: pre-classifier adjectives

2. Licensing pre-classifier adjectives


2.1. Pre-classifier adjectives before counting classifiers
As we saw in chapter 3, Cheng and Sybesma (1998) argue that preclassifier adjectives are allowed only before a certain group of classifiers,
namely, before count classifiers but not before mass classifiers. See the
contrast in (2) and (3):
(2) a. yi da zhang zhi
one big CLpiece paper
a big piece of paper
b. na yi
xiao xiang shu
that one small CLbox book
that small box of books

[mass classifiers]

(3) a.# yi
one
b.# yi
one

[count classifiers]

da
big
da
big

wei
CL
zhi
CL

laoshi
teacher
gou
dog

Contra Cheng and Sybesma (1998), I showed in Chapter 3, it is possible


for pre-classifier adjectives to appear before count classifiers, such as the
specific classifier dong for house in (4a) and the general classifier ge in
(4b).
(4) a. ta ziji
ye gai le yi
da dong fangzi.
he himself also build PFV one big CL house
He also built a big house by himself.
b. wumao-qian mai le
yi da ge mangguo.
fifty cents buy PFV one big CL mango
Fifty cents bought a big mango.
My hypothesis is that adjectives can precede classifiers only when the
classifiers are interpreted with a counting reading. Namely, pre-classifier
adjectives are possible before [+C, -M] and [+C, +M] classifiers in appropriate contexts. Three things follow from this.
First, it is predicted that [-counting] classifiers, that is, [-C, +M] Cls
(measure classifiers) and [-C, -M] Cls (kind classifiers), are not acceptable

Licensing pre-classifier adjectives

177

with pre-classifier adjectives. Confirming this, the examples in (5) and (6)
are judged to be completely unacceptable by the native speakers I consulted.
(5) a.* ta
he
b.* ta
he

he
drink
zou
walk

le
PFV
le
PFV

yi
one
yi
one

(6) a.* yi da zhong


one big CLkind
b.* yi da lei
one big CLkind

da
big
da
big

jing
baijiu.
CLpound liquor
gongli
lu.
CLkilometer road

dongwu
animal
1
zhiwu
plant

[-C, +M]

[-C, -M]

Secondly, [+C, -M] classifiers are by default counting classifiers, but


can have coerced measure readings in some contexts. Pre-classifier adjectives can appear in front of [+C, -M] classifiers on the counting reading but
not on the coerced measure reading.
Look at (7). The classifier phrase yi ge pingguo one Cl apple in (7a)
refers to an atomic apple. It is a counting reading. The ClP in (7b) means
that the quantity of apple put into the cake amounts to that of a normalsized apple. It is a coerced measure reading.
(7) a. ta ba yi ge pingguo toutou shai dao ziji koudai li. [Counting]
he OM one CL apple secretly put to self pocket in
He put an apple into his pocket secretly.
b. zuo zhe ge pingguo-pai, wo zongong yong le you yi ge
make this CL apple pie
I in:total use PFV have one CL
pingguo, ban ge zuotian-de, ban ge jintian-de.
[Measure]
apple
half CL yesterday-Mod half CL today-Mod
To make this apple pie, I used an apple altogether, half of yesterdays leftover and half of todays leftover.
I find that it is only possible to insert the adjective da big before the
classifier phrases in the examples in (7) on the default counting reading but
not on the coerced measure reading. Look at the relevant examples in (8):
1

Note that the phenomenon I am interested in the presence of da/xiao big/small


before kind classifiers should be distinguished from the distribution of the lexical
noun of da-lei major kinds.

178

Adjectival modification in classifier phrases: pre-classifier adjectives

(8) a. ta ba yi da ge pingguo toutou shai dao ziji koudai li.


he OM one big CL apple secretly put to self pocket in
He put a big apple into his pocket secretly.
b.#zuo zhe ge pingguo-pai, wo zongong yong le you yi da
make this CL apple pie
I in:total use PFV have one big
ge pingguo,ban ge zuotian-de,
ban ge jintian-de.
CL apple half CL yesterday-Mod half CL today-Mod
Intended: To make this apple pie, I used a big apple altogether, half
of yesterdays leftover and half of todays leftover.
As said earlier, the pre-classifier adjectives like da/xiao entail the existence of atomic entities. Therefore, in a measure context as in (8b), it is
inappropriate to use preclassifier adjectives.
Thirdly, I predict that when a [+C, +M] classifier phrase is preceded by
a pre-classifier adjective, the classifier phrase can only have a counting
reading. Look at (9):
(9) ta he
le
yi da ping kele.
he drink PFV one big CLbottle coke
He drank a big bottle of coke.
(9) expresses that there was one and only one bottle and I drank one
large quantity of coke from it. In other words, the classifier ping bottle
can only have a counting reading, referring to a concrete bottle, which is
filled with coke. (9) is not true if you drank the same (large) amount of
coke from two different bottles. Thus, (9) cannot have a measure reading.
Predictably, when the [+C, +M] ClP with a pre-classifier adjective refers to a plural entity, it can be co-referential with a distributive determiner,
which, as argued in Chapter 6, is a test for counting contexts. This is illustrated in (10):
(10) ta gang he le san da wan yao.
mei wan dou hen ku.
he just drink PFV three big CLbowl medicine each CLbowl all very bitter
He just drank three big bowls of Chinese (herb) medicine. Each bowl
of it was very bitter.
The distributive determiner mei each is co-referential with san da wan
zhongyao three big bowls of Chinese medicine. The sentence means that
there were three individual bowls of Chinese medicine, and he drank them
one by one. And every time he drank one, he felt the bitterness of the med-

Licensing pre-classifier adjectives

179

icine. (10) is not true if he put the three bowls of medicine into a big container and drank it in one go. Clearly, then, container classifier wan bowl
has a counting reading.

2.2. Concrete portion reading (Partee and Borschev 2012)


Some may wonder whether the examples in (9) and (10) are not true counting contexts. The pre-classifier adjectives da/xiao big/small play two
roles. On the one hand, it entails that there exist an atomic individual as
expressed by the classifier, such as container. On the other hand, it emphasizes the quantity of the stuff. Therefore, it seems to be a blended case of
counting and measure readings. This special reading as exemplified in (9)
is similar to the concrete portion reading discussed in Partee and Borchev (2012). Look at their example in (11):
ti dva
stakana
moloka.
(Russian)
(11) vypej
drink-IMPER these two-ACC glass- GEN milk-GEN
Drink these two glasses of milk.

The Russian example (11) is preferably interpreted with a concrete portion reading. It emphasizes the quantity of stuff, but meanwhile it requires
the existence of two concrete containers, as expressed by the demonstrative
ti. According to Partee and Borchev (2012:473), this concrete portion
reading arises from the reference to the container by a form of metonymy
which has become conventionalized for all container words, from container
to contents. It is not immediately clear whether the concrete portion reading has a counting reading or a measure reading. In Partee and Borchevs
analysis (2012: 474), it is treated as a special kind of counting reading in
our terms.
The Concrete Portion reading is close to the measure readings and involves
the same syntax but does not express measure. We think that it might be
better to have an analysis in which the Concrete Portion reading could be
subsumed under the Container+Contents reading (i.e. the counting reading:
the authors note).

I take the same position as Partee and Borschev (2012) in treating Chinese classifier phrases with pre-classifier adjectives with a counting interpretation. Different from them, I also will give them a counting syntax and
a counting semantics in the latter discussion.

180

Adjectival modification in classifier phrases: pre-classifier adjectives

The examples in (12) and (13) bring out the contrast clearly that preclassifier adjectives go with counting classifiers but not measure classifiers
in Mandarin Chinese.
(12) a. ta shou shang na
zhe
yi ping futejia. [Counting]
he hand on
carry DUR one CLbottle vodka
He carried a bottle of vodka in his hand.
b. ta shou shang na
zhe yi da ping
futejia.
he hand on
carry DUR one big CLbottle vodka
He carried a big bottle of vodka in his hand.
(13) a. ta-de jiu-liang
shi yi ping futejia.
[Measure]
his capacity-for-liquor be one CLbottle vodka.
His capacity for liquor is one bottle of vodka.
b.# ta-de jiu-liang
shi yi da ping
futejia.
his capacity-for-liquor be one big CLbottle vodka.
Intended: His capacity for liquor is a large bottle of vodka.
The classifier ping bottle in (12a) has a counting reading: (12a) refers
to a concrete bottle filled with vodka. In this case, we can insert a preclassifier adjective da/xiao big/small, as in (12b), which expresses that
the bottle of vodka is a big entity from a perspective chosen by the speaker.2 However, in (13), the classifier phrase ping bottle is interpreted with
a measure reading and it denotes an abstract measure unit, say, 750mls,
where the insertion of adjective da/xiao before the classifier is impossible,
as in (13b).
In sum, when pre-classifier adjectives precede classifiers, the classifier
can only have a counting reading. Note that this does not mean that preclassifier adjectives are freely possible with [+Counting] classifiers in
counting contexts. I discuss the question of which counting contexts are
possible in section 3.

The claim made by the sentence that the bottle of vodka is a big entity can be
understood at least in two ways: it can mean that the weight of the vodka in the
bottle is too much for someone to carry, or it can mean that the volume of vodka in
the particular bottle is too much for someone to drink. More about this later.

The modification relation of pre-classifier adjectives

181

3. The modification relation of pre-classifier adjectives


The question I ask in this section is: what does the pre-classifier adjective
modify in the sequence Num-Adj-Cl-N: the classifier, the noun, or something else?

3.1. Adjectival modification in pseudo-partitives


In pseudo-partitive constructions in English, i.e. Num-N1-of-N2, N1 can be
modified by adjectives (Jackendoff 1977). This phenomenon is also found
in Dutch, Greek and many other European languages (see Vos 1999, Alexiadou et al 2007, among others). I use the English examples to illustrate, as
given in (14).
(14) a. a blue bottle of water
b. a delicious cup of coffee
c. an interesting collection of LPs

(Alexiadou et al 2007:421)

The adjectives before N1 exhibit varied modification relations. In (14a),


the adjective blue modifies N1, and the phrase means that the bottle itself
is blue. The only possible interpretation of (14b) is that the coffee is delicious. In other words, N1 is transparent and the adjective can cross it to
modify N2. This is called transparent phenomenon (Jackendoff 1977). In
(14c), the LPs themselves may be awful and the collection as such might
be interesting in terms of size, range, period etc. So in this case, the adjective interesting modifies N1-N2 as a single constituent.
According to Vos (1999: chapter 6), the modification of N1 by adjectives is dependent on the nature of N, as being lexical or functional. If N1
is functional, it rejects adjectival modification; if it is lexical it accepts it.
As the Dutch example in (15) shows (Vos 1999: 168), the morpheme boel
lot of is functional and it cannot be modified by attributive adjectives
such as nice, new etc:
(15) een *leuke/*nieuwe/*grote/*rode/*houten/*Franse boel poppen
a
nice/ new
/big /red /wooden/French lot (of) doll
However, I think that this account is inadequate to explain two facts reflected by the examples in (14). First, it is unable to explain why container
nouns can sometimes be modified by adjectives (14a) but sometimes can-

182

Adjectival modification in classifier phrases: pre-classifier adjectives

not (14b), although they are used in the same syntactic context. Does that
mean that container nouns are sometimes lexical and sometimes functional?
This is not plausible. Second, it fails to give a clear answer whether N1 in
(14c), i.e. collection, is lexical or functional.
I think that size adjectives should be treated separately from other attributive adjectives. A more serious problem is concerned with the modification of size adjectives before N1. Morphemes like aantal in Dutch are
functional, but they can be modified by an evaluative adjective like leuk
nice (which in this case has the meaning of a size adjective). (16a) has
the same meaning as (16b).
(16) a. een leuk aantal
poppen
a nice number dolls
b. een grote aantal
poppen
a large number dolls
a large number of dolls
(Dutch: Vos 1999: 169, cf. Alexiadou et al 2007: 418)
I may find similar examples in English. For example, the noun number
is more likely to be treated to be functional, but it can be modified by the
attributive adjectives like decent or good (meaning sizable in this case).
Look at (17):
(17) A decent/good number of students attended the lecture.
It is clear from the examples of (16) and (17) that the nature of N1 being lexical or functional is not a decisive factor of using size adjectives like
big/small before N1 in pseudo-partitive constructions.
A more radical case regarding size adjectives is found in Greek. (18)
shows that the noun kuti box in Greek cannot be modified by the attributive adjective metaliko metal (18a), but it can be modified by size adjectives, such as meghalo big (18b).
(18) a.*ena
a/one
b. ena
a

metaliko kuti
metal
box
meghalo kuti
big
box

sokolates
chocolates
sokolates
chocolates
(Greek: Alexiadou et al 2007: 419)

The modification relation of pre-classifier adjectives

183

If we follow Vos (1999), shall we treat kuti box as functional? Alexiadou et al (2007: 419) emphasize that the proper reading of (18b) is that
of a big number of chocolates, not that of saying that the size of the box
itself is big. So the adjective meghalo big does not modify the container
as expressed by the noun kuti. In othe words, size adjectives, unlike other
attributive adjectives before N1, which either express properties of N1 or
N2, do not behave like attributive adjectives to express properties of N1 or
N2.
I suggest that size adjectives before N1 in pseudo-partitive constructions be treated as a separate phenomenon, to be distinguished from other
attributive adjectives before N1. The presence of (size) adjectives has nothing to do with the lexical/functional make-up of N1.

3.2. Pre-classifier adjectives in Mandarin


I said in chapter 3 that in Mandarin only size adjectives like da/xiao
big/small can be used before classifiers in (19), but not other attributive
adjectives like lan blue, pianyi cheap, wuwei tasteless in (20):
(19) a. yi chang tiao xianlian
one long CL necklace
a long necklace
b. yi xiao kuai mianbao
one small CLpiece bread
a small piece of bread
(20) a. # yi wuwei
bei
kafei
one tasteless CLcup coffee
Intended: a tasteless cup of coffee
b.# yi lan ping
shui
one blue CLbottle water
Intended: a blue bottle of water
c. # yi pianyi bei kafei
one cheap cup coffee
Intended: a cheap cup of coffee
Therefore, if there is any similarity between pre-clasifier adjectives in
different languages, pre-classifier adjectives in Chinese are expected to
behave like those in (16) but not like those in (14) in Dutch.

184

Adjectival modification in classifier phrases: pre-classifier adjectives

In this section, I will show that any account based on the lexicalfunctional properties of the classifier is not able to capture the licensing
condition of pre-classifier adjectives, since, as I said above, preclassifier
adjectives do not express properties of N1 or N2. I will review Cheng and
Sybesma (1998), Zong (2009) and Zhang (2012).
Cheng and Sybesma (1998) argue that pre-classifier adjectives modify
what they call mass classifiers, and take this as evidence that mass classifiers are inherently nouns, which are borrowed to be classifiers. In their view,
pre-classifier adjectives cannot modify count classifiers, since these are
inherently functional.
In contrast, Zong (2009) claims that pre-classifier adjectives are available for both individual and non-individual classifiers, but pre-classifier
adjectives modify the noun complement if the classifier is an individual
classifier (adjective lowering in the sense of Zhang 2012), and the classifier itself if the classifier is a non-individual classifier, e.g. a container
classifier or a group classifier.
Here, I argue that (i) while pre-classifier adjectives appear in front of
classifiers in surface structure, they do not stand in any direct modification
relation to the classifier, and that (ii) pre-classifier adjectives cannot cross
the classifier head to modify the noun in Chinese.
In what follows, I will examine the modificational relation of preclassifier adjectives in [+C, -M] and [+C, +M] classifier phrases in section
3.2.1 and 3.2.2 respectively.

3.2.1. Rejecting pre-classifier adjectives modifying mass classifiers


Example (21) contains a [+C, +M] classifier phrase (a container ClP):
(21) a. ta zhen le
yi
xiao bei
putaojiu.
he pour PFV one small CLglass wine
He poured a small glass of wine.
[from J.M. Lu 1987: 60]
b. ta zhu le yi
da guo fan.
I cook PFV one big CLpot rice
I made a big pot of rice.
According to Cheng and Sybesma (1998), classifiers like bei glass and
guo pot are directly modified by the adjective xiao small or da big,
because these classifiers are inherently nouns. For them these adjectives
modify the mass classifier in its abstract function as a container. To

The modification relation of pre-classifier adjectives

185

avoid ambiguity, I use the word container to refer to the concrete physical
containers referred to by container classifiers, and the word volume to refer
to the capacity of containers. So what Cheng and Sybesma call container,
I call volume.
(21a) expresses that there is a small quantity of wine contained in the
glass and (21b) that there is a big quantity of rice contained in the pot. According to Cheng and Sybesma (1998), large containers imply large quantities: the quantity of stuff contained in the big or small containers depends
on the actual volume of the container.
Against this, I note that the pre-classifier da/xiao does not express a size
property (i.e. actual volume) of the container, because, as shown in (22),
xiao small can be used, when the container is directly asserted to be big:
(22) ta jingchang yong na ge da bei he
jiu.
he always
use that CL big glass drink wine
buguo meici
dou zhi he
yi xiao bei.
but
everytime all only drink one small CLglass
He always uses that big glass to drink wine, but every time he only
drinks a small glass.
The classifier phrase yi xiao bei a small glass of wine can be used to
describe a situation where the glass itself is very small, so that even if the
whole glass is filled, there is only a small quantity of wine in it. But it can
also describe a situation where a big glass was used, but only filled with a
little bit of wine. And this is the situation described by (22). The word bei
glass occurs twice in (22). On the first occurrence, in da-bei, it is a noun,
referring to a concrete big sized glass. On the second occurrence, it is in
the classifier phrase yi xiao bei a small glass, where it does not refer to a
concrete glass but to the quantity. And (22) means that the quantity of wine
contained in the glass is small, not that the glass or its fixed volume, is
small.
In (23) I look at the abstract volume of containers. Suppose there are
two bowls of the same size and they are filled with the same amount of rice,
say, 200 grams. These two bowls of rice are served to a three-year old kid
and a basketball player. Each of them gets one bowl of rice. Generally
speaking, 200 grams of rice is a lot (too much) for a three-year old, but
maybe not so much (not enough) for a basketball player. We can use the
sentences in (23a) and (23b) to describe the kids and the athletes cases
respectively (note that I ignore the case where the child is extremely hungry and the athlete has no appetite at all).

186

Adjectival modification in classifier phrases: pre-classifier adjectives

(23) The stewardess in the airplane handed each passenger a bowl of rice:
a. na ge san shui de xiaohai chi le yi da wan fan.
that CL three-year Mod kid
eat PFV one big CLbowl rice
That three-year old kid ate a big bowl of rice.
b. na ge yundongyuan zhi chi le yi xiao wan fan.
that CL athelete
only eat PFV one small CLbowl rice
That athlete only ate a small bowl of rice.
If pre-classifier adjectives modify the classifier as a concrete container,
then there should be a big bowl in (23a) and a small bowl in (23b). But this
is not the case. If pre-classifier adjectives modify the classifier as an abstract volume, then the two bowls should contain different quantities of
rice, a large quantity for the kid, a small quantity for the basketball player.
This is also not the case. What the relevant expressions, yi da/xiao wan fan
a big/small Cl-bowl rice in (23) mean is that the rice served in the big bowl
is a large/small quantity for the relevant eater in view of his/her consumption ability in the context. This kind of quantity information is independent
of the actual size or volume of the container as expressed by the classifier.
Therefore, in [+C, +M] classifier phrases, pre-classifier adjectives do
not modify the classifier, regardless of whether the classifier refers to a
concrete container or an abstract volume.

3.2.2. Rejecting pre-classifier adjectives modifying nouns


Unlike Cheng and Sybesma (1998), Zong (2009) and Zhang (2012) accept
the examples with individual classifiers which go with pre-classifier adjectives. As for the modificational relation, they assume that pre-classifier
adjectives modify the noun in the sequence Num-Adj-indiviudal Cl-N. It is
similar to the transparent phenomenon discussed in Jackendoff (1977),
such as the example of a delicious cup of coffee.
In (24) the classifier phrase has a [+C, -M] classifier (individual Cls)
and it is preceded by the adjective xiao small:
(24) ni bie xiaokan zhe ji
xiao li
huangdou, zhe shi
this be
you not belittle these several small CL grain soybean
women hua le duoda xinxue cai peiyu dedao de liangzhong.
we
spend PFV much effort Foc cultivate gain Mod selected seed
Do not belittle these small grains of soybeans. They are the selected
seeds that were cultivated with painstaking effort. [J. Lu1987: 55-56]

The modification relation of pre-classifier adjectives

187

According to Zong (2009) and Zhang (2012), in (24), it is the soybeans


themselves that are small. In other words, the two examples in (25) are
equivalent to each other. Specifically, Zhang assumes that it is an operation
of adjective lowering, that is, the sequence in (25a) is derived by lowering the adjective xiao small from a pre-classifier position to an adnominal
position (25b).
(25) a. ji
li
xiao huanggou.
several CLgrain small soybean
b. ji
xiao li
huangou
several small CLgrain soybean
several grains of small soybeans
I disagree with them. The foremost reason to argue against the adjective
lowering proposal is that adnominal adjectives and preclassifier adjectives
express distinctive modificational relations and different interpretations.
Speficially, it is only adnominal adjectives that are used to express the
entity itself being big or small, but not pre-classifier adjectives. Adnominal
adjectives can either have the bare adjective in (25a) or the adjectival form
with the modification marker de in (26).
(26) ji
li
hen xiao de
huangou
several CLgrain very small Mod soybean
several grains of very small soybeans
The differences between (25a) and (26) between these two types of adnominal adjectives have been discussed in the literature (e.g. D.X. Zhu
1984, Sproats and Shih 1991, Duanmu 1997, Paul 2005). For instance, Paul
(2005) argues that de divides the NPs into two different syntacticosemantic domains: a modifier outside the scope of de - the case of bare
adjectives - has the meaning of a defining charactistic, whereas a modifier in the the scope of de - the case of Adj-de is interpreted as an accessory
property. I take her term defining property to mean classifying property, a property which sorts out entities denoted by N into different subkinds.
Accessory property for her means a property that is not instrumental in
establishing a new (sub)type of N.
Following up on this, I propose that the bare adjective xiao small in
(25a) combines with the noun huangdou soybean directly into a single
lexical/phonological item. As argued, I assume that Chinese nouns denote
kinds, and I propose here that bare adjectives before N are classifying ad-

188

Adjectival modification in classifier phrases: pre-classifier adjectives

jectives, and that the complex adj-N denotes a subkind. Hence, xiaohuangdou in (25a) refers to a subkind of soybean, the kind mini-soybean.
On the other hand, I propose that complex adjectives, like the one in (25b),
modify the set of instantiations of the kind, not the kind itself. So hen xiaode huangdou in (25b) means that the individual soybeans are small. In both
cases the adnominal da/xiao big/small specifies the physical size of the
entity/ies in the denotation of the head N: the difference lies in what entities these are.
Now, if I compare the ClP yi xiao li huangdou several small Clgrain of
soybeans in (27) with the ClPs in (25a) or (26), it is easy to see that the
interpretations are completely different.
(27) ni bie xiaokan zhe
ji
xiao li
huangdou.
you not belittle these several small CL grain soybean
Do not belittle these small grains of soybeans.
(27) does not mean that the actual size of the soybeans is small, but that,
say, in view of the painstaking efforts you devoted to growing them, what
you finally got was much smaller than should have been. Even if the soybeans that were cultivated are the biggest soybeans in the world, they
might be considered as little given the effort put in. Thus pre-classifier
adjectives before individual classifiers do not stand in a modification relation to the noun: they express personal evaluations of the denotation of
Cl+N, personal evaluation that is independent of the actual size of the
denotation of Cl+N.
Note that for non-individual classifiers, it is also impossible that preclassifier adjectives cross the classifier to modify nouns. The example in
(28) show that da/xiao big/small cannot cross the [+C, +M] classifier
head to modify the N complement:
(28) a. yi
da
ping shui
one big CLbottle water
a big bottle of water
b. yi da xiang xiao pingguo
one big CLbox small apple
a big box of small apples
The referent of the N complement shui in (28a) is a homogeneous entity
and it is impossible to say *da-shui big water or *xiao-shui small-water
(see the discussion on natural atomicity in Chapter 4). (28b) is felicitous: it

Pre-classifier adjectives modifying Cl+N

189

is the apples in the box that are small. Clearly, I cannot assume that preclassifier da big modifies the N as well, because that would be infelicitous, like *big small apples. And that is not what (28b) means.
More examples will be provided on the modificational relation of preclassifier adjectives with both [+C, -M] and [+C, +M] classifiers in section
4.
I conclude this section by arguing that pre-classifier adjectives in Chinese modify neither the classifier nor the noun, no matter whether the classifier is lexical or functional, individual or non-individual Cls.
I will show that pre-classifier adjectives modify Cl+N in section 4.

4. Pre-classifier adjectives modifying Cl+N


In this section, I first introduce various contexts of using pre-classifier
adjectives and then explicate the syntactic relation between pre-classifier
adjective and classifier.

4.1. Contexts of using pre-classifier adjectives


I now examine the different types of contexts in which pre-classifier adjectives can be used and what the modificational relation is in each context.
There are (at least) three such types: the consumption context, the context
of significance and the contrastive context.

4.1.1. Consumption contexts


[+Counting] classifiers with pre-classifier adjectives can be used in consumption contexts, which involve verbs like chi eat, he drink. Consider
the examples in (29).
(29) a. Xiaoling chi le
yi da ge mantou.   [+C, -M]
Xiaoling eat PFV one big CL steamed bun
Xiaoling ate a big steamed bun.
b. wucan ta chi le yi da ping suannai.
[+C,+M]
lunch she eat PFV one big CLbottle yogurt
She ate a big bottle of yogurt as lunch.

190

Adjectival modification in classifier phrases: pre-classifier adjectives

Given that the pre-classifier adjective requires a counting interpretation


for the classifier, all the examples in (29) entail the existence of an atomic
entity in the denotation of the Cl+N: (29a) entails the existence of an
atomic steamed bun of which I ate part; (29b) entails that there is a concrete bottle, out of which I ate some yogurt.
While the ClPs with pre-classifier adjectives entail the atomicity of entities in the denotation of Cl+N, the pre-classifier adjectives da/xiao
big/small in (29) do not express that the physical size of the relevant
atomic entities is big or small. (29a) does not mean that the atomic steamed
bun is big, but that the steamed bun I ate is a big quantity for me. (29b)
does not mean that the concrete bottle filled with yogurt is big, but that the
yogurt that she ate out of a bottle is a large quantity for her. Thus preclassifier adjectives like da/xiao big/small express that the quantity of the
atomic entity referred to by Cl+N is big or small in relation to the appetite or the consumption-ability of the eater.
Note that in the cases in (29), as long as what is consumed is a large
quantity for the eater at the moment in question, it is felicitous to use
one+da+Cl+N; it does not matter whether the subject finishes the whole
atomic entity in the denotation or not. However, this is independent from
the semantics of the pre-classifier adjective: when you say I ate one cucumber in Chinese, it doesnt mean that you finished the whole cucumber.
This is due to the special meaning encoded by the perfective le (Soh and
Kuo 2005).
Once again I use the three-year old and the athlete to illustrate the dependency of the pre-classifier adjective on the personal evaluation perspective .This time, we serve each of them a super-large steamed bun, the same
size for each. When finished, the kid says (30a), the athlete (30b):
(30) a. the kid told her mum:
wo chi le yi da ge mantou,
I eat PFV one big CL steamed bun
xianzai shenme ye bu xiang chi.
now
what FOC not want eat
I just ate a big steamed bun, and I do not want to eat anything at the
moment.
b. the athlete told his wife:
wo zhi chi le
yi xiao ge mantou,
xianzai hai e.
I only eat PFV one small CL steamed bun now
still hungry
I only ate a small steamed bun a moment ago, and I am hungry now.

Pre-classifier adjectives modifying Cl+N

191

The same item on the menu is considered to be a big entity by a three


year old and a small entity by an adult basketball player. Again, da/xiao
big/small expresses an evaluation of the quantity as big or small with
respect to the eaters consumption ability.

4.1.2. Contexts of significance


This section looks at the context of significance for pre-classifier adjectives. Consider (31):
(31) a. yi ren
fa le
yi guan xiao yagao.
one person give PFV one CLtube small toothpaste
b. yi ren
fa le
yi xiao guan yagao.
one person give PFV one small CLtube toothpaste
Each is given a small tube of toothpaste. [from J.M. Lu 1987: 60]
(31a) and (31b) characterize two different types of nominal modifiers.
In (31a), the adjective xiao small directly precedes the head noun yagao
toothpaste and is an adnominal modifier, while in (31b), the adjective
xiao small is a pre-classifier adjective. (31a) means that the tube of
toothpaste is small, of a small type. (31b) means that the tube of toothpaste
is an insignificant thing that does not have much value.
The following is a scenario for (31b). The boss wanted to award the excellent employees in the company. He finally awarded each of them with a
tube of toothpaste, which comes in the biggest size tube in the world. The
employees had expected a more interesting present (the hope was for a
BMW automobile). One of them complained: each person got yi xiao
guan yagao as an award. The use of the pre-classifier xiao small highlights the fact that a tube of toothpaste is of low degree of significance or
value etc. in this context. The actual size of the tube of toothpaste is irrelevant.
(31) was an example with a counting classifier and a non-naturally
atomic noun. (32) shows the same contrast with a naturally atomic noun:
(32) a. mei-ge
yuangong dou jiang le
yi ge da zuanshi.
every-CL employee all award PFV one CL big diamond
b. mei-ge
yuangong dou jiang le
yi da ge zuanshi.
every-CL employee all award PFV one big CL diamond
Every employee is awarded with a big diamond.

