CORE LINGUISTIC PROPERTIES

Developmental psycholinguists have demonstrated young children's mastery of a variety of linguistic principles. These principles include ones that draw upon core syntactic and semantic properties. For a linguistic property to be a viable candidate to be considered part of core grammar, one expectation is that it should tie together several apparently unrelated linguistic phenomena. We will illustrate this hallmark of core linguistic properties using a syntactic principle, Principle C of the Binding Theory (Chomsky, Reference Chomsky1981).

The Binding Theory is a set of three principles that govern the anaphoric relations that can be established between different kinds of noun phrases and their potential antecedents (other noun phrases). Principle C is the third principle of the Binding Theory, which is about the anaphoric relations that pertain to referring expressions. Referring expressions include proper names, such as Max, as well as definite descriptions, such as the director. Principle C asserts that referring expressions must be ‘free’, where ‘free’ means that referring expressions must be distinct in reference from any antecedent noun phrase that bears a particular structural relation with them. The structural relation is called c-command. Roughly, one noun phrase A c-commands another noun phrase B if there is a path (in constituent structure) that proceeds upwards from A to the first branching node, and then downwards to B. Setting details aside, the sentences in (1) can be used to illustrate where Principle C does and does not apply.

1. (1)

   1. a. He said Max was drinking wine.

   2. b. Max said he was drinking wine.

   3. c. When he walked in, Max was drinking wine.

Principle C applies in example (1a). Here, Principle C governs the referring expression, Max, because the pronoun he c-commands the name Max in the example. Since Principle C is operative, he and Max must pick out different individuals. So, the pronoun he in (1a) cannot refer to the person named Max. Notice that Principle C establishes when a referring expression can not pick out the same individual as some antecedent noun phrase does. Negative statements like this are often referred to as constraints. Principle C is a constraint that enforces disjoint reference (or non-co-reference).

This means that when Principle C is not operative, as in examples (1b) and (1c), the name Max is permitted to be anaphorically linked to the pronoun he. In these examples, the pronoun he and the name Max may be used to pick out the same individual, because the pronoun he does not c-command the name Max in either of the examples. But, in addition, the pronoun he can be used to pick out some male individual who is salient in the context, or who has been introduced earlier. The consequence is that examples (1b) and (1c) are ambiguous. These examples allow two interpretations. On one, the pronoun he and the name Max are said to be anaphorically linked (or co-referential); on the other, he and Max are said to have disjoint reference (or non-co-referential). By contrast, example (1a) is unambiguous, due to Principle C. This illustrates the ‘constraining’ property of Principle C. It eliminates an interpretation that could otherwise be assigned to the sentence (1a), in the absence of Principle C.

As mentioned, one hallmark of a core linguistic property is that it unifies apparently disparate phenomena. Principle C meets this criterion. It is manifested in several diverse constructions. As well as applying to statements, as in (1a), Principle C governs the interpretations that can be assigned to wh-questions, as in (2). And it also governs the interpretation of statements with quantificational expressions, as in (3) (these examples are adapted from Chomsky, Reference Chomsky1981).

1. (2) Who did he say Max had criticized (for which x, he said

   Max had criticized x)

1. (3) He said Max had criticized everyone (for every x, he said

   Max had criticized x)

Notice that the pronoun he in examples (2) and (3) must refer to a single male individual, and cannot be interpreted as a ‘bound’ pronoun. By comparison, consider examples (4) and (5). In these examples, the wh-word who and the quantificational expression everybody can be anaphorically linked to the pronoun he because Principle C is not operative in these examples.

1. (4) Who said he had criticized Max (for which x, x said

   x had criticized Max)

2. (5) Everybody said he had criticized Max (for every x, x said

   x had criticized Max)

In short, Principle C provides a unified account of a range of apparently disparate linguistic phenomena. It applies to declaratives, wh-questions, and to quantificational expressions.

Two other hallmarks of core linguistic properties were noted earlier, universality and early emergence. These aspects of Principle C have been investigated in the literature, and Principle C has been found to be operative early in the course of language development, and across languages (see Crain & McKee, Reference Crain, McKee, Berman, Choe and McDonough1985; Crain & Thornton, Reference Crain and Thornton1998; Conroy & Thornton, Reference Conroy, Thornton and Otsu2005; Kazanina & Phillips, Reference Kazanina, Phillips, Anna, Do and Johansen2001; Kiguchi & Thornton, Reference Kiguchi and Thornton2004, among others).

