Production of Cantonese classifiers in young heritage speakers and majority language speakers

Abstract

Aims and Objectives:

This study investigated the production of classifiers in two groups of bilingual speakers: young heritage speakers of Cantonese in the USA and peers in Hong Kong, where Cantonese is the majority language. The role of language background factors in the heritage speakers’ performance was also examined.

Methodology:

The participants included 72 heritage speakers (3;10–11;3) and 61 majority language speakers (5;3–12;4). A picture-naming task was used to elicit the production of noun phrases containing six different target classifiers. A language background questionnaire was administered.

Data and Analysis:

The participants’ production was scored in terms of whether a classifier was produced in each obligatory context, and whether an appropriate form of classifier was used. The scores were analysed using multiple regression.

Findings:

Both groups were able to produce a classifier where required, but the heritage speakers were less accurate in selecting the appropriate classifier. There was a large range of abilities among the heritage speakers, and those with more Cantonese experience and a later age of arrival obtained higher scores.

Originality:

This study found both similarities and differences between heritage speakers and majority language speakers, and identified a potential trend of development in heritage speakers after age 10.

Significance/Implications:

This study showed that the classifier is acquired in heritage speakers, but its functional scope was narrowed to cover only grammatical function.

Keywords

Classifiers Cantonese heritage language bilingual development language acquisition

Introduction

Heritage speakers (HSs) are bilinguals who grow up speaking a language at home that is not the majority language (Pascual y Cabo & Rothman, 2012). HSs can be exposed to the majority language as early as from birth, or later when formal schooling begins. After this, HSs typically become dominant in the majority language gradually, often at the expense of the heritage language (HL; Montrul, 2016). HSs often differ systematically from monolinguals, and it has been proposed that HSs’ divergent grammar stems from delays in the developmental process (Benmamoun, Montrul, & Polinsky, 2013; Montrul, 2008). Since the HL is used only in limited contexts, HSs do not have the same opportunities as monolinguals to develop in their HL, particularly as soon as a child socialises outside the home or attends school. Some parents or older siblings may even use the majority language at home. Attrition can also play a role, where acquired linguistic knowledge is weakened or altered as speakers shift away from the HL (Flores, 2012), or if input providers themselves undergo attrition (Sorace, 2004). Due to inherently different acquisition contexts, the grammar developed in HSs is naturally unlike that of monolinguals (Montrul, 2016; Pascual y Cabo & Rothman, 2012; Pires & Rothman, 2009).

The competence of HSs varies greatly (e.g. Kupisch, Akpinar, & Stöhr, 2013; Montrul & Bowles, 2009), and contributing factors include the quantity and quality of input. More input in more domains has been linked to higher levels of proficiency (e.g. Gathercole & Thomas, 2009; Jia, 2008; Unsworth, 2013), and children with later age of arrival (AOA) benefit from receiving more input while in the homeland. They also immigrate with a more advanced and stable knowledge of their HL, which contributes to its maintenance (Allen, 2007). Differing degrees of HL literacy mean that HSs do not all benefit from exposure to written input (Pires & Rothman, 2009; Rinke & Flores, 2014), and sociolinguistic factors, including social economic status, cultural identity and family language policies, also play important roles (e.g. Jia, 2008; Jia & Paradis, 2015; Leung & Uchikoshi, 2012).

Not all morphosyntactic features pose problems for HSs (e.g. Pires & Rothman, 2009; Santos & Flores, 2016), but inflectional morphology has proven to be particularly difficult (e.g. Benmamoun et al., 2013; Song, O’Grady, Cho, & Lee, 1997). HSs have been reported to reduce and simplify morphological systems by lowering the number of distinctions and naturalising marked forms in favour of defaults (e.g. Montrul & Bowles, 2009; Polinsky, 2008; Polinsky & Kagan, 2007). While the studies above have revealed much about the underlying linguistic knowledge of HSs, many of them target inflectional features. Therefore, the present study examines classifiers in Cantonese, which is analytic and lacks inflections.

Cantonese classifiers

Cantonese is commonly spoken in southeastern China, including in Hong Kong as the majority language, and in other areas of Guangdong Province. There are also large overseas populations that speak it as a HL. The classifier (CL) is a morpheme involved in enumeration and quantification. When used with verbs, classifiers indicate how many times an event has occurred (cf. Chao, 1968). Numeral classifiers are used with nouns (Wu & Bodomo, 2009), and different nouns require different classifiers. Numeral classifiers are further categorised into sortal and mensural classifiers (Lyons, 1977; Matthews & Yip, 2011; Wu & Bodomo, 2009; see also Cheng & Sybesma, 1998, for the use of a count–mass distinction).

Typically, numeral classifiers are used before nouns (N) and following numerals, demonstratives and quantifiers (e.g. (1)), although bare nouns can be used when the reference is generic.

By more conservative estimates, there are around 60 numeral classifiers in Cantonese (Erbaugh, 2002; Li & Lee, 2001). Mensural classifiers classify nouns according to quantity, and are more similar to collective nouns in English. For example, 兜 dau1 is used for a basinful of any liquid, and 疊 daap6 for a pile of printed material. Sortal classifiers individuate nouns according to their features, such as shape or animacy. Most nouns are associated with only one sortal classifier, and using a different classifier is usually unacceptable.

The semantic content of classifiers often reflects certain characteristics of co-occurring nouns (Allan, 1977; Killingley, 1982, 1983; Tsang & Chambers, 2011). However, some classifier–noun pairings are opaque and cannot be learnt via semantics or prototypicality, to the extent that they appear arbitrary (Erbaugh, 1986; Matthews & Yip, 2011). For example, 條 tiu4 is used with long, thin objects, including snakes and queues but it would be considered ungrammatical to say *一條刀 jat1 tiu4 dou1, even though knives are long and thin. In most cases, classifier–noun associations cannot be violated for pragmatic effect, and deviant use of classifiers is highly conventionalised and infrequent (Killingley, 1986).

