2.1 Classifiers in Mandarin Chinese

As mentioned in Section 1, classifiers are crucial to the mass-count distinction in Mandarin Chinese. Mandarin Chinese is a classifier language, in which numerals are not allowed to directly modify count nouns, as opposed to number-marking languages such as English. Instead, the presence of a numeral requires the obligatory use of classifiers (see Li & Thompson 1981), which are words that classify the noun depending on the type of its referent.

In the Chinese literature, two general types of classifiers are distinguished: sortal classifiers and mensural classifiers (Allan 1977; Tai 1992; 1994; Tai & Wang 1990; among many others). Sortal classifiers are also referred to as count classifiers. They classify nouns such as shu ‘book’ or ma ‘horse’, which name things with naturally existing quantificational units, or boundaries, illustrated in (3). Mensural classifiers, which are also known as measure words or mass classifiers on the other hand, do not seem to have a rigid association with sortal nouns. As mass classifiers do not denote naturally existing quantificational units, they can co-occur with both nouns such as shu ‘book’ and nouns such as shui ‘water’, as exemplified in (4).

To better understand the classifier system in Mandarin, we illustrate a number of differences between the two types of classifiers on the basis of (3) and (4) (following Li & Thompson 1981; Tai 1992; 1994; Cheng & Sybesma 1998; Cheng 2012). First, count classifiers belong to a closed class whereas mass classifiers are members of open categories. Count classifiers do not convey concrete lexical meanings but fulfil a grammatical function in the language. Mass classifiers, on the other hand, do have lexical meanings and can therefore – in addition to their classifier function – also appear on their own as lexical items (Tai 1994; Cheng & Sybesma 1998). For instance, when not used as a mass classifier, wan ‘bowl’ (see (4a)) has a denotation similar to its English counterpart bowl.

A second difference between the two classifier types concerns their relation to the noun they classify. Count classifiers, due to their function of indicating naturally existing counting units of sortal nouns, display a fixed and rigid association with the noun they classify (Tai & Wang 1990; Tai 1992; Ahrens 1994). As illustrated in (3), different nouns require different count classifiers. Although semantic properties of sortal nouns, such as animacy, solidity, and function may be cues for language speakers to use one count classifier but not another, the knowledge of count classifiers is argued to be rote-learnt (e.g., Li et al. 2010). On the contrary, the relation between mass classifiers and the noun they classify is considered contingent and temporal (Tai 1992; 1994; Ahrens 1994). Nouns do not seem to require a specific mass classifier. The same noun can be compatible with different mass classifiers as long as an appropriate discourse context is provided.

Finally, as count classifiers behave more like grammatical elements than like lexical items, there is a default, namely ge (Cheng & Sybesma 1998). As the default, or the general count classifier, ge is not specified for any shape information, nor does it seem to provide any information about animacy or function. Thus, the examples of sortal nouns we gave in (3) are both compatible with the default ge, although the use of ge may sound informal or less literate.

Employing two diagnostics, Cheng & Sybesma (1998; 1999; 2005; see also Cheng 2012) argue that count and mass classifiers display different syntactic behaviours, which they consider evidence for a syntactic mass-count distinction in Mandarin Chinese. We start with the (im)possibility of co-occurrence with de, a modification marker according to the traditional view (Li & Thompson 1981). Cheng & Sybesma demonstrate that in Mandarin, only mass classifiers, which are also referred to as massifiers in their work, can co-occur with the modification marker de (see (5)), whereas the co-occurrence of count classifiers with de gives rise to ungrammaticality (see (6)).

Secondly, count and mass classifiers behave syntactically differently with respect to the insertion of BIG.SMALL adjectives in classified noun phrases (henceforth CNPs). As is illustrated in (7), CNPs with massifiers allow the insertion of adjectives like xiao ‘small’ or da ‘big’; whereas the insertion of such adjectives leads to ungrammaticality in CNPs with count classifiers, as illustrated in (8).