DOWNWARD ENTAILMENT IN CHILD LANGUAGE: PREVIOUS RESEARCH

Previous child experiments on downward entailment provide separate reports on children's interpretation of disjunction and their understanding of the licensing condition of negative polarity items like any. A series of cross-linguistic studies have investigated children's interpretation of disjunction in the scope of downward entailing operators, including not (Crain, Gardner, Gualmini & Rabbin, Reference Crain, Gardner, Gualmini, Rabbin and Otsu2002; Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004; Jing, Crain & Hsu, Reference Jing, Crain, Hsu and Otsu2005), none (Gualmini & Crain, Reference Gualmini, Crain, Skarabela, Fish and Do2002), before (Notley, Zhou, Jensen & Crain, Reference Notley, Zhou, Jensen and Crainin press) and the subject phrase of every (Boster & Crain, Reference Boster, Crain and Smith1993; Gualmini, Meroni & Crain, Reference Gualmini, Meroni, Crain and Weisgerber2003; Su & Crain, Reference Crain and Bavin2009). It has been found, across typologically different languages, that young children generate the conjunctive entailment when disjunction appears in the scope of downward entailing operators, whereas they assign disjunctive truth conditions to disjunction when it appears in non-downward entailing contexts (see Crain, Reference Crain2008, for a review).

One source of empirical evidence of children's knowledge of downward entailment comes from studies of three- to five-year-old children's understanding of disjunction words in simple negative sentences. Studies have been conducted in English, Japanese and Mandarin Chinese (Crain et al., Reference Crain, Gardner, Gualmini, Rabbin and Otsu2002; Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004; Jing et al., Reference Jing, Crain, Hsu and Otsu2005). Interestingly, in two of these languages, Chinese and Japanese, children's interpretations differed from those of adults. For adult speakers of Japanese and Chinese, words for disjunction are not interpreted as being within the scope of negation in simple negative sentences (see footnote 1). For example, in the Jing et al. study (Reference Jing, Crain, Hsu and Otsu2005), Mandarin-speaking adults were found to accept the sentence Tanglaoya meiyou juqi zhuozi huozhe dianshi ‘Donald Duck didn't lift the table or the TV’ as a description of a situation in which Donald Duck lifted the table but not the TV. That is, adults interpreted the test sentence as if it meant It is the table or the TV that Donald Duck didn't lift, where disjunction takes scope over negation. By contrast, twenty out of twenty-one Mandarin-speaking children responded like English-speaking children and English-speaking adults (see Crain et al., Reference Crain, Gardner, Gualmini, Rabbin and Otsu2002, for the English study). They consistently rejected the test sentence by pointing out that Donald Duck had lifted the table. So, for children, the meaning that was assigned to the sentence can be paraphrased as Donald Duck lifted neither the table nor the TV (see Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004, for similar results in an earlier Japanese study). In short, it appears that, across languages, young children initially adhere to de Morgan's law, sometimes in apparent disregard to the properties of adult language (Crain, Goro & Thornton, Reference Crain, Goro and Thornton2006; Crain & Khlentzos, Reference Crain and Khlentzos2008; Reference Crain and Khlentzos2010). Findings like these are not expected according to experience-based models of language development, which contend that child language matches the input provided by adult language users (Pullum & Scholz, Reference Pullum and Scholz2002; Tomasello, Reference Tomasello2003).

Recently, Su and Crain (Reference Crain and Bavin2009) assessed three- to five-year-old Mandarin-speaking children's knowledge of the different truth conditions for the disjunction word huozhe in the subject phrase versus the predicate phrase of the universal quantifier mei ‘every’. On one trial, there were five aliens purchasing vehicles. Two aliens bought airplanes and chose stars as gifts. Two aliens bought cars and chose books as gifts. Finally, one alien bought a boat and chose a suitcase as a gift. The test sentence was Mei-ge mai-le feiji huozhe xiaoqiche de waixingren dou xuanze-le shu ‘Every alien who bought an airplane or a car chose a book’. It was reported that twenty child subjects rejected such test sentences 91% of the time. Children justified their rejections by pointing out that two aliens who bought airplanes chose stars, not books. This suggests that children assigned a conjunctive entailment to huozhe when it appeared in the subject phrase of mei. Therefore, they interpreted the test sentence as entailing that every alien who bought an airplane chose a book and every alien who bought a car chose a book.