The sortal classifier 個 go3 occurs frequently, and is considered a general or default classifier (e.g. Chao, 1968; Erbaugh, 1986; Li & Thompson, 1981; Stokes & So, 1997). An equivalent exists in Mandarin and Taiwanese and holds a similar status (e.g. Erbaugh, 1986; Myers & Tsay, 2000). Go3 is associated with nouns with different characteristics, including those denoting small, roundish objects (e.g. apples, balls), human beings and abstract entities (e.g. ideas, dreams) (Matthews & Yip, 2011). It is the first acquired classifier in children, and even after other classifiers start emerging, go3 is still often preferred and overused past age 5–6, whereas the overuse of other classifiers (e.g. zek3) is infrequent (Mak, 1991; Poon, 1980; Szeto, 1998; Tse, Li, & Leung, 2007). However, unlike a truly general classifier, go3 cannot be used with all nouns (e.g. not with animals or utensils). Adult speakers also occasionally overuse go3, especially in informal speech, although such use is not perceived as ‘proper’ (Killingley, 1986; Li & Lee, 2001).

For Mandarin, Erbaugh (2002) found that even highly educated speakers overuse the equivalent ge frequently. Mandarin and Cantonese have largely similar classifier systems, but since Mandarin allows a more specific classifier to be substituted after the first mention of a noun while Cantonese does not, the results of Erbaugh’s (2002) study cannot be directly applied to Cantonese without further evidence. Go3 could be viewed as serving a placeholder function, especially during the early stages of acquiring classifier systems, when a learner focuses on mastering the relevant syntactic structures (Hu, 1993). Some researchers differentiate between the overuse of go3, which is considered non-standard but less unacceptable, and the overuse of other classifiers (e.g. Chan, 2011). Based on this contrast, in this study, a specific classifier will refer to any sortal classifier that is not go3 (this term has elsewhere been used for sortal classifiers that can be used with only one to two nouns; Myers & Tsay, 2000; Matthews &Yip, 2011).

Monolingual acquisition of sortal classifiers

In monolinguals, systematic use of classifiers in spontaneous speech is observed from around age 2. Syntactic errors, including omission and double use, are rare and disappear by age 3 in normally developing children (Poon, 1980; Szeto, 1998; Tse et al., 2007). Example (2) shows the double use of classifiers, produced by a child aged 2;04;30 (Szeto, 1998).

Initially, children use only a limited number of classifiers, such as zek3 and go3 (Szeto, 1998). The classifier repertoire expands rapidly after age 3, and accuracy in selecting classifiers also improves (Mak, 1991; Poon, 1980; Szeto, 1998; Tse et al., 2007). By age 6, children use at least 27 sortal classifiers productively, although selection of classifiers is not yet adult-like (Poon, 1980; Tse et al., 2007). As far as the author is aware, no study has shown when children converge on adult-like performance. Different theories have examined whether there is an order of acquiring classifiers, based on, for example, semantic diversity (Mak, 1991) and perceptual characteristics (Erbaugh, 1986; Tse et al., 2007). However, no definitive order has been established so far. The one common finding has been that go3 is the first and most frequent classifier to be used productively (e.g. Mak, 1991; Poon, 1980; Tse et al., 2007). For example, three children aged 2;07, 2;11, 4;05 in Poon (1980) used only go3 in an 89-item picture-naming task. Tse et al. (2007) found that children aged 3–6 used go3 more than 85% of the time during free play. Such overuse likely stems from having a syntactic motivation to use a classifier but possessing only an incomplete classifier system (Mak, 1991).

Bilingual acquisition of classifiers

Classifiers share some characteristics with other grammatical properties that HSs struggle with, including being a late morphosyntactic feature, requiring the integration of information from both the syntax and semantics domains, and involving unpredictable form–meaning mapping, in this case between classifiers and nouns (Laleko & Polinsky, 2013; O’Grady, Kwak, Lee, & Lee, 2011). Classifiers and nouns do not always co-occur, which compounds the negative effects of low input: classifiers can be used without the noun (like in (3)), and some noun phrases do not require a classifier at all (e.g. for an indefinite reading) (in (3), the second classifier still has to be appropriate for the dropped noun).

In HSs of Cantonese, classifier repertoires tend to be small, and speakers overuse or even use only go3 (Chan, 2011; Li & Lee, 2001). There are syntactic errors, including omission and using double classifiers, even in HSs who are past the age where monolinguals make these errors. However, older participants show more target-like use of classifiers, indicating that development can still occur in later childhood (Chan, 2011; Li & Lee, 2001). Classifier knowledge varies between individual HSs. For example, two participants in Li and Lee (aged 5 and 9) had only limited productive ability in Cantonese and did not produce a single classifier. Some of Chan’s participants made use of perceptual features to select classifiers, while others showed little awareness of the semantic nature of the classifiers. Chan also found that more frequent Cantonese use at home and positive attitudes towards one’s heritage contributed towards better performance.

Based on the above findings, the current study aims to further investigate the acquisition of Cantonese classifiers by comparing HSs to majority language speakers. The first research question asks whether the two groups have similar classifier knowledge. Since the syntax structure of classifiers is acquired earlier, HSs are expected to be more target-like in their syntax than in their classifier selection. The second research question asks what language background factors play a role in HSs’ knowledge of Cantonese classifiers. Previous research found that AOA (Jia & Paradis, 2015) and language experience (Chan, 2011) mediate the performance of individual speakers. This study will seek to corroborate these findings.