As observed in Doetjes (1997), among others, classifiers in Mandarin Chinese are only obligatory when nouns are modified by numerals. In the absence of numerals, nouns can appear without classifiers. Some examples of such unclassified nouns are provided in (9) and (10).

Thus, even though the mass-count distinction in Mandarin Chinese is reflected in the syntax, the syntactic reflection is not always obligatory. This gives rise to the question as to what renders a volume-based (corresponding to a mass noun) or a number-based (corresponding to a count noun) interpretation of an unclassified noun in Mandarin. Bearing in mind this question, we will now present three approaches to mass-count in Mandarin Chinese: a lexico-syntactic approach (Doetjes 1997; 2012; Cheng & Sybesma 1998; 1999; 2012; Cheng 2012), a syntax-driven approach (Borer 2005), and a hybrid approach (Pelletier 2012).

2.2 Relevant approaches to the mass-count distinction in Mandarin Chinese

The lexico-syntactic approach allows lexical marking for countability. According to Cheng & Sybesma (1998; 1999; 2012; see also Sybesma 2007; Cheng et al. 2008; Cheng 2012), nouns come out of the lexicon either with or without counting units. If a noun has a discrete counting unit in its lexical-semantic denotation, it is interpreted as count; if, on the contrary, its lexical-semantic denotation does not include such a counting unit, its interpretation is mass. Allowing similar lexical markedness, Doetjes (1996; 1997) distinguishes four types of nouns in the lexicon. They are count singular nouns (e.g., bird), count plural nouns (e.g., birds), count mass nouns (e.g., furniture), and mass mass nouns (e.g., water), which are agued to have different domains of denotation. Count singulars refer to a set of singularities e.g., a, b, c, etc., and count plurals refer to the set of pluralities that are formed on the basis of the set of atoms, e.g., {a, b}, {a, c}, {a, c}, {a, b, c}, etc. Count mass nouns may refer to both the set of singularities and the set of pluralities built on it, e.g., a, b, c, {a, b}, {a, c}, {a, c}, {a, b, c}, etc. Finally, mass mass nouns do not have atomic reference but denote portions of matter, which have, what she calls, the part of-relation, with each other.² In Doetjes’ view, only the three count noun types (count singulars, count plurals, count mass nouns) come out of the lexicon with counting units. In contrast, mass mass nouns are not marked for countability in the lexicon.

According to the lexico-syntactic approach, the distinction between count and mass nouns in terms of countability at the lexical level must be syntactically visible. In other words, lexically marked countability must agree with syntactic marking of countability. In number-marking languages such as English, this syntactic marking is realized by, for instance, plural morphology. In classifier languages such as Mandarin, which lack plural morphology, it is achieved through count classifiers. Thus, for Cheng, Sybesma, and Doetjes, the presence of count classifiers in Chinese is merely a syntactic realization of the lexically marked countability of certain nouns. This means that the interpretation of unclassified nouns in Mandarin should rely on the semantic information that is already present in their lexical denotation.

Different from the lexico-syntactic approach presented above, Borer (2005) argues that nouns do not contain any semantic information in the lexicon: they are unmarked for either count or mass. This implies that ontology does not play a role (contra Quine 1960 and Jackendoff 1993). Instead, Borer proposes that the mass-count distinction is purely structural. In the presence of what Borer calls a dividing structure, realized as a classifier phrase (henceforth CLP) in the syntax, nouns become count and are assigned a number-based interpretation. In contrast, the absence of a CLP gives rise to a mass noun, with a volume-based interpretation. Languages differ as to how they project CLPs. Borer argues that in number-marking languages such as English, CLPs are projected by, for instance, the plural morpheme. In classifier languages such as Mandarin Chinese, CLPs are headed by classifiers. Thus, in Borer’s view, both classifiers and plural morphology count as classifier inflection, projecting CLPs, and being responsible for a number-based interpretation of nouns. However, not all classifiers can project CLPs. Adopting the two types of classifiers distinguished in the literature on Mandarin (see Subsection 2.1), Borer further argues that only count classifiers have the grammatical function to head CLPs, while massifiers do not. Syntactically, massifiers behave the same as measure words in English (see Borer 2005: 96–97). This means that all nouns, across all languages, become mass nouns and receive a volume-based interpretation in the absence of a CLP. For Mandarin Chinese, this means that nouns receive a number-based reading only when modified by a count classifier. When appearing with a massifier, or, crucially for our study, without any classifiers, Mandarin nouns must only have a volume-based interpretation.