On another condition, houzhe appeared in the predicate phrase of the universal quantifier mei. In a typical story, there were four princesses with dogs, and one with a bird. Two of the princesses with dogs took some jewels from a witch's house, two other princesses with dogs took rings, and the one princess with a bird took a flower. The same children accepted the test sentence Mei-ge dai-zhe gou de gongzhu dou tou-le baoshi huozhe jiezhi ‘Every princess who carried a dog stole a jewel or a ring’ 95% of the time. This finding provides evidence that children assigned disjunctive truth conditions to huozhe in the predicate phrase of mei, a non-downward entailing context. These experimental findings resemble those reported in earlier English studies conducted by Boster and Crain (Reference Boster, Crain and Smith1993) and Gualmini et al. (Reference Gualmini, Meroni, Crain and Weisgerber2003).

Taken together, the findings of previous research provide evidence that across typologically different languages, children are aware of the different linguistic properties of the downward entailing subject phrase versus the non-downward entailing predicate phrase of the universal quantifier (e.g. Mandarin mei and English every). The evidence is that they assign different truth conditions to disjunction in these different structural positions. Despite the compelling evidence that young children generate the conjunctive entailment of disjunction in the scope of several downward entailing operators, to our knowledge no study has tested children's linguistic knowledge of the interpretation of disjunction in the predicate phrase of Mandarin meiyouren. The present study is designed to supplement the previous findings by examining Mandarin-speaking children's knowledge that the predicate phrase of meiyouren is downward entailing and, therefore, licenses the conjunctive entailment of disjunction.

What about children's knowledge of the other diagnostic property of downward entailment – the licensing of negative polarity items? There are only a handful of child studies investigating the development of polarity sensitivity, mainly focusing on children speaking English and Dutch (O'Leary & Crain, Reference O'Leary and Crain1994; Koster & van der Wal, Reference Koster, van der Wal, Verrips and Wijnen1995; van der Wal, Reference van der Wal1996). One study by Koster and van der Wal (Reference Koster, van der Wal, Verrips and Wijnen1995) investigated seven- to fourteen-year-olds, as well as young adults. In this experiment, children and young adults were asked for judgements about the grammaticality of the licensing environments of the German negative polarity item hoeven ‘have to’. The grammaticality judgement task revealed that children's understanding of these licensing environments improved with age. To our knowledge, however, there have been no studies adopting such grammaticality judgement tasks to assess younger children's knowledge of negative polarity items, because children younger than six have been found to be unable to make reliable meta-linguistic judgements about the grammaticality of sentences (Saywitz & Cherry-Wilkinson, Reference Saywitz, Cherry-Wilkinson and Kuczaj1982; Edwards & Kirkpatrick, Reference Edwards and Kirkpatrick1999; Cairns et al., Reference Cairns, Schlisselberg, Waltzman and McDaniel2006).

Knowledge of where negative polarity items are and are not licensed has been investigated using other methodologies, when the subjects were younger than six to seven years old. In previous studies, these methodologies include elicited production tasks and act-out tasks (Koster & van der Wal, Reference Koster, van der Wal, Verrips and Wijnen1995; O'Leary & Crain, Reference O'Leary and Crain1994). For example, Koster and van der Wal (Reference Koster, van der Wal, Verrips and Wijnen1995) conducted an elicitation and acting-out experiment with a group of three-year-olds, testing their knowledge of the licensing situations of hoeven. The results showed that even very young children's grammatical knowledge of restrictions on hoeven was essentially adult-like, although there were some particular deviations from the adult usage.

Another study conducted by O'Leary and Crain (Reference O'Leary and Crain1994) adopted the Truth Value Judgement task with an elicitation component. In the task, Kermit the frog uttered false descriptions of events that had happened in the stories witnessed by children. After children's rejection of the puppet's assertions, they were asked to explain to Kermit ‘what really happened?’ It was found that children's negative statements, as in (22), typically contained the negative polarity item anything. By contrast, children produced the positive polarity item something, rather than anything, in the predicate phrase of their statements with pre-subject every, as in (23).

1. (22) kermit: Every dinosaur found something to write with.

   child: No, this one didn't find anything to write with.

1. (23) kermit: Only one of the reindeer found anything to eat.

2. child: No, every reindeer found something to eat.

The findings suggest that four-year-old English-speaking children use the negative polarity item any in downward entailing contexts, but refrain from using any in non-downward entailing contexts.

There are no studies of children's knowledge of the licensing condition of negative polarity items in Mandarin Chinese. To fill in this gap, the present study investigated whether Mandarin-speaking children know that the predicate phrase of meiyouren, a downward entailing context, allows the non-interrogative interpretation of wh-words like shenme. By contrast, the appearance of shenme in the predicate phrase of meigeren, a non-downward entailing context, leads to the question reading.