Methodology

Participants

HSs

Seventy-two HSs were recruited from three primary schools in New York City. They were all students in an after-school programme led by Chinese staff, and an initial survey to parents confirmed that predominantly Cantonese was used at home. The participants were aged 3;10–11;3 (mean = 8;7, SD = 1;8) (see also Supplemental Table 1 for the number of participants in each age group). The majority of the HSs were born in the USA (n = 44). Twenty-three were born in China, four in Hong Kong, and one in Mexico. Among those born in China, all who specified a town or province were born in areas where Cantonese is commonly spoken (e.g. Guangzhou). The AOA for participants born outside the USA (with data provided by 24 of 28 participants) ranged from 0;6 to 9;5, with a mean of 4;5 (SD = 2;8). The HSs’ Cantonese listening and speaking abilities were rated by their parents on a scale of 1–6, with 1 being ‘unable to understand/speak any Cantonese words’ and 6 being ‘able to understand everything appropriate to their age easily’ or ‘able to speak fluently in every situation’. On average, they received moderate ratings, with a mean of 4.22 (SD = 1.61) for listening and 3.93 (SD = 1.28) for speaking.

The HSs’ parents were all native speakers of Cantonese and had high self-reported proficiency; on the same scale of 1–6, they scored 5.59 (SD = .81) and 5.63 (SD = .34) for listening and speaking, respectively. The parents used mainly Cantonese with their children, with an average proportion of 78.85% (SD = 27.68%) by fathers and 74.64% (SD = 31.06%) by mothers. The participants themselves also used mainly Cantonese with their parents, on average 76.92% (SD = 30.40%) of the time with their fathers and 69.57% (SD = 30.58) with their mothers.

Twenty-six HSs reported speaking or hearing languages/varieties other than Cantonese at home, including English (n = 15), Mandarin (n = 9) and Taishanese/Hoisanwaa (n = 7). English was the main language used at school, with 55 HSs reporting that teachers used only English, and 47 reporting that they used only English with other students. Twenty-one participants attended extra-curricular Chinese classes, and 28 reported being literate in Chinese, defined in this study as being able to read/write simple texts of more than a few words.

Hong Kong speakers

Sixty-one children in Hong Kong were tested as a comparison group. They were all born in Hong Kong, apart from one who was born in China and had moved to Hong Kong at 0;5. They were aged 5;3–12;4 (mean = 9;3, SD = 1;10), and they were overall older than the HSs (t(122.36) = 2.27, p = .02 (see also Supplemental Table 1). They received high ratings for their Cantonese abilities from their parents, with a mean of 5.81 (SD = .71) for listening and 5.71 (SD = .70) for speaking. Their parents were all native speakers of Cantonese. At home, Cantonese was used most of the time, with parents using Cantonese on average 86.42% of the time (SD = 16.61%) and children using Cantonese 89.41% of the time (SD = 16.48%). Thirty-one spoke or heard languages/varieties other than Cantonese at home, including English (n = 20), Mandarin (n = 15), Teochew (n = 3) and Fujianese (n = 2).

Children in Hong Kong usually start learning English in kindergarten, and the study of English is compulsory throughout primary and secondary education. The majority of the Hong Kong participants were taught mostly or only using Cantonese at school (n = 42), and the majority used mostly or only Cantonese with other students (n = 57).

Written consent for testing and use of data collected was obtained from the parents of all participants before testing.

Instruments

Picture-naming task

A picture-naming task was used to elicit responses with the NUM + CL + N structure. Six sortal classifiers were targeted for their high frequency of use in children and association with frequent nouns (Tse et al., 2007). Only sortal classifiers were tested, as these classifiers are the least similar to collective nouns in English, and the noun phrases that contain them are concrete and can be easily represented in images.

Each target classifier was elicited using three objects, resulting in 18 different target objects (Table 1). These were all common objects likely to be known to young children and easily recognisable from an image. Related objects were avoided as much as possible in order to construct a more diverse set of stimuli. There is some variation in which classifiers are acceptable with a given noun, particularly because of regional differences. The target classifiers in Table 1 are not always the only ones acceptable for the respective target nouns. All acceptable classifiers, as judged by three native adult speakers, were scored as correct.

Table 1.

Target classifiers and associated objects.

Classifier	Description (Matthews & Yip, 2011)	Target object (English)	Target object (Cantonese)
個 go3	General and neutral classifier	Apple	蘋果 ping4gwo2
	General and neutral classifier	(Foot)ball	足球 / 波 zuk1kau4 / bo1
		(Paper) bag	(紙)袋 (zi2)doi2
張 zoeng2	For flat, horizontal objects	Bed	床 cong4
	For flat, horizontal objects	Chair	凳 dang3
		Table	檯 toi2
粒 lap1	For small objects	Peanut	花生 faa1sang1
		Strawberry	士多啤梨 si6do1be1lei2
		Button	鈕 nau2
隻 zek3	For animals and one of a pair	Hand	手 sau2
		Horse	馬 maa5
		Butterfly	蝴蝶 wu4dip2
條 tiu4	For long, thin objects and certain animals	Fish	魚 jyu2
		Trousers	褲 fu3
		Key	鎖匙 so2si4
把 baa2	For tools and instruments	Scissors	鉸剪 gaau3zin2
		Knife	刀 dou1
		(Beach) umbrella	(沙灘)遮 (saa1taan1) ze1 / (太陽)傘 (taai3joeng4) saan3

The task was presented on an 8” screen tablet using Opensesame (Mathôt, Schreij, & Theeuwes, 2012). In each trial, a circle and a square appeared side-by-side, each containing two, three or four colour images of the target object (the images were from Brodeur, Dionne-Dostie, Montreuil, & Lepage, 2010; Moreno-Martínez & Montoro, 2012). Participants were instructed to describe what was in the circle. The number of objects in the circle was different from the number in the square, so as to elicit a numeral in the response (Eisenbeiss, 2009), which in turn required a classifier. As an example, in a trial the screen might show a square containing two butterflies, and a circle containing three butterflies. The target response would be 三隻蝴蝶 saam1 zek3 wu4dip2 ‘three butterflies’. Participants were instructed to tap the screen once they had completed their response, and the next trial would begin.