A more recent account of the mass-count distinction is proposed in Pelletier (2012). Pelletier’s approach to the mass-count distinction involves the interaction between four features at two levels: two syntactic features, namely, +COUNT_syn and +MASS_syn, and two semantic features, namely, +COUNT_sem and +MASS_sem (and hence its hybrid nature). In line with Borer, but contra Quine and Jackendoff, ontology plays virtually no role in Pelletier’s view. Furthermore, similar to Borer, Pelletier argues that nouns are unspecified in the lexicon for their count or mass denotations. However, a crucial difference with Borer is that for Pelletier nouns are semantically both count and mass, which is realized by the presence of the two semantic features on the noun. In other words, all nouns are in principle flexible. They can refer to individual objects with clear boundaries as well as substances without clear boundaries, just as rope or chocolate. In contrast, Pelletier argues that nouns are lexically unspecified for syntactic count and mass features. When nouns enter the syntax, either +COUNT_SYN or +MASS_SYN can be introduced. When +COUNT_syn or +MASS_syn is introduced, the opposite semantic feature on the noun (i.e., +MASS or +COUNT) is deleted, resulting in a number-based or volume-based interpretation, respectively. Acknowledging the syntactic relevance of the mass-count distinction at the level of classifiers in Mandarin Chinese, Pelletier suggests that the syntactic features +COUNT_SYN and +MASS_syn can only be introduced at the entire level of classified noun phrases (henceforth CNP). In particular, CNPs headed by a count classifier introduce +COUNT_syn, and those headed by a massifier introduce +MASS_syn. Thus, Mandarin nouns get a number-based reading in the presence of a count classifier and receive a volume-based interpretation in the presence of a massifier. As such, Pelletier’s account makes the prediction that in the absence of classifiers, Mandarin nouns can get either a number-based or a volume-based interpretation (since neither +COUNT_sem nor +MASS_sem is deleted).

2.3 Previous empirical findings

Although there are several experimental studies on the (L1 acquisition of the) mass-count distinction in number-marking languages such as English, Dutch, Hebrew, etc. (Barner & Snedeker 2005; Hacohen 2010; van Witteloostuijn 2013; van Witteloostuijn & Schaeffer 2014; this volume; among others), there are very few on classifier languages such as Mandarin Chinese. Most experimental studies on the mass-count distinction in Mandarin Chinese that we know of aim to investigate whether Mandarin speakers, both adults and children, are sensitive to the syntactic cues to the mass-count distinction that Mandarin employs, namely, classifiers (Li et al. 2008; Cheung et al. 2009), and how the presence of classifiers may influence the speakers’ interpretation as mass or count (Li et al. 2010; Cheung et al. 2012).

To the best of our knowledge, there is only one study in which the interpretation of Mandarin nouns without classifiers is investigated, namely, Study 1 in Cheung et al. (2012). In this study, the authors examined how adult speakers of Mandarin Chinese from Taiwan interpreted nouns of different types with respect to their number-based or volume-based denotations in a Quantity Judgment Task (QJT; Barner & Snedeker 2005). Four noun types were included: count nouns such as balls, mass nouns such as water, and two types of flexible nouns, namely, nouns such as apple(s), and nouns such as string(s). Each of these four conditions was administrated to a different group of 14 adult native speakers of Mandarin Chinese. While being presented with photographs of two characters, one with two large individual objects or two large portions of substances and the other with four small individual objects or four small portions of substances, the participants were asked to answer a question in Mandarin Chinese of who had more. As illustrated in (11), the question contained a bare noun, which appeared without any classifiers.