192

Adjectival modification in classifier phrases: pre-classifier adjectives

(32a) with adnominal da big means that the diamond itself is big w.r.t.
a certain standard, say, being 1.0 carats is considered to be a big diamond;
(32b) with pre-classifier da big means that the diamond is considered to
be a significant award for the employees in the eyes of the speaker, but the
diamond itself can be small, say, 0.3 carats.
Thus, in the context of significance, pre-classifier da/xiao big/small
express the speakers personal evaluation of the entity denoted by Cl+N
with regard to the significance of the entity in the context.

4.1.3. Contrastive contexts


The examples in (33) illustrate the use of pre-classifier adjectives in contrastive contexts.
(33) a. zhe yi xiao fu hua
hua le
ta ban-nian shijian.
this one small CL painting cost PFV him half year time
This small painting cost him half year to finish.
[from J. Lu 1987: 55]
b. ta wuli jiaju
tai duo, baifang-de ye bu jiangjiu.
his home furniture too much arrange
Foc not carefully
Yi jin
men youbian yi da zhang jiu-shi mu-chuang,
once enter door right
one big CL old-style wooden bed
zuo-bian yi ge da ligui,
zhuozi, yizi,
left-side one CL big wardrobe table chair
shafa ba wuzi gaode yongyubukan.
sofa OM house made crammed
He has too much furniture at home and it was not arranged properly.
Once entering, on the right side lies a big old style wood bed and on
the left lies a big wardrobe, a table and chairs, which make the house
look crammed.
[from Y.L. Luo 1988: 8]
In (33a), the pre-classifier adjective xiao small in the ClP of yi xiao fu
hua a small Cl painting does not refer to the size of the painting as small
in comparison to other paintings, but expresses that from a particular perspective it is small. For example, given that he spent three years painting it,
one could have expected something bigger.
Typically, we can choose to evaluate the painting from another perspective, and then it will be yi da fu hua a big painting. For example, realizing
the complexity of the techniques used in the painting, we may regard it as a

Pre-classifier adjectives modifying Cl+N

193

big painting. Thus, the adjective expresses a subjective evaluation from the
perspective of the evaluator.
In this context, the pre-classifier adjective highlights the contrast between the size of the painting and the evaluation norm.
Similarly, in (33b), the pre-classifier adjective da big highlights the
contrast between the small space of the house and the big size of the bed.
Given that there is not enough space for so much furniture, the bed is a
gigantic object for such a small room, it makes the room look even more
crowded. The pre-classifier adjective does comment on the beds actual
size: in a study of 8 m2, a small bed is big (and (33b) can be used), and it
makes the room look extremely crowded.
To sum up, pre-classifier adjectives like da/xiao big/small assign a
low or high value to the atomic entity in the denotation of Cl+N on a
certain scale like a scale of quantity, significance etc., from a particular
subjective perspective. Such a subjective/personal choice of context implies that pre-classifiers will show great variability regarding what subjective dimension the pre-classifier relates to.

4.2. Syntactic structure


Tang (1990) is one of the few to explicitly discuss the syntactic status of
the adjective and the classifier. She notes that pre-classifier adjectives differ from attributive adjectives in adnominal or predicative positions in that
they can only be used in bare form and cannot be followed by the modification marker de and modified by degree modifiers like hen very (also
see Tsou 1976, Paris 1981). Look at (34):
(34) a. yi
da
xiang
one
big
CLbox
a big box of books
b. *yi
da
de
one
big
Mod
c. *yi
hen
da
one
very big

shu
book
xiang
CLbox
de
Mod

shu
book
xiang shu
CLbox book

Tang (1990:419) claims that adjective and classifier forms a single constituent, that function as a compound, a zero-level category, rather than as a
phrase. Tang proposes that the ill-formedness of (34b-c) can be explained

194

Adjectival modification in classifier phrases: pre-classifier adjectives

by the Lexical Integrity Hypothesis which says that no syntactic operation may affect only part of a lexical item.
The Lexical Integrity Hypothesis successfully captures the property of
the Cl+N phenomenon (the bare form in (34)), but it cannot explain
another constraint of pre-classifier adjectives, since making Adj+Cl a compound has nothing to say about which adjectives can enter into this compound.
More seriously, the proposal cannot account for the enriched meanings,
and the interpretational variability of pre-classifier adjectives in the three
contexts I discussed. If I treat Adj+Cl as a compound in (31b), xiaoguan
would refer to a small tube. But, as we saw, (31b) does not refer to a tube,
but a(n atomic) tube of toothpaste and the pre-classifier xiao small makes
the evaluative statement that a tube of toothpaste is a small (insignificant)
thing. Given this, it is not plausible to assume that the pre-classifier adjective and the classifier form compounds, as suggested in Tang (1990).
Given that pre-classifier adjectives in Chinese express properties of the
atomic entity referred to by Cl+N, I suggest that they modify the whole
constituent Cl+N, but not the classifier or the noun. It is shown in Section 2 that pre-classifier adjectives are only possible before [+counting]
classifiers. I suggested in Chapter 6 that on the counting reading, the classifier first takes NP as complement to form ClP, which is then taken as
complement by NumP. I suggest that pre-classifier adjectives are located in
the Spec of the ClP3, as in (35).
(35)

NumP
Num ClP
yi one
AdjP
Cl
da big
Cl
NP
ping bottle
N
shui water

Liu (2010) discusses the use of adj+Cl in post-copula positions in the southern
Min dialect. He analyzes adj+Cl as an adjective compound. It is good to point out
that Adj+Cl cannot be used in post-copula position in Mandarin, and that the
Adj+Cl construction under discussion here differs syntactictically and semantically
from the construction Liu discusses.

Semantics of pre-classifier adjectives

195

I said earlier that the classifier phrase with pre-classifier adjectives in


Chinese is similar to the concrete portion reading discussed in Partee and
Borschev (2012). However, our structure in (35) is different from what
they suggest for the concrete portion reading. It is a counting structure per
se. According to Partee and Borschev (2012:474), the concrete portion
reading is close to the measure readings and involves the same syntax but
does not express measure. On the concrete portion reading, two glasses of
milk is a predicate true of milk that fills two glasses; unlike the two measure readings, on this reading there could be a big glass and a small glass.
Our counting structure proposed in (35) can account for the constraint
that pre-classifier adjectives are restricted to dimensional adjectives like
da/xiao big/small (recall the examples in (19) and (20)). Intuitively, I may
simply assume that atoms are by nature individuated entities and have size
and quantity properties, this is why they can be expected to be modified
da/xiao or other dimensional adjectives like chang, long (James C.-T.
Huang p.c.).
From the theoretical perspective, on the counting reading, the classifier
denotes a function from kind to a set of atoms. Sometimes, the classifier
phrase is also called atom Phrase (Chierchia 2010). I may assume that
the Cl0 has the head feature [+atomic], and the element in its specifier position may check atomicity (or the Spec feature and the Head feature have to
agree).

5. Semantics of pre-classifier adjectives


I now propose a semantics for pre-classifier adjectives making use of the
notion of the expressive dimension discussed by Potts (2007) and
Schlenker (2007).

5.1. Introduction to expressives


Potts (2007) identifies six characteristics of expressives: independence,
non-displaceability, perspective dependence, descriptive ineffability, immediacy, and repeatability. Pre-classifier da/xiao big/small have all these
features, but I will here only concentrate on the first three properties.
Independence: the expressive content contributes a dimension of
meaning that is separate from the regular descriptive content.

196

Adjectival modification in classifier phrases: pre-classifier adjectives

Nondisplaceability: the expressive content predicates something of the


utterance situation.
Perspective dependence: the expressive content is evaluated from a
particular perspective, in general the speakers, but this can vary.
I discuss these concepts in an example by Potts (2007).
(36) That bastard Kresge is famous.
The independence property means that we can change or remove the
expressive content of a phrase without affecting its descriptive content
(Potts 2005:3.6.3): we can assent to the descriptive content of (36) the
proposition that Kresge is famouswithout thereby assenting to the characterization of Kresge as a bastard.
Nondisplaceability means that expression bastard is indexical to the
utterance situation: the expressive proposition is valid for the utterer at the
time and place of utterance (Potts 2007).
Perspective dependence means that the expressive is a predicate of personal taste. Such predicates depend on a contextual parameter of a judge
(Potts 2007). Expressives express the judges personal attitude toward a
certain individual or a current state of affair in the speech-act context. In
(29) the speaker is also the contextual judge, bastard is evaluated from the
speakers perspective, and the contextual judge is sharing his attitude toward Kresge with us, his personal dislike of Kresge.
Schlenker (2007) proposes a propositional analysis of expressives. For
Schlenker (2007:237), expressives are indexical (evaluated with respect to
a context), attitudinal (predicating something of the mental state of the
agent in the context) and sometimes shiftable (in that the context of evaluation need not be the context of the actual utterance). The differences between Potts (2007) and Schlenker (2007) are not very important in the
present context; I adopt here the basic assumptions concerning expressives
that they share, assuming that Schlenkers notion of indexical is roughly
equivalent to Pottss notion nondisplaceable, and that Schlenkers notions
of attitudinal and shiftable are roughly equivalent to Pottss notion of
perspective dependence.

Semantics of pre-classifier adjectives

197

5.2. Pre-classifier adjectives as expressives


The three contexts where ClPs with pre-classifier adjectives appear are
repeated in (37). (37a) is a consumption context, (37b) a context of significance, and (37c) a contrastive context.
(37) a. ta na / he
le
yi da
ping
futejia.
he carry/drink PFV one big CLbottle vodka
He carried/drank a big bottle of vodka.
b. yi ren
fa le
yi xiao guan yagao.
one person give PFV one small CLtube toothpaste
Each is given a small tube of toothpaste.
c. zhe yi xiao fu hua
hua le
ta ban-nian shijian.
this one small CL painting cost PFV him half year time
This small painting took him half year to finish.
(37a) means that the wine in the bottle is a large quantity for the subject;
(37b) means that the tube of toothpaste is an insignificant object in the eyes
of the evaluator; (37c) means that the painting is considered to be small in
view of the long time he spent painting it.
I use (37a) to show that pre-classifiers satisfy the characteristic properties of expressives. In (37a), yi da ping futejia a big bottle of vodka
means that the vodka in the bottle has a large value on a quantitative scale.
At least two different scales are natural possibilities here. The preclassifier da can mean that the bottle of vodka contains a heavy quantity of
vodka. Suppose that a three-year old kid is trying to help his dad to carry a
bottle of vodka, and he is staggering his way from the kitchen. Obviously, a
bottle of vodka is too heavy for him. In this situation we can use (37a)
(with carry, of course, and not with drink). The classifier in (37a) can also
mean the bottle of vodka contains a voluminous quantity of vodka in relation to the subjects capacity for liquor. Suppose that Tom is not a good
drinker and he can only drink one glass of vodka maximally. But on the
night that he was dumped by his girlfriend, he drank a whole bottle of vodka. Also in this situation I can use (37a). In this case, yi da ping vodka implies that Tom drank too much alcohol.
Pre-classifier adjectives are shiftable. We saw above that the preclassifier da big can express the speakers personal evaluation of the size
of the bottle of vodka from two different perspectives, weight and capacity
for liquor. In both cases, the speaker shifts the context of utterance to the
context of personal evaluation.

198

Adjectival modification in classifier phrases: pre-classifier adjectives

Pre-classifier adjectives are attitudinal. In the context of the drinking


example, it is the speakers personal opinion that one bottle of vodka is a
large quantity for Tom, exceeding his maximal capacity for liquor. Tom
himself may feel different about that. We saw the same in the example of
the steamed bun in (30) where the speaker thinks that the quantity of the
steamed bun is big for a small child, but small for an athlete.
Pre-classifier adjectives are indexical. The truth of (37a) depends on the
utterance context. For example, in the drinking situation, the speaker thinks
that a bottle of vodka is too much for Tom in that particular situation. This
does not mean that a bottle of vodka is always too much for Tom. If Tom
continues to drink, he will become a drunk who can drink 3 bottles of vodka easily. In that case, if Tom drank a bottle of vodka, I would say that he
drank yi xiao ping vodka a small bottle vodka. Therefore, the personal
evaluation expressed by pre-classifier adjectives is indexical to a particular
context/situation.
I conclude that pre-classifiers da/xiao big/small are expressives. I will
not try to fit pre-classifiers into Pottss (2007) semantics of expressives.
Our aim here is modest: I want to show how pre-classifier adjectives behave differently from attributive/predicative adjectives.
When the adjective precedes the classifier, the classifier in ClPs has a
counting interpretation. The semantics of classifiers in the counting reading, which I argued for in Chapter 6, is illustrated in (38):
(38) NCl NN = x.1(x)(kN k) (1(x)) 2(x) = k
The set of pairs of type dk whose first element is an atomic individual
instantiation of kind kN in context k with property and whose second
element is k.
Following Kennedy (1997), I assume that gradable adjectives denote
measure functions from objects to abstract representations of measurement,
is sets of degrees. I take the degree adjective of da big in its normal interpretation to be a predicate of individuals of type <d, t> as in (39). normBIG
is the degree that is the norm for what counts minimally as big.
(39) NdaN= x. MEAS(x) > normBIG
the set of individuals that measure higher than the norm.

Semantics of pre-classifier adjectives

199

In English, when the adjective big modifies NPs, it shifts to its modifier
interpretation at type <<d,t>,<d,t>> or <<dk,t>,<dk,t>>, depending on
whether it modifies mass or count nominals (see Rothstein 2010).
In the case of pre-classifier adjectives, da/xiao big/small syntactically
modify the count Classifier Phrase Cl+N. Semantically, I take the preclassifier interpretation to be a count predicate of type <dk, t>. I represent
the evaluative function of the pre-classifier adjectives by restricting the
norm of comparison to be dependent on the perspective of the judge in
context k. judgek,BIG is the minimal degree for what counts as big according
to the contextually provided judge in context k. da-Ev is da in its evaluative
function:
(40) NdaEvN =x.MEAS (1(x))>judgek, BIG
I show how the semantics in (40) explains the interpretation of preclassifier da-Ev in (37a). The interpretation of big or small in (40) depends
on judgek,BIG, the minimal value for what counts as big according to the
evaluation perspective chosen by the judge. In one perspective we understand yi da ping futejia a big bottle of vodka as concerning weight, from
the other perspective we understand it as concerning volume. If we choose
to evaluate the bottle of vodka from the perspective of weight, we have to
further identify the norm of evaluation. In our example, the norm of evaluation is a three-year olds carrying ability. If we choose to evaluate the
bottle of vodka from the perspective of volume, the norm of evaluation will
be different, in our example it was subjects capacity for drinking liquor. I
assume that all these decisions are encoded in the parameter judgek,BIG. I
give the following semantics for (37a):
(37) a. ta na / he
le
yi da
ping
futejia.
he carry/drink PFV one big CLbottle vodka
He carried/drank a big bottle of vodka.
(41) a. NpingN = kx.1(x)(BOTTLEk)CONTAIN(1(x),k)2(x) = k
b. Nping futejiaN=NpingN (NfutejiaN)
=x.1(x)(BOTTLEk)CONTAIN(1(x),vodka)2(x) = k
c. Nda-Ev ping futejiaN=Nda-EvN(Nping futejiaN)
=x.1(x)(BOTTLE k)CONTAIN(1(x),vodka)2(x) = k
MEAS(1(x)) > judgek, BIG

200

Adjectival modification in classifier phrases: pre-classifier adjectives

5.3. Pre-classifier adjectives and plurality


Most of the examples with pre-classifier adjectives I have discussed so far
have only yi one as Num. However, ClPs with pre-classifier adjectives
are also possible with other numerals. According to Lu (1987), of 328 examples, 88.1% take yi one. The next most frequent quantifier is ji several (4.6%). The numerals together form 6.4%.4 I have also shown some
examples with plural ClPs, namely (24), as repeated in (42).
(42) ni bie xiaokan zhe
ji
xiao li
huangdou, zhe shi
you not belittle these several small CLgrain soybean
this be
women hua le duoda-de xinxue cai peiyu dedao de liangzhong.
we spend PFV much effort Foc cultivate gain Mod selected seed
Do not belittle these small grains of soybeans. They are the selected
seeds that were cultivated with painstaking effort.
According to the semantics in (40), the pre-classifier da/xiao are predicated of a set of atomic individuals, which can be modified by numbers and
be pluralized. This correctly predicts that in plural Adj+Cl+N, the preclassifier adjective has a distributive reading, in which it distributes over
the individual entities in the denotation of Cl+N. (42) means that each soybean is an insignificant entity in the eyes of the speaker. Accordingly, I
suggest that the pluralization operator, as notated PL*, scopes over the
adj+Cl+N.
I gave the semantics of pluralization in Chapter 4. Here I give pluralization for a predicate P of type <dk,t>: *(P) is given in (43):
(43) *(P) = x.1(x)*1(P)2(x)=k
The semantic derivation of xiao li huangdou small CLgrain soybean is
given in (44) and (45).
(44) Nli huangdouN
= x.1(x) (>?SOYBEAN k) 2(x)=kGRAIN (1(x))

Note that in these statistics, Lu (1987) did not distinguish between individual and
non-individual classifiers.

Conclusions

201

(45) Nxiao li huangdouN= Nxiao-EvN(Nli huangdouN)


=x.1(x)(>?SOYBEANk)2(x)=kGRAIN(1(x))
MEAS(1(x))> judgek, SMALL
Pluralization applies to this predicate and gives the set of pluralities of
atomic soybeans, where each member of the plural entity is counted as an
insignificant entity:
(46) x.1(x)*(>?SOYBEANk)2(x)=kGRAIN(1(x))MEAS(1(x))>
judgek, SMALL
This expression can be modified by numerical adjectives in the normal
way.

6. Conclusions
I examined the syntactic and semantic properties of pre-classifier adjectives. I showed that pre-classifier adjectives can only appear in the counting contexts, before [+C, -M] and [+C, +M] classifiers. I saw that preclassifier adjectives are not unrestricted and can only be used in certain
contexts, most prominently: the consumption, the contrastive and the significance contexts. I argued that pre-classifier adjectives modify neither classifiers nor nouns, but Cl+N. Syntactically, they are the Spec of the ClP. I
proposed a semantics of pre-classifier adjectives as expressives; with
interpretations sensitive to a contextual judges personal evaluation or attitude in a situation to the entities in the denotation of Cl+N.

Chapter 8
Modification marker de in classifier phrases

1. Introduction
It is commonly agreed that the particle de is a modification marker which
takes a complement phrase and that [XP [de]] has the function of modifying the noun it is a sister of. Here, the variable X can be adjectives, relative
clauses, possessors etc..
(1) a. yi ge [piaoliang
de]
one CL beautiful
Mod
a beautiful girl
b. [mai
huoche de] xiao
sell
matches Mod little
the little girl who sells matches
c. [xiaowang
de]
shu
Xiaowang
Mod book
Xiaowangs book(s)

nhai
girl

[Adj-de-N]

nhai
girl

[RC-de-N]

[Poss-de-N]

The syntactic status of de is under debate. Zhu (1961), Paris (1979) and
Li and Thompson (1981) argue that de is a nominalizer and XP-de is a
nominal phrase. A. Li (1990), Sproat and Shih (1991) and Den Dikken and
Singhapreecha (2004) propose that [XP [de]] is best treated as a relative
clause headed by de, while Paul (2007, 2010) argues that the particle de is
a non-root complementizer and XP-de is CP, where the head de can take
any phrase as a complement. I will not attempt to choose between these
analyses here. What is important to us is that [XP [de]] is a modifier, that is,
a predicate expression.
In this chapter, I study the particle de occurring after classifiers, as in
the sequence of Num-Cl-de-N. The phenomenon to be examined are illustrated in (2) and (3):

Introduction

203

(2) a. ta he
le
san ping
(de) jiu.
he drink PFV three CLbottle DE wine
He drank three bottles of wine.
b. fuwusheng kai le
san ping
(#de) jiu.
waiter
open PFV three CLbottle DE wine
The waiter opened three bottles of wine.
(3) a. yanchanghui lai
le
shang qian
ge (de) guanzhong.
concert
come PFV above thousand CL DE audience
In the concert came more than one thousand audience.
b. zhe ge xuexiao zhi you wu ge (#de) laoshi.
this CL school only have five CL DE teacher
In this school, there are only five teachers.
The modification marker de can follow both non-individual and individual classifiers, such as ping bottle in (2a) and ge in (3a). However, the
use of de is not without constraint. It is not appropriate to use de in (2b)
and (3b), though they have the same classifiers as those in (2a) and (3a).
In this chapter, I will address the following issues concerning the syntax
and semantic interpretation of de after classifiers in Chinese:
(a) What is the syntactic structure of Num-Cl-de-N? Can it be said to fit
in with other occurrences of de, where de occurs roughly in the structure
[XP de]? I will argue that indeed even on a pseudo-partitive interpretation,
san ping de has the syntactic structure [[san ping] de], and that it is a predicate expression with a modificational interpretation.
(b) Does the presence of de after Num-Cl have any interpretational effect? I will argue that it requires a measure interpretation of the classifier,
and the predicate phrase [[san bang] de] denotes a measure or dimensional
property.
(c) What kind of classifiers does de appear with? Since it requires the
classifier to be interpreted as a measure expression, de naturally follows
explicit measure expressions such as gongjin kilo. I show that it can also
occur with a counting classifier, but only in that it can induce a measure
interpretation of the classifier.
In section 2, I review three proposals about the licensing condition of
de after classifiers, namely, Cheng and Sybesmas (1998) massifier proposal, Tangs (2005) information weight account and Hsiehs (2008)
indeterminacy account. Section 3 reports the findings made in Li and
Rothstein (2012). In Section 4 and 5, I formulate proposals for the semantics of de after measure and counting classifiers respectively. I propose that

204

Modification marker de in classifier phrases

the post-classifier de forces the Num and the classifier into a constituent,
with a modifier meaning, expressing measurement. Section 6 concludes
this chapter by explaining the two puzzles pertaining to [+Counting] classifiers with de.

2. Licensing the post-classifier de


This section reviews three proposals about the occurrence of de in NumCl-de-N: the massifier account given in Cheng and Sybesma (1998), the
information weight proposal in Tang (2005), and Hsiehs (2008) indeterminacy account.
Before reviewing these proposals, I make a distinction between two
types of readings expressed by de-marked classifier phrases in Chinese, but
I only focus on one of the readings, namely, the pseudo-partitive reading in
this research.
Jiang (2008) makes use of Schwarzschilds (2006) characterization of
the distinction between measure phrases in pseudo-partitives and in attributive modifiers and shows that the phrase Num-Cl-de-N in Chinese has two
readings: the attributive reading (4a) and a pseudo-partitive reading (4b).
(4) san bang de
yingtao
three CLpound MOD cherry
a. three-pound cherry
b. three pounds of cherries

[attributive reading]
[pseudo-partitive reading]

In the attributive reading, san bang de is an adnominal modifier and behaves like a classifying adjective, modifying the head noun yingtao, with
the paraphrase in (4a). Jiang argues that in this reading, (4) denotes a cer1
tain type of cherry, the three-pound type of cherry . In this reading, [XP
de] has the same syntactic position and functions as the adjectival modifier
[AP de], illustrated in (1), and I assume that this is prima facie evidence
that here [XP de] is a predicate with a predicative interpretation.

It is unclear whether for Jiang, this means that each individual cherry has to weigh
three pounds, or whether three-pound cherry is a more general name of a kind of
cherry. While this is an important question semantically, it is not necessary for us to
solve it here, since on either reading, san bang de is an attributive modifier in the
sense of Schwarzschild (2006).

Licensing the post-classifier de

205

The pseudo-partitive reading, expressed in English by (4b), is associated with the normal classifier use of san bang, and in this reading, the
Num-Cl-de sequence expresses information about the quantity of entities.
On this reading, (4) denotes a set of pluralities of cherries whose overall
quantity is three pounds.
Jiang (2008) claims that the two readings of san bang de yingtao as
shown in (4) have two different syntactic derivations and structures. She
proposes that on the attributive reading, Num-Cl is a degree phrase, which
is treated as a relative clause, and which is taken as complement by the
complementizer de, as in (5a), and that the pseudo-partitive reading of
Num-Cl-de-N is derived by moving the degree phrase of Num-Cl to the
specifier position of a null classifier phrase, as in (5b). While not necessarily accepting the details of her analysis, I will follow it in assuming that
the attributive readings and the pseudo-partitive readings have two different structures.
(5) a. attributive reading
NP
CP
SC
NP
ti

N
yingtaoi cherry

C
de
DegreeP
san bang
three pound

b. pseudo-partitive reading
ClP
san bang

Cl

Cl

NP
CP

SC
NP
ti

N
yingtao
de cherry

DegreeP

san bang
three pound

In this research, I will not discuss the attributive uses of Num-Cl-de. I


will focus on the pesudopartitive use of Num-Cl-de, the reading that is
discussed in Cheng and Sybesma (1998), Tang (2005) and Hsieh (2008).

206

Modification marker de in classifier phrases

2.1. De with mass classifiers (Cheng and Sybesma 1998)


Cheng and Sybesma (1998) make two independent claims about postclassifier de. First, they claim that the post-classifier de only follows what
they call mass classifiers like ping bottle and xiang box in (6), but not
count classifiers like the general classifier ge and the classifier for animals tou head in (7).
(6) a. san bang (de) rou
three CLpound DE meat
three pounds of meat
b. liang xiang (de) shu
two CLbox DE book
three boxes of books
(7) a. jintian lai
le
san ge (*de) laoshi.
today come PFV three CL DE teacher
Today three teachers came.
b. women you ba
tou
(*de) niu.
we
have eight CLhead DE cow.
We have eight heads of cows.
However, as shown in Chapter 3, as well as in Tang (2005) and Hsieh
(2008), there are counterexamples to the claim that de only follows mass
classifiers. For example, in (8), classifiers like ke and li are typical individual classifiers in Chinese, but they allow a post-classifier de:
(8) a. yi nian yue
zhongzhi-le yi-bai-sishi-duo-wan
one year about plant-PFV one-hundred-forty-more-million
ke de shumu.
CL DE tree
(They) planted more than 1,400,000 trees a year.
b. yi ge yi sui de
yiner mei-ci
zhi neng
one CL one year Mod baby every time only can
chi 1/3 li de ganmao-yao.
eat 1/3 CL DE cold pill
A one year old baby can only take 1/3 of a cold pill every time.
The second claim made by Cheng and Sybesma (1998) about de concerns its semantics. They argue that the post-classifier de introduces a

Licensing the post-classifier de

207

measure reading for mass classifier phrases. According to them, de-less


mass ClPs have by default an individuating reading, and de-marked ClPs
have a measure reading. The examples in (9) and (10) are taken from their
paper.
According to Cheng and Sybesma (1998), Num-Cl-N san wan tang
three bowls soupin (9a) has a counting reading. It means that there are
three individual bowls, each of which is filled with soup and stands on the
table. (9b), with Num-Cl-de-N, i.e. san wan de tang three bowls de soup
has a measure reading. It means that the soup is spilled all over the table
and there are no bowls.
(9) a. zhuo-shang you
san wan tang.
table on
there-be three CLbowl soup
There are three bowls of soup on the table.
b. zhuo-shang you
san wan de tang.
table on
there-be three CLbowl DE soup
There is enough soup on the table to fill in three bowls. 
As for the contrast in (10), they said that (10a), without de, is infelicitous, since bei glass without de, induces the default reading that the wine
is consumed from an actual glass. In (10) bei glass is followed by de, and
(10b) merely expresses that the quantity of the wine consumed was three
glass-units.
(10) a.# ta yong xiao-wan he
le
san
bei
jiu.
he use small-bowl drink PFV three CLglass liquor
b. ta yong xiao-wan he
le
san bei
de jiu.
he use small-bowl drink PFV three CLglass DE liquor
He used a small bowl to drink three glasses of liquor.
I agree with Cheng and Sybesma that Num-Cl-de-N in (9b) and (10b)
has a measure reading. However, I showed with ample evidence in Chapter
6 that Num-Cl-N, as those in (9a) and (10a), are ambiguous between count2
ing and measure readings. The relevant examples are reproduced in (11).
2

I argued in chapter 6, that [+C, +M] classifiers are equally open for counting and
measure interpretations, that [+C, -M] classifiers have by default a counting reading,
and measure readings by coercion only, that [-C, +M] classifiers have by default a
measure reading, and a counting reading by coercion, while [-C, -M] classifiers, i.e.
kind classifiers, can neither count nor measure individuals.

