THE SSM

The first scope preference model we will consider is the SSM (Notley et al., Reference Notley, Zhou, Jensen and Crain2012). The SSM is based on the earlier semantic subset principle (SSP), introduced by Crain, Ni, and Conway in 1994. The SSM differs in certain crucial respects from the SSP, as we outline below. The SSP was originally introduced in light of the observation that children could potentially encounter a semantic learnability problem when faced with ambiguous sentences like (1) above. This is because the two possible meanings of these kinds of sentences form a special relationship: one reading asymmetrically entails the other. For example, reading (1a) is only true in one context, that in which none of the horses jumped over the fence. By contrast, reading (1b) is true in this circumstance, but also in other circumstances. That is, sentence (1) is true on reading (1b) if none of the horses jumped over the fence, but the sentence is also true on this reading if some of the horses, but not all of them, jumped over the fence. In short, if none of the horses jumped over the fence, then not all of them did. However, the reverse does not hold. If not all the horses jumped over the fence, it is not necessarily true that none of them did.

When one interpretation of an ambiguous sentence asymmetrically entails another interpretation, then we will call the interpretation that makes the sentence true in the narrowest set of circumstances, and entails the other interpretation, the “strong” or “subset” reading. We will call the interpretation that makes the sentence true in the broader range of circumstances the “weak” or “superset” reading. For sentence (1), interpretation (1a) is the strong reading and interpretation (1b) is the weak reading. The problem identified by Crain et al. (Reference Crain, Ni, Conway, Clifton, Frazier and Rayner1994) is the following: suppose that in some language, adult speakers only assign a strong reading to a sentence like (1), but in another language, adult speakers assign both the strong and weak readings. Suppose further (contrary to what has been observed), that children initially hypothesize that (1) has the weak reading, whereas adult speakers allow only the strong reading. In this situation, as Crain et al. observed, children would not be able to acquire a grammar that is equivalent to that of adults (with only the strong reading), at least not on the basis of direct positive evidence provided by adult speakers.

For example, suppose that English-speaking adults only interpreted Every horse didn't jump over the fence to mean that none of the horses jumped over the fence (every > not). Now, suppose that a child acquiring English were to initially hypothesize that the same sentence meant not all of the horses jumped over the fence (not > every). The problem is that the child's reading (not > every) is consistent (true) in circumstances in which none of the horses jumped over the fence. Of course, for these children, the same sentence would also be true in circumstances in which it was not true for adults when some, but not all, of the horses jumped over the fence. The problem is: how will the child find out that such statements are only true, for adults, in one circumstance: when none of the horses jumped over the fence? In order to revert to the strong reading (every > not) from their initial hypothesis (not > every), children would have to record that adults do not use statements like (1) in certain circumstances in which these sentences are acceptable for children; that is, in circumstances in which (not > every) is true, but in which (every > not) is false. According to Crain et al., observing and mentally recording what does not happen (e.g., the circumstances in which adults do not use particular sentences) is likely to be beyond the cognitive capacities of most adults, much less children. We know that even when negative evidence of this type is made explicit to adults, only a small percentage of them actually manage to make use of such information in learning scenarios (Hovland & Weiss, Reference Hovland and Weiss1953).

One way out of this dilemma would be to assume that adults provide appropriate feedback when children produce (1) in a circumstance that makes the sentence true on the weak reading but false on the strong reading. In this scenario, if a child produced Every horse didn't jump over the fence in a circumstance in which only one horse managed to do so, then the adult could come to the rescue and provide feedback informing the child that what he or she said was false, because in the circumstance at least one horse had jumped over the fence. Although conceivable, Crain et al. argued that it could be risky to assume that all children who hypothesized a weak reading of such a sentence could count on corrective feedback from adults to set their grammars straight. Any child who was not fortunate enough to have the necessary parental interactions, in sufficient quantities, would not converge on an equivalent grammar to that of adult speakers of the local language. However, as we know, this never happens. With few exceptions, if any, all normally developing children converge on a system that assigns the same scope preferences as those of adults. This suggests that whatever input is required by children is available in abundance for all language learners. Negative feedback of exactly the right kind by adults, therefore, seems to be an unlikely explanation of children's universal mastery of scope assignments (for a similar argument, see Lasnik & Crain, Reference Lasnik and Crain1985).