To summarize, although independent investigations have been undertaken on children's knowledge of (a) the interpretation of disjunction and (b) the licensing of negative polarity items, few experimental studies in any language have addressed the ‘unification’ feature of downward entailment by testing these two linguistic phenomena under the same linguistic contexts. In the following sections, we report two experiments investigating Mandarin-speaking children's knowledge of these two phenomena of downward entailment, using the quantificational expressions meiyouren versus meigeren. These experiments were conducted to fill a crucial missing link in previous research, i.e. whether or not children have the knowledge that downward entailment unifies several superficially unrelated linguistic phenomena, as a ‘core’ property in human languages.

METHOD

Results and discussion

Before we report the findings, let us review the main experimental predictions. The experimental hypothesis was that children would make a distinction between the downward entailing predicate phrase of meiyouren and the non-downward entailing predicate phrase of meigeren, such that they would be expected to access the different truth conditions of disjunction in these two structures. So, children would reject test sentences like (24), with huozhe in the predicate phrase of meiyouren. On the other hand, children were expected to accept test sentences like (25), with huozhe appearing in the predicate phrase of meigeren. The results confirmed the experimental hypothesis. When huozhe was in the predicate phrase of meiyouren, children rejected the test sentences 94% of the time (90/96). By contrast, when huozhe was in the predicate phrase of meigeren, children accepted the test sentences 90% of the time (86/96).Footnote ⁶ Children also justified their negative judgements of the test sentences for the right reasons. For example, in justifying their rejections of (24), children consistently pointed out that two of the cars lifted kangaroos. The control group of sixteen Mandarin-speaking adults correctly rejected the test sentences with disjunction in the predicate phrase of meiyouren 98% of the time (63/64), and they accepted the test sentences with disjunction in the predicate phrase of meigeren 95% of the time (61/64). Moreover, children accepted the control sentences with huozhe in the predicate phrase of meiyouren (e.g. (26)) 96% of the time (46/48) and they rejected the control sentences with huozhe in the predicate phrase of meigeren (e.g. (27)) 98% of the time (47/48). In addition, children made correct judgements for the filler sentences 99·5% of the time (382/384).

A comparison of the adult data between the previous Jing et al. (Reference Jing, Crain, Hsu and Otsu2005) study and our study reveals two different interpretive patterns for the disjunction word huozhe. Although both studies tested disjunction in negative contexts, adult subjects in the Jing et al. (Reference Jing, Crain, Hsu and Otsu2005) study did not appear to adhere to de Morgan's law, whereas adults in our study did show adherence to it. We attribute this discrepancy to the two different negative structures under investigation, i.e. (i) negation and disjunction appearing as clause mates (as in the Jing et al. (Reference Jing, Crain, Hsu and Otsu2005) study) versus (ii) negation appearing in higher clauses than disjunction (as in our study). As noted, in the Jing et al. (Reference Jing, Crain, Hsu and Otsu2005) study, the adult interpretation of disjunction in simple negative sentences is subject to cross-linguistic variation (Szabolcsi, Reference Szabolcsi, Kenesei and Siptár2002; Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004; Crain, Reference Crain and Bavin2009; Notley et al., Reference Notley, Zhou, Jensen and Crainin press). Consequently, Mandarin-speaking adults allowed disjunction to take scope over negation and failed to follow de Morgan's law. However, in our study, when the negative operator meiyouren appeared in the subject position, disjunction was not able to take scope over meiyouren. It is not surprising that adults always assigned the conjunctive entailment to huozhe in our study, because in this case disjunction remains in the scope of a downward entailing operator (Gualmini & Crain, Reference Gualmini and Crain2005).

No matter how complicated adults' interpretations might be, young children across languages are found to follow the similar interpretive pattern of assigning the conjunctive entailment to disjunction in the scope of downward entailing operators (Crain, Reference Crain2008). In the Jing et al. (Reference Jing, Crain, Hsu and Otsu2005) study, children observe de Morgan's law despite abundant counter-examples from adult data. In our study, the findings clearly demonstrated that Mandarin-speaking children are aware of the different truth conditions of huozhe in the predicate phrases of meiyouren versus meigeren. The four- to five-year-old Mandarin-speaking children we tested demonstrated understanding that huozhe generated a conjunctive entailment when it appeared in the predicate phrase of meiyouren, a downward entailing context. The same children assigned disjunctive truth conditions to huozhe when it appeared in the predicate phrase of meigeren, a non-downward entailing context. In combination with previous cross-linguistic child studies (see, e.g., Crain et al., Reference Crain, Gardner, Gualmini, Rabbin and Otsu2002; Gualmini & Crain, Reference Gualmini, Crain, Skarabela, Fish and Do2002; Gualmini et al., Reference Gualmini, Meroni, Crain and Weisgerber2003; Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004; Jing et al., Reference Jing, Crain, Hsu and Otsu2005; Su & Crain, Reference Crain and Bavin2009; Notley et al.,Reference Notley, Zhou, Jensen and Crainin press), these results provide additional support for the conclusion that young children master one of the diagnostic properties of downward entailment, i.e. the conjunctive entailment of disjunction, at an early age, and across languages.