208

Modification marker de in classifier phrases

(11) a. wo ling
le liang ping
jiu,
I carry PFV two CLbottle wine
zuo shou yi ping, you shou yi ping.
left hand one CLbottle right hand one CLbottle
I carried two bottles of wine, one in the left hand and the other in
the right hand.
b. wo zonggong he
le
you yi ping putaojiu,
I altogether drink PFV have one CLbottle wine
ban ping bai-de ban ping hong-de
half CLbottle white half CLbottle red
I drank a bottle of wine in total, half bottle of a white wine and
half bottle of a red wine.
(11a) shows the counting reading of the classifier phrase liang ping jiu
two bottles wine. And (11b) shows the measure reading in which the
classifier ping bottle measures the quantity of wine as one bottleful, say,
750 ml.
According to the native speakers we consulted, (9a) and (10a) can have
a measure interpretation. (10a) is not regarded as very bad, and, they claim,
it is completely acceptable, if you know the volume of the bowl from the
context.
It seems, then, that whether the classifier phrase can get a measure reading, depends on whether the classifier is typically used as a standard measure unit for the kind of entities in the noun denotation, and whether enough
information can be inferred from the context about the capacity of the container. Glass is a frequently used container for liquor, but it is not a standard or original measure unit for liquor, since we can think of glasses with
different sizes. Therefore, when we use a glass of liquor without any
further contextual information, it is difficult to get a measure reading (cf.
(10a)). But if we use the classifier ping bottle (instead of bei glass) to
modify the noun liquor, then we get a measure reading easily, as in (12a):
(12) a. ta yong xiao wan he le san ping hongjiu.
he use small bowl drink PFV three CLbottle red wine
He drank three bottles of red wine with a small bowl.
This is because the classifier bottle is a prototypical package/container
for liquor, like barrel for oil, cup for coffee, and volume for book. A conventional combination of classifier and noun facilitates measure readings.
For example, the conventional combination of bottle and wine makes the

Licensing the post-classifier de

209

quantity of 750 ml salient, because wine bottles conventionally take 750 ml.
Accordingly, (12a) has a measure reading, and means that he used a small
bowl to drink 3*750 ml of wine.
Similarly, it is conventional to put bubble tea (a kind of popular beverage in China) in a plastic cup of 700 cc. Because of this, (12b) can have a
measure reading, asserting that he uses a small bowl to drink 3*700 cc of
bubble tea.
b. ta yong xiao wan he
le san bei nai-cha
he use small bowl drink PFV three CLcup milk-tea
He drank three cups of bubble tea with a small bowl.
Therefore, while de forces a measure interpretation, the absence of de does
not force a counting interpretation.

2.2. Information weight (Tang 2005)


In chapter 3, we discussed Tangs (2005) criticism of Cheng and Sybesma
(1998). We repeat some of her arguments here. Contra Cheng and Sybesma,
Tang (2005) develops ideas set out first in Tang (1990) and argues that it is
possible for de to follow both individual (13) and non-individual classifiers
(14).
(13) a. liang ben
de shu
[individual classifier]
two CLvolume DE book
books that are sorted in accordance with two in number
b. wu ge de pingguo
five CL DE apple
apples that are sorted in accordance with five in number
(14) a. liang mi de bu
[non-individual classifier]
two meter DE cloth
cloth that is sorted in accordance with two meters
b. liang bang de rou
two pound DE meat
meat that is sorted in accordance with two pounds
However, the examples in (13) and (14) do not embody the same phenomenon as what Cheng and Sybesma (1998) discuss. According to the

210

Modification marker de in classifier phrases

translation of (13) and (14) provided by Tang (2005), Num-Cl-de seems to


behave like a classifying adjective, which expresses properties that are able
to establish subtypes of entities. These classifier phrases seem to be of the
kind which Jiang (2008) calls attributive. However, Cheng and Sybesmas examples of Num-Cl-de-N concern the measure readings, which are
similar to what Jiang calls pseudo-partitives.
It is important to see that these pseudo-partitive and attributive uses of
classifier phrases really are syntactically different and occur in syntactically different positions, as argued by Jiang (2008). Num-Cl-de-N in the attributive reading is generated within the NP (as shown in (5a)), while NumCl-de-N in the pseudo-partitive reading is generated outside the NP. This
correctly predicts that a classifier can appear both in the attributive and the
classifier head positions in the same NP, as illustrated in (15).
(15) wo mai le yi xiang de liang ce (zhuang) de shu.
I buy PFV one CLbox DE two CLvolume pack DE book
I bought a box of two-volume books.
Impossible: I bought a box with two books in it.
In the possible reading here, we have no idea how many volumes are in
the box, since it is a box of books-which-come-in-two-volumes (e.g. a box
of dictionaries each consisting of two volumes, one English-Mandarin, and
the other Mandarin-English). The impossible reading is the reading in
which the pseudo-partitives are stacked: it is impossible to use (15) to assert that I bought a box containing two books.
While Tang (2005) mainly discusses the attributive type of Num-Cl-deN, she also gives a couple of examples, in which Num-individual Cl-de-N
do not have an attributive meaning, but a measure reading:
(16) a. mingtian de huodong xuyao yi bai
zhang de fangzuozi.
tomorrow Mod activity need one hundred CL DE square talbe
Tomorrows activity needs one hundred square tables.
b. yi nian yue zhongzhi le
yi-bai-sishi-duo-wan
one year about plant
PFV one hundred forty more million
ke de shumu.
CL DE tree
(They) planted more than 1.4 million trees a year.
In (16a), yi bai zhang-de one hundred-Cl-de expresses that the number
of individual tables is one hundred, and in (16b), yi-bai-sishi-duo-wan-zhi-

Licensing the post-classifier de

211

de expresses that the number of individual trees is 1.4 million. The interpretations of these examples in (16) are similar to the cases discussed in
Cheng and Sybesma (1998). But the classifiers in (16), such as zhang and
ke, are uncontroversially individual classifiers, and the corresponding classifier phrases in (16) are uncontroversially pseudopartive. Thus, the examples in (16) are true counterexamples against Cheng and Sybesma (1998).
Unfortunately, Tang (2005) does not distinguish the examples in (16)
from the attributive ones in (14) and (15). She treats all of them as complex
NPs with Num-Cl-de as modifiers. Tang suggests that the felicity of the
examples in (16) is related to the information weight of the modifiermodifiee relation: weighty modifiers facilitate the measure interpretation.
Num-Cl phrases like yi bai zhang one hundred picees and yi-bai-sishiduo wanb ke one million and forty hundred thousand Cl are complex and
heavy modifiers, and they can have the particle de after the individual classifier. In contrast, wu ke five Cl is, according to Tang, a simple modifier,
and wu ke de shu five Cl de trees is ungrammatical.
Tangs information weight account works for the examples in (16),
but does not extend to the example in (17). In (17), the numeral yi bai lin yi
one hundred and one is a complex number, but it does not allow the modification marker de.
(17) * mingtian de huodong xuyao yi-bai-lin-yi zhang de fang-zuozi.
tomorrow Mod activity need 101
CL DE square table
Tomorrows activity needs one hundred and one square tables.

2.3. The indeterminacy account (Hsieh 2008)


Hsieh (2008) takes the discussion much further. She gives more examples
where de follows individual classifiers to get a pseudo-partitive reading.
The examples provided by Hsieh in (18) and (19) are taken from the Academia Sinica Corpus.
(18) a. jin yi-bai
wei de qianjiu renyuan
close one-hundred CL DE rescue worker
close to one hundred rescue workers
b. hao ji
bai
tiao de haishe
quite several hundred CL DE sea snake
quite several hundreds of sea snakes

(A.S. Corpus)

(A.S. Corpus)

212

Modification marker de in classifier phrases

(19) a. pai-le
wu-qian
zhang de yizi (A.S. Corpus)
put:in:order-Perf five-thousand CL DE chair
put 5 thousand chairs in order
b. wu-bai-wan
zhi de yazi
(A. S. Corpus)
five-hundred-ten:thousand CL DE duck
5 million ducks
c. tongshi
you peiyang le
23 wei de zhongzi jiangshi
meanwhile again cultivate PFV 23 CL DE seed lecturer
at the same time 23 seed lecturers were educated (A.S. Corpus)
According to Hsieh (2008), the examples in (18) and (19) are characterized with two different contexts. She proposes that in (18), the de-marked
individual ClP expresses a non-fixed or indeterminate quantity, as marked
overtly by approximators. In (18a) the NumP is modified by an approximation modifier jin close to, and in (18b), the Num is ji several, which
expresses an unfixed or indeterminate quantity. The context in (18) is
called an indeterminacy context.
In contrast, the examples in (19) do not need an approximation modifier
and the numeral can be precise. (Note that (19c) is not acceptable for me
and for other native speakers I consulted with.) Hsieh calls the context in
(19) an emphasis/contrastive context. By emphasis/contrastive, Hsieh
means that the de-marked ClP emphasizes large quantities as in as many
as Num N. This account is in some respects similar to Tangs (2005) account of information weight.
Li (2007) observes independently that individual classifiers can be followed optionally by de in some cases, as shown in (20). He argues that de
requires a context of quantification of aboutness or approximation, not
a context of exactness. This is similar to Hsiehs examples in (18).
(20) a. ta peng-zhe shi duo ben de shu
he carry-DUR ten more CL DE book
I was carrying more than 10 books.
b. ta yilian
xie
le
liang-bai
duo feng de xin
she continuously write PFV two-hundred more CL DE letter.
She wrote more than 200 letters continuously.
Hsieh, following Lyons (1976) and Tang (2005), assumes that there are
two lexically different types of classifiers in Chinese, mensural and sortal classifiers. Hence, according to Hsieh, the post-classifier de can be
used in the following three contexts:

Licensing the post-classifier de

213

(i) After a mensural classifier


(ii) After a sortal classifier, when the quantity is non-fixed
(iii) After a sortal classifier, when there is emphasis or contrast.
Hsieh (2008) proposes a unified syntactic account for these three contexts. She argues that Num-Cl in Num-Cl-N is a single constituent, which
heads #P or NumP (Number Phrase). The NumP enters into a Spec-Head
relation with the head noun. She also assumes that Numerals such as yi
one or ji several are merged in NumeralP and that demonstratives such
as zhe this or na that are merged in DemP. This is illustrated in (21).
(21)

#P
DemP

#
NumeralP

Num
ge/ping

According to Hsieh (2008), the structure in (21) is a unified structure


for individual and non-individual classifiers. In other words, the NumP
can be occupied either by classifiers or massifier. In either case, the classifier combines first with the number to form a complex classifier and then
Num+Cl modifies the noun. However, as we showed in Chapter 6, there is
good reason to assme that this is not the case. In this, her proposal also
differs from proposals that assume that mass and count classifier phrases
have distinctive structures, i.e. Cheng and Sybesma (1998), and proposals
based on that, like Borer (2004) and Watanabe (2006).
Hsieh (2008) also proposes the feature taxonomy for the Num head in
(18) in terms of the two features: [PL], which is the singular/plural feature, and [Ind] which is the indeterminite/determinite feature, which she
assumes is a sub-feature of [+PL].
(22)

Num
[-PL]

[+PL]
[-Ind]

[+Ind]

214

Modification marker de in classifier phrases

She proposes that de requires the feature of [+Ind], i.e. de is necessary


to identify the [+Ind] feature under the node of [+PL]. This implies that in
Hsiehs analysis, the particle de is part of NumP and the presence or absence of the particle de after the classifier does not correspond to a difference in syntactic structure of the nominal phrase.
There is a serious problem with this analysis. I agree that when de follows an individual classifier, the examples are better with approximators
like jin nearly or duo more. However, such modifiers are not needed
when de follows a non-individual classifier. In that case exact quantities
are unproblematic, as in three bottles de wine. If de requires [+Ind], these
cases are a mystery.
In addition, as we will show in section 3, when the Num is a fraction, it
is also possible to use de after individual classifiers. Fractions are generally
perceived as pretty exact, so they seem to invalidate Hsiehs [+Ind] proposal again. We will deal with this issue below.
To sum up, both individual and non-indiviudal classifiers can be followed by de in appropriately construed contexts, but not in all cases. It is
impossible to associate the use of de after a classifier with a certain lexical
class of classifiers.

3. Unsolved problems (Li and Rothstein 2012)


Li and Rothstein (2012) make three new observations concerning the optionality of de after classifiers. None of the approaches we reviewed in
section 2 is able to account for the constraints observed by Li and Rothstein. The three observations will be introduced respectively.
First, the classifier ping bottle is a mass classifier or a non-individual
classifier, but it cannot be followed by de in all cases. As we showed in
chapter 6, only in the measure context can de follow the classifier ping
bottle, but not in the counting context, as contrasted between (23a) and
(23b) (also see Li and Rothstein 2012).
(23) a. wo kai le
san ping (#de) jiu.
[Counting]
I open PFV three CLbottle DE wine
I opened three bottles of wine.
b. wo-de wei
neng zhuangxia san ping (de) jiu. [Measure]
my stomach can hold
three CLbottle de wine
My stomach can hold three bottles of wine.

Unsolved problems (Li and Rothstein 2012)

215

(23a) is a counting or individuating context, in which the classifier


phrase denotes three individual bottles of wine, each of which the waiter
opened. In this context, de is impossible. In (23b) we have a measure context in which we indicated how much wine was drunk, and here de is allowed but not obligatory. An account which says that ping is lexically a
mensural classifier or massifier, and that this lexical property is what licenses post-classifier de, cannot explain the contrast.
Secondly, de can follow individual classifiers when the Num is a (contextually) high round number.
Against Cheng and Sybesmas (1998)s claim about (7) (repeated here
as (24)) that the ungrammaticality of the examples with de is not due to the
classifiers, but to the Num. The Nums in these examples (wu five and san
three) are low numbers. If we change the number to a high round number,
like wubai five hundred or qian thousand, these sentences are grammatical with de, as in (25).
(24) a. jintian lai
le
san
ge (*de) laoshi.
today come PFV three CL DE teacher
Today three teachers came.
b. women you ba
tou (*de) niu.
we
have eight CLhead DE cow.
We have eight heads of cows.
(25) a. women qing le
jin qian
wei de laoshi.
we
invite PFV near thousand CL DE teacher
We invited nearly one thousand teachers.
b. women you babai tou de niu.
we
have 800 CL DE cow
We have 800 cows.
The real generalization seems to be that individual classifiers can be
followed by the particle de if the Num is a (contextually) high round number or a fraction. Some more examples are given in (26).
(26) a. sanshi duo ge de pingpangqiu-tai daduo dou
30
more CL DE pingpang table most all
you ren
zai
da.
have people PROG play
Most of thirty-some ping pong tables are being used by people.
(from PKU corpus)

216

Modification marker de in classifier phrases

b. nabian bian zhong le


qi
ba ke,
there then plant PFV seven eight CL
shi lai
ke (de) juzi shu.
ten around CL DE mandarin tree
On that side were planted seven or eight, or around ten mandarin
trees.
(from PKU corpus: prose by Yu Pingbo)
Thirdly, de can also follow individual classifiers when the Num is a
fraction. The examples in (27) show the other case where the Num slot is
filled in by a fraction like , , .
(27) a. san-fen-zhi-yi li de ganmao-yao
one-third
CL DE cold-pill
one third of a cold pill
b. na zhi gou zhan le
ban zhang de shafa.
that CL dog occupy PFV half CL DE sofa
That dog occupied half of the sofa.
The discussion so far shows that it is not possible to give a list of classifiers after which de is allowed. The examples in (23) above indicate that
de is licensed after ping bottle in a measure interpretation, but not in a
counting interpretation, indicating that it is not the lexical classifier itself
which licenses or does not license de, but the use to which the classifier is
being put. De is also allowed after a count classifier with an approximative
marker or with a high round number.
Since, as we shall show, it is plausible to analyze these classifier constructions as occurring in measure contexts, we will argue that it is generally the measure context which allows de to follow the classifier.

4. Semantics of Num- measure Cl-de-N: as much as


4.1. Measure classifiers and the particle de
Cheng and Sybesma (1998) and Hsieh (2008) argue that mass or mensural classifiers like xiang box, ping bottle can be followed by the particle
de. I propose here that the modification marker de always induces a measure reading in the classifier phrase. The following facts naturally follow.

Semantics of Num- measure Cl-de-N: as much as

217

First, [-C, +M] classifiers like gongli kilometer, mi, meter etc can
have an optional de:
(28) a. wo zou le
san gongli
(de) lu.
I walk PFV three kilometer DE road
I walked three kilometers of road
b. ta mai le
liang bang (de) rou.
she buy PFV two pound DE meat
She bought two pounds of meat.
Measure words like gongli kilometer and bang pound do not have
corresponding naturalistic objects like containers; they denote measure
units only. For example, san gongli lu three kilometers road does not
refer to a particular section of road; it just means that the length of road is
three kilometers. The insertion of de after the measure word does not result
in a drastic reinterpretation of the phrase: san gongli de lu three kilometers de road in (28a) has the emphasized meaning that the distance that I
walked is as much as three kilometers.
Secondly, [+C, +M] classifiers (with low precise numbers) can be followed by de when they denote measure units but not when they denote
counting units, as in (29):
(29) a. wo kai le
san ping
(de) jiu.
[Counting]
I open PFV three CLbottle DE wine
I opened three bottles of wine.
b. wo-de wei
neng zhuangxia san ping de jiu. [Measure]
my stomach can hold
three CLbottle DE wine
My stomach can hold three bottles of wine.
In (29a), san ping jiu three bottle wine has a counting reading; there
are three individual bottles, which are opened one by one. In this case, it is
possible to insert de after ping. In (29b), san ping jiu three bottle wine
has the measure reading: it is the amount of wine that my stomach can hold
that is as much as three bottles. Ping cannot be followed by de here.
Note that the insertion of the particle de after the classifier in (29b) does
not affect the truth value of the sentence: (29b) means the same as (29a).
As in (29b) de has an emphatic meaning: it emphasizes the quantity of
three bottles: as much as three bottles.

218

Modification marker de in classifier phrases

4.2. Semantics of Num-Clmeasure-de-N


I argued in chapter 6 that for the measure reading, Num-Cl-N has the structure [Num-Cl [N]], in which Num and Cl form a complex modifier to modify the noun. Num-Cl expresses the quantity of entities in the denotaiton of
N to be n units. Along this line, it is quite reasonable that de occurs optionally after Num-Cl on the measure reading, as in the sequence [[[Num-Cl]
de] N], if we assume that the particle de is a modification marker, which
either explicitly marks or makes [[XP] de ] as predicate modifiers.
In what follows, I will provide a semantic account to explain why de
can naturally follow [+Measure] classifiers.
We assume that the basic meaning of a MEAS head is a measure function nx. MEAS(x) = <n,U> of type <n, <d,t>>. A MEAS head combines
with a number to give a predicate of type <d, t>. We assume that [-C, +M]
classifiers like gongjin kilo and [+C, +M] classifiers like ping bottle
have interpretations as measure heads. In (30) we give their interpretations
and combination with numerals.
(30) a.N gongjinN =nx.MEAS (x)= <n,KILO>
b. NpingN =nx.MEAS (x) = <n,BOTTLE>
c. N liang gongjinN = x.MEAS (x) = <2,KILO>
d. Nsan pingN = x.MEAS (x) = <3,BOTTLE>
These Num-Cl measure predicates can be used in prototypical predicate
positions, like after the verb zhong weigh or you have/reach, as in (31):
(31) a. zhexie pingguo zhong liang gongjin.
those apple
weigh two kilos
These apples weigh two kilos.
b. ta he
de jiu you san ping.
he drink Mod wine have three CL bottles
The wine he drank reaches three bottles.
This measure head, Num-Cl, shifts its basic measure reading when it is
used in a classifier position, i.e. in Num-Cl-N. The reason for this is that
classifiers map kinds onto sets of entities or quantities. Hence, in classifier
position, the head gongjini and ping are of type <n<k,<d,t>>>. Namely,
they apply to a number to give a complex classifier, which applies to a kind
and gives a set of instantiations of the kind with the appropriate measure
properties. This interpretation is given in (32b). Liang gongjin two kilo is

Semantics of Num- measure Cl-de-N: as much as

219

of type <k, <d, t>> and its interpretation is given in (32c). The interpretation of (32a) is in (32d).
(32) a. liang gongjin pingguo
two kilo apple
two kilos of apples
b. NgongjinN =nkx.xkMEAS(x)=<n,KILO>
c. Ngongjin <n <k<d,t>>>N(NliangnN) = k x.xkMEAS(x)=<2,KILO>
d. Nliang gongjin<k<d,t>>N(NpingguokN)=
x.x>?APPLEMEAS(x)=<2,KILO>
I propose that [+C, +M] classifier phrases in their measure reading undergo the same process. We shift the measure interpretation of the [+C, +M]
classifier to the right type as in (33b):
(33) a. ta he
le
san ping
jiu.
he drink PFV three CLbottle wine
He drank three bottles of wine.
b. Derivation I: san ping shui three bottles of water
Nping N = nx. MEAS(x) = <n,BOTTLE>
N pingcl N=nkx. >k(x) MEAS(x) = <n, BOTTLE>
Nsan pingN =kx. x>k MEAS(x) = <3,BOTTLE>
Nsan ping shuiN=x.x >?WATER MEAS(x) = <3,BOTTLE>
Now we come to the measure examples with de. When a basic measure reading is available, the particle de can be unproblematically inserted
after the measure head, as in (34a) with gongjin kilo, and in (34b) with
the non-individual classifier ping when it has its measure reading.
(34) a. tamen chi le liang gongjin de pingguo.
they eat PFV two CLkilo DE apple
They ate as much as two kilos of apples.
b. ta he
le san ping
de jiu.
he drink PFV three CLbottle DE wine
He drank as much as three bottles of wine.
Semantically de applies to a predicate of type <d,t> and turns it into a
modifier of type <<d,t>, <d,t>>. The particle de takes the meaning of liang
gongjin two kilos, given (30c) as input, and turns it into a modifier of the

220

Modification marker de in classifier phrases

head noun. Since the head noun denotes a kind, we shift it to the instantiations of the kind:
(35) Derivation II: liang gongjin de pingguo two kilos of apples
Nliang gongjinN =x.MEAS(x) = < 2, KILO>
Nliang gongjin deN=Px.P(x)MEAS(x) = <2,KILO>
Nliang gongjin de pingguoN=
x. SHIFT(?APPLE)(x)MEAS(x)=<2,KILO>
= x.>?APPLE(x) MEAS(x) = <2,KILO>
In san ping de shui, san ping is of type <k,<d,t>> which is the wrong
type for de to apply to. However, if ping is interpreted as a measure head of
type <n, <d, t>> and not as a classifier, it can apply to Num to give an expression of type <d,t>, which de can take as a complement, as given in
derivation III.
(36) Derivation III: san ping de shui three bottles of water
NpingN= nx.MEAS(x)=<n, BOTTLE>
Nsan pingN=x.MEAS(x)=<3,BOTTLE>
Nsan ping deN=Px. P(x) MEAS(x) = <3,BOTTLE>
Nsan ping de shuiN
=x.SHIFT(?WATER)(x)MEAS(x)=<3, BOTTLE>
= x.>?WATER(x) MEAS(x) = <3,BOTTLE>
In other words, I derive the same meaning as in Derivation I and II, but
from a different structure. These two derivation processes reflect the fact
that Num-Cl-N and Num-Cl-de-N express the same meaning: the quantity
of the denotation of N amounts to the value expressed by Num-Cl. The
latter with de present focuses on the predicate phrase san ping and adds the
pragmatic emphasis interpretation that the quantity of N is as much as the
value of Num-Cl.

5. Num-count Cl-de-N: as many as


I am now concerned with constructions where de follows counting classifiers, [+C, -M] classifiers and [+C, +M] classifiers on the counting reading.

Num-count Cl-de-N: as many as

221

5.1. Counting classifiers in measure phrases


I will argue here that when the classifier denotes a counting unit, it can be
incorporated into a measure phrase forced by the particle de. However, in
this measure reading, Num-Cl-de-N expresses the meaning of as many as
not as much as This process affects [+C, -M] Cls and [+C, +M] on their
counting interpretation.
I first look at [+C, -M] classifiers followed by de. See the examples of
(37), both of which are taken from the Peking University Corpus.
(37) a. nabian bian zhong le
qi
ba ke,
there then plant PFV seven eight CL
shi lai
ke de juzi
shu.
ten around CL DE mandarin tree
On that side were planted seven or eight, or around ten mandarin
trees.
(from PKU corpus: prose by Yu Pingbo)
b. ge shengshi
yue you shisi yu ge de wenyi tuanyi.
each province-city about have forty more CL DE art troupes
Each province or municipality has about more than forty art
troupes.
(from PKU Corpus)
Take (37a) for example. (37a) means that the number of mandarin trees
planted is around ten. The presence of de emphasizes the value expressed
by Num-Cl and stresses that the overall quantity of mandarin trees planted
is estimated to be as many as ten.
It is important to point out that the individual classifier ke denotes in
(37a) a counting unit and not a measure unit. (37a) provides a tree-planting
context, where individual trees are planted one after another, so the individual classifier ke can only refer to individual trees. However, in the example in question, the counting unit is not used in a counting context to
count the numbers of trees one by one; the context is a measure context in
which the total number of trees, the overall quantity, is measured. We
relate the quantity of trees to a numerical value via a measurement mechanism like estimation. (37a) means that the overall number of trees planted is
estimated to be equal to ten individuals.
In this measure reading, the Num of the classifier phrase must be a
(contextually) high round number. Compare the examples in (38) with (37).

222

Modification marker de in classifier phrases

(38) a. tamen yi nian zhong le wu ke (*de) shu.


they one year plant PFV five CL DE tree
They planted as many as five trees within one year.
b. tamen yi nian zhong le wu-bai-lin-yi
ke (*de) shu.
they one year plant PFV five-hundred-and-one CL DE tree
They planted as many as five hundred and one of trees within a
year.
c. tamen yi nian zhong le
wu-bai
ke (de) shu.
they one year plant PFV five hundred Cl DE tree
They planted (as many as) five hundred of trees within one year.
When the Num is a precise number like wu five in (38a) or wubai-linyi five hundred and one in (38b), de is unacceptable. If the number is a
high round number like wubai five hundred in (38c), de can follow classifier ke. We propose that what goes on is similar to what we saw in the
previous example: we use a high round number as a method of estimation
or approximation. That is, we can use the high round numbers to let the
ClP express a measure phrase expressing an estimated measure value. (For
a general discussion of using high round numbers to express estimation,
see Kadmon 1987).
Next I show that [+C, +M] Cls can also be followed by de when denoting counting units. Also in this case, they express the measure meaning of
as many as I take the container classifier ping bottle as an example.
Look at (39):
(39) a. na-ge fuwusheng kai le liu-bai
ping
de hong jiu.
that-CL waiter
open PFV six-hundred CL bottle DE red wine
That waiter opened as many as six hundred bottles of red wine.
b.#na-ge fuwusheng kai le liu ping de hong jiu.
that-CL waiter
open PFV six CLbottle DE red wine
Intended: That waiter opened as many as six bottles of red wine.
c.# na-ge fuwusheng kai le liu-bai-lin-yi ping de hong jiu.
that-CL waiter open PFV 601
CLbottle DE red wine
Intended: That waiter opened as many as 601 bottles of red wine.
When classifier ping denotes a counting unit, it refers to a concrete bottle as in (39a). (39a) means that the number of actual bottles the waiter
opened is estimated to be approximately 600 bottles, it doesnt mean that
what he opened was as much as 600 bottles. In this case, although the classifier ping denotes a counting unit, it has been incorporated into a measure

Num-count Cl-de-N: as many as

223

phrase, which measures the overall number of individual entities, as many


as 600 bottles.
The interpretation of as many as for [+C, +M] classifiers is subject
to the same constraint as that for [+C, -M] classifiers: the measure reading
of as many asis available only when the Num of [+C, +M] classifier
phrase is a (contextually) high round number as in (39a), but not in the
cases of (39b) and (39c).
Predictably, the measure interpretation of as many as is impossible
for true measure words, [-C, +M] classifiers like gongjin, kilo and gongli
kilometer. This is because they are true measure words and denote intrinsic measure units; they do not correspond to actual naturalistic objects like
bottle or cup. 500 gongjin (de) dami 500 kilos rice does not imply that
there are five hundred individual kilo units of rice.

5.2. Semantics of Num-Clcount-de-N


I start by specifying our syntactic assumptions about the construction. In
section 4, I argued that, when the classifier denotes a measure unit, de
takes Num+Cl as complement to form a modifier of the head noun. Since
in the measure reading Num-Cl-N has the structure [[Num+Cl]N], the semantic interpretation is direct.
However, when the classifier has a counting interpretation, the Cl and
the N form a constituent which is modified by the Num. I propose that the
presence of de after the classifier forces syntactic re-analysis of the counting structure into the measure structure. In other words, I assume that no
matter whether the classifier denotes a counting unit or a measure unit, the
presence of de after classifier always force Num and Cl to be a constituent.
(40) shows that some modifiers like zhengzheng appear in front of the
whole nominal phrase, and do not modify the whole phrase or the noun but
Num+Cl.
(40) a. ta zhong le
zhengzheng wubai ke de shu.
he plant PFV as-a-whole 500 CL DE tree
He planted 500 hundred trees in total.
b.* ta zhong le wubai zhengzheng ke de shu.
he plant PFV 500 as-a-whole CL DE tree
c.* ta zhong le wubai ke de zhengzheng shu.
he plant PFV 500 CL DE as-a-whole tree

224

Modification marker de in classifier phrases

I now turn to the semantics of Num-Cl-COUNT-de-N.