There were two blocks of trials, and each target object was presented once in each block. There were two pseudo-randomised lists for each block to counterbalance the order of presentation across participants. The lists ensured that the target number, the target classifier and the position of the circle on the screen were not repeated in consecutive trials. A training task was conducted using a different set of stimuli. Responses were recorded using a desktop microphone connected to a handheld digital recorder.

Language background questionnaire

Data on the participants’ language background was collected using a language background questionnaire (LBQ) for parents, written in Chinese. The questionnaire was adapted from the BiLingual Language Experience Calculator (BiLEC) (Unsworth, 2013), and was distributed to participants to take home after testing. The questionnaire targeted information on the participants’ background, parents’ background and the participants’ language use. Some questions were only applicable to HSs, such as AOA and whether or not they were literate in Chinese. A shorter, English, oral version of the questionnaire was also administered to the participants. Eleven HSs did not return their questionnaire, so their own responses were used in the analysis.

Results

Scoring

The participants’ performance was evaluated based on two aspects of using classifiers. Firstly, grammatical accuracy referred to whether a classifier was used in an obligatory context (i.e. when a response contained a number and a noun). Responses containing non-Cantonese nouns were excluded from the calculation. Selection accuracy referred to whether, when a classifier was used, it was appropriate for the noun. Responses containing non-target Cantonese nouns were included as long as the appropriate classifier for the actual response was the same as for the target noun. For example, the utterance 三隻牛 saam1 zek3 ngau4 ‘three cows’ in response to a picture of three horses was included because ‘horse’ and ‘cow’ require the same classifier (zek3). If a classifier was used with a non-Cantonese noun, the Cantonese equivalent was used to determine the appropriate classifier.

Multiple regression analyses were conducted to compare the grammatical and semantic accuracy of the two groups of participants. Binomial generalised linear mixed models were fitted using the lme4 package (Bates, Maechler, Bolker, & Walker, 2015) in R (R Core Team, 2016). Participants were entered as Subject, and the different target objects as Object, both as random variables. The dependent variable was accuracy on each trial. Models were fitted by adding predictors one by one and retaining only significant predictors. Likelihood ratio tests and Akaike information criteria (AICs) were used to obtain final models with the best fit.

Grammatical accuracy

In terms of grammatical accuracy, the Hong Kong group (HK group) performed at ceiling, with only two participants not scoring 100% (each omitting one classifier). The mean score for the HSs was also near ceiling (97.07%, SD = 10.28%), but the range of scores was much larger. Forty-one out of 72 participants scored 100%, 30 scored between 63% and 97% and one scored 31%. One HS omitted classifiers in half of his responses, but the proportion of omission was lower for other participants (at most 23%). In total, the HSs produced 45 responses without classifiers, that is, only NUM + N. Regression analyses revealed no significant effect of Group on grammatical accuracy scores (p = .17), indicating no overall difference between the two groups. As the HK group was overall older than the HSs, the two groups were also compared using age-matched sub-groups, with 52 participants from each group. The average age difference between each age-matched HS–HK pair was 2 months 23 days. Again, regression analyses revealed no significant differences between the two groups in terms of grammatical accuracy (p = .11).

Some HSs used classifiers with nouns in Mandarin or other Chinese varieties, which have similar classifier rules. Classifiers were also used with English nouns, some of which were inflected for plural (e.g. (4) by a HS). One HK participant used English in five trials, while 16 HSs used English nouns in 81 trials.

Selection accuracy

Table 2 shows the selection accuracy scores of the two groups. Overall, the HK group performed well, with eight participants scoring 100% and 27 scoring 91.43–97.22%. The HSs had lower scores, with only 12 scoring above 50%.

Table 2.

Selection accuracy scores (%).

	Heritage speaker group				Hong Kong group
	Mean	SD	Min	Max	Mean	SD	Min	Max
Overall	32.52	20.45	9.09	93.33	90.16	10.51	55.56	100
go3	98.92	5.60	60	100	96.17	9.33	50.00	100
zoeng1	21.13	35.74	0	100	94.81	13.10	33.33	100
baa2	17.74	30.29	0	100	93.72	17.26	0	100
tiu4	18.50	33.24	0	100	86.01	20.91	0	100
lap1	5.12	16.19	0	83.33	75.30	28.20	0	100
zek3	19.69	29.31	0	100	94.54	9.96	66.67	100

Figure 1 shows that performance was not equal for the different classifiers. Almost all the HSs scored 100% for go3 but 0% on all specific classifiers. The HSs scored the lowest (between 0% and 10%) on all three target objects requiring lap1, while the HK participants scored lowest (<80%) on ‘strawberry’ and ‘peanut’, which required lap1, and ‘key’, which required tiu4. These results suggest that knowledge of lap1 was particularly poor, regardless of target object.

Figure 1.

Distribution of selection accuracy scores.

The selection accuracy of the two groups was compared using regression analysis. The HK group was on average older and there was a significant correlation between semantic accuracy and age of testing (AOT) in the HK group (r = .40, p < .001), but not in the HSs (p = .18). Therefore, only age-matched sub-groups were used for this analysis. The final model is shown in Table 3 (conditional R² = .36, marginal R² = .82), and the results showed that HSs were less accurate at selecting classifiers (β = −5.33, SE = .39, p < .001).

Table 3.

Model for selection accuracy.

	Estimate	Std. error	z-value	p-value
(Intercept)	3.98	.64	6.21	<.001
Group	−5.33	.39	−13.83	<.001

Classifier substitution

Table 4 shows the types of substitutes used in the responses containing non-target classifiers. The substitute classifiers used by each group are listed in Supplemental Table 2.

Table 4.

Number of responses containing different types of substitution.