METHOD

Results and discussion

Here are the experimental hypotheses. If children know that downward entailing contexts license wh-words as negative polarity items, then they would be expected to interpret ones with shenme in the predicate phrase of meiyouren as negative statements; and they would be expected to interpret ones with shenme in the predicate phrase of meigeren as questions. To illustrate, children should reject the puppet's statement for (28), and to justify their ‘No’ response by pointing out that every panda ate some fruit. Moreover, if children interpret the question word shenme as a question word in (29), then they should produce the answer ‘strawberry’, because every panda ate a strawberry. On the other hand, if children do not understand that the predicate phrase of meiyouren licenses shenme as a negative polarity item in (28), then they would be expected to interpret shenme in the predicate phrase of meiyouren as making a wh-question. In short, children would be expected to provide answers to both (28) and (29). Besides, if children don't interpret sentence (29) with shenme in the predicate phrase of meigeren as a question, then they would be expected to say either ‘Yes’ or ‘No’ to the puppet's utterance. This could happen, for example, when children merely associate the level intonation of all the test sentences with the statement reading.

The main finding was that children and adults analyzed shenme as a negative polarity item in the predicate phrase of meiyouren 100% of the time, so they interpreted (28) as a statement, and rejected it on the grounds that every panda ate some fruit. By contrast, the same children, as well as adults, interpreted shenme in the predicate phrase of meigeren (e.g. (29)) as a question-marker 100% of the time, so they provided answers to this type of test sentences.Footnote ⁷ It should be pointed out that neither children nor adults interpreted shenme as having scope over the negative subject meiyouren. That is, for sentences like (28), children did not assign a non-adult interpretation such as what fruit did no panda eat? This is further evidence that meiyouren is a strong negative quantifier in the subject position, which blocks the movement of both huozhe (in Experiment 1) and shenme (in Experiment 2) to take scope over meiyouren.

Recall that all the test sentences were uttered by using level intonation. It becomes important to ask whether or not the subjects depended on this intonation pattern in reaching the statement interpretation for shenme in sentences with meiyouren, as in (28). The concern is that children might treat shenme differently just on the basis of the intonation of a sentence. More precisely, one strategy children might take is to treat a sentence with a rising intonation as a question and so they interpret shenme as ‘what’; whereas when children encounter a sentence with a level intonation, they know that shenme can't be ‘what’ and they might ignore shenme in the sentence. Following this strategy, if shenme in (28) is omitted, then the sentence would be ‘No panda ate fruit’, which shares the same truth value as the sentence with shenme in the present task. Hence, an alternative explanation of the data we need to consider is that children might offer their replies to (28) by simply omitting shenme and drawing upon the association between the level intonation and the statement reading. As for this explanation, we admit that it may be possible that the level intonation has assisted children in reaching the statement reading of shenme in sentences with meiyouren. So, as one reviewer suggests, a better way to remove the subjects' intonation strategy is to utter both sentences with rising intonation, so that the subjects wouldn't use the level intonation to access the statement reading of shenme in the predicate phrase of meiyouren. However, we would like to raise two other relevant questions: Do children always assign the statement reading to a sentence with level intonation? Do children simply ignore shenme in sentences with level intonation? We hold a more conservative view towards these questions. To present a counter-example, let us take a close look at sentence (29) with level intonation, in which shenme appears in the predicate phrase of meigeren. Suppose that children ignore shenme in (29) and interpret this sentence with level intonation as a statement (i.e. every panda ate fruits), then they should say ‘Yes’ to (29). This type of response, however, was not observed in the experiment, because all the subjects provided answers indicating the fruits that each panda had eaten. The data seem to suggest that children don't just associate the level intonation of a sentence with the statement reading, nor do they ignore shenme in the test sentences. So, it is highly likely that children also processed the wh-word shenme when it appeared in the predicate phrase of meiyouren, as in (28). This indicates that children made a distinction between the predicate phrases of meiyouren vs. meigeren, when they assigned different interpretations to shenme in these two structures.