In chapter 6, I argued, following Rothstein (2010) that the basic interpretation of counting classifiers like ke is as function mapping kinds to sets
of individuals that are atomic relative to counting context k, a function of
type <k,<dk,t>>. We illustrate this in Derivation V:
(41) Derivation V: wubai ke shu five hundred trees
NkeN =kx. 1(x) (>k k) (x) 2(x)=k
Nke shuN=x. 1(x)(>?TREEk)UNITtree(x)2(x)=k
Nwubai ke shuN
=x. 1(x)*(>? TREE k)*UNIT-tree(x)2(x) =k|1(x)|k = 500
(pluralizing in the context of applying ke shu to wubai)
When de follows the counting classifier, de forces Num and individual
Cl to form a single constituent of type <d, t>, which can then shift to the
modifier type. Even though a measure reading is imposed, the counting
meaning of the classifier is not discarded.
I propose the following: de looks for a predicate meaning of type <d,t>.
The classifier shifts to a measure function of type <n,<d,t>>. The meaning
it shifts to is that of an estimation function, EST, which estimates the overall number of a plurality. I suggest that Num-ClCOUNT-de-N expresses the
estimated value of a plurality, measured in terms of a quantity of counting-units.
(42) a. Basic counting meaning of the classifier:
kx. 1(x) (k k) (x) 2(x)=k
b. Shifted measure meaning of the counting classifier:
nx. x *k EST(x) = <n,Unatural unit>
The function that maps number n onto the set of members of the pluralization of the context set k whose estimated value is n.
c. This applies to the round number 500:
x. x *k EST(x) = <500,Unatural unit>
The set of pluralities in *k, whose cardinality is estimated to be approximately equal to 500 natural units.
In (42c) ke denotes atomic plant units and the Num-Cl shifts to the set
of pluralities of atomic plant-type entities whose estimated value is 500.

Num-count Cl-de-N: as many as

225

Thus, estimation is a contextual operation introduced as a counting measure, and the numbers are round, because that is what they are in estimation.
This measure phrase can be used as a predicate in copula position as in
(43):
(43) ta zhong de shu you wubai ke.
he plant Mod tree have 500
CL
The trees he planted reached 500.
When the measure head is used in a classifier position, as in (43), its interpretation is given in Derivation IV:
(44) a. ta zhong le wubai ke de shu.
he plant PFV 500 CL DE tree
He planted five hundred trees.
b. Derivation VI: wubai ke de shu five hundred trees
Step 1: N500 keN= x.x*kEST(x) = <500,Unatural plant unit>
Step 2: N500 ke deN= Px.*P(x)x*k
EST(x) = <500, Unatural plant unit>
Step 3: N500 ke de shuN
=x. *>?TREE(x) x *k EST(x) = <500,Unatural plant unit>
The set of pluralities that are sums of instantiations of the kind tree,
whose estimated cardinality is around 500.
The present semantics accounts for the constraint that in the readings
discussed here the Num must denote a large round number. As Krifka
(2002, 2009) argues, round numbers in measuring contexts tend to have
round interpretations. Furthermore, according to Krifka (2002:446-447)
short expressions have a preference for vague interpretations, () long
expressions have a preference for precise interpretations.
Given this, estimation will favor round numbers in short expressions.
Moreover, it will favor high numbers, because, arguably, one normally
doesnt need to estimate a small number, one just checks.
We saw that Hsieh proposed that de was sensitive to a feature of indeterminateness. I propose that the inderminateness is a by-product of the
semantic reinterpretation of the counting classifier as an estimation measure.

226

Modification marker de in classifier phrases

6. Conclusions: two puzzles about [+counting] classifiers


So far, I have made use of the distinction between counting and measure
operations to account for the licensing of de after classifiers. I argued that
the use of de after a classifier cannot be correlated with a certain lexical
type of classifier, but rather it depends on whether the classifier denotes a
measure operation or a counting operation.
I have made two claims concerning the licensing of de: (i) classifiers
that can be interpreted with a measure reading can be followed by de, in
which case the measure reading is obligatory; (ii) classifiers which are
normally associated only with a counting interpretation can be followed by
de if the numeral is a (contextually) high round number or the numeral is
modified by approximative modifier. This is because approximation is a
kind of measure operation, and in these constructions the counting classifier is reanalyzed as a measure expression, and Num-Cl-de-N always has the
syntactic structure [[[Num-Cl]de] N].
I propose that de denotes a function from predicates to predicate modifiers (is of the type <<d, t>,<<d, t>, <d, t>>>). It combines with a predicate expression of type <d,t>, and yields an expression which modifies
another predicate. While it is beyond the scope of this paper to give a semantic analysis of other uses of de as a modification marker, my analysis
suggests that a unified account of de modification is possible, one that incorporates both Num-Cl-de sequences as well as AP-de, de- marked relative clauses and so on.
There are still two puzzles remaining. The first puzzle is that classifiers
with a default counting interpretation can be coerced into a measure reading with low cardinals in recipe type contexts, but, nonetheless, de cannot
follow the classifier in these contexts. The examples are given in (45).
(45) zuo zhe ge pingguo-pai wo zongguo yong le
make this CL apple-pie
I altogether use PFV
you yi/san
ge (*de) ping.
have one/ three CL DE apple
To make this apple pie, I used one/three apple(s) altogether.
I assume that in this context, there is a pragmatic shift from the standard
meaning of yi ge pingguo to a reading roughly paraphrasable as to make
this apple pie, I used one apples-worth of apple. I assume that this pragmatic reanalysis occurs later in the derivation after the classifier has combined with the NP, and thus does not result in a structural reanalysis of the

Conclusions: two puzzles about [+counting] classifiers

227

classifier phrase. Therefore, de cannot follow the classifier in this case.


While this is a somewhat ad hoc explanation, it is supported by the fact
that recipe contexts are crosslinguistically distinguished from standard
counting and measuring contexts. Thus for example, in English, recipe
contexts are the only context in which of can be dropped from a measure
partitive, e.g. two kilos flour versus two kilos of flour. Given that these
exceptions are limited to a very specific and exceptional context, it seems
to us clear that the distribution of de is generally associated with a measure
interpretation of the classifier, which is reflected in the structural properties of the classifier phrase.
The second puzzle is that when Num is a fraction, such as , , or a
decimal, such as 10%, 15%, it is possible to use de after the classifier to
get a measure reading, as illustrated in (46) (examples from Li and Rothstein 2012).
(46) a. yi ge yi shui de yin'er mei-ci
one CL one year DE baby every time
zhi neng chi 1/3 li de ganmao-yao
only can eat 1/3 Cl DE cold pill
A one year old baby can only take 1/3 of a cold pill every time.
b. 1/4 ge de xigua
keyi zha ban bei guozhi zuoyou.
1/4 CL DE watermelon can crush half CLcup juice approximately
1/4 of a watermelon can be crushed into half cup of juice.
Her and Hsieh (2010: 539-540) independently observe that fractions
of a number, including those with a value smaller than one, drastically
increase acceptability. Their examples are given in (47a-b) and our own
example is given in (47c).
(47) a. ba-fen-zhi-yi ke de gaolicai
one-eight
CL DE cabbage
one-eighth of a cabbage
b. si-fen-zhi-yi ke de
yangcong
one-fourth CL DE onion
one fourth of an onion
c. ba-fen-zhi-san
gongjin de dami
eight-portion-Mod-three (three eighths) CLkilo DE rice
three eights of a kilo of rice

228

Modification marker de in classifier phrases

The syntax and semantics of fractional expressions has not yet been
studied in any detail, but it is clear that crosslinguistically fractions have a
different grammar from cardinal natural numbers. In English, they are necessarily partitive as in a third of a pill, a quarter of a kilo of flour, a half of
an apple (although (a) half an apple is also acceptable), and in English, as
in French, the fractional expression is based on the ordinal numeral. In
Mandarin, the fraction is expressed by a complex expression using the
word fen portion. Si fen zhi yi one-fourth has the compositional structure in (48), and literally means one of four portions, with the classifier
giving the unit which is divided into portions.
(48) si fen
zhi
four portions Mod
one fourth

yi
one

An analysis of the semantics of fractions is beyond the scope of this paper, given that these constructions are highly complicated, both in Mandarin and crosslinguistically. However, it is clear that fractions have a different interpretation from natural number predicates. A natural number such
as san three can be interpreted either as a name for a number of type n, as
in a measure context, or as a predicate of plural entities, in which case it
modifies a Cl-N in a counting context (see discussion in Rothstein 2012).
Fractions clearly cannot be predicates of plural entities, since they do not
count atomic entities. Similarly, the classifier cannot have its usual interpretation, in which it denotes function from kinds into atomic instantiations
of the kind. Instead, the role of the classifier is to give the name of the unit
which the fraction divides into portions. I thus suggest that in these contexts, the fraction and the classifier combine to form a fractional predicate,
and thus form a predicate constituent which can be the complement of de.
Thus syntactically, we suggest, Num combines immediately with the classifier and the insertion of de is thus possible after the [N-Cl] constituent.
This should not be taken as an analysis of these constructions, but a hypothesis to be explored in further research. Nonetheless, what does seem
clear is that the role of the classifier in these fractional constructions is
different from its role in counting constructions which use natural numbers.
This can be seen in the contrast between (49a) and (49b).
(49) a. ta chi le san-fen-zhi-yi
ge de pingguo.
he eat PFV san-portion-Mod-one CL DE apple
He ate one third of an apple.

Conclusions: two puzzles about [+counting] classifiers

229

b. ta chi le san-fen-zhi-yi
de pingguo.
he eat PFV san-portion-Mod-one DE apple
He ate one third of the apples.
OR He ate one third of an apple.
In (49a), where the fraction is followed by ge, the sentence must mean
he ate one third of a single apple. Thus the classifier does not have its
usual use in which it combines with N and the resulting Cl+N denotes a set
of atomic individual apples. Instead it indicates what is the single entity
which is portioned by the fraction. In (49b) where there is no classifier, we
see that the sentence means he ate one third of some quantity of apples,
either a third of a plurality of apples or one third of a single apple. The
examples in (49) are sufficient to show us that the use of fractions is very
different from other numerical structures: unlike the natural numbers, the
fraction does not require a classifier, and when the classifier is present it
has a very different meaning from its usual one. So while we do not yet
understand either the syntax or semantics of fractional constructions, we
can see that these constructions are sufficiently different from normal
counting contexts not to constitute a counterexample to the account that we
have presented in this chapter.

Part III:
Definiteness in classifier languages

Chapter 9
Definite classifiers in southern Chinese languages

1. Introduction
In the present chapter, we study aspects of the meaning of classifiers
beyond the counting and the measure readings, and beyond the language of
Mandarin Chinese.
It is found that in mainland Southeast Asian languages, like Vietnamese,
Hmong, Thai etc, classifiers are able to mark definiteness (see Bisang 1996,
1999, Simpson et al 2011, among others). According to Bisang (1993), in
Hmong, Cl+N can have a definite interpretation and the classifier works
like a (quasi-) definite article. For example, Cl+N in (1a) means the
widow but not a widow. Lbel (1996, 2000) observes that in Vietnamese,
the classifier in Cl+N can either express definiteness or indefiniteness in
a sentence like (1b). Qu cam in (1b) either means the orange or an
orange.
(1) a. ces nyob nyob tus pij-ntsuag txawm
then one day CL widow
then
yug
tau
ib
tug
me-tub
give:birth PFV one
CL
son
Then one day, the wodow gave birth to a son.
b. ti mua
qu cam.
I
buy
CL orange
I bought the/an orange.

[Hmong]

[Vietnamese]

In many southern Chinese languages, such as Wu, Cantonese, Xiang, Gan


etc., classifiers can also mark definiteness, especially in the Cl+N con1
struction. Some of the examples are illustrated in (2).
1

Definite classifiers are rarely found in northern Chinese languages, like standard
Mandarin. According to Wang and Gu (2006), Jianghuai Mandarin are the northern
most dialect groups that have definite classifiers. Chinese languages to the north of
Jianghuai Mandarin often do not have definite classifiers.

234

Definite classifiers in southern Chinese languages

(2) a. pun34 sy55 hou34 thi35.


[Cantonese: Guangdong]
CL
book good look
The book is interesting.
b. p53-55 s53 h34 53 p53-55 v12 li23. [Wu: Shanghainese]
CL book good that CL
not good
This book is good, tht one is not good.
c. thi35 35 55 th3 1 t lie. [Jianghuai Mandarin: Lianshui]
CL ox PASS steal RVC PRT
The/an ox was stolen.

In these three languages, classifiers can be used independent of numerals, as in the construction Cl+N, to express definiteness. In Cantonese,
pun3 sy55 in (2a) means the book, referring to some particular book that is
familiar with the interlocutors. In Shanghainese, Cl+N, such as p5355 53
s in (2b) also expresses definiteness. It is used in contrast with the (distal) demonstrative phrase, so it also expresses proximal deixis. P53-55s53
can be interpreted this book (Pan and Tao 1999: 33-34). In (2c), in Jianghuai Mandarin, Cl+N can be interpreted either as definite or indefinite,
and thi35 35either means an ox or the ox, depending on the context
(Wang 2005). We consider classifiers in Cl+N that mark definiteness to
be definite classifiers.
In this chapter, I will discuss definite classifiers in Cantonese and Wu
Chinese (the Fuyang variant), with a comparison with Mandarin. All the
three languages I am to investigate are classifier languages, in which numerals cannot modify nouns without a classifier, as in the sequence
Num+Cl+N. In all three, classifiers can be used independently of the
Num, as Cl+N. The languages differ with respect to the distribution and
interpretation of Cl+N construction. Look at (3):
(3) a. (*ge) laoban mai le
liang che.
CL boss
buy PFV CL
car
The boss bought a car.
b. k lpan ma l
bu tshots.
CL boss buy PFV CL car
The boss bought a car.
c. go louban maai zo ga
ce.
CL boss
buy PFV CL car
The boss bought a/the car.

[Mandarin]

[Wu:Fuyang]

[Cantonese]

Introduction

235

As in (3a), in Mandarin, the Cl+N construction is only found in postverbal positions and has an indefinite reading. In Cantonese and Wu Chinese (the Fuyang dialect), the construction is available in both preverbal
and postverbal positions. In Wu Chinese, as in (3b), Cl+N has a definite
reading, when appearing preverbally; indefinite when appearing postverbally (Li and Bisang 2012). In Cantonese (the Hong Kong variant), in (3c),
preverbal Cl+N has a definite reading, while postverbal Cl+N is either
definite or indefinite (see Cheung 1972).
Cl+N has been discussed intensively in the literature in many Chinese
dialects (e.g. Cheung 1972, Shi and Liu 1985, Cheng and Sybesma 1999,
2004, Wu and Bodomo 2009, Simpson et al 2011, Li and Bisang 2012),
without reaching agreement about the syntactic status and the semantic
function of the classifier. Many questions are still left open. I am particularly interested in the following:
(i) What are the factors that constrain the distribution of indefinite and
definite Cl+N?
(ii) What is the syntax of indefinite and definite Cl+N phrases? Specifically, is indefinite Cl+N a classifier phrase or a numeral phrase? Is
definite Cl+N a definite phrase?
(iii) Is the Cl+N construction derived from the counting reading or the
measure reading of the classifier?
(iv) What is the semantic function of the classifier in indefinite and definite Cl+N constructions?
Concerning the first question, I argue that the interpretation of the classifier in Cl+N as definite or indefinite is pragmatically constrained by
information structure. Since Chinese languages are topic-prominent languages, in which preverbal nominals tend to function as topics or secondary topics (Li and Thompson 1976, 1981), preverbal nominals are by default interpreted as definite. That is why definite Cl+N are prototypically
found in preverbal positions.
My answer to the second question is that in the indefinite reading of
Cl+N, the classifier heads the maxiamal projection of ClP, and Cl+N
is inherently predicative. In the definite reading, the classifier undergoes
Cl-to-D raising and heads a definite DP.
As for the third question, I claim that the Cl+N construction is only
available when the classifier is interpreted with a counting function and not
a measure function. The counting function of classifiers can be extended to
mark (in)definiteness, since counting classifiers pick out a set of atomic

236

Definite classifiers in southern Chinese languages

individuals instantiating the kind denoted by the noun, rather than a set of
quantities. If definiteness is analyzed in terms of old/new information, I
make the plausible claim that individuals can be old or new relative to the
information, but this distinction makes little sense for quantities.
With respect to the semantic functions of classifiers, I propose that
Cl+N has a predicative meaning, and that indefinite readings of Cl+N result from default existential closure over the VP interpretation. Definite
interpretations of Cl+N result from Cl-to-D- raising. This is semantically
interpreted as existential closure of Cl-N in the indefinite meaning. Following Landman (2004), definiteness is a semantic constraint added to the
existential meaning derived in this way. Thus, though I use Landmans rule
of argument formation, I do not claim like Landman (2004) that definites in
Chinese start out at type d. Instead, I argue that definites are derived by
raising from predicates to the generalized quantifier reading, and that definiteness is a semantically expressed familiarity constraint, like the exactly effects in Landman (2004) or weak familiarity la Roberts (2003).
This chapter is structured as follows. Section 2 presents the distribution
of indefinite and definite Cl+N in Mandarin, Wu, and Cantonese. In
Section 3, I account for the distribution patterns of Cl+N in terms of
(in)definiteness related to information structure. Section 4 and 5 discuss
the syntax of indefinite and definite Cl+N respectively. Section 6 examines the semantics of Cl+N. Section 6 gives a semantic representation
for indefinite and definite Cl+N, with a correlation with the counting
function of classifiers. Specifically, a familiarity-based semantics for definite Cl+N is provided. The chapter is wrapped up in section 7.

2.

Cl+N in three Chinese languages

In the previous chapters, I discussed classifiers in the Numeral-Classifier


construction Num+Cl+N. We saw that in Mandarin numerals cannot
modify nouns directly without the assistance of classifiers. The same is
true in Wu and Cantonese:
(4) a. i  *(ts) kiu
one
CL dog
one dog
b. ian *(k) in
two CL man
two people

[Wu: Fuyang]

Cl+N in three Chinese languages

(5) a. leuhng *(jek) gau


two
CL dog
two dogs
b. yat  *(chaan) faahn
one  CL
meal
one meal

237

[Cantonese]

(from Matthews and Yip 1994: 93-97)

On the other hand, all the three languages have Classifier+Noun constructions without a numeral. For example, Mandarin has postverbal
Cl+N, Wu and Cantonese can have both preverbal and postverbal
Cl+N. Recall the examples in (3).
I start with my discussion with Cl+N in Mandarin.

2.1. Cl+N in Mandarin


In chapter 5, I showed that bare nouns in Mandarin can occur in preverbal
and postverbal positions. Bare nouns in preverbal positions have a definite
reading, as in (6a); those in postverbal positions can either be definite or
indefinite, as in (6b). Moreover, bare nouns are number-neutral, so shu in
(6) have both singular and plural readings.
(6) a. shu
bujian
le.
book missing PRF
The book is missing.
OR The books are missing.
b. wo mai le
shu.
I buy PFV book
I bought book(s).
OR I bought the book(s).

[Definite]

[Definite or Indefinite]

Unlike bare nouns, the Cl+N construction in Mandarin is not possible


in preverbal positions like the subject position in (7a), and it is only found
in postverbal positions, as in (7b). The postverbal Cl+N in (7b) only has
an indefinite reading, meaning a book. The speaker of (7b) simply asserts
that he was engaged in an activity of book-buying.

238

Definite classifiers in southern Chinese languages

(7) a.* ben


shu
bujian le.
CLvolume book missing PRF
b. wo
mai le
ben
shu.
I
buy PFV CLvolume book
I bought a book.
In contrast to the examples with bare nouns in (6), there is an implication of singularity for Cl+N. As shown in (8), mei ben each volume in
the continuation clause requires a plural NP as antecedent: the bare noun
shu book can provide that, Cl+N cannot.
(8) a. wo mai le shu. Mei-ben dou hen haokan.
I buy PFV book each-CL all very interesting
I bought books. Each is interesting.
b. wo mai le
ben
shu. # Mei-ben dou hen haokan.
I buy PFV CLvolume book each-CL all very interesting
# I bought a book. Each is interesting.
(7a) shows that Cl+N cannot be used as subject. One way of making
it grammatical is to put existential you there be before Cl+N, as in (9a):
(9) a. you
ben
shu bujian le.
there-be CLvolume book missing PRF
There is a book missing.
In this case, ben shu Cl book introduced by the existential you keeps
its singular meaning, but gets an indefinite interpretation: it refers to a
book, not books, and not the book. So, if there were ten books on my desk
last night, but now I noticed that there are only nine books left and I am not
sure which one is missing, I can report this event by using (9a) with
Cl+N as an indefinite phrase. Huang (1982, 1987) calls the sentence
initial you an existential verb. Semantically, it existentially closes the indefinite (see Cheng 1990, Tsai 1994). Since you is an existential verb,
Cl+N in (9a) can be considered to be located in postverbal position as
well.
Besides canonical object positions (as in (7b)) and as the complement
of you-sentences (in (9a)), Cl+N is also found in post-copula positions,
as in (9b) and (9c), where Cl+N expresses properties like studenthood
and cathood.

Cl+N in three Chinese languages

(9)

b. ta
shi
ge
he
be
CL
He is a student.
c. zhe
shi
zhi
this
is
CL
This is a cat.

239

xuesheng.
student
mao.
cat

I can then make the generalization that Cl+N in Mandarin has an indefinite interpretation and appears in a lexically governed position.
However, this generalization seems to be challenged by the ba construction. Mandarin is an SVO language, where the postverbal position is the
base-generated position for objects (Li and Thompson 1974, Huang 1982,
Sun and Givon 1985, Wu 2003). But when the object is definite or specific
and the verb has a causative meaning, the object can be moved from the
base-generated postverbal position to a preverbal position, and is marked
by the object maker ba. This is called the ba construction. Generally speaking, only bare nouns and demonstrative phases are allowed to occur after
ba, in which bare nouns have a definite reading, but indefinite expressions
like Num-Cl-N are not allowed after ba. One of the few cases in which BA
allows an indefinite object is the Cl-N construction. (10b) is assumed to be
derived from (10a):
(10) a. Jintian Lao Zhang mai le tou niu.
[SVO]
today Lao Zhang sell PFV CL cow
Lao Zhang sold a cow today.
b. Jintian Lao Zhang ba tou niu gei mai le.
[SOV]
today Lao Zhang OM CL cow give sell PRF
The peasant had a cow sold today.
Both in (10a) and (10b), the Cl+N phrase, i.e. tou niu head cow, has
an indefinite reading, a cow. However, Cl-N in (10a) is preferably interpreted as a non-specific reading. In (10b), Cl-N tends to be understood
as specific. The speaker of (10b) knows which cow of LaoZhang he talks
about, but the hearer has no relevant information to identy which cow the
speakers talks about. Therefore, it is specific indefinite.
Therefore, my generalization still holds: in Mandarin Chinese, Cl+N
is only possible in lexically governed positions, such as after the main verb,
after the existential you, or after the object marker ba, in which it receives
an indefinite reading. It cannot be used in preverbal positions like subject
and topic.

240

Definite classifiers in southern Chinese languages

2.2. Cl+N in Wu
2.2.1. Preverbal Cl+N
In Wu,Cl+N can occur unproblematically in preverbal positions, like the
subject position and the (sentential) topic position. The examples in (11)
illustrate the use of Cl+N as subject.
(11) a. ts kiu s-i die.
CL dog die
PRT
The dog died.
b. ts kiu kuan m-po tha dzan.
CL dog CL tail
too long
The tail of the dog is too long.
The Cl+N ts kiu the dog in (11a) refers to a particular dog that is
identifiable in the context by the interlocutors, or some dog that is familiar
to the interlocutors based on their shared background knowledge. In this
case, Cl+N is read definite: the dog. The subject of (11b) is a complex
noun phrase, in which the head noun Cl+N, kuan m-po Cl tail, is modified by another Cl+N ts kiu Cl dog.2 Both Cl+N phrases are interpreted with a definite meaning, so the phrase means: the tail of the dog.
The preverbal Cl+N in (11) should be distinguished from the Cl+N
construction that is introduced by the existential quantifier iu there be.
Compare (11a) with (12a):
(12) a. iu
ts kiu s-i die.
there-be CL dog die PRT
There died a dog.
b. iu
i ts kiu si-i die.
there-be one CL dog die PRT
There died one dog.
2

There are two nominal phrases before the main verb. The construciton is called
double nominative constructions by some researchers (e.g. Teng 1974, Li and
Thompson 1981). The two nominal phrases are called big subject and small
subject respectively. Other researchers (e.g. Xu and Liu 1998) suggest that the first
nominal phrase is a dangling topic and the second is the subject of the clause. We
simply consider the two nominals be a single complex nominal phrase, in which the
first nominal modifies the second.

Cl+N in three Chinese languages

241

In (12a), ts kiu Cl dog can only have an indefinite meaning, meaning


a dog. (12a) is similar to (12b), which has a NumP as subject. I assume
that iu there be is verbal, like you there be in Mandarin, and hence
Cl+N in (12a) is in a postverbal position.
In contrast, the Cl+N subject in (11) can stand on its own in argument
position, without requiring existential iu, and in fact it is read with a definite meaning.
We now consider the topic use of Cl+N. (13) is an example of
Cl+N in sentential initial position, the default position for the topic of
the sentence.
(13) k biku, thi-i  die.
CL apple I eat-finish PRT
The apple, I ate it.
In (13), the pronoun I is assigned with the theta-role of agent and it
is the subject of the clause. The sentence initial nominal biku apple is
the syntactic object of the sentence, but it is also the topic of the sentence.
The entity denoted by Cl+N, i.e. k biku Cl apple, is part of the proposition that is being talked about, so it must be interpreted as definite, the
apple.
Wu has a construction similar to the ba construction in Mandarin. In
Wu (the Fuyang dialect), the object marker is kh catch, so I call this
construction the kh construction. The most striking difference between
the ba construction in Mandarin and the kh construction in Wu is that
Cl+N after object marker kh gets a definite reading and not an indefinite reading.
(14) a. l tsan
kh ts iu
Lao Zhang OM CL cow
LaoZhang sold the cow.
b. i
kh pn y
m-i
he OM CL book lost
He lost the book.

ma-i die.
sell
PRT
die.
PRT

Cl+N in (14a-b) refers to a particular cow or a particular book identifiable or known by both the speaker and the hearer. Hence, it is definite.
For example, ts iu in (14a) may refer to the only cow that owned by
Laozhang, i.e. Laozhangs cow.

242

Definite classifiers in southern Chinese languages

2.2.2. Postverbal Cl+N


Cl+N is also found in postverbal positions in Wu, and there it has an
indefinite reading, like in Mandarin.
Two types of postverbal positions can be distinguished: base-generated
object positions and post-copula positions. (15) is an example with Cl-N in
object position:
(15) a. thi l
k piku. iuz tyosa la k.
I eat PFV CL apple just desk
on CL
I ate an apple, the one on the table.
b. sa k yo
ma l
bu
tshots.
I last CL month buy PFV CL
car
h
h
h
go
tso ?

ts -ts a-k an z
you
guess
be
what car
I bought a car last month. Can you guess what car it is?
In (15), apple and car do not denote entities known or identifiable by
the addressees: Cl car and Cl apple simply mean a car and an apple
respectively. It is only continuation in (15a) that identifies the apple; similarly, in (15b) the question presupposes the indefiniteness of the car in the
first part of the sentence.
Post-copula nominals are predicational and express properties of the
3
subject. Post-copula Cl+N are not referential in most cases. For example, ts kiu Cl dog in (16a) only expresses the doghood of the subject
and it does not refer to a particular dog.
(16) a. k ts z ts kiu.
this CL be CL dog
This is a dog.
b. ia
ban-iu z k akuoin.
his/her friend be CL foreigner
His/her friend is a foreigner.
In Wu, if you want to use a definite nominal as object, it must always be
moved to a preverbal position, via the kh construction or by topicalization.

I leave aside copula sentences with definite NPs, such as He is the student that I
mentioned to you last time or He is Mr. Smith (see Partee 1987).

Cl+N in three Chinese languages

243

Thus, if I want to make definite reference to an apple or a car in (15a-b),


I would use the sentences in (17) and (18) with the kh construction. (Note
that the marker kh does not always have to be realized overtly, as shown
in (18)).
(17) kh k binku thii die.
I OM CL apple eat
PRT
I ate an apple, the one on the table.
(18) sa k yo
bu tshots ma l ul die.
buy PFV back PRT
I last CL month CL  car
I went to buy the car last month.
In sum, in Wu, the definiteness of Cl+N is closely related to its syntactic position being preverbal or postverbal. Preverbal Cl+N have a
definite interpretation and postverbal Cl+N have an indefinite interpretation. More importantly, it is a syntactic requirement to prepose definite
expressions from a postverbal position to a preverbal position in Wu.

2.3. Cl+N in Cantonese


As noted in the literature (e.g. Cheung 1972, Hashimoto 1993, Matthews
and Yip 1994, Cheng and Sybesma 1999, and many others), in Cantonese,
Cl+N can occur preverbally and postverbally: preverbal Cl+N is unambiguously definite, but postverbal Cl+N is ambiguous between an indefinite and a definite reading.
(19) a. ji
bat hou hou se.
CL
pen good good write
This/that pen is good to write with.
b. keuih maai-zo gaa ce.
he
sell-PFV CL car
He sold a car/ the car.
[from Matthews and Yip 1994: 93]
In (19a), ji bat is definite: this/that pen (an even better gloss may be
the pen, since Cl+N has a definite reading, but not a deictic reading.). In
(19b), Cl+N, i.e. gaa ce Cl car occurs in object position. It can have a
definite meaning, where it means: a car identifiable for the addressee in

244

Definite classifiers in southern Chinese languages

the context, or an indefinite meaning, where we only express that a car


selling event took place.
There is in Cantonese a construction similar to the ba construction in
Mandarin, with the marker jeung take, but it is not very productive.
For instance, the two examples in (20) are ungrammatical, no matter
whether we use bare noun or Cl+N after jeung. According to our Cantonese informants, they would use SVO sentences to express the same information. Note that the counterparts in Mandarin of the cases in (20) are
totally grammatical with bare nouns, as in (21).
(20) a.* ngo jeung (go) pingguo sik zo.
I OM CL apple
eat PRT
Intended: I ate the apple(s).
b.* keuih jeung (go) yahn da zo.
he
OM CL man beat PRT
Intended: He beat the man/the men.

