Language Sciences 25 (2003) 353–373

www.elsevier.com/locate/langsci

Individuals, kinds and events:

classifier coercion of nouns
Chu-Ren Huanga,*, Kathleen Ahrensb
a
Institute of Linguistics, Academia Sinica, No. 128, Sec. 2, Yanjiouyuan Road,
Nangang, Taipei, Taiwan
b
Graduate Institute of Linguistics, National Taiwan University, 1 Roosevelt Road, Section 4,
Taipei, Taiwan

Accepted 14 December 2001

Abstract
This paper challenges the traditional view that nominal classifiers classify individuals.
Instead, we suggest that classifiers coerce nouns to refer to kinds and events as well as to
individuals. This finding argues against the view that nouns refer only to entities, and suggests
that classifiers do not simply agree with a noun, but instead coerce a particular meaning from
it. Moreover, the Mandarin classifier system creates a taxonomic system involving events,
kinds and individuals respectively. Within each classifier type an independent classification
system of the collocating noun type is created. These findings are important first because they
emphasize that the understanding of the semantics of nouns involves more than simple refer-
ence to an individual entity. Second, it is the first time that the previously abstract semantic
distinctions among kinds, individuals and events, as well as within kinds and within events,
have been found to be instantiated in a particular system of a natural language grammar,
namely, the classifier system. # 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Chinese; Classifier; Coercion

1. Introduction

The notions of individuals, kinds and events are three useful, albeit abstract,
semantic distinctions found in formal semantic theories. These concepts usually
involve separate domains of inquiry within formal semantics, but Lyons (1995),

* Corresponding author. Fax: +886-2-2785-6622.

E-mail address: [email protected] (C.-R. Huang).

0388-0001/03/$ - see front matter # 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0388-0001(02)00021-9
354 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

summarizes the prevailing viewpoint concerning these concepts in the two passages
below.

Throughout this book I have adopted the viewpoint of naı̈ve realism, according
to which the ontological structure of the world is objectively independent both
of perception and cognition and also of language. . ..According to the viewpoint
adopted here, the world contains a number of first-order entities (with first-
order properties) which fall into certain ontological categories (or natural
kinds); it also contains aggregates of stuff or matter (with first-order proper-
ties), portions of which can be individuated, quantified, enumerated—and thus
treated linguistically as entities—by using the lexical and grammatical resources
of particular natural languages. All natural languages, it may be assumed,
provide their users with the means of referring to first-order entities and
expressing propositions which describe them in terms of their (first-order)
properties, actual or ascribed, essential or contingent: such languages have the
expressive power of first-order formal languages (p. 325).

In the above passage, Lyons argues for the existence of kinds and individuals, as
well groups of individuals that can be apportioned. These concepts are available to
all human languages, and all languages that have these concepts are considered to be
first-order formal languages.
Lyons is less certain about the concept of event being present in all natural lan-
guages, as the following passage shows:

Whether all natural languages have the greater expressive power of various
kinds of higher-order formal languages is a more controversial, and as yet
empirically unresolved, question. But some natural languages certainly do; and
[such languages enable] users to reify, or hypostatize, the properties of first-
order entities, the relations that obtain among them, and the processes, activ-
ities, and states of affairs (and other kinds of situations) in which they are
involved (p. 325).

Higher-order languages, then, allow users to understand abstract concepts as con-

crete (i.e. through metaphorization). Higher-order languages can also describe the
events in which first-order entities are involved. Lyons argues that English is a
higher-order language.
It is interesting to note that Lyons rejects the notion that there is an underlying
linguistic or cognitive motivation for the concepts of individuals, kinds and events
(i.e. ‘the ontological structure of the world is objectively independent both of per-
ception and cognition and also of language’). Moreover, these three concepts are
treated as distinct by formal semanticists, and not as part of a unified system. But we
will show in this paper that these concepts are part of a coherent and conceptually
motivated grammatical system. We will show that individuals, kinds, and events are
reflected in the classifier system of Mandarin Chinese, a linguistic system based on
cognitive principles.
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 355

This finding is relevant for the following reasons: first, it challenges the traditional
treatment of individual as one of the basic semantic concepts. In other words, when
individuals, kinds and events share the same level of linguistic/cognitive taxonomy, a
treatment that puts one of them (individuals) as basic and the others derived needs
to be re-examined. Second, this classification allows a fresh look at the complex
semantic concepts of nouns, and at the interaction and coercion that takes place
between classifiers and nouns. Third, it explains why the kind reading in Mandarin
Chinese occurs with a subset of classifiers in addition to its function as a generic
reading typical of bare nominals.1 Lastly, for each noun type (i.e. individual, kind
and event nouns) the corresponding classifier type creates a taxonomic system. The
fact that the more abstract noun types (i.e. events and kinds) can be classified by a
grammaticalized system offers the strongest empirical support for their conceptual
felicity, since it is the first time that previously abstract semantic distinctions
between kinds, individuals and events is found to be instantiated as a particular
system of a natural language grammar. It is important to note that it is well known
that linguistic devices can be used to mark the distinctions of individual, kind, and
event. However, there are two crucial differences between these cases and the
Chinese classifier system that we are discussing now. First, previous cases all
involve grammatical devices that are independently motivated and are taken from
separate grammatical levels or systems, while the classifier system we discuss here is
a single integral grammatical system. Second, previous cases involve representa-
tional clues that typically, but not always, mark the semantic distinction, while the
Chinese classifier system entails the exact grammatical function to categorize these
semantic distinctions. In other words, there are two possible levels of grammatica-
lization. A distinction is grammaticalized if the contrasting elements can be dis-
tinguished by some representational clues. However, a system of categorization is
grammaticalized only when the whole system is given corresponding grammatical
representations.2
This paper is organized as follows: in Section 2 we distinguish classifiers from
measure words in Mandarin Chinese and we give an overview of the data on which
our study was based. In the third section, we present evidence for the individual
readings of nouns and demonstrate that nouns can occur with more than one indi-
vidual classifier, and that these occurrences coerce different meanings from the
noun. In the fourth section, we discuss the kind readings of classifiers and demon-
strate that kind classifiers select a particular class of nouns. In the fifth section, we
demonstrate that the Mandarin classifier system also contains classifiers that can
coerce an event reading from a nominal. In the last section, we summarize our
findings, giving examples of nouns that can be coerced to all three readings, indivi-
dual, kind and event, and suggest future areas for study.
1
Note that in this paper, the term kind loosely follows Carlson’s (1977) definition. In this reading, kind
refers to a (contextually relevant) subset of individuals to which the noun applies. For instance, the sen-
tence Cats are smarter than dogs gets translated into the reading of The cat kind is smarter than the dog
kind. This reading cannot be refuted by an individual case of a certain dog having higher IQ than a certain
cat. See discussion of the kind reading in Mandarin Chinese in Huang (1987).
2
Thanks to a reviewer for pointing this out.
356 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