Noun type	Number of responses/Type of substitute				Non-classifier words	Double classifier construction
	Total	go3	Specific classifier	Mensural classifier	Non-classifier words	Double classifier construction
Heritage speaker group	1021	932 (91.28%)	18 (1.76%)	4 (.39%)	4 (.39%)	63 (6.17%)
Cantonese
Non-Cantonese Chinese	29	28 (96.55%)	1 (3.45%)
Hong Kong group
Cantonese	212	176 (83.02%)	32 (15.09%)	4 (1.89%)

Both groups relied heavily on go3, while substitution using specific classifiers was infrequent. As the results above showed that most HSs did not produce any classifier other than go3, it appears that they were using go3 indiscriminately regardless of which classifier was appropriate for the context.

Twenty-one HSs produced a total of 63 responses containing double classifier constructions (e.g. (5), (6)), with go3 inserted before the target specific classifier. This construction was restricted to seven target objects, including ‘knife’ and ‘umbrella’ (requiring baa2), ‘horse’ (requiring zek3), ‘fish’ (requiring tiu4) and ‘bed’, ‘chair’ and desk’ (requiring zoeng1).

In a minority of cases, a specific classifier was used as a substitute for the target classifier. In 18 of the HK participants’ responses and seven of the HSs’, the substitutions can be viewed as extensions of the semantic scope of the chosen classifier. In (7), the HS used zek3 for the target object ‘fish’. Since zek3 is used for animate objects (albeit usually four-legged animals), the participant’s choice of classifier matched an important characteristic of the target object. Other substitutions were less logical semantically. For example, zi1 and zek3 in (8) and (9) (both by HSs) usually refer to long, thin objects and animate objects, respectively, which are not congruent with the target objects.

Nineteen HK participants also used 把 baa2 with ‘key’, which is a pairing used in Mandarin. Since Mandarin is taught in schools in Hong Kong, the HK participants may have transferred this pairing from Mandarin to Cantonese.

Predictors of HSs’ selection accuracy

Multiple regression analyses were used to investigate which language background factors predicted selection accuracy among the HSs. Trials targeting go3 were excluded from this analysis as scores for go3 were at ceiling. The following predictors were considered:

AOT;

Cantonese experience (obtained by taking the mean of four measurements from the LBQ: the proportion of Cantonese used by the participant with each parent, and vice versa; α = .90);

Chinese literacy;

Cantonese listening ability, as rated by parents;

Cantonese speaking ability, as rated by parents.

The final model (Table 5) shows that the HSs who used more Cantonese with their parents were more accurate at selecting classifiers (β = −2.97, SE = 1.44, p = .04), while other tested factors had no predictive effects on scores (ps > .05). The overall model fit was conditional R² = .20, marginal R² = .76.

Table 5.

Final model predicting selection accuracy for heritage speakers.

	Estimate	Std. error	z-value	p-value
(Intercept)	−1.57	1.19	−1.32	.19
Cantonese experience	−2.97	1.44	−2.06	.04 *

Significant at the .05 level.

Considering only the HSs born outside the USA, a separate analysis on AOA and AOT (Table 6) showed a significant effect of AOA only (β = 0.78, SE = .26, p = .003), indicating that later arrivals selected classifiers more accurately independent of their AOT.

Table 6.

Results of multiple regression on age of arrival (AOA) and age of testing (AOT).

	Estimate	Std. error	z-value	p-value
(Intercept)	−3.62	2.48	−1.46	.14
Age of arrival	.78	.26	3.00	.003 **
Age of testing	−.34	.34	−1.00	.32

Significant at the .01 level.

While AOT did not predict selection accuracy, Figure 2 shows that the HSs progressed towards higher accuracy after age 10. In this figure, individual participants’ scores for each target classifier were plotted against AOT, and a loess curve was fitted for each classifier with a span setting of α = .70. In comparison, there was less change in the HK participants’ scores after around age 8. Individual differences can be observed for both groups, with many participants’ scores deviating from the group trend.

Figure 2.

Selection accuracy scores for different classifiers across the age of testing range.

Discussion

This study compared young HSs of Cantonese to peers from Hong Kong, and found that the two groups were similarly accurate in producing a classifier where required, even though the HSs were acquiring Cantonese in a minority language context. Individually, there was a wide range of performance, with one HS scoring as low as 31%. Previous research found that omission occurred only in monolinguals younger than age 5 (Poon, 1980; Szeto, 1998), or age 8 for HSs (Li & Lee, 2001). In the current study, however, HSs up to age 11;1 omitted classifiers, showing that HSs can diverge significantly from majority language speakers in their morphosyntactic development, so that age norms for other speaker populations (even bilingual ones) cannot be readily applied to HSs. It is possible that some HSs may continue to omit classifiers and not converge on the norm of using the NUM + N + CL structure.

Classifiers may be considered relatively salient features due to their frequency and predictability (irrespective of form); therefore, their omission in older children is surprising. Omissions could be explained as performance errors, or be indicative of poor productive ability: consistent omission has so far not been observed in adolescent or adult HSs but is common in very young (monolingual) children (<2 years). For example, a speaker struggling to retrieve the appropriate lexical items while naming an object may seek only to name the essential, meaning-bearing elements of the numeral and the noun. It remains to be seen whether classifier usage is related to vocabulary knowledge.

Double classifiers reflect the opposite problem of omission. Double classifiers were used mostly with utility items or furniture, such as ‘knife’, ‘bed’ and ‘chair’. Since these objects are used in daily life in the generic sense or with definite reference (10), where no numeral is used, the HSs may have heard classifiers used without a numeral so frequently that they interpreted the classifier and noun to form a single noun, and added go3 as the classifier during production (Matthews & Yip, 2011; Poon, 1980).

Fourteen of the 22 HSs who used double classifiers used only go3 otherwise, which indicates an inability to produce specific classifiers. Follow-up conversations revealed that most of the participants who used double classifiers considered the specific classifier as part of the noun representing the relevant objects. As such, the one-noun phrase interpretation is applicable to the present data. In addition, the discovery of both omission and double classifiers in a group of HSs sharing geographical and social proximity reflects the heterogeneity of HS populations.