[Cantonese]

(21) a. wo ba
pingguo chi le.
I OM apple
eat PRF
I ate the apple(s).
b. ta ba ren da le.
He OM man beat PRF
He beat the man/men.

[Mandarin]

Secondly, compared with the ba construction in Mandarin, the jeung


construction has a narrower range of uses. I cite Matthews and Yip (1994:
144): unlike ba, however, it is not used with all transitive verbs, but is
primarily restricted to cases where the motion takes place (as in (22a)). It
is also applicable in metaphorical cases of movement or removal, such as
exchanging places (as in (22b)).
(22) a. keuih jeung di
wujou saam jauwaih pehk.
s/he take CL-PL dirty clothes around throw
He throws his dirty clothes all over my place.
b. yiu jeung di mhoisam ge yeh gut-yat-seng tan-jo
keuij.
need put CL unhappy Mod stuff one-voice
swallow-PFV it
You should take the unhappy things and swallow them in one
gulp.

Information structure and (in)definite Cl+N

245

Thirdly, when the jeung construction with Cl+N is felicitous, the


Cl+N only has a definite reading. For example, the plural classifier
phrase di wujou in (22a) can only mean the dirty clothes but not some
dirty clothes.
I will argue in section 3 that in Cantonese, the preferable position for
definite Cl+N is in preverbal positions, but the lack of a productive jeung
construction paves the way for the expansion of definite Cl+N interpretations into postverbal positions.
So far, I have examined the syntactic distribution and the referential
functions of [Cl+N] in Mandarin, Wu and Cantonese. The referential functions of [Cl+N] in terms of [definite] in each of the four syntactic positions of topic, subject, preposed object (marked by prepositions, such as
BA in Mandarin) and canonical object are summarized in the following
table for each of the three Sinitic languages discussed:
Table 4. Distribution of definite and indefinite Cl+N
[Cl+N]
Mandarin
Wu (Fuyang)
Cantonese

Topic

[+definite]
[+definite]

Subject

[+definite]
[+definite]

Preposed obj
[-definite]
[+definite]
[definite]

Canonic obj
[-definite]
[-definite]
[definite]

As can be seen from the above table, there seems to be a correlation between (in)definiteness and the syntactic position of [Cl+N] relative to the
verb. Indefinite [Cl+N] is restricted to the postverbal position, i.e. the canonical object position, in all three languages, while definite [Cl+N] is
usually found in preverbal positions (topic, subject and the preposed object
position). These general positional rules need two specifications: (i) Mandarin does not allow definite [Cl+N] in topic or subject positions, and (ii)
Cantonese allows postverbal definite [Cl+N].
The next section will show how this distributional pattern can be accounted for.

3. Information structure and (in)definite Cl+N


In this section, I adopt a semantic-pragmatic approach to account for the
distribution of the indefinite and definite readings of Cl+N phrases. I
propose that these interpretations of Cl+N are constrained by pragmatic

246

Definite classifiers in southern Chinese languages

notions, following from the association between word order and information structure.
Li Frances (1971) was one of the first to claim that word order reflects
information structure in Mandarin. Following on this, Li and Thompson
(1976, 1981) argue that Mandarin Chinese is a topic-prominent language in
which the preverbal subject position represents an unmarked topic, while
the postverbal position is associated with focus. In a more recent study, Xu
(2004) shows that the sentence-final position is the default position for
informational focus in Mandarin.
Informally, the topic is what is being talked about, which implies
givenness or high degree of identifiablity. As Lambrecht puts it (1994:262),
a topic constituent must have a referent, and this referent must be identifiable and have a certain degree of pragmatic salience in discourse Thus,
nominals acting as topics tend to be definite. In contrast, the term focus
refers to expressions that are either prosodically or syntactically prominent
and convey new information: typically nominals acting as focus are indefinites, although definite NPs are not excluded from focus position. As Lambrecht claims, a focus constituent is in principle free with respect to the
question of identifiability and activation.
Chinese languages are topic prominent languages, in which preverbal
nominals are usually topics and postverbal nominals are usually focus. As
a consequence, the general tendency in Chinese is that, preverbal nominals
have by default definite interpretations and postverbal nominals indefinite
interpretations.
Even though Chinese languages are all topic-prominent languages, they
exhibit different degrees of topic-prominence. D. Liu (2001) claims that
Wu is a more prototypical topic-prominent language than Mandarin and
Cantonese: it is very easy to topicalize elements like objects, i.e. to move
them to the sentence initial position or the position near the subject. In
other words, Wu has SVO as the basic word order, but SOV and OSV orders can also be derived via topicalization. According to D. Liu (2001) and
S. Tang (2006), Cantonese is the most prototypical SVO language, in
which object-preposing (namely, topicalization) rarely happens; Wu Chinese is the weakest SVO language and it has an alternative OV word order;
Mandarin is a mild SVO language, a language sitting between Wu and
Cantonese in terms of topic-prominence. Therefore, these three languages
can be ranked as follows, according to degree of topic prominence from
high to low: Wu > Mandarin > Cantonese.
I start with Wu. I propose that in Wu Chinese, as the strongest topicprominent language, the distribution of definite and indefinite Cl+N

Information structure and (in)definite Cl+N

247

strictly follows the generalization about the correlation of word order and
information structure. If so, I expect that in Wu, preverbal Cl+N has a
definite reading and postverbal Cl+N has an indefinite reading, since the
former falls into the topic domain and the latter into the focus domain.
The sentence initial position is the default topic position in Chinese
languages, and Wu is no exception. The preposed object occurring between
subject and verb is characterized as secondary topic in Wu, which can be
marked overtly or covertly. According to Hu, Pan and Lee (2003) and Tang
(2006), in Wu Chinese (e.g. the Ningbo and the Shanghai variants), the
secondary topic position is grammaticalized. It is a stable position in Wu
syntax and it is a syntactic requirement to have a secondary topic for some
structures in Wu. As in the case of Fuyang Wu, definite objects are required to be located in the secondary topic position.
The Cantonese data reveals the same processes as in Wu, except for the
possibility of having definite readings in post-verbal positions. So why are
the latter possible?
In the first place, I think that the frequent use of Cl+N in preverbal
positions, such as subject, may help it to acquire the status of a quasidefinite marker, a use which then can be applied to it in other nonpreverbal positions. Secondly, I think that the most important reason for
the possibility of expressing definiteness in postverbal positions is the lack
of a productive object-preposing operation in Cantonese corresponding to
the Mandarin ba construction that we saw above. I propose that because
Cantonese does not have a productive mechanism for preposing definite
objects from postverbal positions, it doesnt have enough easy syntactic
ways of expressing meanings with, say, definite objects, leading to the
extension observed (see Li and Bisang 2012). Therefore, the lack of a productive disposal construction in Cantonese paves the way for the extension
of definite Cl+N from preverbal positions to postverbal positions.
In Mandarin, there is never definite Cl+N, but indefinite Cl+N only
(in lexically governed positions). According to D. Liu (2001), Mandarin is
a less prototypical topic-prominent language than Wu. I propose that, in
contrast to Wu and Cantonese, the Cl+N construction in Mandarin is not
fully grammaticalized: Cl+N is only found in base-generated postverbal
object position, a focus position where Cl+N has an indefinite interpretation. There is the famous ba construction in Mandarin, but it does not
guarantee a definite interpretation of Cl+N. What the disposal construction does is to arrange the object in a certain way according to a particular
criterion (e.g. definiteness), but the construction itself does not impose a
definite reading to the preposed object.

248

Definite classifiers in southern Chinese languages

The question that remains is: why does Mandarin not allow definite
Cl+N as subjects or topics? Although this is currently an open question,
one possibility is that it may be related to a typological difference between
northern and southern Chinese languages with respect to D. Liu (2001)
argues that northern Chinese languages like Mandarin are demonstrativeprominent languages in that they prefer to use demonstrative + N instead
of Cl+N to express definiteness, while southern Chinese languages like
Wu and Cantonese are classifier-prominent languages, in that they prefer to
use Cl+N over demonstrative + N. This is still an empirically-driven
typological observation, however, not a complete explanation. Tang (2006)
offers a syntactic explanation to explain the lack of definite Cl+N in
Mandarin from the perspective of verb movement, and I refer readers to
him for further consideration of this issue.
I now move from the distribution of these expressions to the syntax of
Cl+N.

4. Syntax of indefinite Cl+N


In this section I discuss the syntax of indefinite Cl+N. I address the following questions: (i) is the indefinite Cl+N a reduced form of
one+Cl+N by eliding the numeral one? (ii) What is the syntactic projection of indefinite Cl+N? Is it a NumP (with an empty number) or ClP?

4.1. Indefinite Cl+N as a reduced form of one+Cl+N (L 1944)


L (1944) derives the indefinite Cl+N from one+Cl+N in Mandarin.
L claims that there are two variants of the numeral yi one in Mandarin: a strong yi and a weak yi. The strong yi is stressed and it is the cardinal
one. It emphasizes the singularity of entities, as illustrated in (23).
(23) a. yitong
bei bu
de you *(yi) ge xuesheng, wu ge chefu.
meantime PASS arrest Mod have one CL student five CL carter
Those who were arrested at the same time include one student and
five carters.
b. wo *(yi) ge ren
qu.
I
one CL person go
I will go by myself.

Syntax of indefinite Cl+N

249

The weak yi is not only phonologically weak, i.e. unstressed, but also
semantically weak, i.e. it is not referential or quantity-denoting, shown in
(24):
(24) a. ta chi le
(yi) wan
fan jiu congcong zou le.
he eat PFV one CLbowl rice then hurriedly leave PRF
He left in a rush after taking a bowl of rice.
b. wo shi (yi) ge hen wangu de ren.
I be one Cl very stubborn Mod man
I am a very stubborn man.
L claims that the stressed yi cannot be omitted (as in (23)), but unstressed yi can be omitted in certain contexts (as in (24), and, importantly,
omission of unstressed yi is only possible in postverbal position.
Thus, according to L, Cl+N is a reduced form of unstressed yi + Cl
+ N in postverbal positions.4

4.2. Indefinite Cl+N as NumPs


In their comparative study of Cantonese and Mandarin nominal structures,
Cheng and Sybesma (1999:525-527) discuss the distinction between
Cl+N and yi+Cl+N. They argue, against L (1944), that indefinite
Cl+N cannot be regarded as a phonological reduction of yi+Cl+N by
omitting the numeral yi one.
They show that in Mandarin indefinite Cl+N only has a non-specific
reading, while one+Cl+N is three-way ambiguous between a specific, a
non-specific and a quantity reading. They use bounded predicates and secondary predicates as tests to distinguish Cl+N and yi+Cl+N.
Cheng and Sybesma show that a Cl-N phrase cannot occur as the object
of a bounded predicates, whereas a yi-Cl-N phrase can, as shown in (25):
(25) a. Wo chi-wan-le
yi-kuai
I eat-finish-PFV one-CL
I finished a cookie.
4

binggan.
cookie

L mentions that speakers tend to use Cl+N in the oral discourse, e.g. daily
conversation, but when they are asked to write it down, they usually write it in the
form of yi+Cl+N. I suggest that this phenomenon is to do with the teaching of
prescriptive grammar and the use of Chinese in a formal register in schools.

250

Definite classifiers in southern Chinese languages

b. *Wo chi-wan-le
kuai binggan.
I eat-finish-LE CL cookie
According to Cheng and Sybesma, there is no phonological reason why
yi one could not be suppressed in (25).
Huang (1987) shows that in secondary predicates the object NP (i.e. the
subject of the secondary predicate) must be indefinite and specific. Cheng
and Sybesma show that in this context, yi+Cl+N phrases cannot be replaced by Cl+N:
(26) a. Wo jiao-guo
yi-ge xuesheng hen congming.
I teach-EXP one-Cl student very intelligent
I once taught a student who was very intelligent.
b. *Wo jiao-guo ge xuesheng hen congming.
I teach-EXP CL student
very intelligent
While Cheng and Sybesma argue against the reduction of Cl+N to
one+ClN, surprisingly enough, later in the same paper they propose the
same structure for both: they propose that when Cl+N is interpreted as an
indefinite, it has an empty Num head, and the whole phrase projects into a
NumP:
(27)

NumP
Num

ClP
Cl

NP
N

However, to posit a null numeral for Cl+N predicts that a numeral,


i.e., yi one, can be recovered in Num+Cl+N, and this indeed echoes the
one-deletion analysis of Cl+N which views that there is an omitted numeral yi one in Cl+N.
I think that Cheng and Sybesmas arguments against the reductionist
approach are convincing. One+Cl+N and Cl+N are not identical.
Hence I think their syntactic analysis postulating a null one numeral is
not to be accepted.

Syntax of indefinite Cl+N

251

4.3. Indefinite Cl+N as ClP


In this section, I study some more differences between one+Cl+N and
indefinite Cl+N. With a different set of diagnostics, I reach the same
conclusion as Cheng and Sybesma (1999) that indefinite Cl+N is nonspecific, and one+Cl+N can be quantity-denoting, specific, and nonspecific. I propose that the indefinite Cl+N is a classifier phrase, not a
numeral phrase.
Like L (1944), I distinguish weak yi from strong yi. I will show that
indefinite Cl+N and one+Cl+N differ in distribution, regardless of
whether the numeral yi is stressed or not.
I look at stressed yi first. I show three differences between stressed yi
+CL+N and Cl+N.
First, stressed yi+Cl+N can be coordinated with other NumPs to express the accumulative quantity of different entities, as in (28a). In contrast,
Cl+N cannot be coordinated with full NumPs, as in (28b).
(28) a. wo chi le
yi
ge pingguo he liang
I eat PFV one CL apple and two
I ate one apple and two oranges.
b.* wo chi le
ge pingguo he liang ge
I eat PFV CL apple and two CL
Intended: I ate an apple and two oranges.

ge juzi.
CL orange
juzi.
orange

Secondly, true NumPs, including stressed yi+Cl+N, can occur in subject position with an indefinite interpretation, but, as we have seen, Cl+N
can never be subject in Mandarin:
(29) a. yi ge pingguo bu gou
wo chi.
one Cl apple
not enough I eat
One apple is not enough for me to eat.
b.*ge pingguo bu gou
wo chi.
Cl apple not enough I eat
Intended: An apple is not enough for me to eat.
A. Li (1998) argues that although Chinese generally only allows definite nominals as subject, NumPs can be used as subjects if they denote
pure quantities: in (29a) the predicate being not enough forces a quantity
denotation for the subject, one Cl apple. Such a quantity reading is impossible for Cl+N, and (29b) is ungrammatical.

252

Definite classifiers in southern Chinese languages

Thirdly, temporal expressions like xiaoshi hour or yue month are durational, and can only be used with Num+Cl+N, not with Cl+N:
(30) a. wo hua
le *(yi) ge xiaoshi chifan.
I spend PFV one CL hour eat meal
I spent one hour eating meal.
b. wo dai xainggang dai le *(yi) ge yue.
I at Hong Kong stay PFV one CL month
I stayed for one month in Hong Kong.
Next I discuss the differences between Cl+N and unstressed
yi+Cl+N.
Unstressed yi +Cl+N allows specific and non-specific readings, while
Cl+N can only have a non-specific reading in Mandarin. I show this with
examples involving relative clause modification and creation verbs.
Modification by a relative clause: Zhang (2006) shows that when a
relative clause (RC) modifies a NumP in the order RC+Num+Cl+N, the
NP has a specific reading, while if a relative clause modifies a NP in the
order Num+Cl+RC+N, the NP has both specific and non-specific readings. In (31) we see that yi+Cl+N allows modification with both types of
relative clauses:
(31) a. wo zai zhao [RC xue yingyu de ] yi ge xuesheng. [Specific]
I PROG seek learn English Mod one CL student
ta keneng zai tushuguan li.
he maybe at library in
I am looking for a student who learns English. Maybe he is in the
library.
b. wo zai
zhao yi
ge [RC xue yingyu de ] xuesheng.
I PROG seek one CL learn English Mod student
shui dou keyi, danshi zuihao shi da-si de.
[(Non-)specific]
who all fine but
best
be senior Mod
I want to look for a student who learns English. Any one will do,
but it is better to have a senior student.
(31a) means that the speaker has a specific student in mind, a student
that is not known by the addressee. (31b) allows the same reading, but also
allows a non-specific interpretation, where the speaker has no particular
individual in his mind, and anyone who learns English will be fine for him.

Syntax of indefinite Cl+N

253

In Cl+N, relative clauses can only modify the noun, not the whole
Cl+N phrase, as in (32):
(32) a.* wo xiang zhao [RC xue yingyu de] ge xuesheng.
I want seek study English Mod CL student
I want to look for the student who learns English.
b. wo xiang zhao ge [RC xue yingyu de] xuesheng. [Non-specific]
I want seek CL learn English Mod student
I want to look for a student who learns English.
When the noun of Cl+N is modified by the relative clause as in (32b),
it only has a non-specific reading, where any student who learns English
will do.
Jiang (2012) points out that the test I used here is flawed. She suggested
that it is possible for Cl+N to have a specific reading. She said that if an
appropriate context is provided, Cl+N in (32) can have a specific reading.
The example (c) is taken from Jiang (2012).
c. Zhang jiaoshou zai
zhao XiaoWang.
[Non-specific]
Zhang professor PROG seek Xiaowang
Ta xiang zhao ge [RC xue yingyu de] xuesheng.
he want seek CL
learn English Mod student
Prof. Zhang is looking for Xiaowang. He wants to look for a student who learns English.
According to Jiang (2012), the indefinite Cl+N has a specific reference and it refers to Xiaowang, the student Prof. Zhang is looking for.
However, I do not think that this is correct. In fact, Xiaowang is only one
of the candidates that suits the description expressed by Cl+N. The person that Cl+N refers to can be Xiaowang or anybody else who learns
English. It is still non-specific.
According to Fodor and Sag (1982), a specific indefinite introduces a
new discourse referent such that the speaker has a unique individual in
mind (cf. von Heusinger 2011:11). If CL+N in (32c) is specific and
refers to the individual XiaoWang, then the individual is introduced twice
and is not a new discourse referent. Moreover, as we said above, XiaoWang is not the only candidate that satisfy the description, so the uniqueness requirement for specific indefinite is also violated. We would thus
still consider Cl+N to be non-specific indefinite.

254

Definite classifiers in southern Chinese languages

Creation verbs: one+Cl+N and Cl+N have different interpretations in contexts of creation. Diesing (1992) notices that some verbs place
felicity restrictions on their objects that have to do with presuppositionality.
In particular, verbs of creation are incompatible with objects whose existence is presupposed.
Zhang (2006) reinterprets this constraint as a constraint disallowing
specific interpretations of indefinites in creation contexts. In (33) Cl+N
is acceptable. One+Cl+N is also acceptable, but not with a specific reading: yi ge dangao one Cl cake and ge dangao Cl cake cannot refer to a
particular cake that is identifiable to the hearer, both talk about an unspecific cake-baking event.
(33) a. wo kao le
yi
ge dangao.
I
bake PFV one CL cake
b. wo kao le ge dangao.
I
bake PFV CL cake
Both: I baked a cake.
We see that stressed yi +Cl+N has a quantity reading, unstressed
yi+Cl+N a specific and a non-specific reading, while Cl+N only has a
non-specific reading.
I propose the structure in (34) for indefinite Cl+N in Chinese, where
ClP is the maximal projection of the indefinite Cl+N and there is no
other functional projection above it.
(34)

ClP
Cl
NP
beni volume
N
shu book

5. Syntax of definite Cl+N


In this section I discuss the syntax of definite Cl+N. Our main question
here is: does the definite Cl+N project into a ClP (Cheng and Sybesma
1999) or a DP (Simpson 2005)?
In section 5.1 I review a phonological reduction approach to the definite
Cl+N (Shi and Liu 1985). In section 5.2, I will examine Cheng and Sy-

Syntax of definite Cl+N

255

besmas (1999) proposal that definite Cl+N are classifier phrases headed
by classifiers. I make our own proposal in section 5.3.

5.1. From Dem+Cl+N to definite Cl+N


The definite Cl+N construction has been discussed by many Chinese
dialectologists in different Chinese dialects. Some Chinese dialectologists
hold the view that definite Cl+N is derived from demonstrative phrases
by omitting the demonstrative (e.g. Shi and Liu 1985, Wang 2005).
Shi and Liu (1985) claim that in the Suzhou dialect of Wu, the definite
Cl+N is derived by deleting the distance-neutral demonstrative /g/
before Cl+N. Distance-neutral demonstrative refers to that whose
deixis is dependent on the context in which the expression is used and it
can either be distal or proximal. For example, when /g/ is contrasted with
a distal demonstrative, it indicates the entity to be close to the speaker;
when it is contrasted with a proximal demonstrative, it indicates the entity
to be far from the speaker. When it is used by itself, it is neutral in terms of
distance.
According to Shi and Liu (1985), definite classifiers are quite similar to
distance-neutral demonstratives in terms of their semantic function. On the
definite interpretation, Cl+N simply refers to identifiable entities and it
does not distinguish distance. To put it explicitly, Cl+N does not emphasize quantity, and carries no contrastive meaning, and has a weak deictic function (Shi and Liu 1985:160). Based on this similarity, Shi and Liu
assume that Cl+N originates from the demonstrative phrase of
/g/+one+Cl+N.
Shi and Liu (1985) also suggest that there is phonological evidence for
this analysis. They observe that in the Suzhou dialect, when the Cl+N construction has a definite reading, the (monosyllabic) classifier has the tone
sandhi from its base tone to a secondary high level tone (tone 44). This
tone is consistent with that of the classifier after the neutral demonstrative
/g/. For example, the classifier bu has the same tone of 44 when it appears after the demonstrative, as in (35a) and when it is used in Cl+N, as
in (35b).

256

Definite classifiers in southern Chinese languages

(35) a. ge44 bu44 thi tso


Dem CL car
this car
b. bu44 thi to
CL
car
the car
There are two problems with this proposal. First, if definite Cl+N is
derived from Dem+Cl+N by eliding the demonstrative, it is unclear how
the definite meaning is transferred from the demonstrative to the classifier.
Take the case of Cl+N in the Suzhou dialect discussed above. If Cl+N
is derived from Dem+Cl+N as proposed above, they ought to have the
same meaning. But this does not seem to be the case. In Shi and Lius paper, I find the examples from the Suzhou dialect in (36).
(36) a. ts
ts-d sa
z-ho lE- g ?
CLpiece paper what place PRT
Where is the piece of paper?
b. g ts
ts-d sa
z-ho lE- g ?
this CLpiece paper what place PRT
Where is that piece of paper? (Note: that has a distance neutral use)
While at a first glance, the subject ts ts-d piece paper in (36a) and
g ts ts-d this piece paper (36b) have the same meaning: both refer to
a particular paper that is identifiable in the context. However, the sentences
are in fact not synonymous. In (36a), the interlocutors know which paper is
under discussion; the definiteness of Cl+N relies on the context and
there is no contrastive (deictic) meaning involved. (36b), on the other hand,
has a salient deictic meaning. It is quite possible that the referent wasnt
mentioned in the previous discourse, and expresses new information and a
contrastive meaning.
Secondly, languages that have definite Cl+N do not necessarily have
a neutral demonstrative in their language system. For example, Cantonese
has the definite Cl+N construction, but it does not have a neutral demonstrative. Based on the current literature on Cantonese, there was no such a
distance-neutral demonstrative either in this history. This is unexpected
based on Liu and Shis analysis.
In view of these two problems, I assume that definite Cl+N and
Dem+Cl+N are different constructions. It is impossible that there is any
derivational relation between definite Cl+N and demonstrative phrases.

Syntax of definite Cl+N

257

More seriously, for language with both definite and indefinite Cl+N,
such as Fuyang Wu, it is very unlikely that the bare classifier construction
Cl+N has two different sources: indefinite Cl+N is derived from
one+Cl+N by omitting the numeral one if we follow L (1944), and definite Cl+N is derived from demonstrative+Cl+N by omitting the demonstrative if we follow Shi and Liu (1986). We reject both reductionist
approaches. The bare classifier construction Cl+N is a distinctive construction from one+Cl+N and demonstrative+Cl+N.

5.2. Definite Cl+N as ClP


Cheng and Sybesma (1999) propose that in languages with articles/determiners, the deictic function in the nominal phrase is taken care of
by the article/determiner (D0), while in languages without articles / determiners like Chinese, some of those functions are performed by the Cl0,
including the deictic function. Specifically, Cheng and Sybesma (1998)
assign the classifier in definite Cl+N the same meaning as the definite
article in English, i.e. being an iota operator.
Cheng and Sybesma propose (37) as the structure of definite Cl+N in
Cantonese, where the classifier projects as the head of ClP and does not
project any high nodes like NumP or DP.
(37)

ClP
Cl

NP
N

Thus, they treat indefinite Cl+N as a NumP with an empty Num, and
definite Cl+N as ClP without a Num projection.
This proposal has been criticized in Simpson (2005). Simpson argues
that the analysis of definite Cl+N as a lower projection than indefinite
Cl+N gives the wrong interpretation of true numeral classifier phrases of
Num+Cl+N. I use the Wu data to illustrate the problems pointed out by
Simpson. (38) shows that Cl+N can be modified by numerals and that
Num+Cl+N has an indefinite reading.

258

Definite classifiers in southern Chinese languages

(38) a. bu tshots phai die.


CL car
broken PRT
The car is broken.
b. ma le
ia bu tshots.
I buy PFV two CL car
I bought two cars.

[Wu: Fuyang]

Simpson (2005:14) points out that if the classifier has the same interpretation as the definite article in English, i.e. being a genuine definiteness
marker, one would expect (38b) to either have a definite interpretation, i.e.
the two cars or a partitive reading, i.e. two of the cars. However, this is
not the case, and neither reading is possible. We can only get an indefinite
reading for (38b), i.e. two cars.
For Cheng and Sybesma (1999), the definite reading is never available
for Num+Cl+N. They stipulate that NumP is always indefinite, and that
definite nominals are always projected into ClPs.
The problem with that stipulation is that, with it, the analysis does not
extend beyond Cantonese. In Wu (Shanghainese and the Fuyang dialect)
the Num+Cl+N can have a definite reading (see H.Wang 2008 for a general discussion on definite numerals in Wu dialects). For example, in the
Fuyang dialect, the morpheme ian is an imprecise quantifier meaning
several and can be used in the numeral phrase of ian+Cl+N. ian +Cl+N
has an indefinite reading in postverbal position, as in (39), and a definite
reading in preverbal position, as in (40).

ian k in l pa . [Indef NumP]


(39)  in
I search several CL people come help I
I am looking for some people to help me.
(40) ian k in t ga-i ?
several CL people at where
Where are the people?
Impossible: Where are some people?

[Definite NumP]

Note that the morpheme ian has a high-level tone when it expresses an
imprecise number (the) several; it has a low-rising tone, when it means
the exact number two. The definite reading is possible only in the imprecise number use.

Syntax of definite Cl+N

259

In view of these issues, I think that it is problematic to assume that the


definite Cl+N is a classifier phrase, with a less articulated projection than
the indefinite Cl+N.

5.3. Definite Cl+N as DP


I now propose a DP analysis for definite Cl+N (cf. Simpson 2005 for a
similar proposal).
A. Li (1999) argues that Chinese determiner phrases have the DP structure: [DP D[NumP Num[ClP Cl[NP N]]]. I propose that when Cl+N has a
definite reading, the classifier heads the projection of DP. I propose that
due to the lack of lexical elements in Num, and the lack of a lexical Delement, the Cl undergoes Cl-to-D raising and becomes the head of DP, as
in (41):
(41)

DP
D
pni volume

NumP
Num

ClP
Cl
ti

NP
N
y book

I discuss two arguments for assuming this structure.


Argument 1: One piece of syntactic evidence that support the analysis
is the fact that classifiers can modify proper names, as in (42b) (note that
Cl+proper name (42b) refers to the same individual as the bare proper
name does (42a)):
(42) a. i-ua
kints mi l zapan.
[Wu: Fuyang]
XiaoWang today not come work
XiaoWang did not come to work today.
b. k i-ua
kints mi l
zapan.
CL XiaoWang today not come work
The Xiaowang did not come to work today.