2. Background

In the following section we define classifiers as contrasted with measure words,

and we discuss the scope and source of classifiers used in our study.

2.1. Classifiers and measure words

Two traditions of study of classifiers exist in Mandarin Chinese. The classical view
does not differentiate measure words from classifiers. For example, Chao (1968:
584–620) refers to classifiers as individual measures, and subsumes them under the
rubric of ‘‘measure words’’. Li and Thompson (1981: 106) state that ‘‘any measure
word can be a classifier’’.
However, another tradition is based on the premise that measure words are dis-
tinguishable from classifiers on the basis of a simple syntactic test (Kuo, 1998; Tai
and Chao, 1994): Can the genitive de particle be inserted between the classifier/
measure word and its noun? In (1) below, de cannot be inserted, so ben is a classifier.
But in (2), de can be inserted, thus kuai is a measure word.

(1a) yi ben shu

one CL book
‘one book’

(1b) *yi ben de shu

one CL GEN book
‘one book’

(2a) yi kuai rou

one piece meat
‘a piece of meat’

(2b) yi kuai de rou

one piece GEN meat
‘a piece of meat’

Tai (1990) points out that there is also an important semantic distinction between
the two notions in that classifiers can only classify over a limited and specific group
of nouns, while measure words can be used as a measure for a wide variety of nouns.
His definition is as follows:

A classifier categorizes a class of nouns by picking out some salient perceptual

properties, whether physically or functionally based, which are permanently
associated with the entities named by the class of nouns; a measure word
does not categorize but denotes the quantity of the entity named by a noun.
(1990: 312).
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 357

For example, in (2b) above, ‘kuai’ is a measure word because it does not pick out
any salient properties of ‘meat’ nor does it categorize a particular type of meat.
‘Ben’, on the other hand, picks out the salient properties of ‘book’ (i.e. a bound
volume with content).
Underlying the concept that a classifier categorizes over a class of nouns based on
permanent perceptual properties is the idea that the basic semantic function of
nouns is to refer to classic individuals. And although it is a main function of classi-
fiers in Mandarin to select individuals, in what follows we will show that it is
inadequate to only view nouns as referring to classic individuals, and that instead
nouns can be coerced by different types of classifiers to refer to kinds and events as
well as to individuals. Before we discuss coercion to kind and event types (Sections 4
and 5), however, we will first explore coercion within the individual reading to
demonstrate how different classifiers can emphasize different salient properties of the
same noun (Section 3).
In this study, while acknowledging the conceptual validity of the classifier/
measure dichotomy, our emphasis in on how the ‘classifier’ system correlates with
the semantic properties of a noun. In other words, we follow Lyons (1995) and
simply treat ‘mensural classfiers’ as a sub-category of classifiers. Hence, there is no
need to make any a priori cutoff between the two possible sub-categories.

2.2. Source and scope of data

Our data are based on the Mandarin Chinese Classifier Dictionary (Chang et al.,
1994; Huang et al., 1995b). The dictionary itself was compiled based on data from
the Academia Sinica Balanced Corpus of Modern Mandarin Chinese (abbreviated
as Sinica Corpus hereafter). The Sinica corpus is a grammatically tagged and
balanced corpus containing a total of five million words (CKIP, 1995). It reflects the
Mandarin Chinese spoken in Taiwan.
The dictionary contains definitions and examples of 427 classifier and measure
words. One hundred twenty of these are found in elementary school textbooks and
the other 307 were selected based on their high frequency in the Sinica Corpus. The
dictionary also has a second part that lists 1910 noun ending morphemes (and
their associated 12,352 nouns) and lists the classifiers that can co-occur with these
morphemes.
In the dictionary, the classifiers and measure words are classified into seven types.
(The dictionary uses the term liang4ci2 to refer to both classifiers and measure
words.) The first three types correspond to our definition of classifier based on the
syntactic and semantic definitions given in the previous section and that latter four
types correspond to measure words. The seven types are (1) general classifiers (what
we call individual classifiers), (2) event classifiers, (3) kind classifiers, (4) proximation
measure words, (5) container measure words, (6) standard measure words and (7)
activity measure words. There are 174 general classifiers listed in the dictionary, as
well as 35 event classifiers and 14 kind classifiers. In the following section we will
look at how the general classifiers can coerce a particular semantic reading based on
their semantic properties.
358 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

3. The individual reading

The individual reading is the most common aspect of classifier usage. Previous
work (Tai and Chao, 1994; Ahrens, 1994) has emphasized that the classifier picks
out salient features of the noun. Thus each noun will have associated classifier(s),
that are learned along with that noun. For example, in (3) the classifier tiao classifies
for long, cylindrical, flexible objects (Tai and Wang, 1990).