A more reliable way out of the potential dilemma was advanced by Crain et al. (Reference Crain, Ni, Conway, Clifton, Frazier and Rayner1994). These researchers postulated that children adhere to a learnability principle called the SSP. The SSP orders children's linguistic hypotheses such that children initially hypothesize the strong (subset) readings of the kinds of scope ambiguities where the alternative readings are in an asymmetric entailment relationship. This ensures that children will always have the requisite positive evidence to converge on the grammar of the local language, regardless of the reading that is favored by adult speakers of the local language. Because the child initially assigns the strong reading to sentences such as (1), if adult speakers assign the strong reading, then the child has already converged on the adult grammar. Conversely, if adult speakers assign both the strong and the weak readings, then children will witness “unexpected” circumstances in which only the weak reading of (1) is true. Such circumstances constitute positive evidence that the local language allows the weak reading in addition to the strong reading, and children can add this reading to their grammars.Footnote ¹

A learnability dilemma of this type was expected to arise for sentences like (1). In English, sentences of this form can receive both strong and weak readings. By contrast, when such sentences are translated into some other languages (e.g., Mandarin Chinese or German), only the strong reading is possible (or at least highly preferred). It has been argued that languages like Mandarin Chinese have a structural constraint on scope relations that enforces an isomorphism between surface syntax and semantic interpretation (Aoun & Li, Reference Aoun and Li1989; Huang, Reference Huang1982; Lee, Reference Lee1986). The SSP would predict that children acquiring all languages should initially assign the strong (Mandarin or German) interpretation to the sentences under consideration, such that the English sentence Every horse didn't jump over the fence should initially be judged by English-speaking children to be true only in the circumstance in which no horse jumped over the fence. This would enable English-speaking children to add the weak reading to their grammars, based on circumstances in which adult usage makes the sentence true only on the weak reading, that is, where not all horses jumped over the fence. In this way, children always have the positive evidence necessary to converge on grammars equivalent to those of adults.

The original finding that children seemed to prefer the isomorphic reading of sentences like (1) was in line with the predictions of the SSP, because the isomorphic reading of (1) is also the strong subset reading. However, just as subsequent research called into question the observation of isomorphism, the finding that the nonisomorphic reading of sentences like (1) was available to young children was seen to challenge the SSP. Because the nonisomorphic reading is the weak superset reading, the SSP does not anticipate that children would have access to this reading early in the course of language development. To the contrary, weak readings are expected to emerge in children's grammars only after the requisite positive evidence is encountered. Of course, the amount of evidence children require in order to add meanings to their grammars is an open question, but on the face of it, the SSP is undermined, at least to some extent, by such experimental findings.

Nonetheless, these findings are completely consistent with a reformulation of the SSP, dubbed the SSM (Notley et al., Reference Notley, Zhou, Jensen and Crain2012). The SSM recasts the learnability principle as a maxim for assigning preferences to alternative interpretative options, rather than as a categorical constraint on children's initial hypotheses. The SSM is expected to apply whenever two (or more) scope-bearing elements appear in a sentence and where the scope interaction between the operators results in both strong and weak readings that are in an asymmetrical entailment relationship.

The first difference between the SSM and the SSP is that the SSM anticipates that both scope assignments will be available to children, even in languages that display a strong preference for a particular scope assignment. The disfavored scope assignment is available to both children and adults, according to the SSM, but it is often only elicited when it is “marked” by prosodic stress and/or accompanied by contextual support. The second difference between the SSM and the SSP is that the SSM predicts a default preference for the strong reading of the relevant sentences initially, whereas the SSP predicts the categorical absence of the alternative weak reading in children's grammars.

On the SSM, the assumption remains that children are faced with a semantic subset problem when scope ambiguities result in asymmetric entailments. However, the nature of the problem is slightly different than the one the SSP was designed to solve. The task for children becomes to determine which of their available readings is preferred by adult speakers of the local language. If children initially favor the strong reading, they will be able to converge on the adultlike scope preferences using positive evidence, following the same logic as the SSP. That is, children who witness circumstances in which a particular sentence is used on the weak superset reading will be able to adjust their preferences, adding weight to this way of resolving the relevant ambiguities. In contrast, supposing that children were to initially favor the superset reading, then the majority of the input would be consistent with that interpretation, and the alignment with adult preferences would be much more prolonged. Adhering to the SSM will prevent this potential delay in the acquisition of the adultlike preferences.