Therefore, we contend that these experimental findings provide supporting evidence that three- to five-year-old Mandarin-speaking children are aware that downward entailing contexts license wh-words as negative polarity items. Whereas children treated the wh-word shenme as a negative polarity item in the predicate phrase of meiyouren, a downward entailing expression, the same children analyzed shenme as a question word in the predicate phrase of meigeren, a non-downward entailing context. Taken together, the findings clearly demonstrate that Mandarin-speaking children master one of the diagnostic properties of downward entailment, which underlies the appropriate appearance of negative polarity items.

GENERAL DISCUSSION

This study investigated children's knowledge of downward entailment from a cross-linguistic perspective. Specifically, we tested children's knowledge of two diagnostic properties of downward entailment: first, the conjunctive entailment of disjunction, and second, the licensing of negative polarity items. We conclude that, by four years old, Mandarin-speaking children are sensitive to the different truth conditions associated with disjunction in the predicate phrases of meiyouren versus meigeren. In particular, Mandarin-speaking children assigned a conjunctive entailment to disjunction when it appeared in the predicate phrase of meiyouren, which is downward entailing; and they assigned the disjunctive truth conditions of disjunction when it appeared in the predicate phrase of the non-downward entailing quantifier meigeren. Moreover, Mandarin-speaking children interpreted shenme as a negative polarity item in the downward entailing predicate phrase of meiyouren, whereas they interpreted the wh-word shenme as a question-marker in the non-downward entailing predicate phrase of meigeren. Taken together with the results of previous cross-linguistic studies (Boster & Crain, Reference Boster, Crain and Smith1993; O'Leary & Crain, Reference O'Leary and Crain1994; Crain et al., Reference Crain, Gardner, Gualmini, Rabbin and Otsu2002; Gualmini & Crain, Reference Gualmini, Crain, Skarabela, Fish and Do2002; Gualmini et al., Reference Gualmini, Meroni, Crain and Weisgerber2003; Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004; Jing et al., Reference Jing, Crain, Hsu and Otsu2005; Su & Crain, Reference Crain and Bavin2009; Notley et al., Reference Notley, Zhou, Jensen and Crainin press), the present findings invite the conclusion that the semantic property of downward entailment applies across languages and appears early in the course of language development. In addition, our study invites the inference that Mandarin-speaking children understand that the semantic property of downward entailment unifies two superficially unrelated linguistic phenomena, i.e. the conjunctive entailment of disjunction and the licensing of wh-words as negative polarity items. Altogether, these features make downward entailment a viable candidate to be considered as a ‘core’ linguistic property in human languages.

There are two competing accounts of how young children demonstrate competence of these ‘core’ linguistic properties in the initial stage of language development. According to one account, children extract linguistic knowledge from the linguistic environment, using domain-general strategies. These strategies are used by children to form statistical generalizations based on distributional cues. This is the experience-dependent account (Pullum & Scholz, Reference Pullum and Scholz2002; Tomasello, Reference Tomasello2003, Goldberg, Reference Goldberg2003; Reference Goldberg2006). According to the experience-dependent account, core properties are ones with ‘maximal generality’ (Tomasello, Reference Tomasello2003), which are in fact more general, regular and frequent within a given language, as compared to peripheral phenomena. Hence, language learners are expected to have an even easier time learning these core properties (Goldberg, Reference Goldberg2003; Reference Goldberg2006; Tomasello, Reference Tomasello2003). On the alternative account, based on Universal Grammar (UG), the core linguistic properties are shared by all human languages and these properties express regularities. But different from the experience-dependent account, such regularities are ‘deep-seated’ (Chomsky, Reference Chomsky1965), rather than simply occurring with greater frequency than other linguistic phenomena. More precisely, these core properties run ‘deep’ because they tie together a number of linguistic phenomena which, on the surface, appear to be unrelated. For instance, advocates of UG ask the question: How do children across languages understand that the linguistic environments governing the conjunctive entailment of disjunction appear to track other features of natural languages, such as the appearance of negative polarity items? To answer this question, proponents of UG contend that human children come to the task of language acquisition already equipped with the core principles of human languages (such as downward entailment), and they draw upon a priori knowledge of these universal principles as part of the human genome (Crain & Pietroski Reference Crain and Pietroski2001; Crain, Gualmini & Pietroski, Reference Crain, Gualmini, Pietroski, Carruthers, Laurence and Stich2005; Crain & Thornton, Reference Crain, Thornton, Traxler and Gersbacher2006; Crain et al., Reference Crain, Thornton and Khlentzos2009; Pietroski & Crain, Reference Pietroski, Crain, Margolis, Samuels and Stichin press).