In terms of classifier selection, while the HK participants had high accuracy, most HSs were only able to produce go3, and used specific classifiers infrequently or not at all. There was an overall difference between the two groups, even after matching for age. The HSs’ high scores on trials targeting go3 may be because go3 was not just the default, but in fact the only form of classifier acquired. The reduction in the number of different forms, in some cases to only the default, is not unlike that observed in HSs for other morphosyntactic systems, including inflectional ones (e.g. Polinsky, 2008; Song et al., 1997). Some HSs also used mensural classifiers, showing a type of cross-category substitution that has not been observed in the HK participants in this study or in monolinguals in previous research (Tse et al., 2007).

Within the age range tested, there was no indication that classifier knowledge became lost or inaccessible (Figure 2), possibly since there was no directly competing system in English. After age 10, sufficient input might have been accumulated from various sources, including family members and Chinese classes, for development to ‘resume’. Therefore, although AOT had no predictive effect on the HSs’ selection accuracy, the HSs may eventually become more like their HK group peers at a later age.

Input

In this study, the HSs were similar to the HK group only in using (any) classifiers where required, but not in selection. The earlier-developing grammar may have been acquired by the time the HSs attended school, whereas the classifier repertoire was still expanding and was therefore affected by the shift towards English; faced with reduced input, HSs often develop more slowly, leading to non-target-like acquisition (e.g. Rinke & Flores, 2014; Unsworth, 2013). Even though the HSs in this study used mainly Cantonese at home, just like the HK speakers, there were other sources of input differences, such as the English-dominant society in the USA.

Reduced input affects the development of classifiers strongly, because the form of classifier is not always predictable from the perceptual features of the following noun, so large amounts of input are required to form classifier–noun pairings. The HSs might also have received input that was qualitatively divergent from the homeland variety, if attrition had occurred in the Cantonese of their parents or other input providers (Sorace, 2004). Parents’ classifier knowledge was not investigated in the current study, but Lo and Nagy (2016), using conversational data from the Heritage Language Variation and Change in Toronto Corpus (Nagy, 2011), reported no simplification pattern in first and second generation immigration speakers compared to homeland speakers. However, a simple analysis of 10 speakers’ data in this corpus (five from each generation) revealed that the majority of instances where a classifier was required called for only the default go3 or the plural di1, so the results are not directly applicable.

The effects of input also operated at a within-group level, as the HSs with more Cantonese exposure, whether through use with their parents or by virtue of immigrating later, selected classifiers more accurately. Concerning AOA, the pre-immigration period of living in a majority language environment offers maximal Cantonese input or even schooling in Cantonese, leading to advantages in acquiring a frequency-dependent grammatical structure. These results are consistent with previous findings that show the benefit of more input and later AOA for target-like acquisition of HLs (e.g. Gathercole & Thomas, 2009; Jia & Paradis, 2015; Montrul, 2008; Unsworth, 2013).

Cross-linguistic influence

The lack of sortal classifiers in English did not appear to lead to loss in Cantonese. Perhaps collective nouns in English, which are similar to mensural classifiers in Cantonese, even ensured the maintenance of classifiers. However, there was an unexpected case of cross-linguistic influence, where the participants applied a typically Mandarin classifier–noun pairing to Cantonese. Some schools in Hong Kong have started teaching Chinese in Mandarin as opposed to in Cantonese, so young speakers in Hong Kong could have gained proficiency in Mandarin and transferred this choice of classifier. In contrast, adult speakers in Hong Kong seemed less inclined to accept the use of baa2 with ‘key’; following the findings above, informal interviews were conducted with 18 adult native Cantonese speakers in Hong Kong, and while speakers older than 40 years (n = 8) categorically rejected the use of baa2 with ‘key’, younger adults (n = 10) preferred tiu4 but accepted baa2 in writing. This suggests cross-generational differences due to increased influence from Mandarin, but it is also possible that the HK participants had merely confused 鎖 so2 ‘lock’ (which requires baa2) and 鎖匙 so2si4 ‘key’.

For the HSs who also spoke or heard some Mandarin or Taishanese at home (n = 15), there was a possibility that Mandarin or Taishanese classifier choices had been transferred to Cantonese. Among the responses containing inappropriate classifiers, none of the CL + N combinations produced are permitted in Mandarin, although substituting a specific classifier with the default is common in native Mandarin speakers (Erbaugh, 2002). Six HSs who used some Taishanese at home produced two CL+N combinations acceptable in Taishanese: go3 with 士多啤梨 si6do1be1lei2 ‘strawberry’ (alternatively 草莓 cou2mui2 ‘strawberry’) as well as go3 with 馬 maa5 ‘horse’, so these HSs’ lower selection accuracy (average: 37.84%) could be explained by direct transfer. Since using go3 with these two nouns as a default substitute for specific classifiers is indistinguishable on the surface from selecting it due to influence from Taishanese, it is not conclusive that cross-linguistic influence in this manner had affected selection accuracy. Nevertheless, it is important to consider what other languages and varieties HSs are exposed to, and to investigate whether they can be a source of divergence in HL production.

Conclusion

If the semantic content of specific classifiers is viewed as a reflection of the characteristics of associated nouns (and so as somewhat redundant), then the default classifier can be used in place of specific classifiers without compromising the overall conveyed meaning, and perhaps even reduce the cognitive burden of processing. Such a possibility is particularly important for HSs, who might not all receive adequate input to learn unpredictable classifier–noun pairings. Morphosyntactic features have been found to pose difficulty for HSs, and the current study has found classifiers to be no exception.

While the data so far indicates that the HSs diverge from majority language speakers in terms of classifier repertoire, their ability to use classifiers where required reflects a degree of mastery of Cantonese, which would enable them to communicate with other speakers. In addition, further development at a later age should not be precluded, especially in speakers who pursue studies in the HL or who move to the homeland.