(3a) yi tiao shenzi

one CL rope
‘rope’

(3b) yi tiao she

one CL snake
‘snake’

(3c) yi tiao yu
one CL fish
‘fish’

(3d) yi tiao shayu

one CL shark
‘shark’

Tai argues that properties entailed in the classifiers can be deduced from examin-
ing the perceptual similarities among nouns that they occur with. For example, in
addition to the analysis given in Tai and Wang (1990) for tiao, Tai and Chao (1994)
proposes that zhang classifies over flat, rectangular, two-dimensional, horizontal
objects (i.e. it classifies for paper, tables, chairs etc.).
In addition to occurring with classifiers that contain specific perceptual properties,
nouns can, in addition, occur with the classifier ge. This classifier has been viewed as
a neutral classifier, because it can occur with nouns of many different semantic
classes. Meyers (2000) argues that ge is more correctly viewed as a semantically
vacuous morpheme that is inserted by a default rule, as opposed to being linked
with nouns through analogy as other classifiers are. This may be the case; how-
ever, Ahrens (1994) demonstrates that the critical point is that the occurrence of
ge (whether by a default rule or by analogy) does not apply uniformly across
nouns.
Her production experiments with native speakers demonstrate that if the classifier
has a unique and elaborative relationship (cf. Langacker, 1987) with its noun, it is
less likely to occur with the neutral classifier ge. On the other hand, the less unique
and elaborative the relationship, the more likely the neutral classifier ge will occur in
place of the more specific individual classifier. For example, in an experiment where
subjects were asked to describe the pictures they saw, tiao occurs with shenzi ‘rope’
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 359

83% of the time, but occurs only 47% of the time with she ‘snake’, only 23% of time
with ‘fish’ and only 16% of the time with ‘shark’.
Ahrens attributes these results to the fact that although ‘snake’ embodies the
qualities of ‘thin, cylindrical and flexible’, and therefore has an elaborative rela-
tionship with its classifier, the animal classifier zhi is competing for occurrence with
this noun, and so it does not involve a unique relationship with the noun. ‘Fish’ not
only has the animal classifier competing to occur with the classifier zhi, it also does
not necessarily entail the quality of cylindricality. ‘Shark’ has an animal classifier
competing with tiao and in addition, does not embody ‘thinness’ nor ‘cylindricality’.
Thus, although the system of individual classifiers is based on cognitive principles,
the occurrence of a noun with a particular classifier is dependent upon the unique-
ness of the relationship with the noun as well as the possibility of other competing
classifiers.
This possibility of other competing classifiers is what we will turn to next. In
addition to individual nouns occurring with the semantically neutral ge, nouns can
also occur with different specific classifiers, although this fact has not been examined
closely in the previous literature. For example, in (4) the use of the classifier ben is
the usual specific classifier used for books. There is a contrast, however, in meaning
between (4a) and (4b).

(4a) Zhangsan mai le san ben shu/bijiben

Zhangsan buy ASP three CL_IN book/notebook
‘‘Zhangsan bought three books/notebooks.’’

(4b) Zhangsan mai le san bu shu/*bijiben

Zhangsan buy ASP three CL_IN book/notebook
‘‘Zhangsan bought three (different) books/*notebooks.’’

Ben refers to individual books, while bu refers to both individual books as well as to
the content of the individual books. Thus, in the case of (4a) the three books may (or
may not) be three copies of ‘‘War and Peace’’, or they may be notebooks when
bijiben ‘notebook’ is used. However, in the case of (4b) it is that the three books
must all differ in terms of their content, as seen from the fact that the classifier bu
cannot occur with bijiben ‘notebook’ because notebooks do not necessarily involve
content.
Another interesting example of individual classifiers creating a semantic type
shifting is given in (5).

(5a) bangongshi li you san ju dianhua

office IN has three CLS_IN telephone
‘‘There are three telephone sets in the office. [i.e. three pieces of machinery]’’

(5b) bangongshi li you san xian dianhua

office IN has three CLS_IN telephone
‘‘There are three telephone lines in the office. [i.e. three telephone numbers]’’
360 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