260

Definite classifiers in southern Chinese languages

Longobardi (1999) argues with the data in (43) that in Italian proper
names are generated in N and raised to D, except when the D is already
filled by a determiner.
(43) a.* Antica Roma
b. Roma Antica
c. Lantica Roma
In the same spirit I propose that in (42a), the Chinese proper name
XiaoWang is raised from N to D, whereas in (42b), the proper name must
remain in the position of NP and cannot undergo N-to-D raising, because
the D position is filled by the classifier k.
Simpson (2005) gives a similar argument on the basis of data from
Southeast Asian languages like Vietnamese, Thai and Cantonese. He
shows that in Vietnamese a second general classifier element can occur
preceding the regular classifier, resulting in sequences with clear definite
interpretations. The examples in (44) are from Simpson (2005:15). (Note
that we do not find a similar construction in Chinese languages.)
(44) a. con dao [anh cho toi muon ], no that sac.
[Vietnamese]
CL knife you give me borrow, it real sharp
The knife you gave me is really sharp.
b. cai con dao [anh cho toi muon ], no that sac
CL CL knife you give me borrow, it real sharp
The knife you gave me is really sharp.
In (44a), Cl+N has a definite interpretation; this definite interpretation
is preserved in in (44b), where Cl+N is modified by the general classifier
cai. This follows, if we assume with Simpson that in (44a) the classifier is
raised by Cl-to-N, while in (44b), the general classifier is in the D0 position.
Argument 2: I mentioned in Section 5.1 that in some dialects, like the
Suzhou dialect, the classifiers in the definite Cl+N construction have
tone sandhi. In the Fuyang dialect of Wu, I also observed tone sandhi for
the definite use of classifiers as in (45):
(45) a.  thi l i
I
eat
PFV one
I ate a bowl of rice.

uan432 van.
CLbowl rice

Syntax of definite Cl+N

261

b. uan432-55 van,  thi-i  die.


CLbowl rice I eat- finish
PRT
The bowl of rice, I finish it.
When the classifier uan bowl is used in normal classifier positions, it
has its base tone of 432 (falling and rising). However, when it is used in
the definite Cl+N, the CL changes its base tone 432 to a high tone 55. I
suggest that the changed tone can be seen as an overt realization of D at the
PF level.
One of the advantages of the DP proposal for definite Cl+N is that it
predicts that it is possible for NumP to have a definite reading if the Num
can undergo Num-to-D raising. I propose that this is what happens in (40),
repeated here:
(40) ian k in t ga-i ?
several CL people at where
Where are the people?

[Wu: Definite NumP]

I propose the following structure of the definite NumP in (40):


(46)

DP
D

NumP

ClP
Numti
iani several
Cl
k

NP
N
in people

To sum up, in this section, I argued for a DP analysis of definite


Cl+N where the classifier in Cl+N undergoes Cl-to-D raising to get the
definite reading. This proposal can not only account for definite Cl+N
but also definite Num+Cl+N in Wu (Num raises from Num0 to D0).

262

Definite classifiers in southern Chinese languages

6. Semantic interpretation of Cl+N


In this section, I will argue that the classifier in the Cl+N construction is
interpreted with a counting reading. I will use this to formulate a semantics
for indefinite and definite Cl+N.

6.1. Cl+N with a counting reading


I begin by examining in Mandarin and in Wu, which types of classifier can
be used in the construction of Cl+N.
I start with [-C, +M] classifiers like gongjin kilo, li mile, and bang
pound, measure words that are by default only associated with measure
interpretations. These classifiers require numerals (see also Yip 2008). As
(47) shows, measure words cannot form a Cl+N construction in Mandarin:
(47) a.* wo zou le
li
lu.
I walk PFV mile road
b. wo zou
le
yi
li
lu.
I walk PFV one mile road.
I walked one mile.

[Mandarin]

The Wu data in (48) show that [-C, +M] classifiers cannot form Cl+N
in subject position either:
(48) a. * di
lu
tso le
pan k tiodei.
CLmile road walk PFV I half CL hour
The mile of road took me half an hour.
ian k.
b. */??? thi s
CLliter water cool PRT
The liter of water is cool.

[Wu]

Next I check [+C, -M] classifiers, classifiers that are by default associated with counting readings. We have already seen many examples of these,
they can be used without numerals in postverbal positions in Mandarin:
(49) a. wo mai le ben shu.
I buy PFV CL book
I bought a book/#books.

[Mandarin]

Semantic interpretation of Cl+N

263

b. zhuo-shang fang-zhe ge pingguo.


table-on
put-DUR CL apple.
On the table lies an apple/# lie apples.
In both examples, Cl+N refers to individual entities. (49a) means that
there is an individual book that I bought, and (49b) means that there is a
single atomic apple lying on the table, not apple stuff or a plurality of apples.
The examples in (50) show that in Wu Chinese, Cl+N with [+C, -M]
classifiers can be used in the subject position with a definite reading.
(50) a. pn
y
z -ko.
CLvolume book be mine.
The book is mine.
b. k piku lani die.
CL apple rotten PRT
The apple is rotten.

[Wu]

I next look at [+C, +M] classifiers like ping bottle, xiang box, qun
group, dui pile etc in Mandarin. Those classifiers are equally open to
counting and measure interpretations. These classifiers can only be used in
Cl+N when the classifier has a counting reading. Look at the following:
(51) a. wo shou shang na zhe ping jiu.
[Mandarin]
I hand on
take DUR CLbottle wine
I am carrying a bottle of wine in my hand.
b.# ta-de wei
neng zhuang xia
ping jiu.
his stomach can contain down CLbottle wine
His stomach can contain a bottle of wine.
The container classifier phrase in (51a) has a counting reading, which
implies the existence of a concrete bottle filled with wine. The Cl+N, i.e.
ping jiu bottle of wine can only be indefinite. (51b), on the other hand,
talks about the capacity of the stomach or his drinking ability, which triggers a measure interpretation. The only available reading for (51b) is the
absurd interpretation that he has a concrete bottle in his stomach, i.e. the
reading where Cl+N ping jiu bottle wine is interpreted with a counting
reading, we are then forced to put a concrete bottle in his stomach. This is,
of course, infelicitous.

264

Definite classifiers in southern Chinese languages

In Wu Chinese, [+C, +M] classifiers can appear in the definite Cl+N


construction, when those classifiers are interpreted with a counting reading.
See the contrast between (52a) and (52b).
(52) a. thi ko bi
tiu
ikuo
tinkhiu ko.
[Wu]
I eat Mod CLbottle wine England import PRT
The bottle of wine that I drank was imported from England.
b.# ti ko bi
tiu
tele u-li
ta
s
la.
I eat Mod CLbottle wine at stomach-in Prog burn PRT
Intended: The bottle of wine that I drank is burning in my stomach.
The generalization is that a Cl+N construction is possible only when
the classifier is interpreted with the counting function. According to Yip
(2008), this generalization holds in Cantonese as well.

6.2. Semantics of indefinite Cl+N


In section 5.2, we argued that the indefinite Cl+N construction has the
maximal projection of ClP (as in (53)). This structure is consistent with our
analysis of the structure corresponding to the counting reading, where the
classifier first takes NP as complement and then the whole element can be
taken as complement by NumPs (cf. chapter 6).
(53)

ClP
Cl
beni volume

NP
N
shu book

The structure of (53) can form the basis of an indefinite NumP as in


(54a), or a definite DP with a demonstrative D as in (54b), or be the complement of Num, and Num+Cl+N the complement of DP, as in (54c).
(54) a. liang ben shu
two CL book
two books
b. zhe ben shu
this CL book
this book

Semantic interpretation of Cl+N

265

c. zhe liang ben shu


these two CL book
these two books
Following the observation that the classifier in Cl+N only has a
counting reading, we propose that indefinite Cl+N denotes a set of
atomic entities, a set of singularities. In Chapter 6, we proposed the following meaning for the [+Counting] classifiers:
(55) kx. 1(x)(>k ?k)(1(x))2(x)=k
The classifier in (55) takes a kind as input and maps it onto set of the
atomic instances of the kind in counting context k, that have the lexical
property expressed by the classifier. This is the basis for the indefinite
reading of Cl+N:
(kind-denoting)
(56) a. NshuN =?BOOK
>?
b.Nben shuN=x. 1(x) ( BOOK k) Volume(1(x)) 2(x)=k
This semantics for Cl+N makes two predictions.
First, it predicts that the indefinite Cl+N is available in postverbal positions in all the three Chinese languages. The semantics of Cl+N in (55)
interprets Cl+N as a predicative nominal phrase of type <dk, t>. This is
the right indefinite semantics for noun phrases in post-copula positions
(Partee 1987). For object position, I make the assumption, which is widely
made in the literature following Heim (1982), that object position is in the
scope of an operation of default existential closure, and I assume that it is
this operation that gives the Cl+N its indefinite interpretation. The
interpretation is shown in (57):
(57) a. Wo mai le
ben
shu.
I buy PFV CLvolume book
I bought a book.
b. Nben shuN=x.1(x) (>?BOOK k) Volume(1(x)) 2(x)= k
c. x
e[BUY(e) Ag(e)=I Th(e)=x
1(x)(>?BOOKk)Volume(1(x)) 2(x)=k]
However, unlike in English, I assume that default existential closure has
scope over the VP and not over the subject position (e.g. Diesing 1992).

266

Definite classifiers in southern Chinese languages

This explains why indefinite Cl+N cannot occur preverbally, unless they
are under the scope of the explicit existential operator you as was shown in
(9a).
Secondly, the semantics in (55) predicts that Cl+N can only be interpreted as a singular indefinite, not a plural indefinite. Cl+N denotes a set
of atoms, a set of singularities, and existential closure expresses that this
set is not empty.

6.3. Semantics of definite Cl+N: from counting to definiteness-marking


In this subsection, I propose a semantics for the definite interpretation of
the classifiers
Lyons (1999) claims that the semantic-pragmatic concept of definiteness exists in all languages, but that its grammatical realization may differ
cross-linguistically. While languages pick their notion of definiteness from
the family of concepts that relate to inclusiveness, identifiability and referentiality, they may differ in what element they choose. So the crucial
question for us is: what does definiteness of classifiers mean in Chinese
languages?

6.3.1. A uniqueness-based approach of definiteness


Cheng and Sybesma (1999) assume that classifiers with a definite interpretation have the same meaning as the English definite article: if we maintain that Chinese nouns are predicates, classifiers are like Ds in that (a)
they are type-shifters, changing predicates into arguments, and (b) they
yield the definite interpretation (comparable to an iota operator). (Cheng
and Sybesma: 1999: 520-521).
Wu and Bodomo (2009) argue against this. They show that the definite
reading of Cheng and Sybesmas Cantonese examples has restrictions that
English definite phrases headed by the do not have. Consider (58).
(58) a. keoi maai-zo gaa ce.
he sell-ZO CL car
He sold the car.
(Cheng and Sybesma 1999:524, (25b))
b. Wufei jam-jyun
wun /di
tong la.
Wufei drink-finish CLbowl /CLPL soup PRF
Wufei finished the soup. (Cheng and Sybesma 2005:270, (24b))

Semantic interpretation of Cl+N

267

Wu and Bodomo (2009:497) argue that (58a) is only appropriate when


the hearer already knows that a third person owned a car (and only one car),
and that (58b) requires that the hearer knows that a person called Wufei
was served a particular bowl of soup or some particular soup. Without
these assumptions, you do not get a definite reading, but an indefinite reading, where the bare classifier gaa ce in (58a) is understood as a car, and
wun/di tong in (58b) as a bowl of soup/some soup. The definite interpretation of definite Cl+N depends on the context to a great exent.
I will show here that definite classifiers in Wu shouldnt be identified
with English style determiners either.
First, the+N in English obviously does not have the same distribution
as definite Cl+N in Wu. As shown earlier, in Wu, definite Cl+N is
restricted to preverbal positions, while the+N in English does not have
such a restriction.
Secondly, the definite classifier in Wu does not have a uniqueness presupposition. Uniqness is neither a necessary nor a suffcient condition for
the use of definite classifiers in Wu.
Following Sharvy (1980) and Link (1983), the English definite article
the is interpreted as a sigma operation (generalizing Russells 1905 iota
operation): an operation that maps a set on the sum of the elements in that
set, if that sum is itself in the set, and is undefined otherwise. Now, I assume a predicative interpretation for Cl+N, and a DP interpretation for
the whole phrase, so it is not the classifier itself that is interpreted as the
sigma operation, but, presumably, the D position. But the sigma operation
has a uniqueness presupposition built into its meaning. Li and Bisang
(2012) argue that this is not appropriate for Wu.
In the first place, note that uniqueness is found in Chinese languages
like Wu in expressions that are not classifier phrases at all.
Lbner (1985) proposes that across languages, there exists a special
group of nouns which refers to unique entities, independently of the particular situation referred to, nouns like sky and sun. Those nouns tend to
occur with the definite article in English: the sky and the sun. In Wu, these
nouns occur as bare nouns, as in (59).
(59) a. thi z lan ko.
[generic]
[Wu]
sky be blue PRT
The sky is blue (in general).
fa-kua-fa-i.
b. tha-ian i-tin-v-tin ko
sun
non-stop
Mod emit-light-emit-heat
The sun emits light and heat non-stop.

268

Definite classifiers in southern Chinese languages

Note that the examples in (59) are generic: (59a) means that the sky is
blue in general and (59b) means that the sun has the properties ascribed to
it in a permanent way.
Similarly, proper names referring to places or locations can also be used
to make unique reference independent of context. Again, in English, they
take the definite article: the Great Wall, the Capital. In contrast, the counterparts in Wu occur as bare nouns:
ko ?
(60) a. dzan tshn ko zkua z
Great Wall what time
build PRT
When was the Great Wall built?
b. [Looking at the map of China]
iudo
t ga-i?
Capital at where
Where is the Capital?
Secondly, the examples with definite Cl+N in Wu that I have discussed do not generally presuppose uniqueness. Look at the examples in
(61):
(61) a. [In a room with three doors, one of which is open]
uan mn b kuan- i.
CL door help me close
it
Please help me close the door.
b. kints ul pnl la, ts leth thi pha i die.
today back halfway on Cl tyre go broken PRT
Today, on the way back, the tyre was flat.
(61a) is an immediate situational use: three doors are part of the background, and I do not need to assume that Cl+door must refer to a door
that is presupposed to be unique. There is reference to a single door, because of the semantics of the classifier given, but the actual utterance will
identify the correct one without presupposition requirements. In other
words, if accommodation goes on here (as predicted by Kadmon 1987), it
is so subtle that native speakers do not notice it.
(61b) is an associative use or a bridging cross-reference use situation. A
singular tyre is mentioned of a car which is known to the hearer and the
speaker. Not enough information is given to determine which of the four
tyres it is, i.e. not enough information is given to uniquely identify the tyre.
And such information is not required either.

Semantic interpretation of Cl+N

269

These cases are not atypical for Wu at all, nor for Cantonese: mostly the
uses of definite Cl+N are like the cases in (59), uniqueness plays no role.

6.3.2. A familiarity-based approach of definiteness


While I have argued that uniqueness plays no role in the semantics of definite [Cl+N], familiarity does. Familiarity accounts of definiteness are given
in the literature by Christopherson (1939), Hawkins (1978), Heim (1982),
and many after that. The idea is that definites are used to pick out referents
that are in some sense familiar to the discourse participants. For example,
Christophersen (1939) argues that the use of the in English directs the
hearer to the referent of a noun phrase, by indicating that this referent is
familiar to hearer as well as speaker. Heim (1982) proposes that the use of
a noun phrase is familiar if it links to a discourse referent about which
there is already information in the local context of interpretation, which
she formalizes as a file of information held in common by the interlocutors
in the discourse.
A recent proposal concerning familiarity is made by Roberts (2003).
Roberts (2003) distinguishes two kinds of familiarity: strong familiarity
and weak familiarity. Strong familiarity refers to discourse-anaphoric
uses of definites, which essentially require a definite to be anaphoric to a
preceding linguistic expression. In contrast, the notion of weak familiarity allows for a number of ways in which something can be familiar: by
being perceptually accessible to the discourse participants, by being globally familiar in the general culture, or by being derivable through contextual existence entailments (Roberts 2003: 304). Weak familiarity only requires that the existence of the relevant entity be entailed by the
interlocutors common ground (Roberts 2003: 306).
I propose that definiteness in the Chinese languages be identified with
Roberts notion of weak familiarity. I propose that definite Cl+N refer to
entities that are directly involved in the situation or are presupposed to be
familiar or identifiable by interlocutors, as part of the background information.
What contexts allow the presence of Cl+N? In general, the definite
Cl+N construction is more frequently found in daily conversations than
in descriptive contexts like, for instance, story-telling. The most natural
situations for their use are immediate situation uses like we have seen in
(59): on-the-spot uses, the bridging use etc. Anaphoric uses are not very

270

Definite classifiers in southern Chinese languages

natural. In what follows, I will use the data of Wu Chinese (the Fuyang
dialect) to illustrate the possible contexts for definite Cl+N.
Context 1 the on-the-spot use: definite Cl+N naturally refers to a
perceptually visible entity.
(62) khua kul, bu tshots p
ia kua-i die. [Wu: Fuyang]
quick come CL car PASS they scratch PRT
Come over! The car was scratched by someone.
Here the speaker is reporting what he/she found in the scene. Suppose
that there are two cars in their family and they are parked next to each
other. The speaker finds that one of them is scratched and calls the rest of
the family members. The hearers will only identify which car is scratched
after entering the immediate situation.
The expression k in Cl man can even refer to the speaker or the addressee if (and only if) they are on the spot:
[Wu: Fuyang]
(63) a. kints k in man thii.
today CL man very tired
Today, the person (the speaker) is very tired.
b. km
k in ka l ian man.
currently Cl man so old look PRT
Lately, the person (the addressee) looks so old.
Cl+N in (63a) refers to the speaker himself, and (63b) refers to the
hearer on the spot. A scenario for (63a) is the following: a husband returns
home after working for a long day. When he sees his wife in the kitchen,
he says (63a) to describe his tiredness to his wife. A scenario for (63b) is
the following: two old friends have not seen each other for a few years.
One day they meet in the street. One of them says (63b) to the other, to
comment on the addressees physical appearance.
Context 2the familiarity use: definite Cl+N refers to the entity
that is assumed to be known by interlocutors in a local context.
(64) a-bi, ts kiu iankhan san mbi die.
A-Ping, CL dog seem
get sick PRT
A-Ping, the dog seems to get sick.

[Wu: Fuyang]

In (64), the speaker is telling the hearer about the sickness of the dog. In
reporting this event, he assumes that the hearer knows which dog he is

Semantic interpretation of Cl+N

271

talking about. In most cases, the definite refers to the individual in their
immediate environment, e.g. their own dog at home. But the dog need not
be on the scene.
In the two uses discussed so far, the definite can easily be a new definite, not mentioned before in the discourse.
Context 3the bridging use: the definite CL+N can be modified by
different modifiers, e.g. nouns, pronouns, adjectives, relative clauses.
[Modifier+CL+N] always implies definiteness.
(65) a. [ts kiu] [ts tia] tni die.
CL dog CL leg fracture PRT
The leg of the dog was fractured.
die.
b. fakan [k khkuan] uai
room CL switch
broken PRT
The switch of the room does not work.

[Wu: Fuyang]

In these examples, the head noun has a relational interpretation, such as


the leg of etc.. Clark (1977) calls this the bridging phenomenon. I
assume that the modifiers provide relevant contextual information to help
to identify relevant referents in the denotation of N.
As I mentioned, definite Cl+N are not naturally used in cases of
strong familiarity, i.e. as discourse anaphors. In the English example in
(66a) (from Heim 1982), the glass is strongly familiar, since it is a discourse anaphor to the indefinite antecedent a glass. In Wu, it is preferable
to use demonstratives or pro-drop in such cases, as in (66c-d).
(66) a. A wine glass broke last night. The glass had been very expensive.
b. ? khpha l
ts pts. [ts pts] mn ku.
I break PFV CL glass CL glass very expensive
I broke a glass. The glass had been very expensive.
ts pts. mn ku
ko.
c. khpha l
I break PFV CL glass very expensive PRT
I broke a glass. (It) had been very expensive.
d. khpha l ts pts. [k ts pts] man ku.
I break PFV CL glass DEM CL glass very expensive
I broke a glass. This glass had been very expensive.
There are two facts that a familiarity-based proposal can easily explain.

272

Definite classifiers in southern Chinese languages

First, definite Cl+N has an episodic reading, while definite bare


nouns have a generic reading:
(67) a. (*k) tha-ia z iuan ko.
Cl sun
be round PRT
The sun is round.
b. kints *(k) tha-ia man d.
today CL sun
very big
Today, the sun is really strong.

[Generic]

[Episodic]

(67a), with the bare noun tha-ia sun, means that the sun has the generic property of being round. In contrast, (67b) with the definite Cl+N
k tha-ia means that the sun is really strong at a particular moment.
I propose that bare nouns like tha-ia have a uniqueness requirement
built into their semantics and are used naturally in contexts where we express what the characteristic properties of this unique object are. On the
other hand, the definite reading of Cl+N k tha-ia implies familiarity in
a context here and now: we are likely to use it to express properties that are
instantiated here and now, episodic properties.
The second fact is why there is a subject-object asymmetry for definite
Cl+N in Wu:
(68) a.
khpha le [ts uan].
1SG break PFV CL bowl
I broke a bowl.
NOT I broke the bowl.

[SVO]

b. kh [ts uan ] thi khpha die.


1SG OM CL bowl go break PRT
I broke the bowl.
c. [ts uan], khpha ko.
CL bowl 1SG break PRT
The bowl, I broke (it).
NOT A bowl, I broken (it).

[SOV]

[Topicalization]

As explained, in Chinese, topics are usually found in preverbal positions and express hearer-old information. In contrast, foci are usually in
postverbal positions and express hearer-new information. But, of course,
familiarity is a hearer-old notion. Hence we expect definite Cl+N to
occur in topic positions, hence preverbal positions.

Semantic interpretation of Cl+N

273

6.3.3. Semantics of definite Cl+N


I now formulate the semantics of definite CL+N.
The semantics I gave for indefinite CL+N, such as ben shu CL book
in (69), is repeated in (70):
(69) Wo mai le
ben
shu.
I buy PFV CLvolume book
I bought a book.
(70) a. Nben shuN=x.1(x)(?>BOOK k)Volume(1(x)) 2(x)= k
b. x
e[BUY(e)Ag(e)=ITh(e)=x1(x)(?>BOOKk)
Volume(1(x)) 2(x)=k]]
I derive a predicative meaning for Cl+N, which gets existentially
closed by default existential closure in postverbal positions like object
position.
I assume the same predicative interpretation for definite Cl+N to start
with. I proposed a different syntax for definite Cl+N, which involves Clto-D raising. I claim that this syntactic operation is semantically interpreted
much like the operation of argument formation AF in Landman (2004).
Landmans AF operation is in essence a definiteness type shifting rule: it
maps a predicate interpretation on a generalized quantifier which is the
result of existentially closing and maximalizing the predicate interpretation.
Now maximalization corresponds to Sharvys sigma operation, and I have
argued in this chapter that the sigma operation is not appropriate for Chinese. So I want to replace it by what is appropriate for Chinese, a condition
of weak familiarity. I formulate argument formation as:
(71) AFWEAK-FAM: P.x[(x) WEAK-FAM(x) P(x)]
I propose, then, that AFWEAK-FAM is the interpretation of Cl-to-D raising.
This means that it will take the predicative meaning of Cl+N (in (72a)
into a generalized quantifier meaning (in (72b):
(72) a. x.1(x) (>k k) (1(x)) 2(x)= k
b.Px[1(x)(>kk)(1(x))2(x)=kWEAK-FAM(1(x))
P(1 (x))]
I work this out for the definite Cl+N in (73):

274

Definite classifiers in southern Chinese languages

(73) kints man lan. k thuado la pi ti


man die.
today very cold CL window on ice freeze full PRT
Today is very cold. The window is full of ice.
(74) Denotation of Cl+N at type <dk, t>
a. Predicate interpretation:
Nk thuadoN=x.1(x)(>?WINDOW ?k)Unit(1(x))2(x)=k
b. Definite interpretation:
Px[1(x)(>?WINDOW k)Unit(1(x))2(x)=k
WEAK-FAM(1(x))P(1 (x))]
With this semantic expression, the window will need to satisfy the requirements of weak familiarity discussed above.

7. Summary
This chapter examined the distribution, syntax and semantics of the indefinite and definite Cl+N construction in three Sinitic languages. I showed
that the distribution of definite and indefinite Cl+N is constrained to
different degrees by the information structure in the three languages I examined. In particular, definite Cl+N are typically found in preverbal
positions, which are usually topics, and indefinite Cl+N in postverbal
positions, which are usually foci.
I argued that indefinite Cl+N are classifier phrases whose indefinite
semantics in object position are derived from default existential closure
over the internal arguments of the VP, (following in essence Heim 1982),
while definite Cl+N are DPs, in which the classifier undergoes Cl-to-D
raising, a process which is semantically interpreted as argument formation
(following Landman 2004). Argument formation produces a definite
generalized quantifier interpretation. Whereas in English the definiteness
condition involved is taken to be a maximalization condition (following
Sharvy 1980), for Chinese this condition is taken to be weak familiarity
(following Roberts 2003).

Chapter 10
Definite classifiers and their modifiers

1. Introduction to modified Cl+N


In chapter 9, I showed that in some southern Chinese languages, classifiers
are able to undertake the role of definiteness-marking in the ClassifierNoun construction. Specifically, I argued that definite classifiers in Cl+N
in Wu and Cantonese are not fully-grammaticalized definite articles and
they are quasi definite articles: (i) they are able to instantiate the D head
(via Cl-to-D raising) and head a DP projection; (ii) definiteness expressed
by definite classifiers is characterized with familiarity but not uniqueness (cf. Li and Bisang 2012). This goes against Bokovis (2010:13)
claim that obligatory numeral classifier systems occur only in NP languages if a language has an obligatory classifier system, it does not have
DP.
In the current chapter, I extend the research from the bare form of
Cl+N to non-bare Cl+N, i.e. the complex phrase composed of Cl+N
and its preceding elements, such as adjectives, demonstratives etc. In particular, I make a case study on non-bare Cl+N in Wu (the Fuyang dialect). I will show that the peculiar syntactic and semantic properties of
non-bare Cl+N in Wu can only be explained with an appeal to the DP
proposal of definite classifiers.
As discussed in chapter 9, both indefinite and definite readings are
available for Cl+N in Wu: preverbal Cl+N with a definite reading and
postverbal ones with an indefinite reading. These two uses are represented
in the example (1).
[pn
(1) [k iasn] iits thi ma l
CL student yesterday go buy PFV CL
The student went to buy a book yesterday.

y].

book

In addition to that, Cl+N in Wu can also be used in non-bare forms.


That is, they can be preceded by adjectives, demonstratives and possessors

276

Definite classifiers and their modifiers

etc. Unlike bare Cl+N that is ambiguous between indefinite and definite
readings, non-bare Cl+N is unambiguously definite. Consider (2):
[ts kiu]
(2) a. i k
small Mod CL dog
the small dog
b. ta [ts kiu]
that CL dog
that dog
c. a
[ts
kiu]
you
CL dog
your dog

[Adjective/RC]

[Demonstrative]

[Possessor]

It is even possible to have multiple occurrences of those elements before Cl+N in Wu, as in (3). But they are subject to a certain ordering
restriction. For example, the demonstrative must stand closest to Cl+N,
as in (3a) and it cannot precede a modified Cl+N, as in (3b).
santshl k
(3) a. in
new
born
Mod
that newly born dog
b.#ta in santshl k
that new born
Mod

ta [ts kiu]
that CL dog
[ts kiu]
CL dog

In contrast, in Mandarin, there is only indefinite Cl+N but no definite


Cl+N. The sentence (4a) is ungrammatical with Cl+N as subject. It is
sufficient to use a bare noun, such as bi pen as its subject, meaning the
pen(s). However, Mandarin Cl+N can be used in postverbal positions
without any problem, where it has an indefinite reading, as in (4b).
(4) a. * [zhi bi]
huai
le.
CL pen broken PRF
The pen was broken.
b. wo jintian mai le
[zhi bi].
I
today buy PFV CL pen.
I bought a pen today.

[Mandarin]

Moreover, Mandarin does not have non-bare Cl+N either. It is impossible to find Adjective/Relative clause/Possessor/+Cl+N in Manda-

Syntax of modified Cl+N in Wu

277

rin and the only exception is the demonstrative phrase Dem+Cl+N.


Compare (5a-b) with (5c).
(5) a.* xiao
de
zhi
gou
small
Mod CL
dog
Intended: the small dog
b.*wo
de
zhi
gou
I
Mod CL
dog
Intended: the dog of mine
c. na
zhi
gou
that
CL
dog
that dog

[Adj/RC]

[Possessor]

[Demonstrative]

The following questions are raised from the data exhibited in (1-5):
a. What is the underlying reason why Wu allows Cl+N to be modified and Mandarin does not? Is it possible to propose a unified DP
structure for determiner phrases in Mandarin and Wu?
b. What is the ordering restriction regarding the multiple occurrence
of modifiers before Cl+N in Wu? What does that tell us about
their syntactic status?
c. Why can those elements preceding Cl+N induce a definite reading in Wu? How can Cl+N and its modifiers be interpreted in a
compositional way?
The rest of the chapter is structured as follows. Section 2 discusses the
syntactic relations of definite classifier and its preceding elements. I argue
that modified Cl+N is a definite expression, in which the classifier heads
a DP projection and demonstratives are located in [Spec DP] and other
preceding elements are DP modifiers. In Section 3, I propose a compositional semantics for modified Cl+N la Bach and Cooper (1978). The
last section is the conclusion of the chapter.

2. Syntax of modified Cl+N in Wu


In this section, I will look at the syntax of definite classifiers by examining
the relation between Cl+N and its preceding elements. The following
two questions will be examined: (i) is non-bare Cl+N an indefinite
phrase with a specific reading or a definite phrase? (ii) if non-bare Cl+N

278

Definite classifiers and their modifiers

is a definite phrase, then what is the syntactic relation between the definite
classifier and its preceding elements?