The linguistic phenomena under investigation (i.e. the interpretations of huozhe and shenme in the predicate phrases of meiyouren versus meigeren) support the account of language development offered by the theory of UG, and they seemingly pose a serious challenge for the experience-dependent account. The experience-dependent account focuses on how children use statistical learning mechanisms to learn distributional patterns of morphological items. Based on such patterns, it might be possible to account for how children acquire the licensing conditions for negative polarity items such as English any or Mandarin renhe. For example, if children are extremely good record keepers, they might keep track of the fact that renhe appears in the predicate phrase of meiyouren (e.g. Ban shang meiyouren yanzou renhe yueqi ‘Nobody in this class plays any musical instrument’), but not in the predicate phrase of meigeren (e.g. Ban shang meigeren dou yanzou *renhe yueqi ‘Everybody in this class plays *any musical instrument’). However, in learning the interpretation of disjunction, what children learn is not a distributional property. Rather, what they learn is that disjunction is assigned different truth conditions when it appears in different linguistic environments. Similarly, in learning the interpretation of the Mandarin wh-word shenme, what Mandarin-speaking children learn is that the same word, e.g. shenme, is interpreted as a negative polarity item in downward entailing linguistic contexts, but it functions as a wh-marker in other linguistic contexts. It is not the distribution of shenme that is at issue for learners, but its interpretive properties. So, even children who had learned the distribution of other negative polarity items, such as English any or Mandarin renhe, based on distributional facts, would be hard-pressed to use the same learning algorithms to establish the connection between downward entailing expressions and the functions of the wh-word shenme. Even if children are excellent record keepers, distributional learning will not suffice. To elaborate this point, our next step is to investigate whether the input does provide decisive evidence for the facts that children must acquire.

As we have seen, children master the different truth conditions of disjunction in the predicate phrases of meiyouren versus meigeren. It turns out, however, that there is a great paucity for the disjunction word huozhe in the input, making these two interpretations rarely accessible for Mandarin-speaking children. In a survey of total 43 289 Mandarin adult utterances in the CHILDES database (MacWhinney, Reference MacWhinney2000),Footnote ⁸ we only found eleven adult utterances with huozhe. None of these utterances resemble the test sentences. Moreover, among these few cases that contain huozhe, adult utterances generally suggest alternative choices or express mutual exclusivity. Here are two examples from the input to children in the CHILDES database: Jiang waiyu gei ma ma ting huozhe lai ge feiwen ba ‘Speak a foreign language to mum or give (me) a kiss’; Shuo hao huozhe buhao ‘Say good or not good’. These examples suggest that adults' daily usage of disjunction often carries an implicature of exclusivity, by assigning disjunction an exclusive-or interpretation in non-downward entailing contexts (see footnote 2). It is hence difficult for language learners to observe the conjunctive entailment of disjunction in the input, which only occurs in downward entailing structures. Moreover, if children construct their grammars solely on the basis of the adult input, they are likely to erroneously conclude that disjunction is exclusive-or (Morris, Reference Morris2008). As a consequence, language learners could easily fail to capture the relationship between the property of downward entailment and the conjunctive entailment of disjunction, which holds only if disjunction is interpreted inclusively. In short, on the experience-dependent account, we are owed an explanation of how children master the different truth conditions of disjunction, and especially the conjunctive entailment of disjunction, given that the relevant linguistic evidence is not readily attested in the primary linguistic data.

Our next concern is whether statements containing shenme in the predicate phrases of meiyouren vs. meigeren are available in sufficient quantity in the input to ensure that every child converges on a grammar that allows these different readings of shenme. The transcripts of adult input reveal that there is no exact match with shenme in the predicate phrase of a negative quantificational expression like meiyouren, which takes the subject position. At the same time, we found only two critical examples with shenme in the predicate phrase of the universal quantifier mei, e.g. shuaya yihou, meigeren dou zai mang shenme ne? ‘What is everybody doing, after brushing his teeth?’ Such low frequencies of relevant input make it unlikely that every child encounters the requisite evidence by the age at which they are found to know the different interpretations of shenme in corresponding linguistic contexts.