The predictions of the SSM are confirmed in much previous work, looking at various operator combinations and across various languages. In the absence of contextual manipulations, we have seen that children appear to have a default preference for the strong reading of sentences containing the universal quantifier “every” and negation like Every horse didn't jump over the fence. This effect has also been shown in Mandarin for sentences like Mei zhi xiaotuzi dou bu zai chi huluobo, “Every rabbit is not eating carrots” (Fan, Reference Fan2010). Moreover, it has been shown that even when the negation operator appears higher in the sentence structure than the universal quantifier (e.g., The strong guy didn't put every elephant on the table; Bu-shi mei-ge xiaonühai dou chuan qunzi, “It is not every girl who is wearing a skirt”), English-speaking and Mandarin-speaking children display a preference for the strong reading of these sentences, as opposed to adults (Fan, Reference Fan2010; Musolino & Lidz, Reference Musolino and Lidz2006).Footnote ²

The results of studies looking at children's interpretation of sentences containing negation and an existential quantifier like “some” or “a” also support the predictions of the SSM (e.g., The detective didn't find some guys). Actually, for English-speaking adults, sentences like The detective didn't find some guys usually only have one reading; “some” must take scope over “not.” The resulting weak meaning is that there were some guys who the detective did not find (but, by implication, he probably also did find some guys). This is because “some” is a positive polarity item (PPI) in English. PPIs typically must not occur in the scope of negation. However, if children are not initially sensitive to this constraint on PPIs, then they could access two hypothetically possible meanings of a sentence like The detective didn't find some guys. The first reading is the adultlike one (that there were some guys who the detective did not find). This is the weak reading because it is true if the detective did find some guys and not others, and also if the detective found no guys. The second reading is one in which “not” takes scope over “some.” In this case, the sentence means that the detective did not find any guys. This is the strong reading. Musolino et al. (Reference Musolino, Crain and Thornton2000) tested children on sentences like these and found that children initially display a preference for interpreting “some” within the scope of negation, thereby assigning a strong interpretation to the test sentences (something adults never do).Footnote ³ This effect has also been found in languages like Dutch for sentences like De jongen heft een vis niet gevangen, “The boy didn't catch a fish” (Kramer, Reference Kramer2000),Footnote ⁴ and in Kannada for sentences like Avanu biskit tinn-al-illa, “He didn't eat a cookie” (Lidz & Musolino, 2005/Reference Lidz and Musolino2006).

Finally, the SSM is supported by studies investigating children's interpretation of sentences like The pig didn't eat the carrot or the pepper and The dog reached the finish line before the turtle or the bunny across languages. These sentences can be grouped together because they contain a downward entailing operator (“not” or “before”) and a disjunction (“or”). When a downward entailing operator takes scope over disjunction, disjunction is assigned a conjunctive interpretation. For example, when negation is assigned scope over disjunction in the sentence The pig didn't eat the carrot or the pepper, the interpretation is that the pig did not eat either vegetable. In contrast, if disjunction is assigned scope over negation, the interpretation is that the pig either did not eat the carrot or did not eat the pepper, and it is not clear which. The strong reading is the one on which the pig ate neither vegetable, because if this is true, then it is certainly true that the pig did not eat one vegetable. English speakers have a preference for assigning the strong reading to sentences like this. Conversely, Japanese speakers have a preference for assigning the weak reading. Nonetheless, it has been shown that 3- to 5-year-old English-speaking and Japanese-speaking children pass through a stage at which they prefer the strong reading (Crain, Gardner, Gualmini, & Rabbin, Reference Crain, Gardner, Gualmini, Rabbin and Otsu2002; Goro & Akiba, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004a, Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004b; Gualmini, Reference Gualmini2005; Gualmini & Crain, Reference Gualmini, Crain, Skarabela, Fish and Do2002, Reference Gualmini and Crain2005).