Both ju4 and xian4 are individual classifiers. However, ju selects machinery while xian4
selects a line-like object, including the more abstract meaning of lines of communication.
The selection of a noun by a classifier is limited however. ‘‘Book,’’ for example,
cannot occur with the animal classifier zhi to refer to, for example, books about
animals, or a child’s book in the shape of an animal. The classifier must select a
property of the entire class of nouns, and not just a particular noun. Thus, all tele-
phones have a physical receiver as well as a phone line. All books have a physical
manifestation of paper (or what appears to be paper in a computer) and have con-
tent, while notebooks only necessarily have the former. It is the classifier that selects
the relevant properties of the noun and coerces the appropriate meaning.
Pustejovsky (1995) proposes the concept of qualia structure to encode a ‘‘set of
semantic constraints by which we understand a word when embedded within the
language’’ (p. 86). These constraints may be Constitutive, Formal, Telic, or Agentive
and are defined as follows: Constitutive constraints involve the relationship between
an object and its parts and include material, weight and components. Formal con-
straints distinguish the object within a larger domain and involve orientation, mag-
nitude, shape, dimensionality, color and position. Telic constraints involve the
purpose and function of the object, and Agentive constraints are ‘‘the factors
involved in the origin of an object’’ (Pustejovsky 1995: 86). For instance, book is
Constitutively defined to contain information; it is Formally defined to be made of
bound sheets of paper; its Telic constraint defines its typical purpose as being read;
and finally, its Agentive constraint states that it is created by writing.
Thus we can see that with the noun dianhua ‘telephone’, the classifier ju4 in (5a)
has the Formal role of telephone as an object while the classifier xian in (5b) repre-
sents the Telic role of telephone as a tool to connect to telephone lines. In the case of
shu ‘book’, the classifier ben has the Formal role of a bound volume while the clas-
sifier bu involves the Constitutive role of complete and structured knowledge. Thus,
individual classifiers can coerce nominal semantic types, and semantic coercion can
be predicted through a well encoded qualia structure.
In terms of the representation of meaning, Pustejovsky (1995) implies that the
complete qualia structure represents a single nominal sense that facilitates genera-
tion of different interpretations. In our theory (Ahrens et al., 1998), we make the
crucial distinction of allowing a sense to be instantiated as different meaning facets.
These different meaning facets (of the same sense) can be derived either by Puste-
jovsky’s type-shifting rules or Allen’s (1977) pragmatic rules. Following our earlier
attempt to define the unit of sense, our focus of the current study is on how these
disparate semantic objects can be coherently represented as a sense. We assume
without further elaboration that when different semantic types (such as individuals,
sets of individuals, properties, sets of properties, etc.) are involved, a Chierchia et al.
(1989) style type-shifting rule will apply. However, with or without type-shifting,
contextual information will ‘coerce’ a sense by eliminating other possible inter-
pretations in a richly encoded but under-specified lexicon. Hence, what is important
to us is the fact that, very often, the semantic shifting is applicable to an entire class
of nouns (Ahrens et al., 1998). For example, the classifier duo refers to ‘bud’ of a
plant or flower, while zhu refers to the plant itself, as in (6).
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 361

(6a) yi duo hua

one CL flower/plant
‘one flower’

(6b) yi zhu hua

one CL flower/plant
‘one plant’

This same shifting in instantiated meaning is applicable to all nouns that end in the
morpheme hua ‘flower’, such as meiguihua ‘rose’ or meihua ‘plum flower’, etc. In
these cases, it is differences in the Formal properties of the noun that are being
distinguished.
We have demonstrated above that classifiers can coerce nouns to have a particular
individual reading depending on the information entailed in the classifier itself. The
classifier can vary in the Constitutive, Formal, Telic or Agentive roles that it carries.
The classifier then forces the noun to undergo a semantic type-shifting, so that the
salient properties that the classifier entails are then viewed as the salient properties
of that noun (even if they weren’t viewed as salient previously).
In what follows we will show that this semantic type-shifting is not limited to the
individual reading of nouns. Classifiers can also type-shift nouns to a kind reading
and to an event reading.

4. The kind reading

4.1. Kind classifiers and their semantics

A kind classifier explicitly marks that the nominal element that it selects and gives
it a kind reading, as in (7).3

(7) Nei zhong gou hen huai

that CLS_KND dog very bad
‘‘That kind of dog is bad.’’

If there was only one kind classifier, then one could argue that the notion of kind,
like shape or dimension, is just one of the salient perceptual concepts utilized in the
classifier system of the language. In this scenario, the kind reading would be just one

3
Downing (1996) does refer to kind-classifiers and quality-classifiers in her study of the Japanese clas-
sifier system. However, she uses the term to refer to classifiers that pick out natural kinds, which are often
already present in noun classes. In her study, the crucial distinction is whether a category is inductively or
deductively defined. In other words, Downing’s kind-classifiers actually pick out a category ‘given by the
world’. This is very different from our kind classifiers that categorizes different kinds and coerce kind
readings from the noun. Her kind-classifiers would simply be a subset of our individual classifiers.
362 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

Table 1
Kind classifiers

of the nominal semantic features picked by the classifier. In fact, we will show that
there is a whole set of different kind classifiers selecting different semantic kinds. In
other words, this is a case where a sub-system of classifiers is devoted to the classi-
fication of the concept of kinds. This fact suggests that kind is treated as a primary
semantic type that is ascribed to nominals and allows sub-typing selection by clas-
sifiers. Thus the Mandarin classifier system will offer the first known evidence that
the semantic notion of kind is grammaticalized in a language.
In addition to the mostly commonly used zhong3, 13 additional kind classifiers are
listed in Table 1.
All these kind classifiers have the following common grammatical characteristic:
they select a broad class of nouns, unlike the highly idiosyncratic selection of the
individual classifiers. Semantically, this follows from the fact that kind classifiers select
the kinds represented by nouns, not the individuals referred by them. Hence, the nat-
ural kind delimited by a certain noun class is selected by the appropriate kind classifier.
Thus lei4 selects the kinds defined by properties over concrete objects; lei4 can also refer
to a kind defined by a collection of smaller kinds as in (8). In (9), yang4 selects the kinds
defined by shape and appearance. Shi4 in (10) is similar to yang4 with an emphasis on
appearance; however, in this case the appearance must be an artificial styling. Kuan3 in
(11) selects the kinds defined by intensive design; and lastly, ma3 in (12) and dang3 in
(13) select kinds of events. This leaves zhong3 as the neutral kind classifier, not unlike
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 363

the neutral (or default) individual classifier ge, but even more prominently so since it co-
occurs with virtually all nouns, concrete or abstract.

(8) zhei san lei shiwu dou shi richang bixu de

this three CLS_KD food all be daily required DE
‘‘All these three types of food are daily requirements.’’

(9) ta daile san yang shuiguo lai kan ni

s/he bring-PERF three CLS_KD fruit come see you
‘‘S/He brought three kinds of fruit when she came to see you.’’

(10) canzhuo shang bai man ge shi caiyao

dining-table on place full every CLS_KD dish
‘‘The dining table is full of all different kinds of dishes.’’