The results showed that number-based judgments were most frequent for count nouns, i.e., 100%, and least frequent for mass nouns, i.e., 0%. When confronted with the two flexible noun types, the number-based judgments were in-between: 75% for apple-type flexible nouns and 62.5% for string-type flexible nouns.

Cheung et al. interpret these results in light of their previous findings on the mass-count distinction in English, a number-marking language, and Japanese, a classifier language (Barner et al. 2009: Experiment 1). These findings with English and Japanese monolinguals were obtained in QJTs as well, in which participants were asked to answer the question Who has more X by giving a quantity judgment in the similar experimental setting as Cheung et al.’s Study 1. The critical question in the Japanese QJT is presented in (12), adapted from Inagaki & Barner (2009: (4)). The Japanese critical question contained a bare noun too, with no classifiers.

The English QJT was administrated to a total of 64 English monolinguals, who were equally divided into four quantity judgment conditions, representing four noun types: count nouns like pens, mass nouns like bread, flexible count nouns like eggs, and flexible mass nouns like egg. Since Japanese is a classifier language and does not have obligatory plural morphology similar to Mandarin Chinese, the Japanese QJT did not distinguish between flexible count and flexible mass nouns (eggs vs. egg, respectively), and hence only involved three test conditions: count, mass, and flexible. A total of 63 native speakers of Japanese participating in the Japanese experiment were divided into one of these three test conditions. Results of the English and Japanese QJTs are summarized in Table 1 (data adopted from Barner et al. 2009).

Comparing their Mandarin QJT results with those obtained in English and Japanese as reported in Table 1, Cheung et al. (2012) conclude that the semantics of Mandarin nouns do not fundamentally differ from that of nouns in number-marking or other classifier languages. The authors argue that their quantity judgment results show that semantic differences at the lexical level drive syntax, and thus support the lexico-syntactic approach.

However, there are several limitations of the above-presented QJTs in the two classifier languages, namely Mandarin and Japanese, which make it impossible to draw disentangle the three theoretical approaches described in Subsection 2.2. A major concern is that the noun types included in these studies (four types in the Mandarin QJT, and three types in the Japanese QJT) were not administered to the same group of participants. If each participant only received one noun type, he or she could develop an automatic quantity judgment strategy, giving rise to one and the same interpretation of that noun type over and over again. This prevents comparison between the different noun types within the same participants. Another shortcoming is that no child participants took part in the QJTs. Thus, no conclusions can be drawn on possible interpretational patterns of unclassified nouns in classifier languages during acquisition or explanations in this respect. Third, nouns of the object-mass type (e.g., furniture), which represents a crucial ontological type in the mass-count literature, was not included in Cheung et al.’s Mandarin QJT or Barner et al.’s Japanese QJT. Moreover, the quantity judgment question in both the Mandarin and the Japanese QJTs was presented in written, rather than oral form as in Barner & Snedeker’s (2005) original QJT. We consider the mode of stimuli presentation highly relevant to examining the interpretation of unclassified nouns in classifier languages because written language (e.g., characters in Mandarin Chinese) may convey meaning through their iconicity, which can potentially influence participants’ quantity judgments. Such influence is unavoidable in Mandarin. In Japanese, it depends on the character type. If logographic kanji was used in the Japanese quantity judgment question, similar influence was expected as in Mandarin Chinese since kanji is adopted from Chinese characters. The mode of stimuli presentation is moreover crucial to the current study due to the following reasons. First, compared to written stimuli, orally presented stimuli generally render more spontaneous and natural speech, which is what we aim to investigate in the current study. Second, stimuli in written form cannot be used with young children.

Filling the methodological gaps of the previous QJTs in two classifier languages, the current study explores empirical data to differentiate between various theoretical approaches to the mass-count distinction in Mandarin Chinese. In the next section we present the predictions that the three theoretical approaches make for adult and child Mandarin Chinese regarding the interpretation of unclassified nouns.