Similar results have been reported for English- and Mandarin-speaking children's interpretation of sentences like The dog reached the finish line before the turtle or the bunny. English has a strong preference for assigning a strong reading to these sentences, on which the dog reached the finish line before both other participants. Mandarin allows for the same sentences to be assigned a weak reading, on which the dog arrived before one of the other participants, about 25%–40% of the time. However, again, children acquiring either English or Mandarin have been shown to pass through a stage at which they prefer to assign the strong reading (Notley et al., Reference Notley, Zhou, Jensen and Crain2012).

In sum, there appears to be a wealth of evidence supporting the predictions of the SSM as a default scope interpretation strategy in children. In addition, given that the SSM assumes that both interpretations of a scopally ambiguous sentence are available to the child, it is not surprising that certain contextual changes can affect which scope interpretation he or she accesses. However, it then becomes reasonable to ask whether the SSM is necessary, or whether children's scope assignment preferences can be explained entirely in terms of the context in which the sentences are presented. An alternative scope preference model that attempts to do just that is the QAR model (Gualmini, Reference Gualmini, Zeijlstra and Soehn2007a, Reference Gualmini and Otsu2007b, Reference Gualmini2008; Gualmini et al., Reference Gualmini, Hulsey, Hacquard and Fox2008; Hulsey et al., Reference Hulsey, Hacquard, Fox, Gualmini, Csirmaz, Gualmini and Nevins2004).

THE QAR MODEL

The QAR model takes as its starting point that every statement in a conversation is understood as the answer to a question, made salient in the context. The full model outlined by Hulsey et al. (Reference Hulsey, Hacquard, Fox, Gualmini, Csirmaz, Gualmini and Nevins2004) makes two predictions. The first prediction is that children (and adults) base their preferred interpretation of a scopally ambiguous sentence on whether it is a good answer to the salient “question under discussion” (QUD) in the context. For example, for a yes/no-question, a good answer is defined as being an assertion that entails either the “yes” or “no” answer to that question. The second prediction is that if both possible scope interpretations address the relevant question, then children (and adults) are expected to choose the interpretation that makes the sentence true in the context, following the conversational principle of charity (for conversational principles, see Grice, Reference Grice, Cole and Morgan1975). The QAR model purports to be able to explain all the data accumulated so far on children's scope preferences (without the need for an isomorphic account or the SSM). For example, in the original studies investigating children's interpretation of sentences like (1), we could imagine that, perhaps, the QUD was something like whether or not any of the horses would succeed in jumping over the fence, as in (3).

1. (3) QUD: Did any of the horses jump over the fence?

2. (4) Every horse didn't jump over the fence.

   a. None of the horses jumped over the fence. (every > not)

   b. Not all of the horses jumped over the fence. (not > every)

If the statement Every horse didn't jump over the fence is the speaker's attempt to answer this unspoken question, then assigning the universal quantifier wide scope over negation results in a good “no” answer to the QUD, along the lines of (4a). By contrast, assigning negation wide scope over the universal quantifier results in a circuitous answer to the question, along the lines of (4b). The interpretation in (4b) could entail either a “yes” or “no” answer to the QUD, because it is true if some horses jumped over the fence or if no horses jumped over the fence. According to this model, (4a) therefore represents the only good answer to the QUD. In the contexts tested by Musolino et al. (Reference Musolino, Crain and Thornton2000), (4a) was false (because some horses had jumped over the fence). This is thought to explain why children rejected test sentences like (4) in the original studies: because (4a) was the only good answer to the QUD, not because (4a) was the strong (or isomorphic) reading.

In subsequent studies, the QUD was supposedly set up to be whether or not all of the horses would succeed in jumping over the fence, as in (5).

1. (5) QUD: Did all of the horses jump over the fence?

If the same answer (4) is given to this question, both scope interpretations (4a) and (4b) constitute acceptable answers to the question. Both interpretations now entail a “no” answer to the QUD. In this situation, children are expected to choose the scope interpretation that makes (4) true. In a context in which some horses did not jump over the fence, and some did, this is interpretation (4b). This is thought to explain why children accepted test sentences like (4) in the studies designed to satisfy the felicity conditions of negation (e.g., Gualmini, Reference Gualmini2005). Thus, according to the QAR model, the most important factor driving children's scope interpretations is whether a possible interpretation of a proposition represents a good answer to the supposed QUD. Although adults are assumed to be driven by the same considerations as children, adults are thought to be better able to accommodate a question that differs from the one made salient by the context. This leads adults to select (4b) as their interpretation of (4) in both kinds of experimental settings.