(11) fengtian jinnian tuichule liang kuan xinche

Toyota this-year push-out-PERF two CLS_KD new-car
‘‘Toyota has brought out two lines of new cars this year.’’

(12) xunlian he bisai wanquan shi liang ma(zi) shi

training and competition total be two CLS_KD matter
‘‘Training and actually playing are two totally different matters.’’

(13) qiuxue he jiaoxue wanquan shi liang dang(zi) shi

studying and teaching totally is two CLS_KD matter
‘‘Studying and teaching are two totally different things.’’

The kind classifiers in the above six instances can be replaced by the neutral classifier
zhong3, but not by the neutral individual classifier ge. This further supports the
position that kind classifiers form a semantically motivated grammatical class by
themselves and are different from individual classifiers. Moreover, like other
Mandarin classifiers, the instances of kind classifiers that we give in (8)–(13) involve
the numeral-classifier construction, which shows that kinds are individuated and
enumerated.

4.2. The bare NP reading

Bare NPs in Mandarin Chinese have a wide range of possible interpretations.

Among these interpretations the kind reading is one of the most difficult to account
for (see example 14 below).

(14) gou bi mao da

dog compare cat big
‘‘Dogs are bigger than cats. [preferred reading]’’
OR ‘‘This dog is bigger than this cat.’’
364 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

With the kind reading exemplified in (14), the NPs do not refer to any specific indi-
vidual, nor do they refer to all instances in within the species. Hence (14) does not
logically entail that a specific dog is bigger than a specific cat, contrary to the pre-
diction of an account where the bare NP refers to either an individual animal or
some subset of the animals. What is interesting is the additional reading. It says a
specific dog is bigger than a specific cat. This is an impossible interpretation for the
(English) generic reading. However, if we take the kind reading in Chinese as pick-
ing up the contextually defined subset of individual(s) to which the noun applies,
then it is possible to see contexts where exactly one individual will be picked. Hence
the individual reading can be uniformly accounted for (Huang, 1987).
The same semantic effect is achieved when the kind interpretation is marked by a
classifier. Although the kind reading is previously thought to be one of the generic
readings typical of bare nominals (Carlson, 1977; Chierchia, 1982), this reading can
also occur with the kind classifier in Mandarin as shown in (15).

(15) zhe zhong ma zai huabei hen changjian

this CLS_KD horse at China-north very common-seen
‘‘This kind of horse is very commonly seen in Northern China.’’
(Huang, 1987).

The sentence in example (15) does not entail that any specific individual horse of
the referred kind is in north China, nor does it entail that a majority of this kind of
horse is there. Its interpretation is that a contextually defined subset of this parti-
cular kind of horses is commonly seen in north China. In addition, this data casts
doubt on a structural account, such as in the Government and Binding framework
(e.g. Tang, 1990), where classifiers are assigned to the Specifier position; since whe-
ther the Specifier position is lexically filled or not can no longer be a test of the
specificity/definiteness of the NP.4
The fact that (16) below serves as a paraphrase of (14) also supports our obser-
vation that the existence of classifiers such as zhong3 achieve the same semantic
effect of kind reading.

(16) zhez hong dongwu bi naz hong dongwu da

this CLS_KD animal COMP that CLS_KD animal big
‘‘This kind of animal is bigger than that kind of animal.’’

The free variation of Mandarin bare nominals between kind and individual read-
ings is accounted for with type-shifting (Huang, 1987; Chierchia et al., 1989). It is

4
The relationship between the classifier and the specific interpretation is language-dependent. For
example, in Cantonese a classifier is necessary but not sufficient for a specific interpretation. However, in
Mandarin Chinese a classifier is neither necessary nor sufficient to imply a specific interpretation (Mat-
thews and Pacioni, 1997).
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 365

interesting to observe, however, that the use of the classifier zhong3 in (15) and (16)
entails that there is obligatory type-shifting and the sentence is unambiguously
interpreted as a kind reading.
In sum, we have shown in this section that kind classifiers form an integral class
within the Mandarin classifier system. They individuate and classify the different
kind readings that can be obtained from nominal semantics. Thus we have shown
that the notion of kind is not only crucial to theories of nominal semantics, but is
also attested by the grammatical system of a natural language.

5. The event reading

In this section we will show that the Mandarin classifier system contains a sub-
class that selects another theoretically significant entity: i.e. event-type entities.
Chierchia (1982) was one of the first semanticists to propose that events and activ-
ities could be referred to as entities. He studied nominalization and the English
gerund system and demonstrated that the grammatical system marks the type-
shifting from events to event-type entities (i.e. names of the events.) In (17), that
the gerund ‘‘seeing’’ refers to a certain type of event as is obvious from its verbal
derivation.

(17) Seeing is believing.

Thus the semantics of nominalization is treated by Chierchia as an instance of type-

shifting which individuates the events. In Mandarin, however, not only is no
morpho-lexical marking involved when the event-type nominal is deverbal, it is
also possible to coerce an event type reading from a noun without any deriva-
tion. We will show in this section that the event classifiers coerce just such an
effect. Since there will be no marking on the nouns, it is necessary to develop tests
for event-type entities to prove that the classifier-noun phrase does have an event
reading.
It should be noted that event-type nominals have eventive semantics, including
event structures. The semantic (not syntactic) nature of events is that they are tem-
porally anchored, and that they take arguments. Thus, the semantic tests we have
for event-type nominals are that they occur as temporal delimiters, and that they
allow oblique arguments. Two additional syntactic tests are that they satisfy the
subcategorization requirement of predicates which take event-type arguments, and
that they cannot be replaced by the neutral individual classifier ge.
First, an event cannot take place without a time frame, either a temporal point of
its occurrence or a (relative) time duration of its existence. Thus, temporal reference
is an integral part of the semantics of events, including event-type nominals. The
individual classifiers, on the other hand, do not carry temporal reference in their
semantics. Thus, an important grammatical characteristic of event-type nominals is
that they co-occur with a temporal subordinator, such as yi3hou4 ‘after’ (without
any predicates) to establish the temporal (and causal) sequence of events; (18a)
366 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

shows how such a temporal sequence is established with a full clause. In (18b) we
show that when the classifier tang4 is used, the noun taijiquan ‘‘TaiChi boxing’ can
refer to the event without any overt verbal predicate.