The present study was limited to the production of six sortal classifiers, so further research could investigate comprehension as well as other types of classifiers. However, investigating classifiers or even Cantonese as a HL is not straightforward: HSs grow up among linguistically diverse speakers using different Chinese dialects and varieties, all of which are sufficiently dissimilar, such that it is not easy to attribute certain patterns of language use to, for example, dialectal variation, divergent development or cross-linguistic transfer.

Supplemental Material

IJB808002_Supplemental_Material_REV2 – Supplemental material for Production of Cantonese classifiers in young heritage speakers and majority language speakers

Supplemental material, IJB808002_Supplemental_Material_REV2 for Production of Cantonese classifiers in young heritage speakers and majority language speakers by Rachel TY Kan in International Journal of Bilingualism

Footnotes

Acknowledgements

The author is deeply grateful to Monika S. Schmid for her guidance and feedback in the preparation of this article, and to Naomi Nagy for providing data from the Heritage Language Variation and Change in Toronto Corpus.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by Funds for Women Graduates.

Supplemental Material

Supplemental material for this article is available online.

Author biography

Rachel TY Kan is a Lecturer at the Education University of Hong Kong. Her research focusses on language development in multilingual children and speech sounds in language contact.

References

Allan

(1977). Classifiers. Language, 53(2), 285–311.

Allen

S. E. M.

(2007). The future of Inuktitut in the face of majority languages: Bilingualism or language shift? Applied Psycholinguistics, 28, 515–536. doi:10.1017.S0142716407070282

Bates

Maechler

Bolker

Walker

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1–48. doi:10.18637/jss.v067.i01

Benmamoun

Montrul

Polinsky

(2013). Defining an “ideal” heritage speaker: Theoretical and methodological challenges. Theoretical Linguistics, 39(3–4), 259–294. doi:10.1515/tl-2013-0018

Brodeur

M. B.

Dionne-Dostie

Montreuil

Lepage

(2010). The Bank of Standardized Stimuli (BOSS), a new set of 480 normative photos of objects to be used as visual stimuli in cognitive research. PLoS One, 5(5), e10773. doi:10.1371/journal.pone.0010773

Chan

J. M. S.

(2011). Cantonese classifiers and the British-born Chinese children of Ap Chau (Unpublished MSc thesis). Oxford: University of Oxford.

Chao

Y.-R.

(1968). A grammar of spoken Chinese. Berkeley, CA: University of California Press.

Cheng

L. L.-S.

Sybesma

(1998). Yi-wan tang, Yi-ge tang: Classifiers and massifiers. The Tsing Hua Journal of Chinese Studies, New Series, 28(3), 285–412.

Eisenbeiss

(2009). Contrast is the name of the game: Contrast-based semi-structured elicitation techniques for studies on children’s language acquisition. Essex Research Reports in Linguistics, 57(7), 1–27.

10.

Erbaugh

M. S.

(1986). Taking stock: The development of Chinese noun classifiers historically and in young children. In Craig

(Ed.), Noun classes and categorization (pp. 399–436). Amsterdam/Philadelphia: John Benjamins.

11.

Erbaugh

M. S.

(2002). Classifiers are for specification: Complementary functions for sortal and general classifiers in Cantonese and Mandarin. Cahiers de Linguistique - Asie Orientale, 31(1), 33–69. doi:10.1163/19606028–90000098

12.

Flores

(2012). Differential effects of language attrition in the domains of verb placement and object expression. Bilingualism: Language and Cognition, 15, 550–567. doi:10.1017/S1366728911000666

13.

Gathercole

V. C. M.

Thomas

E. M.

(2009). Bilingual first-language development: Dominant language takeover, threatened minority language take-up. Bilingualism: Language and Cognition, 12, 213–237. doi:10.1017/S1366728909004015

14.

(1993). The acquisition of Chinese classifiers by young Mandarin speaking children (Unpublished doctoral dissertation). Boston, MA: Boston University.

15.

Jia

(2008). Heritage language development, maintenance, and attrition among recent Chinese immigrants in New York City. In He

A. W.

Xiao

(Eds.), Chinese as a heritage language: Fostering world citizenry (pp. 189–204). Honolulu, HI: University of Hawaii Press.

16.

Jia

Aaronson

(2003). A longitudinal study of Chinese children and adolescents learning English in the United States. Applied Psycholinguistics, 24, 131–161. doi:10.1017/S0142716403000079

17.

Jia

Paradis

(2015). The use of referring expressions in narratives by Mandarin heritage language children and the role of language environment factors in predicting individual differences. Bilingualism: Language and Cognition, 18(4), 737–752. doi:10.1017/S1366728914000728

18.

Killingley

S.-Y.

(1982). A short glossary of Cantonese classifiers. Newcastle upon Tyne: Grevatt & Grevatt.

19.

Killingley

S.-Y.

(1983). Cantonese classifiers: Syntax and semantics. Newcastle upon Tyne: Grevatt & Grevatt.

20.

Killingley

S.-Y.

(1986). Normal and deviant classifier usage in Cantonese. Anthropological Linguistics, 28(3), 321–336. doi:10.1177/000134558801600101

21.

Kupisch

Akpinar

Stöhr

(2013). Gender assignment and gender agreement in adult bilingual and second language speakers of French. Linguistic Approaches to Bilingualism, 3(2), 150–179. doi:10.1075/lab.3.2.02kup

22.

Laleko

Polinsky

(2013). Marking topic or marking case: A comparative investigation of heritage Japanese and heritage Korean. Heritage Language Journal, 10(2), 40–64. doi:10.1017/CBO9781107415324.004

23.

Leung

Uchikoshi

(2012). Relationships among language ideologies, family language policies, and children’s language achievement: A look at Cantonese-English bilinguals in the U.S. Bilingual Research Journal, 35(3), 294–313. doi:10.1080/15235882.2012.731588

24.