2.1. Modified Cl+N as a definite expression


As said earlier, a non-modified Cl+N in Wu can be interpreted with either a definite or an indefinite reading in different syntactic contexts (recall
example (1)). However, a modified Cl+N has an unambiguous definite
reading in Wu. Modified Cl+N refers to entities that are assumed to be
known to the interlocutors or are able to be identified in the context. Consider the examples in (6):
y
(6) a.
k
pn
red
Mod CL
book
the red book
b. ta mts k
k iku
wear hat
Mod CL boy
the boy that wears a hat

In (6a), by using the modified Cl+N, the speaker intends the hearer to
identify a particular book, namely, the red book but not the others. The
sentence (6a) presupposes that in the given context, there are more than
one book and the adjective red helps to identify the relevant one, to be
contrasted with the alternatives, say, the black one, the white one etc. We
suggest that a modified Cl+N, such as the one in (6a), expresses a definite reading and not a specific indefinite reading. The example (6b) can be
interpreted in the same way that the hearer is able to identify the relevant
boy with the property expressed by the modifier, i.e. the property of wearing a hat.
I now show syntactic evidence to prove that non-bare [Cl-N] is a definite expression, but not a specific indefinite phrase. I provide two pieces of
evidence in support of this statement.
The first piece of evidence is concerned with the definiteness effect
in existential clauses (Milsark 1987). Existential clauses in Mandarin are
expressed by the existential verb you there be and only indefinite phrases,
such as numeral classifier phrases, are allowed to be the pivot of existential
(see Huang 1987 for Mandarin existentials). Existentials in Wu work in the
same way as Mandarin. They are introduced by the existential verb iu
there be and definite phrases are not allowed in existentials.

Syntax of modified Cl+N in Wu

279

For example, (7a) has the phrase Cl+N, i.e. k in, as pivot of the existential clause, but it can only be interpreted with an indefinite reading,
meaning a person, but not a definite reading the person. It is parallel to
the example with a numeral classifier phrase, as in (7b).
(7) a. xotia
iu
k in
ti
t .
downstairs there:be CL people PROG wait you
There is a person waiting for you downstairs.
b. xotia
iu
i k
in
ti
t .
downstairs there:be one CL
people PROG wait you
There is a person waiting for you downstairs.
On the contrary, modified Cl+N is not allowed to appear in existential,
as shown in (8a). To make this sentence grammatical, we must delete the
existential iu there be, as in (8b). The sentence (8b) is turned into a nonexistential sentence, which has a modified Cl+N as subject.
(8) a. # xotia
iu
so in
k
k in
ti
t .
downstairs there:be send letter Mod CL people PROG wait you
There is the man who sends newspaper is waiting for you downstairs.
b. xotia
so in
k
k in
ti
t .
downstairs send letter Mod CL people PROG wait you
The man who sends newspaper is waiting for you downstairs.
The second piece of evidence is concerned with the possibility of the
recovery of the numeral one.
Cl+N always has a singular interpretation. When Cl+N has an indefinite singular reading, it is possible to recover the numeral i one before it. However, it is impossible to have ONE insertion before definite
Cl+N. For example, in (9a), we can insert an optional ONE before the
postverbal tin i-sa CL shirt, meaning a shirt, but in (9b), it is impossible to have ONE before definite Cl+N at the subject position.
(9) a. a m-ma p ta
l
(i) [tin i-sa] l.
you mum give you bring PFV one CL shirt directional
Your mum (asked me to) bring a shirt to you.
b. (*i) [ts kiu] m-i di.
one CL dog lost
PRT
The dog is missing.

280

Definite classifiers and their modifiers

In the case of modified Cl+N that we discuss here, the operation of


ONE insertion is never possible between Cl+N and its modifiers. The
relevant examples are given in (10). They behave like definite Cl+N but
not indefinite Cl+N.
(10) a. ta
(*i) ts
that
one
CL
that dog
b. i
k
(*i)
small Mod one
the small dog
c. in san tshl k
new born
Mod
the newly born dog

kiu
dog

[Demonstrative]

ts
CL

kiu
dog

(*i)
one

ts
CL

[Adjective]

kiu
dog

[Relative clause]

From these three facts, I conclude that the modified Cl+N is a definite
expression, but not a specific indefinite phrase. The classifier in modified
Cl+N is supposed to have the same syntactic and semantic properties as
its use in non-modified Cl+N.

2.2. Modified Cl+N as DP


Given that modified Cl+N is a definite phrase, we assume that it can be
decomposed into definite bare Cl+N phrases and left pheripheral elements in a simplified way. In this subsection, we will first examine whether
the relevant linguistic facts about modified Cl+N are able to support the
DP proposal of definite classifiers, as proposed in Chapter 9 and in Li and
Bisang (2012). Secondly, we will figure out where the preceding elements
should be located with respect to definite Cl+N in the relevant syntactic
structure. Specifically, we will separate demonstratives from the rest, such
as adjectives, relative clauses, and possessors etc. We will argue that in Wu,
demonstratives are [Spec DP] and adjectives, relative clauses etc. are DP
modifiers.

2.2.1. Definite classifiers as D head


It is argued earlier that in Wu, in definite bare Cl+N, the classifier instantiates the D0, via Cl0-to-D0 raising. One crucial piece of evidence in

Syntax of modified Cl+N in Wu

281

support of the DP hypothesis for definite Cl+N given by Li and Bisang


(2012) is about the use of proper names in Wu. It is shown below:
(11) What about XiaoWang?
a. i-ua
in?
Xiao Wang Q
b. k i-ua
in?
CL XiaoWang Q
Proper names with or without classifier have the same denotation, i.e.
being rigid designators. For example, both k iua and iua refer to the
same individual, XiaoWang. Assuming that both proper name and CLproper name are treated as DP, a plausible account for the data in (21) is
that: proper names are base-generated as N, which may be raised to D position, as in (11a), or remain in N position if the D position is lexically filled,
such as a definite classifier, as in (11b) (see Longobardi 1999 for the discussion on proper names in Italian).
I now provide two independent pieces of evidence to support the DP
structure of modified Cl+N.
The first evidence comes from coordination. When two modifiers are
coordinated and then followed by a classifier, such as
[[Possessor1 +Possessor2]+Cl], it makes reference to a singular entity, as
in (12a). However, if what is coordinated are two modifier-Cl, such as
[[Possessor1+CL]+[Possessor2-CL]], the coordinated phrase makes a plural reference, as in (12b).
(12) a. [iua ts ilo ] pn y
[Singularity]
XiaoWang
and Xiao Lu CL
book
Xiao Wang and Xiao Lus book
b. [DP [iua] pn] ts [DP[ilo] pn] y
[Plurality]
XiaoWang CL and
XiaoLu CL book
XiaoWangs (book) and Xiao Lus book
The test in (12a) suggests that the classifier is not part of the modifier
and they occupy two different positions. The plural referentiality of the
coordinated phrase suggests that the coordination of [Modifier-CL] in (12b)
may be a sort of DP coordination and that the head of the DP is most likely
to be headed by the classifier (see Jenks 2011 for relevant facts in Thai).
The second piece of evidence is about the singular demonstrative phrase.
In both Wu and Mandarin, Dem-Cl-N implies singularity. In Mandarin, an

282

Definite classifiers and their modifiers

optional numeral one can be inserted between Dem and [Cl-N], as in


1
(13a), but such insertion is impossible in Wu, as in (13b).
(13) a. na
that
b. ta
that

(yi) ben shu


one CL book
(*i) pn y
one CL book

[Mandarin]
[Wu]

One of the ways of explaining the contrast in (13) is that in Mandarin,


the classifier in Dem-Cl-N remains at its base-generated classifier position,
i.e. between Num and N, and that in Wu, the classifier is located at a position higher than Num0. It is possibly raised from Cl0 to D0 by crossing the
empty Num head, if we assume that Wu has the DP structure: [DPD[NumP
Num [ClP Cl [NP N]]]].
Based on the two facts exhibited in (12) and (13), we suggest that in
modified Cl+N in Wu, the classifier no longer sits at its base-generated
Cl0 position and it is a determiner head and it heads a DP projection.
Therefore, definite classifiers always act as the head of DP, regardless of
being modified or not.

2.2.2. Dems as [Spec DP]


This section examines the syntactic status of the elements preceding
Cl+N, such as demonstratives, adjectives, possessors etc. I treat demonstratives separately from the other types of modifiers, since they may occupy two different syntactic positions. I claim that demonstratives are specifiers to the DP and the other modifiers are simply DP modifiers.
I deal with the syntactic status of demonstrative first. There are two interesting facts about demonstratives in Wu.
First, demonstratives are able to express definiteness. Numeral classifier phrases are indefinite in Wu in general and a demonstrative is needed to
turn them into definites. However, the phrase Dem-Num-Cl-N can only
express plurality but not singularity. In other wods, the numeral after Dem
can only be larger than one. Compare (14a) with (14b).

In Mandarin, modifiers can modify [Dem-Cl-N] only but not [Cl-N]. Maybe this
is due to the fact that in Mandarin, [Cl-N] is inherently indefinite and Cl0-to-D0
raising is prohibited in general.

Syntax of modified Cl+N in Wu

283

(14) a. ta
san
pn y
that
three CL book
these three books
b.* ta i
pn
y
that one
CL
book
Intended: that book
To express singularity, we simply use Cl+N (15a) or Dem+Cl+N
(15b), but demonstrative cannot directly modify a noun without a classifier
(15c). Therefore, the singularity of (15b) is dependent on the classifier and
the demonstrative itself does not express singularity.
(15) a. ts
kiu
z
ala-ko.
CL dog be
1PL-MOD
The dog is ours.
b. ta
ts kiu z ala-ko.
that CL dog be 1PL-MOD
That dog is ours.
c. *ta
kiu z ala-ko.
that dog be 1PL-MOD
That dog is ours.
The facts in (14) and (15) tell us that although definite classifiers and
demonstratives are both able to express definiteness, they are differentiated
in number: the former are definite singulars, while the latter are preferred
to express definite plurals, though singular reference is also possible.
Second, in expressing definite singulars, a demonstrative always stands
on the left side of classifiers, not vice versa. Namely, Dem-Cl-N is the only
possible order and Cl-Dem-N is not grammatical. Compare (16) with (15b).
(16) * ts
CL

ta kiu z ala-ko.
that dog be 1PL-MOD

How can we account for these two facts? According to Giutsi (1999,
2002), a FP is licensed by (a) making the specifier visible and/or (b) making the head visible. The realization of a functional head is the last resort.
In our case, the D head, i.e. the classifier, has to be visible only when we
want to express singularity, as exhibited in (14-15). A possible arrangement of demonstrative and definite classifier is that demonstratives, as the
leftmost element in the DP domain, are located in the specifier position of

284

Definite classifiers and their modifiers

DP, as headed by definite classifiers.They are responsible for expressing


plural definites.

2.2.3. Adjs/RCs as [Spec DP]


I now move onto the discussion on adjectives, relative clauses and possessors before definite Cl+N. There are several differences between demonstratives and the other elements, such as adj/RC and possessor.
The first difference is concerned with the possibility of the modification
marker /k/. It is obligatory for Adj and RC to take a modification marker
/k/. It is possible for possessors to take /k/, though not obligatory. Demonstratives can never be followed by the modification marker /k/.
(17) a. i *(k)
ts
small Mod
CL
the small dog
b. in
san
tshl
new
born out
the newly born dog
c. a
(k)
ts
1PL
Mod CL
your dog
d. ta (*k) ts
that Mod CL dog
that dog

kiu
dog
*(k) ts
Mod CL

[Adjective]

kiu
dog

[Relative clause]

kiu
dog

[Possessor]

kiu

[Demonstrative]

The possibility of the presence of /k/ separates adjectives, relatives and


possessors from demonstratives. The former are modifiers in nature and the
demonstratives are not, if we take the position that the morpheme /k/ is a
modification marker which obligatorily follows modifiers when modifying
nouns (comparable to the modification marker de in Mandarin, cf. Chapter
8).
The second difference is about the ordering of demonstrative and these
modifiers. Occurrence of multiple preceding elements is possible, but it is
restricted by the ordering constraint that demonstratives must always stand
closest to [Cl-N] and the Adj/RC or possessors stand to the left side of DP.
In other words, we can only have the order of Modifier+Dem+Cl+N, as
in (18a) but not Dem+Modifier+Cl+N, as in (18b).

Semantics of non-bare Cl+N

(18) a. Adj/RC/Poss+Dem+Cl+N

k
ta
ts
black Mod that
CL
that dog with black fur
b.*Dem+Adj/RC/Poss+CL+N
*ta
k
ts
that
black Mod CL

285

kiu
dog

kiu
dog

If our claim is correct that RCs, adjectives and possessors are modifiers
and demonstratives are elements within DP, then the fact in (18) tells us
that those modifiers are DP modifiers.
To summarize, the relevant linguistic facts about non-bare Cl+N in
Wu that we have assembled so far show that (i) non-bare Cl+N is a definite expression, in which the classifier is a determiner head, and (ii) demonstratives are better to be treated as [Spec DP] and (iii) adjectives, relative clauses and possessors are DP modifiers.

3. Semantics of non-bare Cl+N


This section works on the compositional semantics of non-bare Cl+N. As
we saw earlier, nominal modifiers are left-branching in Wu Chinese. If my
claim is correct that elements like adjectives and relative clauses etc. are
DP modifiers, how can they be interpreted in a compositional way?

3.1. Definite classifiers characterized with familiarity


Before working out the compositional semantics, I repeat some of the arguments on definiteness expressed by definite classifier, as proposed in
Chapter 9 and in Li and Bisang (2012).
Lyons (1999) claims that the semantic-pragmatic notion of definiteness
exists universally, but the notion of definiteness may be represented differently, such as exclusiveness, uniqueness, identifiablity etc. According to
Li and Bisang (2012), definite classifiers in Wu Chinese are characterized
with the pragmatic notion of familiarity and not the semantic notion of
uniqueness.
First, uniqueness is not a necessary condition for the use of definite
classifiers in Wu. There is a certain group of nouns that are inherently definite crosslinguistically. They are called culturally unique entities in

286

Definite classifiers and their modifiers

Lbner (1985). They can be expressed either by bare nouns or Cl+N in


Wu, but Cl+N are used in a different context from bare nouns.
z
lan
ko.
(19) a. thin
sky
be
blue PRT
The sky is blue (in general).
b. kints ban
thin
man lan.
today CL
sky
very blue
Today, the sky is very blue.

[Generic]

[Episodic]

The bare noun thi sky in (19a) is used in a generic sentence, and it refers to the unique sky known to all of us. (19a) means that the sky is blue in
general. However, ban thi CL sky in (19b) refers to the sky in a particular situation, which is known to the interlocutors. In this specific example,
the temporal adverbial kints today helps to locate the entity in a very
particular spatial-temporal point.
Second, uniqueness is not a sufficient condition for the use of definite
classifiers. Definite Cl+N does not presuppose uniqueness or exclusiveness, but they imply familiarity or identifiability.
(20) a. ta
ts
kiu
that
CL
dog
that dog
b. in
santshl k
ts kiu
new
born
Mod CL dog
the newly born dog

[Demonstrative]

[Relative clause]

(20a) implies that there is a set of individual dogs, and the use of demonstrative ta that, accompanied with gesture, helps to identify the relevant entity among others in the context. By uttering (20b), the speaker
intends to refer to a particular dog, namely, the newly born one and not
others. This requires not that there be just one dog but that there be just one
dog that was newly born.
Another observation made by Li and Bisang (2012) is that bare Cl+N
is ambiguous between indefinite and definite readings, and that a definite
[Cl-N] is restricted to preverbal positions, which are argued to be (secondary) topic positions in Wu. (cf. the relevant arguments were presented in
chapter 9).
To combine these arguments, it is proposed that definite Cl+N in Wu
is characterized with familiarity in that the entity referred to by definite

Semantics of non-bare Cl+N

287

Cl+N function as the topic of the sentence and it presupposes the entity
referred to to be familiar to the interlocutors.

3.2. Interpret modified Cl-N compositionally


In English, relative clauses behave differently from other modifiers, such
as adjectives, in that they occur on the right side of the head noun. Theoretically speaking, there are two possible ways to analyze the constituent relation of the head noun and the relative clause, namely, [[the N]-RC] and
[the [N-RC]]. In the first case, the relative clause modifies the determiner
phrase and in the second, the relative clause modifies the noun and the
determiner scopes over the whole modified constituent.
According to Partee (1975), [the [N-RC]] in (21b) is preferred over [[the
N]-RC] in (21a) for the modified definite expression in English in terms of
compositionality. In other words, the uniqueness requirement of the definite article the is relative to the extension of N-RC and not to that of N.
The reasoning goes as follows: like adjectives, relative clauses in English
should be treated as predicate modifiers, <<e, t> <e, t>>, which applies to
predicates of type <e, t> and return entities that are of the same type.
(21) a.

NP
NP
Det

b.
CP

NP
Det

N
N

CP

Nevertheless, in our case of non-bare Cl+N in Wu, we claimed that


different modifiers before Cl+N are treated as DP modifiers. They are
parallel to the structure in (21a). The linear order of Modifier-Cl-N seems
to go contrary to Partees compositional semantics based on the structure
(21b), since those modifiers do not modify the noun but the whole definite
phrase Cl+N.
Moreover, the presence of the marker /k/ after Adj/RC and possessors
suggests that those elements preceding Cl+N are predicative in nature
(either being predicate or predicate modifiers). However, definite
Cl+N is analyzed as a generalized quantifier in our early study. Therefore, it is quite surprising that [Cl-N] as a GQ can be modified by these
predicate modifiers. There seems to be an obvious type mismatch.

288

Definite classifiers and their modifiers

The mismatching problem can be solved by taking into account the proposal offered by Bach and Cooper (1978). According to Bach and Cooper
(1978), it is possible to work out a compositional semantics for the structure (22a), i.e. [[the N]-RC]. They argue that noun phrases (DPs in our
term) can optionally take an extra property argument, which is saturated by
the denotation of a high-adjoined relative clause and intersected with the
property contributed by the content of the noun phrase. Its semantics is
sketched in (22).
(22)

NP2(=DP)
NP1(=DP)
Det

S(=CP)
who loves Mary

every
man
a. NNP1N= RP[(x)[man(x) R(x)] P(x) ]
b. NSN = z[love (z, m)]
c. NNP2N = P [(x)[man(x) love (x, m)] P(x) ]
It is clear from the semantics in (22) that a type-shifting is introduced:
the GQ is converted from type <<e, t>, t> to a function of type <<e, t>, <e,
t> t>.
As inspired by Bach and Coopers semantics in (22), we are now ready
to work out the compositional semantics of definite Cl+N and its modifiers.
In the discussion of bare Cl+N, we claimed that definite Cl+N is a
generalized quantifier and it is lifted from the predicative use of Cl+N.
This lifting operation (Partee 1987) may be seen as the corresponding syntactic operation of raising, namely, Cl-to-D raising. They also propose that
a contextual variable C is introduced in the denotation of definite Cl+N,
which represents contextual familiarity.
We, following Chierchia (1998) and Li (2011), assume that bare nouns
in classifier languages are kind denoting, as in (23).
(23) Denotation of bare nouns
NNN= k

Semantics of non-bare Cl+N

289

Indefinite [Cl-N] denotes a set of atomic instantiations of the relevant


kind, as represented by k (k is a variable over kinds). See the representation
in (24).
(24) Indefinite [Cl-N]
NCl-N N = kx. 1(x)(>k ?k)(1(x))2(x)=k
Definiteness is a feature in D, which shifts from the predicate type denoted by [Cl-N] to the GQ meaning denoted by the [Det-Cl-NP]. See (25).
(25) From indefinite [Cl-N] to definite [Cl-N]
NCl-NN= RP.x[P(x)R(x) FAMILIAR(x)]
(kx. 1(x)(>k ?k)(1(x))2(x)=k)
= Px[1(x)(>kk)(1(x))2(x)=k FAMILIAR (1(x))P(1 (x))]
In the case of modified Cl+N, the elements preceding Cl+N, such as
adjectives, RCs and possessors are DP modifiers. We suggest that they
express properties that intersect with the denotation of definite Cl+N and
saturate the contextual variable C. The information expressed by elements
preceding Cl+N is familiar to both the speaker and the hearer. Namely,
the information expressed those modifiers is part of the familiarity required
by definite classifiers in D0 position.
(26) i
k
small
Mod
the small dog

[ts
CL

kiu]
dog

The semantics of modified [Cl-N] is represented in (27):


(27) a. Ni k N= x. small (x)
b. Semantics of bare definite [Cl-N]
Nts-kiuN=Px[1(x)(>DOGk)UNITanimal(1(x))2(x)=k
FAMILIAR (1(x)) P(1 (x))]
c. Semantics of non-bare [Cl-N]
Ni-k ts kiu N= Nts kiuN (NikN)
Px[1(x)(>DOGk)UNITanimal(1(x))2(x)=k
FAMILIAR (1(x)) P(1 (x))] (x. small (x))
=x[1(x)(>DOGk)UNITanimal(1(x))2(x)=k
FAMILIAR (1(x)) small (1 (x))]

290

Definite classifiers and their modifiers

4. Concluding remarks
In this chapter, we investigated the syntactic and semantic properties of
non-bare classifier+noun in Wu Chinese. We showed that it is possible
for classifier languages, such as Wu Chinese, to have definite classifiers.
We argued that in non-bare Cl+N construction, classifiers are used as
definiteness markers and they are grammaticalized as quasi-definite articles.
We proposed that Wu Chinese not only has (quasi)definite articles, which
are able to instantiate the D head, but also it has a very refined DP structure, in which demonstratives are located at [Spec DP] and modifiers like
adjectives, possessors etc are DP modifiers.

References

Abney, Steve
1987
The English noun phrase in its sentential aspect. Ph.D. diss., Department of Linguistics, MIT.
Aikhenvald, Alexandra
2000
Classifiers: A Typology of Noun Categorization Device. Oxford:
Oxford University Press.
Alexiadou, Artemis, Liliane Haegeman, and Melita Stavrou
2007
Noun Phrase in the Generative Perspective. Berlin: Mouton de
Gruyter.
Allan, Keith
1977
Classifiers. Language 53: 285-311.
Aoun, Joseph, and Yen-Hui Audrey Li.
2003
Essays on the Representational and Derivational Nature of Grammar: The Diversity of Wh-Constructions. Mass: MIT Press.
Bach, Emmon
1986
The algebra of events. Linguistics and Philosophy 9: 5-16.
Bach, Emmon, and Robin Cooper
1978
The NP-S analysis of relative clauses and compositional semantics.
Linguistics and Philosophy 2 (1): 145-150.
Bisang, Walter
1993
Classifiers, quantifiers, and class nouns in Hmong. Studies in Language 17 (1): 1-51.
1999
Classifiers in East and Southeast Asian languages: counting and
beyond. In Numeral Types and Changes Worldwide. Gvozdanovicm, Jadranka (ed.), 113-185. Berlin/New York: Mouton de Gruyter.
Bolinger, Dwight
1967
Adjectives in English: Attribution and predication. Lingua 18: 1-34.
Borer, Hagit
2005
Structuring sense. Vol. 1: In Name Only. Oxford: Oxford University Press.
Borschev, Vladimir, and Barbara H. Partee
2004
Genitives, types, and sorts: The Russian genitive of measure. In
Possessives and Beyond: Semantics and Syntax (UMOP 29), J.-Y.
Kim, Y.A. Lander, and B.H. Partee (eds.), 29-43. Amherst, MA:
GLSA Publications.
Bokovi, Zeljko
2010
Phases in NPs and DPs. Ms. University of Connecticut.

292

References

Bunt, Harry
1985

Mass Terms and Model Theoretic Terms. Cambridge: Cambridge


University Press.
Carlson, N. Gregory
1977
Reference to Kinds in English. New York: Garland.
Chao, YuenRen
1968
A Grammar of Spoken Chinese. Berkeley: University of California
Press.
Chen, Matthew
2000
Tone Sandhi: Patterns across Chinese Dialects. Cambridge: Cambridge University Press.
Cheng, Lai-Shen Lisa
1991
On the typology of wh-questions. Ph.D. Diss., Mass.: MIT.
2009
On every type of quantificational expression in Chinese. In Quantification, Definiteness, and Nominalization, Monika Rathert, and
Anastasia Giannakidou (eds.), 53-75. Oxford: Oxford University
Press.
Cheng, Lai-Shen Lisa, and Rint Sybesma
1998
Yi-wangtang, yi-getang: classifiers and massifiers. Tsing-Hua
Journal of Chinese Studies. New Series XXVIII (3): 385-412.
1999
Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry 30(4): 509-542.
2005
Classifiers in four varieties of Chinese. In Handbook of Comparative Syntax, G. Cinque, and R.S. Kayne (eds.), 259-292. Oxford
University Press.
Cheng, Lai-Shen Lisa, Jenny Doetjes, and Rint Sybesma
2008.
How universal is the universal grinder. Linguistics in the Netherlands-2008: 50-62.
Cheung, Samuel
1972.
Xianggang yueyu yufa yanjiu [Research on Cantonese in Hong
Kong]. Hong Kong: Hong Kong Chinese University Press.
Chierchia, Gennaro
1998a
Reference to kinds across languages. Natural Language Semantics
6: 339-405.
1998b
Plurality of mass nouns and the notion of semantic parameter. In
Events and Grammar, Susan Rothstein (ed.), 53-103. Dordrecht:
Kluwer.
2010
Mass nouns, vagueness and semantic variation. Synthese 174: 99149
Christopherson, Paul
1939
The Articles: A Study of Their Theory and Use in English. London:
Oxford University Press.

References

293

Chung, Sandra
2000
On reference to kinds in Indonesian, Natural Language Semantics
8: 157-171.
Clark, Herbert
1977.
Bridging. In Thinking: Readings in Cognitive Science, P.N. Johnson-Laird and P.C. Wason (eds.), 411-420. Cambridge.
Corbett, Greville
2000
Number. Cambridge: Cambridge University Press.
Corver, Norbert
1998
Predicate movement in pseudo-partitive constructions. In Possessors, Predicates, and Movement in the Determiner Phrase, Artemis,
Alexiadou, and Chris Wilder (eds.), 215-257. Amsterdam: John
Benjamins.
Craig, Colette
1992
Classifiers in a functional perspective. In Layered Structure and
Reference in a Functional Perspective, M. Fortescue, P. Harder,
and L. Kristoffersen (eds.) 277-301. Amsterdam: John Benjamins.
Croft, Williams
1994
Semantic universals in classifier systems. Word 45:145-171.
Dayal, Veneeta
2004
Number Marking and (in)definiteness in kind terms. Linguistics
and Philosophy 27(4): 393-450.
Del Gobbo, Francesca.
1999
Nominal phrases in Mandarin and Cantonese. UCI Working Papers
in Linguistics 5:1132.
Den Dikken, Marcel and Pornsiri Singhapreecha
2004
Complex noun phrases and linkers. Syntax 7(1): 1-54.
Denny, J. P.
1976
What are noun classifiers good for? Chicago Linguistic Society 12:
122-132.
1979
The extendedness variable in classifier semantics: universal features and cultural variation. In Ethnolinguistics: Boss, Sapir, and
Whorf Revisited, M. Mathiot (ed.), 99-119. The Hague: Mouton.
Diseing, Molly
1992
Indefinites. Mass: the MIT Press.
Doetjes, Jenny
1997
Quantifiers and selection. Ph.D. Diss., Rijksuniversiteit Leiden.
Dolling, Johannas
1992
Flexible
Interpretationen
durch
Sortenverschiebung.
In
Fgungspotenzen, E. Zimmermann and A. Strigin (eds.), 23-62.
Berlin: Akademie Verlag.

294

References

Duanmu, San
1997
Phonologically motivated word order movement: evidence from
Chinese compounds. Studies in Linguistic Sciences 27(1): 49-73.
Enfield, Nick
2007
A Grammar of Lao, Berlin: Mouton de Gruyter.
Erbaugh, Mary S.
2002
Classifiers are for specification: complementary functions of sortal
and general classifiers in Cantonese and Mandarin. Cahiers de linguistique Asie Orientale 31(1): 33-69.
Fang, Mei
2002
zhishici zhe he na zai beijinghua zhong de yufahua [The grammaticalizaiton of demonstratives this and that in Beijing Mandarin].
Zhongguo Yuwen 4: 343-356.
Fodor, Janet, and Ivan Sag
1982
Referential and quantificational indefinites. Linguistics and Philosophy 5: 355398.
Gerstner-Link, Claudia, and Krifka Manfred
1993
Genericity. In Handbuch der Syntax, Joachim Jacobs, Arnimvon
Stechow, Wolfgang Sternefeld, and Theo Vennemann (eds.), 966978. Berlin: Walter de Gruyter.
Gerstner-Link, Claudia
1988
UberGenerizitat. Generische Nominalphrasen in Singularen
Aussagen und Generischen Aussagen. Ph.D. diss., University of
Munich
Gil, David
2005
Numeral Classifiers, In The World Atlas of Language Structures, M.
Haspelmath, M. Dryer, D. Gil and B. Comrie (eds.), 226-229. Oxford: Oxford University Press.
Gillon, Brendan
1992
Towards a common semantics for English count and mass nouns.
Linguistics and Philosophy 15(6): 597-640.
Giutsi, Giuliana
1999
The functional structure of noun phrases: a bare phrase structureapproach. Working papers in Linguistics. Vol. 9, n.1-2: 105-160.
2002
The functional structure of noun phrases: A bare phrase structure
approach. In Functional Structure in DP and IP, Guglielmo Cinque
(ed.), Vol. 1: 54-90. Oxford: Oxford University Press.
Greenberg, H. Joseph
1972
Numeral classifiers and substantival number: Problems in the genesis of a linguistic type. Working Papers on Language Universals 9:
1-39.
1974
Language typology: a historical and analytic overview. (Janua Linguarum, Series Minor 184.) The Hague: Mouton.