(18a) dale taijiquan yihou ta shenti shufu duo le

playLE Tai_Chi after s/he body comfortable more LE
‘‘S/he feels much better after doing Tai Chi.’’

(18b) san tang taijiquan yihou ta shenti shufu duo le

three CLS_EV Tai-Chi after s/he body comfortable more PERF
‘‘S/he feels much better after (performing) three rounds of Tai Chi.’’

Second, it is well-known that event-type nominals preserve their argument

structure and take (oblique) arguments, such as the English possessive ‘‘John’s pro-
motion’’ or ‘‘Mary’s leaving early’’. In Mandarin, a deverbal noun can take a pos-
sessive (encliticized with de), or preceding oblique arguments, such as the underlined
NP in (19).

(19) Zhangsan dui Lisi de jianyi

Zhangsan TO Lisi DE suggestion
‘‘Zhangsan’s suggestion/advice for Lisi (i.e. Zhangsan advised Lisi)’’

In (20), we show that when an event classifier is used, the nominal head takes an
event reading and the possessor is interpreted as an argument of the head.

(20) zhongtong de yitong dianhua

president DE oneCLS_EV telephone
‘‘A call from the President (i.e. the President made the phone call)’’

Last, it can be shown in Mandarin that the event reading is selected by a few small
classes of verbs, such as light verbs jin4xing2 ‘‘to proceed’’ (Huang et al., 1995a,b),
happenstance verbs fa1sheng1 ‘‘to happen’’, and event-evaluation verbs hua1 ‘‘to
cost’’. For instance, hua1 takes a nominal subject that refers directly to an event
(21a), or a clause describing an event (21b).

(21a) xishi hen hua qian

happy-event very cost money
‘‘Weddings cost a lot.’’

(21b) jiehun hen hua qian

marry very cost money
‘‘It costs a lot to get married.’’
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 367

In (22), we show that the same semantic selection can be satisfied when an event
classifier is used. Thus, it supports our position that event classifiers selects event-
type nominals.

(22) zhechang dianying bu hua qian

thisCLS_EV movie NEG cost money
‘‘It did not cost any money to see this movie.’’

In addition, example (23) also shows that the event reading can be coerced from a
bare NP, similarly to example (21a).

(23) dianying hen hua qian

movie very cost money
‘‘It is very expensive (to make/finance/watch/...) movies’’

However, when a bare NP is type-shifted to an eventive reading, the actual event is

underspecified and may have a wide range of interpretations whereas in (22), we
showed that when the event meaning is coerced by a classifier, it also determines the
event-type. In other words, the semantic coercion fits the classical description of
what a classifier does. The event classifiers force the event reading by classifying the
event into a specific type. In (22), chang3 refers to scheduled and regularly occurring
events.
To sum up, we have demonstrated with two semantic tests and two syntactic tests
involving selectional restrictions that event classifiers do coerce the event readings
on the nouns that they co-occur with. We will discuss in more detail the semantics of
the event classifiers in the next section.

5.1. Event classifiers and their semantics

Pustejovsky’s (1995) theory of a Generative Lexicon proposes that the semantics

of nouns cannot be completely represented without referring to two different event-
structures encoded in the qualia structure of each lexical entry. The Agentive role of
a noun is defined in terms of an event structure describing how this noun ‘‘origi-
nated’’. The Telic role is defined in terms of an event structure describing the func-
tion of the noun. His point is that polysemy cannot be accounted for if these two
aspects of a noun are not considered.
Adopting Pustejovsky’s position, the fact that certain classifiers can coerce event
readings from nouns that are typically interpreted as individuals is not surprising,
since their semantic representation already contains event structure information.
Since these event structures are idiosyncratically encoded, we also correctly predict
that the selection between event classifiers and their nominal heads are more
restrictive than kind classifiers. We will further classify the event classifiers into the
event-type classifiers, which individuate different event structures, and the event-
token classifiers, which individuate each occurrence of an event.
368 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

Table 2
Event classifiers

Huang et al. (1995a,b) list 35 classifiers as event classifiers. They are listed in Table 2.
Among these event classifiers, the event-type classifiers are more specialized since
they select a particular event type. For instance, the event-type classifier chu1 selects
the nouns which contains the stems of either ju2 ‘‘drama’’ in (24a) or xi4 ‘‘play’’ in
(24b).
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 369

(24a) shashibiya daodi yigong xiele ji chu xi?

Shakespeare eventually together write-PREF how_many CLS_EV play
‘‘How many plays did Shakespeare write all together?’’

(24b) bailaohui jin nian zhi yan le yi chu gewuju

Broadway this year only play PERF one CLS_EV musical
‘‘Only one musical (e.g. Cats) was performed on Broadway all year this year.’’

The two sentences above have Agentive (24a) and Telic (24b) readings respectively,
suggesting that these event readings are coerced from the Pustejovskian qualia
structure. The Agentive reading of (24a) specifically refers to the event types of how
a play comes into being. The Telic reading of (24b) refers specifically to the event
types of why a play is produced (to be viewed). They also clearly show that chu1
individuates event-types.
In contrast, when an event-token classifier is used, the same noun will refer to the
occurrences of the event. The event-token classifier chang3 selects a scheduled event
as in (25).