C. N.

Thompson

S. A.

(1981). Mandarin: A functional reference grammar. Berkeley, CA: University of California Press.

25.

Lee

(2001). L1 development in an L2 environment: The use of Cantonese classifiers and quantifiers by young British-born Chinese in Tyneside. International Journal of Bilingual Education and Bilingualism, 4(6), 359–382. doi:10.1080/13670050108667738

26.

Nagy

(2016). Variable use of Heritage Cantonese classifiers. Paper presented at New Ways of Analyzing Variation 45, University of Victoria/Simon Fraser University, Vancouver.

27.

Lyons

(1977). Semantics. Cambridge, MA: Cambridge University Press.

28.

Mak

D. L. W.

(1991). The acquisition of classifiers in Cantonese (Unpublished doctoral dissertation). Reading, England: The University of Reading.

29.

Mathôt

Schreij

Theeuwes

(2012). OpenSesame: An open-source, graphical experiment builder for the social sciences. Behavior Research Methods, 44(2), 314–324. doi:10.3758/s13428–011–0168–7

30.

Matthews

Yip

(2011). Cantonese: A comprehensive grammar. 2nd edition. London: Routledge.

31.

Montrul

Bowles

(2009). Back to basics: Incomplete knowledge of differential object marking in Spanish. Bilingualism: Language and Cognition, 12, 363–383. doi:10.1017/S1366728909990071

32.

Montrul

S. A.

(2016). The acquisition of heritage languages. Cambridge: Cambridge University Press.

33.

Montrul

S. A.

(2008). Incomplete acquisition in bilingualism: Re-examining the age factor. Amsterdam: John Benjamins.

34.

Moreno-Martínez

F. J.

Montoro

P. R.

(2012). An ecological alternative to Snodgrass & Vanderwart: 360 high quality colour images with norms for seven psycholinguistic variables. PLoS ONE, 7(5), e37527. doi:10.1371/journal.pone.0037527

35.

Myers

Tsay

(2000). The acquisition of the default classifer in Taiwanese. Paper presented at the Seventh International Symposium on Chinese Languages and Linguistics, National Chung Cheng University, Taiwan.

36.

Nagy

(2011). A multilingual corpus to explore variation in language contact situations. Rassegna Italiana Di Linguistica Applicata, 43(1/2), 65–84.

37.

O’Grady

Kwak

H.-Y.

Lee

O.-S.

Lee

(2011). An emergentist perspective on heritage language acquisition. Studies in Second Language Acquisition, 33(2), 223–245. doi:10.1017/S0272263110000744

38.

Pascual

Cabo

Rothman

(2012). The (il)logical problem of heritage speaker bilingualism and incomplete acquisition. Applied Linguistics, 33(4), 450–455. doi:10.1093/applin/ams037

39.

Pires

Rothman

(2009). Disentangling sources of incomplete acquisition: An explanation for competence divergence across heritage grammars. International Journal of Bilingualism, 13(2), 211–238. doi:10.1177/1367006909339806

40.

Polinsky

(2008). Relative clauses in heritage Russian: Fossilization or divergent grammar? Formal Approaches to Slavic Linguistics, 16, 333–358.

41.

Polinsky

Kagan

(2007). Heritage languages: In the “wild” and in the classroom. Language and Linguistics Compass, 1(5), 368–395. doi:10.1111/j.1749–818X.2007.00022.x

42.

Poon

E. Y.

(1980). Some aspects of the ontological development of nominal classifiers in Cantonese (Unpublished MA thesis). Hong Kong: University of Hong Kong.

43.

R Core Team. (2016). R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. Retrieved from https://www.r-project.org/

44.

Rinke

Flores

(2014). Morphosyntactic knowledge of clitics by Portuguese heritage bilinguals. Bilingualism: Language and Cognition, 17(4), 681–699. doi:10.1017/S136672891300076X

45.

Santos

A. L.

Flores

(2016). Comparing heritage speakers and late L2-learners of European Portuguese: Verb movement, VP ellipsis and adverb placement. Linguistic Approaches to Bilingualism, 6(3), 308–340. doi:10.1075/lab.14006.san

46.

Song

O’Grady

Cho

Lee

(1997). The learning and teaching of Korean in community schools. In Kim

Y.-H.

(Ed.), Korean language in America (Vol. 2, pp. 111–127). Los Angeles, CA: American Association of Teachers of Korean.

47.

Sorace

(2004). Native language attrition and developmental instability at the syntax-discourse interface: Data, interpretations and methods. Bilingualism: Language and Cognition, 7(2), 143–145. doi:10.1017/S1366728904001543

48.

Stokes

S. F.

L. K. H.

(1997). Classifer use by language-disordered and age-matched Cantonese-speaking children. Asia Pacific Journal of Speech, Language, and Hearing, 2, 83–101. doi:10.1179/136132897805577413

49.

Szeto

K. K.

(1998). The acquisition of Cantonese classifiers (Unpublished MPhil thesis). Hong Kong: The University of Hong Kong.

50.

Tsang

Chambers

C. G.

(2011). Appearances aren’t everything: Shape classifiers and referential processing in Cantonese, 37(5), 1065–1080. doi:10.1037/a0023601

51.

Tse

S. K.

Leung

S. O.

(2007). The acquisition of Cantonese classifiers by preschool children in Hong Kong. Journal of Child Language, 34(3), 495–517. doi:10.1017/S0305000906007975

52.

Bodomo

(2009). Classifiers ≠ determiners. Linguistic Inquiry, 40, 487–503.

53.

Unsworth

(2013). Assessing the role of current and cumulative exposure in simultaneous bilingual acquisition: The case of Dutch gender. Bilingualism: Language and Cognition, 16, 86–110. doi:10.1017/S1366728912000284

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.34 MB