2.4 Predictions

The three theoretical approaches described in Subsection 2.2 make different predictions not only with respect to the interpretation of unclassified nouns in Mandarin Chinese but also regarding possible preferences in terms of interpretation. Empirically testing the interpretation of unclassified nouns in Mandarin-speaking adults and Mandarin-acquiring children will have crucial theoretical implications. In (13), we list our predictions.

The three theoretical approaches also make different predictions regarding speakers’ interpretational consistency. As the lexico-syntactic approach attributes the interpretation of unclassified nouns in Mandarin Chinese to the lexical-semantic information of the nouns (i.e., 13(i)), it predicts consistency in speakers’ interpretations of the same unclassified nouns with respect to their mass-count denotations. Interpretational consistency is also expected when we follow Borer’s syntax-driven approach. Since unclassified nouns in Mandarin Chinese lack a CLP, they should be consistently interpreted as having a mass reading. In contrast, assuming all nouns to be semantically both count and mass at the lexical level, Pelletier’s hybrid approach does not guarantee similar interpretational consistencies. Speakers may show different interpretations when confronted with the same unclassified nouns. This gives rise to another question that we need to address if we do find such interpretational preferences: how do we explain these?

Although Pelletier’s approach allows both number- and volume-based interpretations of the same unclassified nouns, it does not provide an explicit explanation for potential preferences in this respect. We hypothesize that linguistic experience plays a role here. The term linguistic experience is inspired by Borer’s concept of convention, which we interpret as speakers’ experience with nouns appearing in different types of syntactic structures in the input. In particular, we hypothesize that linguistic experience accounts for speakers’ preferences for certain nouns to appear in certain syntactic structures, which, in turn have certain preferred interpretations by the speakers. The more often a noun appears with a count classifier or a massifier, the stronger a speaker’s preference is for a number-based or a volume-based interpretation of the noun, respectively.

Given that children have less linguistic experience than adults do, we predict a difference between adult and child speakers of Mandarin when it comes to the interpretational preferences of unclassified nouns. Adults are predicted to show a preference for either a number-based or a volume-based interpretation, depending on how often they have encountered a number-based or volume-based interpretation of the relevant noun in the input, whereas children are predicted to equally allow number-based and volume-based interpretations of unclassified nouns due to little linguistic experience. We list the predictions regarding interpretational consistency following from the three theoretical approaches in (14).

In the next section, we describe the methods to test these predictions.

4.1 Quantity Judgment Task

For each test condition, we calculated the mean proportion of number-based quantity Judgments per participant group. In order to investigate the effect of age group (adults, two- and three-year-olds, and four- and five-year-olds) and test condition (Count, Mass, Flexible, Object-mass) on the participants’ preferences in their quantity judgments, we performed a two-way ANOVA. Results of the two-way ANOVA show a significant interaction between the effects of these two factors on the participant’s quantity judgment performance as well (F (6, 2367) = 24.090, p = .000). Moreover, there is a significant effect of age group (F (2, 2367) = 49.978, p = .000) and a significant effect of test condition (F (3, 2367) = 244.241, p = .000) on the quantity judgment preferences of the participants. Multiple comparisons based on age groups reveal that there is no significant difference between the two child groups (p = .559). In contrast, we attest a significant difference between the adult group on the one hand, and each of the two child groups on the other (p = .000 in both cases). In other words, our data do not show any significant development by age in Mandarin speaking children below the age of six. We therefore no longer distinguish two child groups but collapse all child results. Figure 1 displays the average proportion of number-based quantity judgments by condition.

Figure 1

Mean proportions of number-based judgments in adult and child Mandarin Chinese per test condition.