(25) bailaohui jin nian gong yanle yibai chang gewuju

Broadway this year together playPERF one hundred CLS_EV musical
‘‘Broadway had one hundred performances of musicals this year.’’

Thus (25) claims that musicals were performed 100 times on Broadway without
referring to whether the same play was shown or not, while (24b) claims that only
one musical was performed without claiming to how many times it was done so.
Even though event-types and event-tokens are semantically distinct entities and
the classification is supported by the two unambiguous cases given above (24b and
25), we do find that in many cases the same classifier will be polysemous with both
event-type (26a) and event token readings (26b).

(26a) Changrong you shi ban feiji fei gaoxiung

EVA has ten CLS_EV airplane fly Kaohsiung
‘‘EVA has ten scheduled flights to Kaohsiung .’’

(26b) Changrong gang feizou le san ban feiji

EVA just fly-away PERF three CLS_EV airplane
‘‘Three EVA flights just took off.’’

Example (26a) has the event structure of ‘‘flights’’, while (26b) individuates each
particular flight.
In sum, our semantic account of the event classifiers in this section suggests that
event classifiers can adopt the event information encoded in qualia structures to
370 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

define both Agentive and Telic roles and coerce the semantics into agentive or telic
events. We also have shown that the semantics of event classifiers can be further
distinguished as referring to either the event type or event tokens.

6. Conclusion

We have demonstrated above that it is inadequate to view the ontology of the

classifier system as based on the classic notion of individuals. We argue instead that
the top level of the ontology of the Mandarin Chinese classifier system is actually tri-
partite: individual, kind, and event. An important motivation in studying this new
ontology of the Mandarin Chinese classifier system is the fact that even seemingly
straightforward referential nominals have complex semantic content, as suggested
by our data as well as recent studies on nominal lexical semantics which refer to
different individual or events (Pustejovsky, 1993, 1995). In fact, the semantics of
nouns are more complicated than one might suspect. As we have shown above,
Mandarin Chinese offers one of the most dramatic illustrations since its nouns can
have very different meanings without any additional morphological markings. Our
main generalization is summarized below. Three of the possible meanings of the
noun dian4hua4 ‘‘telephone’’ and of fei1ji1 ‘‘airplane’’ can be brought out by the use
of different classifiers, as given in Tables 3 and 4 respectively.
The individual classifiers ju and xian can select one of two interpretations relating
to the ‘phone’. The former selects the telephone machinery itself, while the latter
selects the line for the phone, as in ‘line one’ or ‘line two’. The kind classifier coerces
a kind reading so that a particular type of phone (i.e. cordless or cell phone) is
referred to. The event classifier coerces an eventive reading and refers to the com-
pleted act of calling.

Table 3
Classifier and nominal semantic type correlation for dianhua ‘‘telephone’’

Semantic/Classifier type Example Reference

Individual yi ju dian-hua Telephone

one CLS_IN telephone Machinery
‘‘one telephone’’
yi xian dian-hua
one CLS_IN telephone Telephone line
‘‘one telephone line’’

Kind yi zhong dian-hua A particular kind of phone,

one CLS_KD telephone e.g. cordless
‘‘one kind of telephone’’

Event yi tong dian-hua The completed event of calling

one CLS_EV telephone
‘‘one phone call’’
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 371

Table 4
Classifier and nominal semantic type correlation for feiji ‘‘airplane’’

Semantic/Classifier Type Example Reference

Individual yi jia feiji Airplane

one CLS_IN airplane
‘‘one airplane’’

Kind yi zhong feiji A particular kind of airplane, e.g. jet

one CLS_KD airplane
‘‘one kind of airplane’’

Event yi ban feiji The completed event of flying

one CLS_EV airplane
‘‘one scheduled flight’’

In Table 4, jia is the individual classifier that is used to refer to the physical entity
of an airplane, while zhong is used to refer to a particular kind of airplane. Ban is an
event classifier that coerces the concept of a completed event flying from the noun
‘airplane.’
The implications for this new tripartite classification of classifiers are three-fold:
first, it explains why the kind reading in Mandarin Chinese, while it can be under-
stood as a generic reading typical of bare nominals, also occurs with a subset of
classifiers (i.e. the kind classifiers) and is a semantic type in and of itself. Second, the
previously abstract semantic distinctions between kinds, individuals and events are
now found to be instantiated in one particular system (i.e. the classifier system) of a
natural language grammar, and moreover, these three classifier types have their own
taxonomic system. Thus, this suggests that the semantic types of individuals, kinds
and events are useful categorizing tools for humans (and not just formal semanti-
cists), and that these concepts are encoded on a cognitive level. Finally, this classi-
fication allows us to a take a fresh look at the complex semantic contents of nouns,
and at the interaction and coercion that takes place between classifiers and nouns.
In future studies, we hope our findings will help us to better understand the
interaction of semantic meanings among kinds, individuals, and generics (Carlson
and Pelletier, 1995; Chierchia, 1994). In particular, based on the ready availability of
the kind reading, either with bare plurals in all languages or with kind classifiers in
Chinese, we suspect that the kind role should be part of the nominal qualia structure
in the theory of a Generative Lexicon. With regard to the event readings of nom-
inals, even though the present theory of qualia structure does offer an possible
account of how these readings can be obtained from the lexical semantics of the nom-
inals, more thorough studies are needed in order to shed light on whether the Agentive
and Telic events are the only events that are necessarily encoded in the nominal
semantics, as well as additional studies on how these event readings are selected.
Lastly, as Alexandra Aikhenvald has pointed out (p.c. August 2001) event-type
classifiers in other languages have been described, although the term event may not
372 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373

have been used [e.g. Aikhenvald’s (2000) discussion of use of classifiers with abstract
nouns (pp. 335–336)]. It would be interesting to examine if these instances do fall
under event-type classifiers as we have defined them in this paper.
Moreover, if they do, do these languages also have a kind classifier system (in
addition of course to an individual classifier system)? That is, to what extent are the
semantic concepts of individual, kind and event instantiated in the grammars of
other language’s classifiers systems?