References

295

Greenberg, Yael
1998
An overt syntactic marker for genericity in Hebrew. In Events and
Grammar, S. Rothstein (ed.), 125-144. Kluwer Academic Publishers.
Hashimoto, M. J.
1977
The genealogy and the role of the classifier in Sino-Tibetan. Computational Analyses of Asian and African Languages 7: 69-78.
Hashimoto, Yue
1993
Comparative Chinese Dialectal GrammarHandbook for Investigators. Paris: Ecole des Hautes Etudes en Sciences Sociales.
Hawkins, J. A.
1978
Definiteness and Indefiniteness: A Study in Reference and Grammaticality Prediction. London: Croom Helm.
Heim, Irene
1982
The semantics of definite and indefinite noun phrases. Ph.D. diss.,
University of Massachusetts.
Her, One-soon, and Chen-Toen Hsieh
2010.
On the semantic distinction between classifeirs and measure words
in Chinese. Language and Linguistics 11(3): 527-551.
von Heusinger, Klaus
2011
Specificity, referentiality and discourse prominence: German indefinite demonstratives. In Proceedings of Sinn and Bedeutung 15,
Reich, Ingo et al. (eds.), 9-30. Universaar-Saarland University
Press: Saarbrcken, Germany.
Higgins, Roger
1973
The pseudocleft construction in English. Ph.D. diss., MIT.
Hoeskstra, J.
1988
-fol, fol en de sintaksis fan mjitteoantsjuttingen. In Wurdfoarried
and Wurdgrammatika. S. Dyk, and G. de Haan (eds.), 74-114.
FryskeAkademy, Leeuwarden/Ljouwert.
Hsieh, Miao-Ling
2007
Whats number got to do with it? Paper presented at The 5th Conference of the European Association of Chinese Linguistics. MPI,
September 4-7. Leipzig. Germany.
2008
The Internal Structure of Noun Phrases in Chinese. Books Series in
Chinese Linguistics. No.2. Taipei: Crane Publishers.
Hu, Jianhua, Haihua Pan and Peppina Lee
2003
Ningbohua yu Putonghua zhong huati he cihuati de jufa weizhi
[The Syntactic Positions of Topic and Subtopic in Ningbo Dialect
and Mandarin Chinese], In Huati Xinlun. [New Ideas about Topic
and Focus], Xu Liejiong and Danqing Liu (eds.), 164-175. Shanghai: Shanghai Education Press.

296

References

Huang, C.-T. James


1982
Logical relations in Chinese and the theory of grammar. Ph.D. diss.,
MIT.
1987
Existential sentences in Chinese and (in)definiteness. In The Representation of (In)Definiteness, Reuland Eric and Alice ter Meulen
(eds.), 226-253. Cambridge, MA: the MIT Press.
Huang, Aijun, and Thomas Lee
2009
Quantification and Individuation in the Acquisition of Chinese Classifiers. In Proceedings of the 10th Tokyo Conference on Psycholinguistics, Otsu, Y. (ed.), 117-141. Japan: Keio University.
Huang Shuangfan
1981
On the scope phenomenon of Chinese quantifiers. Journal of Chinese linguistics 9: 226-243.
Iljic, Robert
1994
Quantification in Mandarin Chinese: Two markers of plurality. Linguistics 32: 91-116.
Jackendoff, Ray
1977
X-bar Syntax: A Study of Phrase Structure. Cambridge, MA: MIT
Press.
Jenks, Peter
2011
The hidden structure of Thai noun phrases. Ph.D. diss., Harvard
University.
Jiang, Li
2008
Monotonicity and measure phrases in Chinese. Paper presented in
IsCLL-11, Taiwan.
2012
Nominal arguments and language variation. Ph.D. diss., Harvard
University.
Joosten, Frank
2003
Accounts of the count-Mass distinction: A Critical Survey. Nordlyd
31(1): 216-229, University of Troms, Troms, Norway.
Kadmon, Nirit
1987
On unique and non-unique reference and asymmetric quantification.
Ph.D. diss., University of Massachusetts at Amherst.
Koptjevskaja Tamm, Maria
2001
A piece of the cake and a cup of tea: Partitive and pseudopartitive nominal constructions in the Circum-Baltic Languages. In
The Circum-Baltic Languages: Typology and Contact 2, Dahl.
and M. Koptjevskaja-Tamm (eds.), 523-568. Amsterdam: John
Benjamins.
Kratzer, Angelika
1995
Stage-level/individual-level predicates. In The Generic Book, G.N.
Carlson and F.J. Pelletier (eds.), 125-175. University of Chicago
Press.

References

297

Krifka, Manfred
1989
Nominal reference, temporal constitution and quantification in
event semantics. In Semantics and Contextual Expressions. Renate
Bartsch, John van Benthem, and Peter van Emde Boas, (eds.), 75111. Dordrecht: Foris.
1995
Common nouns: A contrastive analysis of English and Chinese. In
The Generic Book, Greg. N. Carlson and Jeff. F. Pelletier, (eds.),
398-411. University of Chicago Press.
2002
Be brief and vague! And how bidirectional optimality theory allows
for Verbosity and Precision. In Sounds and Systems. Studies in
Structure and Change. A Festschrift for Theo Vennemann, David
Restle and DietmarZaefferer (eds.), 439-458. Berlin/New York:
Mouton de Gruyter.
2008
Approximate interpretations of number words: A case for strategic
communication. In Theory and Evidence in Semantics, E. Hinrichs,
and J. Nerbonne (eds.), 109-132. Stanford: CSLI Publications.
Krifka, Manfred, Francis Jeffry Pelletier, Gregory N. Carlson, Alice ter Meulen,
Godehard Link, and Gennaro Chierchia.
1995
Genericity: An introduction. In The Generic Book, Gregory N.
Carlson.and Francis J. Pelletier (eds.), Chicago: University of Chicago Press.
Kripke, Saul
1982
Naming and Necessity. Oxford: Blackwell.
Ladusaw, William A.
1982
Semantic constraints on the English partitive construction. Proceedings of WCCFL 1, 231-242.
Lakoff, George, and Mark Johnson
1980
Metaphors We Live By. Chicago: University of Chicago Press.
Lambrecht, Knud
1994
Information Structure and Sentence Form: Topic, Focus, and the
Mental Representation of Discourse Referents. Cambridge: Cambridge University Press.
Landman, Fred
1989
Groups I/II. Linguistics and Philosophy 12(5)-(6).
2004
Indefinites and the Type of Sets. Oxford: Blackwell.
Landman, Fred, and Susan Rothstein
2009
Incremental homogeneity in the semantics of aspectual for-phrases.
In Syntax, Lexical Semantics and Event Structure, Rapapport Hovav,
Malka, IviSichel, and Edit, Doron (eds.). Amsterdam: John Benjamins.
Lee, Hun-Tek Thomas
1986
Studies on quantification in Chinese, Ph.D. Diss., UCLA.

298

References

Lehrer, Adrienne
1986
English classifier constructions. Lingua 68: 109-148.
Li, N. Charles, and Sandra A. Thompson
1976
Subject and topic: A new typology of language. In Subject and Topic, Charles, N. Li (ed.), 458-489. New York: Academic Press.
1981
Mandarin Chinese: A Functional Reference Grammar. Berkeley:
University of California Press.
Li, Frances
1971
Case and communicative function in the use of ba in Mandarin.
Ph.D. diss., Cornell University.
Li, Linding
1985
Zhuyu de yufadiwei [The grammatical status of subject]. Zhongguoyuwen 1: 62-70.
Li, XuPing
2007
The big/small issue of mass/count classifiers. Paper presented at the
fifth annual conference of EACL (European Association of Chinese
Linguistics). Leipzig, 2007.
Li, XuPing, and Walter Bisang
2012
Classifiers in Sinitic languages: From individuation to definitenessmarking. Lingua 122, 335-355.
Li, XuPing, and Susan Rothstein
2010
Two types of measure readings in Chinese classifier phrases. Paper
presented at the 12th International Symposium on Chinese languages and linguistics. June, Taipei, 2010.
2012
Measure readings of Mandarin classifier phrases and the particle de,
Language and Linguistics, 13(4): 693-741.
Li, Audrey Yen-Hui
1990
Order and constituency in Mandarin Chinese. Dordrecht: Kluwer.
1998
Argument determiner and number phrases. Linguistic Inquiry 29:
693-702.
1999
Plurality in a classifier language. Journal of East Asian Linguistics
8: 75-99.
Lima, Suzi
2012
Numerals and the universal packager in Yudja. Ms. University of
Mass.
Lin, Jo-wang
1996
Polarity licensing and wh-phrase quantification in Chinese. Ph.D.
diss., University of Massachusetts.
1997
Chinese Noun Phrase Structure: DP or NP?, In Chinese Language
and Linguistics III: Morphology And Lexicon, Feng-fu Tsao and H.
Samuel Wang (eds.) 401-434. Academia Sinica, Taipei.

References

299

Link, Godehard
1983
The logical analysis of plurals and mass terms: A lattice-theoretical
approach. In Meaning, Use and the Interpretation of Language, R.
Buerle, C. Schwarze, and A. von Stechow (eds.), 303-323. Berlin/New York: Walter de Gruyter.
Lipton, James
1968
An Exaltation of Larks. New York: Grossman.
Liu, Chen-Sheng
2010
Dimension-denoting classifiers in Taiwanese compound adjectives.
Journal of East Asian Linguistics 19(2): 181-205.
Liu, Danqing
2001
Wuyu de jufa leixing tedian [Two major typological features of Wu
dialects]. Fangyan (4).
Lobeck, Anne
1995
Ellipsis: Functional Heads, Licensing, and Identification. New
York: Oxford University Press.
Lbel, Elisabeth
1989
Q as a functional category. In Syntactic Phrase Structure Phenomena, C. Bhatt, E. Lbel, and C. Schmidt (eds.), 133-158. Amsterdam/Philadelphia: John Benjamins.
1996
Function and Use of Vietnamese Classifiers. In Lexical Structure
and Language Use. Proceedings of the International Conference
on Lexicology and Lexical Semantics Mnster Vol 2, E. Weigandand F. Hundsnurscher (eds.), 293-304. Tbingen: Niemeyer.
2000
Classifiers and semi-lexicality: functional and semantic selection. In
Theorie des Lexikons, N. Corverand H. van Riemsdijk (eds.), 3182. Erschienen.
Lbner, Sebastien
1985
Definites. Journal of Semantics 4: 279-326.
Longobardi, Giuseppe
1994
Reference and proper names. Linguistic Inquiry 25(4): 609-665.
Lu, Jianming
1987
Shuliangci zhongjian charu xiongrongci qingkuang kaocha [A survey of the insertion of adjectives between numeral and classifier].
Yuyan Jiaoxue yu Yanjiu 4: 53-72.
Luo, Yuanlin
1988
Guanyu shuliangci zhongjian charu xiongrongci qingkong de buchong kaocha [A supplement study of the insertion of adjectives
before classifiers]. Hanyu Xuexi 1988(4): 7-12.
L, Shuxiang
1942
Zhongguo wenfa yaolue [Essense of Chinese Grammar]. Beijing:
Commercial Press.

300

References

1990

1980

Reprint. Ge zi de yingyong fanwei, fulun danweici qian yi de tuoluo


[On the usages of ge and the dropping of the number one before
classifiers]. In L Shuxiang wenji [Colletctions of L Shuxiang],
Vol 2: 139-168. Original version: 1944.
Xiandai hanyu ba bai ci [Eight hundred words in Modern Chinese].
Beijing: Commercial Press.

Lyons, John
1976
Semantics. Vol. 1. Cambridge: Cambridge University Press.
Lyons, Christopher
1999
Definiteness. Oxford: Cambridge University Press.
Matthews, Stephen, and Virginia Yip
1994
Cantonese: A Comprehensive Grammar. London: Routledge.
Matthewson, Lisa
2001
Quantification and the nature of cross-linguistic variation. Natural
Language Semantics 9:145-189.
Mller, Ana, and Oliveira Fatima
2004
Bare nominals and number in Brazilian and European Portuguese.
Journal of Portuguese Linguistics 3:11-36.
Muromatsu, Keiko
1998
On the syntax of classifiers. Ph.D. diss., University of Maryland.
de Oliveira, Roberta, and Rothstein Susan
2011
Bare singular nouns in Brazilian Portuguese. Lingua 121(15):
2153-2175.
Pan, Wuyun, and Tao Huan
1999
Wuyu de zhidaici [Pronouns in Wu]. In Daici [Pronouns], Li Rulong and Zhang Shuangqing (eds.), 25-67. Guangzhou: Jinan
University Press.
Paris, Marie-Claude
1981
Problmes de Syntaxe et de Smantique en Linguistique Chinose.
Paris: Collge de France.
Partee, Barbara
1986
Ambiguous pseudo-clefts with unambiguous be, In Proceedings
of NELS l6, S. Berman, J.-W.Choe, J. McDonough (eds.), 354-366.
1987
Noun phrase interpretation and type-shifting principles. In Studies
in Discourse Representation Theory and the Theory of Generalized
Quantifiers, J. Groenendijk, D. de Jongh, and M. Stokhof (eds.),
115-143. Foris, Dordrecht.
1998
Copula inversion puzzles in English and Russian. In Formal Approaches to Slavic Linguistics: The Seattle Meeting 1998, Katarzyna Dziwirek, Herbert Coats, and Cynthia Vakareliyska (eds.), 361395. Ann Arbor: Michigan Slavic Publications.

References

301

Partee, Barbara, and Vladimir Borschev


2012
Sortal, relational, and functional interpretations of nouns and Russian container constructions. Journal of Semantics 29 (4):445-486.
Paul, Waltraud
2005
Adjectival modification in Mandarin Chinese and related issues.
Linguistics 43(3): 757-793.
2007
The insubordinate subordinator DE in Mandarin Chinese. Ms.
EHESS.
2010
Adjectives in Mandarin Chinese: The rehabilitation of a much ostracized category. In Adjectives: Formal Analysis in Syntax and
Semantics, Cabredo-Hofherr P. and Matushansky O. (eds.), 115154. Amsterdam: Benjamins.
Potts, Chris
2005
The Logic of Conventional Implicatures. Oxford: Oxford University Press.
2007
The expressive dimension. Theoretical Linguistics 33(2): 165198.
Pustejovsky, James
1993
Type coercion and lexical selection. In Semantics and the Lexicon,
J. Pustejovsky (ed.), 73-94. Dordrecht: Kluwer Academic Publishers.
Reinhart, Tanya
1987
Specifier and operator binding. In The Representation of
(In)definiteness, Eric J. Reuland and Alice G. B. ter Meulen (eds.),
226-253. Cambridge, Mass.: MIT Press.
van Riemsdijk, Henk
1998
Categorical feature magnetism: the endocentricity and the distribution of projections. Journal of Comparative Germanic Linguistics 2:
1-48.
Roberts, Craige
2003
Uniqueness in definite noun phrases. Linguistics and Philosophy 26:
287-350.
Rothstein, Susan
2001
Predicates and their Subjects. Kluwer: Dordrecht.
2009
Individuating and measure readings of classifier constructions: evidence from Modern Hebrew. Brills Annual of Afroasiatic Languages and Linguistics 1: 106-146.
2010
Counting and the mass-count distinction. Journal of Semantics
27(3): 343-397.
Rullmann, Hotze, and Aili You
2006
General number and the semantics and pragmatics of indefinite bare
nouns in Mandarin Chinese. In Where Semantics Meets Pragmatics.
Klaus, von Heusinger, and Ken, P. Turner (eds.), 175-196. Amsterdam: Elsevier.

302

References

Saito, Mamoru, T.-H. Jonah Lin, and Keiko Murasugi


2008
N'-ellipsis and the structure of noun phrases in Chinese and Japanese. Journal of East Asian Linguistics 17(3): 247-271.
Schlenker, Philippe
2007
Expressive presuppositions. Theoretical Linguistics 33(2): 237245.
Schwarzschild, Roger
2006
The role of dimensions in the syntax of noun phrases. Syntax 9(1):
67-110.
2009
Stubborn distributivity, multiparticipantnouns and the count/mass
distinction, In Proceedings of NELS 39, Suzi Lima, Kevin Mullin,
Brian Smith (eds.).
Selkirk, Elisabeth
1977
Some remarks on noun phrase structure. In Formal Syntax, P. Culicover, T. Wasow and A. Akmajian (eds.), 285-316. New York:
Academic Press.
Sharvy, Richard
1980
A more general theory of definite descriptions. The Philosophical
Review 89: 607624.
Shen, Yang
2013
The quantifying role of adverbs in Chinese. Paper presented at the
International Symposium on the expression and acquisition of
quantification in the Chinese language. May 2013, INALCO, Paris.
Shi Dingxu
1992
The nature of topic comment constructions and topic chains. Ph.D.
diss. Los Angeles: University of South California.
Shi, Rujie, and Danqing Liu
1985
Suzhou fangyan liangci de dingzhi yongfa ji ji biandiao [On the definite use of classifiers and its tone sandhi in the Suzhou dialect.]
Yuyan yanjiu 1985(1).
Shi, Yuzhi
2002
Lun hanyu de jiegou yiyi he cihui biaoji zhi guanxi [The Interaction
between Structural Meanings and Lexical Markers]. Dangdai
Yuyanxue 1: 25-37.
Simpson, Andrew
2005
Classifiers and DP structure in Southeast Asian languages. In
Handbook of Comparative Syntax, Cinque, G, and R. S. Kayne
(eds.), 806-838. Oxford University Press.
Simpson, Andrew, Hooi Ling Soh, and Hiroki Nomoto.
2011
Bare classifiers and definiteness: A cross-linguistic investigation.
Studies in Language 35(1): 168-193.

References

303

Soh, Hooi Ling, and Jenny Kuo


2005
Perfective aspect and accomplishment situations in Mandarin Chinese. In Perspectives on Aspect, H. J. Verkuyl, H. de Swart and A.
van Hout (eds.), 199-216. Springer.
Sproat, Richard, and Shih Chilin
1991
The cross-linguistic distribution of adjective ordering restrictions.
In Interdisciplinary Approaches to Language.Essays in honor of
S.-Y. Kuroda, Georgopoulos, C. and Ishihara, R. (eds.), 565-592.
Dordrecht: Kluwer Academic Publishers.
Stavrou, Melita
2003
Semi-lexical nouns, classifiers, and the interpretation(s) of the
pseudo-partitive construction. In From NP to DP, Coene, M. and Y.
DHulst (eds.), Vol. 1. 329-354. John Benjamins Publishing Company.
Sun, Chaofen, and Talmy Givon
1985
On the so-called SOV word order in Chinese: A quantified text
study and its implications. Language 61(2): 329-351.
Tai, H.-Y. James, and Lianqing Wang
1990
A semantic study of the classifier tiao. Journal of the Chinese Language Teachers Association 25: 3556.
Tang, Jane Chih-Chen
1990
Chinese phrase structure and the extended X-theory. Ph.D. diss.,
Cornell University.
1996
Tamai-le bi shi-zhi and Chinese phrase structure. BIHP 66.
2005
Nouns or classifiers: A non-movement analysis of classifiers in
Chinese. Language and Linguistics 6 (3): 431-472.
Tang, Sze-Wing
2006
Hanyu shoushi huatiju leixing de canshu fenxi [A parametric approach to the typology of subtopics in Chinese dialects]. Yuyan
kexue 5(6): 3-11.
Tao, Liang
2006
Classifier loss and frozen tone in spoken Beijing Mandarin: the +
phono-syntactic conspiracy. Linguistics 44 (1): 91-133.
Teng, Shou-Hsin
1974
Double nominatives in Chinese. Language 50 (3): 455-473.
Tsai, Dylan Wei-Tien
1994
On economizing the theory of A-Bar dependencies. Ph.D. diss.,
MIT.
Tsou, Benjamin
1976
The structure of nominal classifier systems. In Austoasiastic Studies,
vol. 2. Phillip N. Jenner, Stanley Starosta, and Laurence C. Thompson (eds.), 1215-1247. Honolulu, Hawaiii: University of Hawaii
Press.

304

References

Vos, Margaretha
1999
A grammar of partitive constructions. Ph.D. diss., Tilburg University.
Wang, Jian
2005
Suwan Quyu Fanyan Yanjiu Bijiao Yanjiu [A comparative study of
dialects spoken in Suzhou and Anhui]. Ph.D. diss., Department of
Chinese, Beijing University.
Wang, Jian, and Jinsong Gu
2006
Lianshui (Nanlu) hua liangci de teshu yongfa [On special uses of
classifiers in Lianshui (Nanlu) dialect]. ZhongguoYuwen 3: 237241.
Wang, Li
1958
Hanyu shigao [A history of Chinese]. Beijing: Kexue Chubanshe.
1943
Zhongguo xiandai yufa [Grammar on Modern Chinese]. Beijing:
Commercial Presss.
Watanabe, Akira
2006
Functional projections of nominals in Japanese: Syntax of classifiers. Natural Language and Linguistic Theory 24: 241-306.
Wierzbicka, Anna
1988
The Semantics of Grammar. Amsterdam/Philadelphia: John Benjamins Publishing Company.
Wisniewski, Edward
2010
On using count nouns, mass nouns, and pluralia tantum: what
counts? In Kinds, Things and Stuff: Mass Terms and Generics.
Francis Jeffry Pelletier (ed.), 166-190. Oxford University Press.
Wiggins, David
1965
Identity statements. In Analytic Philosophy, R. J. Butler (eds.), Oxford: Blackwell.
Wilhelm, Andrea
2008
Bare nouns and number in Dne Sulin. Natural Language Semantics 16: 39-68,
Wilkinson, Karina
1991
Studies in the semantics of generic noun phrases. Ph.D.diss., University of MassachusettsAmherst.
Wiltschko, Matina
2012
Decomposing the mass/count distinction: Evidence from languages
that lack it. In Count and Mass across Languages, Massam, Diane
(ed.), Oxford: Oxford University Press.
Williams, Edwin
1983
Semantic vs. syntactic categories. Linguistics and Philosophy 6:
423-446.

References
Wu, Fuxiang
2003

305

Hanyu bansuijieci yufahua de leixingxue yanjiu [A typological


study of WITH-preposition in Chinese]. Zhongguo Yuwen 1.
Wu, Yicheng, and Adam Bodomo
2009
Classifiers Determiners. Linguistic Inquiry 40(3): 487-503.
Xu, Liejiong
1995
Definiteness effect on Chinese word order. Cahiers de linguistics
Asie Orientale, 24(1):29-48.
2004
Manifestation of informational focus. Lingua 114: 277-299.
Xu, Liejiong, and Danqing Liu
1998
Huati de jiegou yu gongneng [The structure and functions of topics]. Shanghai: Shanghai Education Press.
Yang, Henrietta
2005
The structure of the DP in Chinese.Ph.D. diss., University of Texas
at Austin.
Yang, Rong
2001
Common nouns, classifiers, and quantification in Chinese. Ph.D.
diss., Rutgers University.
Yip, Chak-Lam
2008
Complicating the oversimplification: Chinese numeral classifiers
and true measures. In Proceedings of the 20th North American
Conference on Chinese Linguistics (NACCL-20). Marjorie K.M.
Chan and Hana Kang (eds.), Vol 1: 285-295. Columbus, Ohio: The
Ohio State University.
Yuan,Yulin
1996
Huatihua ji qi xiangguan guocheng [Topicalization and its relevant
procedures]. Zhongguo Yuwen 4.
Zhang, Niina Ning
2006
Representing specificity by the internal order of indefinites. Linguistics 44 (1): 1-21.
2012
Numeral classifier structures. Manuscript. National Chung Cheng
University.
Zhou, Peng, and Liqun Gao
2009
Scope processing in Chinese. Journal of Psycholinguistic Research
38: 11-24.
Zhu, Dexi
1961
Shuo de [On de]. ZhongguoYuwen 12: 1-15.
1982
Yufa Jiangyi [Lecture Notes on Grammar]. Beijing: Commercial
Press.
1984
Yufa, Xiuchi, Zuowen [Grammar, rhetoric and writing]. Shanghai:
Shanghai Education Publishing House.

306

References

Zimmermann, Ede
1993
On the proper treatment of opacity in certain verbs. Natural Language Semantics 1: 149 - 179.
Zong, Shouyun
2009
Shu+Xing+Liang geshi de fenghua ji qi yuyi yuyong chayi [On the
divergence of Num+Cl+N and its semantic and pragmatic differences]. Xiuci Xuexi 3: 37-42.

Index

accommodation, 268
additive, 137, 154
adjective, 37, 174
adnominal adjective, 174
dimensional adjective, 38, 60, 83,
182, 195
left-peripheral adjective, 174
pre-classifier adjective, 37, 49,
57, 174
animacy scale, 70
appositive, 92
approximation, 212, 226
argumental language, 86, 88
atomization, 161
attributive reading, 62, 204
ba construction, 119, 239
bare argument, 44, 88
bare argument language, 69, 71
bare classifier construction, 154, 235
bare noun, 2, 44, 52, 94, 101, 162,
237
bare plural, 94
Boolean algebra, 101, 164
Cantonese, 5, 234, 243
classifier language, 16
classifier reduplication, 139, 142
coercion, 159
concrete portion, 179, 195
container classifier, 18, 22, 25, 129,
135, 170
container classifiers, 149
conventional classifier, 23
copula clause, 104, 106, 108, 112,

114
count classifier, 45, 127
count mass noun, 68, 73
count noun, 42, 46, 102, 166
count structure, 73
countability, 2, 69, 71
counting, 2, 3, 130, 134, 164, 167,
179, 221, 223
count-to-mass shifting, 78
definiteness, 4, 93, 97, 266
definite bare noun, 116, 121
definite classifier, 136, 234, 240,
278
definite numeral, 258, 261
definite reading, 95, 97, 99
DP structure, 3, 141
demonstrative, 4, 90, 282
Dne Sulin, 71
distributivity, 132
dotted type, 130
estimation, 222, 224
existential closure, 265, 266
existential quantifier, 238
existential reading, 94
expressive, 195
familiarity, 4, 97, 121, 269, 288
weak familiarity, 4, 269
feature analysis, 143
focus, 246
fraction, 65, 215
functional, 20, 33, 43, 133

308

Index

general classifier, 74, 76


generic sentence, 89
group classifier, 17, 25, 74, 76, 149
Hebrew, 130
high countability, 13
high round number, 225
Hmong, 233
identity sentence, 105, 115
indefinite phrase, 94, 95, 98
indeterminate quantity, 212
individual classifier, 15, 17, 21, 144
individuation, 2, 69, 71, 79, 80, 81,
146
information structure, 246
information weight, 211, 212
iota operator, 266, 267
jeung construction, 245
kind, 86, 87, 102
subkind, 91, 112, 114
kind classifier, 19, 92, 150
kind-level predicate, 90, 91
Lao, 15
lexical, 43, 133
Lexical Integrity, 194
mass classifier, 46, 127
mass mass noun, 68, 73
mass noun, 42, 44, 46, 102, 166
measure, 2, 3, 130, 134, 163, 171,
218
measure phrase, 128
mensural
see non-individual classifier
Minangkabau, 16

minimal parts, 68, 73, 75


natural atomicity, 84, 163
nominal properties, 25, 26
nominalizer, 36
non-individual classifier, 15, 34, 35
NP-ellipsis, 40
number agreement, 132
number marking, 35, 71, 72, 75
general number, 45
numeral classifier, 15
object-level reading, 86, 87
ontology, 2, 69, 70, 80, 82
discrete individuals, 46
homogeneous entity, 47
opacity, 93
opaque reading. see opacity
partial-object, 79
particle de, 50, 61, 62, 138, 202
partition classifier, 18, 25, 149
pragmatic shifting, 226
pre-classifier adjective, 175
predicational sentence, 104, 113
predicative, 103
proper name, 260
pseudo-partitive construction, 34, 56,
133, 181
pseudo-partitive reading, 62, 129,
133, 204
relational noun, 40
relative clause, 95, 287
relativization, 52, 53
root noun, 128, 164
round number, 65, 215
Russian, 130

Index

309

scope, 98
intermediate reading, 100
narrow scope, 97, 99
wide scope, 98
secondary topic, 247
semantic shifting, 134, 158
signature property, 42
sortal classifier
see.individual classifier
specificity, 96, 249, 252, 253
standard measure, 19, 147
stubbornly distributive predicate,
140
stuff, 79
syntactic re-analysis, 223

Thai, 14
topic, 118, 119, 241, 246
secondary topic, 119
topic-prominent language, 246
transparent phenomenon, 38, 56, 181
transparent reading, 93
type shifting, 94, 106, 111

taxonomy, 91
temporary classifier, 19, 147

Yudja, 71

uniqueness, 4, 97, 267


universal grinding, 68, 78
Vietnamese, 233
word order, 246
Wu Chinese, 5, 234, 240, 242, 277