A similar line of reasoning can be employed to explain why children sometimes access the weak reading of negative statements with the existential quantifier. As we discussed, it has been shown that English-speaking children appear to prefer to assign a strong interpretation to sentences like The dwarf didn't deliver some pizzas, contrary to adults (Musolino et al., Reference Musolino, Crain and Thornton2000). The strong interpretation is that the dwarf did not deliver any pizzas. However, with some contextual manipulations, Gualmini (Reference Gualmini2005) found that 15 English-speaking children (aged 4;01–5;06) were able to access the weak reading of sentences like this 90% of the time. Essentially, the contextual manipulations set up an expectation that the dwarf should deliver all of the pizzas, but the actual outcome of the story failed to meet this expectation (he only delivered some of the pizzas). Again, the explanation offered by proponents of the QAR model is that in the original study, the QUD may have been understood to be whether the dwarf would deliver any pizzas, but with the contextual changes in the subsequent study, the question was more likely to be whether the dwarf would deliver all the pizzas, as in (6).

1. (6) QUD: Did the dwarf deliver all of the pizzas?

2. (7) The dwarf didn't deliver some pizzas.

   a. There were some pizzas that were not delivered. (some > not)

   b. The dwarf delivered no pizzas. (not > some)

Now, for sentence (7), both possible interpretations are suggested to be good answers to the QUD. They both entail a “no” answer to the question. Because only interpretation (7a) was true in the context, however, the QAR model expects children to adhere to the principle of charity and select this interpretation. This is precisely what Gulamini observed in his 2005 study. A similar result has been obtained for passive sentences like Some pizzas were not delivered, when the same contextual manipulations are used (Gualmini et al., Reference Gualmini, Hulsey, Hacquard and Fox2008). In all these cases, these experiments show children favoring a weak interpretation of the target sentences.Footnote ⁵

STUDY PREDICTIONS

The question now becomes whether all data on children's scope preferences for sentences in which one reading asymmetrically entails the other can be adequately accounted for by the QAR model, or whether the SSM is necessary to capture underlying patterns of behavior.Footnote ⁶ The QAR model and the SSM make identical predictions in all cases in which only one of the two available interpretations of a scopally ambiguous sentence entails an answer to the QUD (when one of the interpretations asymmetrically entails the other). That is, if only one of the two interpretations entails an answer to the QUD, it must be the strong one. This is because, by definition, it is impossible for an answer to be entailed by the weak interpretation unless it is also entailed by the strong interpretation.Footnote ⁷ The SSM assumes, in these cases, that children select the strong interpretation because an underlying preference for the subset interpretation will allow them to converge more rapidly on adultlike preferences in their language on the basis of positive evidence. The QAR model assumes, in these cases, that children select the strong interpretation because it is the only good answer to the QUD. It is logically impossible to tease these reasons apart in such contexts.

Where the QAR model and the SSM diverge, however, is in cases in which contextual manipulations have made both available interpretations of a sentence good answers to the QUD. The QAR model does not assume that children have any underlying preference for one interpretation over the other, so in these cases, children are expected to adhere to the principle of charity and select the interpretation that makes the sentence true in the given context. Moreover, when only one of the available interpretations makes the sentence true, this will always be the weak interpretation. Thus, if this second prediction of the QAR model is correct, we expect to see a preference for the weak interpretation in children's data whenever this interpretation constitutes a good answer to the QUD. In contrast, the SSM assumes that children have an underlying preference for the strong interpretation, and that this preference could potentially surface even when the weak interpretation is a good answer to the QUD. The SSM does acknowledge that children's hypothesized preference for the strong interpretation can be overridden with contextual manipulations. However, exactly what these contextual manipulations need to be is not clear, because no previous research has actually set up a QUD explicitly. Even in the experiments specifically designed to test the QAR model, no question was explicitly asked before the presentation of the test sentences. Therefore, in this study, we deliberately pit the predictions of these two models head to head, and in particular we make our QUD explicit. Our aim is to clearly test whether making the weak interpretation a good true answer to some QUD is sufficient to elicit a preference from children for the weak interpretation of a scopally ambiguous sentence (in which one reading asymmetrically entails the other). If this is the case, then this will be further evidence for the QAR model. If not, then this will be evidence against the QAR model, and in line with the SSM. We turn now to our experiments.