The experience-based approach might come up with other hypotheses to account for children's knowledge of the two interpretive patterns of shenme. For instance, one might ask whether there is a correlation between (i) meiyouren and declaratives and (ii) meigeren and questions, so that children could just pick up the frequency bias of the interpretive patterns of these two linguistic expressions, without considering the interpretation of shenme. With this concern, we searched the Mandarin corpora with the target words meiyouren and meigeren (without the combination of shenme), and the adult data only revealed three instances of meiyouren (two statements and one question) and seven matches with meigeren (four statements and three questions), suggesting no such hypothesized correlation between these two expressions and the corresponding interpretive patterns in the input. We then took the investigation further by searching the negator meiyou and the universal quantifier mei as the target words. This time, the results reveal a reverse pattern from the hypothesized correlation, i.e. the presence of meiyou is associated with more cases of questions, whereas the occurrence of mei is associated with more cases of statements. To be specific, out of a total of 472 adult utterances with the negator meiyou, we found 278 questions and 194 statements; and out of a total of 60 adult utterances with the universal quantifier mei, there are 45 statements and 15 questions. So, it is unlikely that children interpret the test sentences with shenme by simply depending on the statistical regulation of expressions like meiyouren or meigeren. Instead, children must adhere to specific linguistic structures (i.e. shenme in the predicate phrases of meiyouren versus meigeren) and observe the exact ways different linguistic expressions interact with each other (Gulamini & Crain, Reference Gualmini and Crain2005).

It remains to be determined whether or not both interrogative and non-interrogative usages of shenme are abundant in children's linguistic experience, such that children could learn to master these different interpretations of shenme based on the input. As for this question, because it seems impossible for us to sort out the exact number of these two readings out of 8452 adult utterances with shenme in the database, we would like to assume that both these two readings are available in the adult input. However, our research question is how children make a distinction between these two interpretations and how they recognize the relevant linguistic environments for each reading. Exacerbating the problem of learnability is the fact that different licensing situations for the non-interrogative use of shenme differ in their frequency of occurrence in the adult input. For instance, in the database, we found 170 adult utterances with the non-interrogative use of shenme in the antecedent of the Mandarin conditional clauses with jiu ‘then’, 46 utterances with shenme in bare conditionals, 34 utterances with shenme in the restrictor of the Mandarin universal quantifier dou ‘all’ and 8 utterances with shenme co-occurring with the Mandarin licensor ye ‘also’. In our case, however, there is not a single instance of shenme in the predicate phrase of meiyouren. As a consequence, on the experience-based account, children would be expected to master those more frequent linguistic contexts earlier than the other contexts, or they might easily fail in assigning the correct interpretations in linguistic contexts they don't usually encounter. For example, some (perhaps many) children might commit errors, such as interpreting shenme in the predicate phrase of meiyouren as following its normal function as a question, but this is contrary to the findings of our experimental research.

Even given the most optimist assumptions about children's capacities to form and test generalizations, and supposing that children successfully discern (all of) the apparently subtle distinctions amidst the buzzing bloom of conversation, and supposing that they remember all the distributional facts about where the same lexical item shenme is assigned a statement reading and where it functions as a normal question word, it still seems highly unlikely that children could draw upon their knowledge of the two readings of shenme to understand when disjunction generates a conjunctive entailment, and when it does not. Of course, linguists ‘discovered’ the cross-linguistic connection between these and other linguistic phenomena by considering a vast array of data, both positive and negative, and they had access to a vast array of cross-linguistic data, all of which could be examined at once. By contrast, the language learner must master these phenomena in the absence of negative evidence and without any cross-linguistic data. Moreover, as we noted earlier, it is difficult for the language learner to keep detailed records of when disjunction words do and do not generate conjunctive entailment, since the relevant distinction is not readily accessible in the input. Therefore, it seems highly implausible that every biologically normal child, across the globe, has the requisite skills to make an observation of the input data and draw the description of the evidence that ‘the conjunctive entailment is assigned to disjunctive statements if and only if the non-interrogative use of shenme appears in that linguistic environment’.

However, these problems of learnability are circumvented on the account advanced by advocates of UG. From the UG perspective, human children are innately endowed with substantive core principles of grammar. Guided by these principles, children can effortlessly ‘cognize’ seemingly complicated linguistic phenomena, including the semantic properties that are associated with downward entailing expressions. Therefore, on the UG approach, children should demonstrate knowledge of these core linguistic principles from the earliest stages of language development. Mandarin-speaking children's early emergence of knowledge of two apparently disparate linguistic phenomena associated with downward entailment therefore lends credibility to the theory of Universal Grammar.