Acknowledgements

We would like to thank the National Science Council of the ROC for funding
NSC 86-2411-H-194-012 and NSC90-2411-H-002-008-MC. We also would like to
thank our colleagues at CKIP, especially Lili Chang, Keh-jiann Chen and Feng-yi
Chen for their stimulating ideas and discussion. Thanks also go to the participants
at the 1996 Pacific Asia Conference on Language, Information and Computation
where an earlier version of this paper was presented, and to Stanely Starosta for his
comments on that same earlier version. In addition, we would like to thank the
reviewers of this paper and Alexandra Aikhenvald for their comments on a later
version as well as Dora Hsin-yi Lu for help with preparing the final version. All
errors, are of course, our own responsibility.

References

Aikhenvald, A.Y., 2000. Classifiers: A Typology of Noun Categorization Devices. Oxford University
Press, Oxford.
Ahrens, K., 1994. Classifier production in normals and aphasics. Journal of Chinese Linguistics 22 (2),
202–247.
Ahrens, K., Chang, L., Chen, K.-J., Huang, C.-R., 1998. Meaning representation and meaning instan-
tiation for Chinese nominals. Computational Linguistics and Chinese Language Processing 3 (1), 45–60.
Allen, K., 1977. Classifiers. Language 53 (2), 285–311.
Chang, L., Chen, K.-j., Huang, C.-R. 1994. The use of corpus in dictionary compilation. Proceedings of
ROCLING IX. pp. 255–279 (in Chinese).
Carlson, G., 1997. A unified analysis of the English bare plural. Linguistics and Philosophy 1, 413–457.
Carlson, G., Pelletier, F., 1995. The Generic Book. University of Chicago Press, Chicago.
Chao, Y.R., 1968. A Spoken Grammar of Chinese. University of California Press, Berkeley.
Chierchia, G., 1982. Nominalization and Montague grammar: a semantics without types for natural lan-
guage. Linguistics and Philosophy 5, 303–354.
Chierchia, G. 1994. Plurality of mass nouns and the notion of semantic parameter. Department of Cog-
nitive Science Working Papers, Vol. 7, Istituto Scientifico San Raffaele.
Chierchia, G., Partee, B., Turner, R. (Eds.), 1989. Properties, Types and Meaning, Vol. II. Semantic
Issues. Kluwer, Dordrecht.
Chinese Knowledge Information Processing (CKIP), 1995. An Introduction to the Sinica Corpus. CKIP
Technical Report 95-02. Academia Sinica, Taipei (in Chinese).
Downing, P., 1996. Numeral Classifier Systems: The Case of Japanese. John Benjamins Publishing,
Amsterdam/Philadelphia, PA.
Huang, C.-R. 1987. Mandarin Chinese NP de: A Comparative Study of Current Grammatical Theories.
Dissertation, Cornell.
 C.-R. Huang, K. Ahrens / Language Sciences 25 (2003) 353–373 373

Huang, C.-R., Yeh, M., Chang, L.-P. 1995a. Two light verbs in Mandarin Chinese. A corpus-based study
of nominalization and verbal semantics. In: Proceedings of NACCL6, Vol. 1. GSIL, USC, Los Angeles,
pp. 100–112.
Huang, C.-R., Chen, K.-j., Lai, C.-X., 1995b. Mandarin Chinese Classifier and Noun-Classifier Colloca-
tion Dictionary. Mandarin Daily Press, Taipei.
Kuo, Y.-j. 1998. A Semantic and Contrastive Analysis of Mandarin and English Measure Words.
Unpublished MA thesis, National Chung Cheng University, Jiayi, Taiwan.
Langacker, R., 1987. Foundations of Cognitive Grammar, Vol. 1. Stanford University Press, Stanford.
Li, C.N., Thompson, S.A., 1981. A Reference Grammar of Mandarin Chinese. University of California
Press, Berkeley.
Lyons, J., 1995. Linguistic Semantics: An Introduction. Cambridge University Press, Cambridge.
Meyers, J., 2000. Rule vs. analogy in Mandarin classifier selection. Chinese Languages and Linguistics 1
(2).
Matthews S., Pacioni, P. 1997. Specificity and genericity in Cantonese and Mandarin. In: Xu L. (Ed.),
Proceeding of the Symposium on Referential Perspectives of Chinese NPs. Paris, EHESS, pp. 45–61.
Pustejovsky, J., 1995. The Generative Lexicon. MIT Press, Cambridge.
Pustejovsky, J., 1993. Type coercion and lexical selection. In: Pustejovsky, J. (Ed.), Semantics and the
Lexicon. Kluwer, Dordrecht.
Tai, J.H.-Y., 1990. Variation in classifier systems across chinese dialects: towards a cognition-based
semantic approach. In: Proceedings of the First International Symposium on Chinese Languages and
Linguistics, pp. 308–322.
Tai, J.H.-Y., Chao, F-Y., 1994. A semantic study of the classifier Zhang. Journal of the Chinese Language
Teachers Association 29, 67–78.
Tai, J.H.-Y., Wang, L.-Q., 1990. A semantic study of the classifier Tiao. Journal of the Chinese Language
Teachers Association 25, 35–56.
Tang, C.-C. J., 1990. Extended X-bar Theory and the Chinese Phrase Structure. PhD Thesis, Cornell
University.