The adult results show the following pattern. In all conditions but the object-mass condition, adults allow both number-based and volume-based judgments. Nevertheless, there are clear preferences for one of the two interpretations in the count, mass and flexible conditions. In the count (e.g., balls) condition the adult Mandarin speakers clearly prefer number-based quantity judgments (77%). When confronted with unclassified nouns of the ontological type mass (e.g., milk) and those of the flexible type (e.g., stone(s)), the adult participants show a preference for volume-based quantity judgments, as the average proportions of number-based judgments are merely 9% and 18%, respectively. Results of One-Sample t-tests confirm that the preference of the adult control group in none of the four test conditions is explained by chance. The proportions of number-based quantity judgments by the adults significantly differ from 0.5 in all test conditions, as shown in Table 3.

As indicated by the proportion of 98%, the object-mass condition virtually always renders a number-based interpretation for the adults. Turning to the child results, we observe two patterns. On the one hand, as early as the age of three, children already clearly prefer number-based quantity judgments in the count (e.g., balls) and object-mass conditions (e.g., furniture). The mean proportions of number-based quantity judgments in these two test conditions are both above 80%. One-sample t-tests revealed that the children performed significantly above chance (0.5) in both conditions (Table 4), which resemble the adult findings in these conditions (see Table 3).

On the other hand, similar adultlike performance is absent with the ontological types mass (e.g., milk) and flexible (e.g., stone(s)). Whereas the adult speakers of Mandarin significantly prefer volume-based quantity judgments in the mass and flexible conditions (see Table 3), the child participants do not appear to have quantificational preferences when presented with unclassified nouns in the two relevant conditions. In both of the two test conditions, the mean proportions of number-based judgments attested with the child participants hover around 50%, i.e., chance, as shown by the results of One-Sample t-tests provided in Table 4.

4.2 Corpus study

In our corpus study, we counted how often the four ontological types corresponding to the conditions of our experiment occurred with a count classifier, with a massifier, or without any classifiers, i.e., unclassified. The results are presented in Table 5.⁷ The reader is referred to Appendix III for results at the item level.

Our CIC results demonstrate two characteristics of the syntactic distribution of Mandarin nouns. One is that unclassified nouns frequently appear in the language, as indicated by the high numbers in the row “unclassified” in Table 5. Of all the nouns coded for the different ontological types, the count type occurs without a classifier 76% of the time (1684 out of 2208), the mass type 80% of the time (2360 out of 2939), the type object-mass 89% of the time (2264 out of 2543), and finally, the flexible type 75% of the time (2422 out of 3250).

A second characteristic is that all nouns in Mandarin, in spite of their ontological basis, appear with both count and mass classifiers, although different ontological types prefer different types of classifiers. Nouns of the ontological type count are attested much more often with a count classifier than with a massifier (i.e., 364 out of 2208 vs. 60 out of 2208); whereas nouns belonging to the mass type and those of the type object-mass are used much more often with a massifier than with a count classifier (i.e., 555 out of 2939 vs. 24 out of 2939, and 250 out of 2543 vs. 29 out of 2543, respectively). As for the ontological type flexible, they are used more often with a massifier than with a count classifier as well, although the difference between the two is smaller than for mass and object-mass type nouns: 622 with a massifier out of 3250 vs. 206 with a count classifier out of 3250. The difference between the four ontological types in their preferences for different types of classifier is statistically significant, as confirmed by the result of a chi-square test (X² (3, N = 2110) = 884.323, p = .000).

4.3 A comparative look: The QJT results and CIC frequency data

In order to compare the interpretational patterns attested in our QJT and the CIC results that may represent the participants’ linguistic experience with mass-count, we combined both the QJT results and the CIC data in Figure 2. On the x-axis, the different ontological categories are distinguished. The y-axis represents the proportions of number-based interpretations.

Figure 2

Comparison of the proportions of number-based interpretations of the participants in the QJT and the syntactic count environments in the CIC.

As for the CIC results, the proportions of number-based interpretations include all nouns with a count classifier with respect to the total number of nouns with a classifier. Unclassified nouns were not included in this calculation, because nouns appearing without any classifiers do not contribute to speakers’ linguistic experience under Pelletier’s account (see Subsection 2.2).