EXPERIMENT 1

Methodology

We used a truth value judgment task (TVJT) to test English- and Mandarin-speaking children's and adult's interpretations of sentences like The elephant didn't eat both the carrot and the capsicum. The TVJT is designed to investigate which meanings children can and cannot assign to sentences (Crain & Thornton, Reference Crain and Thornton1998). The task involves two experimenters: one acting out stories with toy characters and props, and the other playing the role of a puppet who watches the stories alongside the child. At the end of each story, the puppet explains to the child subject what he thinks happened in the story. The child's task is to decide whether the puppet said the right thing or not. If the child informs the puppet that he was wrong, then the child is asked to explain to the puppet what really happened.

Our task was an adaptation of a task used by Goro and Akiba (Reference Goro, Akiba, Chand, Kelleher, Rodriguez and Schmeiser2004a, Reference Goro, Akiba and Otsu2004b). The task consisted of eight short scenarios in which an animal was asked whether it would like to eat a carrot and a green capsicum. If the animal in question decided to eat both vegetables, the child was asked to give the animal a gold medal. If the animal only ate one vegetable, the child was asked to give the animal a silver medal. If the animal ate neither vegetable, the child was asked to give the animal a black cross. Animals that did not want to eat their vegetables had the option of placing these unwanted vegetables in a fridge or rubbish bin. This meant there were never any vegetables left in front of the animals. Once the last animal had finished eating, our puppet then produced a target sentence about each animal. There were four critical trials on which the animal had only eaten one vegetable and had therefore received a silver medal. An example of a critical trial is illustrated in Figures 1–3.

Figure 1.

Elephant presented with choice of vegetables.

Figure 2.

Elephant eats the carrot and puts the capsicum in fridge.

Figure 3.

Elephant receives a silver medal.

On the critical trials, the puppet produced a sentence like (8) for English speakers, or (9) or (10) for Mandarin speakers.

1. (8) The elephant didn't eat both the carrot and the capsicum.Footnote ⁸

2. (9)

   

3. (10)

   

We included the floating quantifier both in our English test sentences in order to increase the rate at which adults demonstrate the weak reading of these sentences (in which negation takes scope over conjunction). Other work testing English-speaking adults’ interpretation of sentences without both, like The smurf didn't jump over the tree and the pond, has found that adults actually reject the sentences in contexts in which the smurf jumped over the tree, but not the pond, over two-thirds (66%) of the time (Goro, Minai, & Crain, Reference Goro, Minai, Crain, Bateman and Ussery2006). A scope reading in which conjunction takes wide scope over negation is very much available to English adults. As will be seen in the Results Section, by adding the operator both to our test sentences, we increase the rate of adults’ acceptance of these kinds of sentences.

To mirror the English sentences, the critical Mandarin sentences needed to contain the quantificational adverb dou (“all” or “both”). However, the presence or absence of dou results in two different word orders in Mandarin. As can be seen in (9), without dou, negation precedes conjunction. In (10), with dou contributing the meaning of the English particle both, conjunction precedes negation. We therefore included both types of sentence to ensure these ordering differences had no effect.Footnote ⁹

The sentences were presented with a neutral intonation pattern, with no contrastive focus on either both/dou or on and/he. This was done to avoid potentially biasing our subjects toward one of the possible scope interpretations. To avoid running our experiment with multiple prosodic conditions (e.g., stress on both and and; stress on and, but not both; stress on didn't and both; etc.), some of which may have favored the weak interpretation and some the strong interpretation, we opted to remove prosodic cues. This ensured that both interpretations of the test sentences remained potentially available to our subjects.Footnote ¹⁰ These test sentences were not preceded by an explicit QUD in Experiment 1.

Sentences like (8)–(10) have two possible interpretations, and one of these possible interpretations asymmetrically entails the other. The interpretations are given in (11).

1. (11)

   1. a. The elephant did not eat both of the vegetables, just one. (or possibly neither) (not > and, the weak reading)

   2. b. It is both the carrot and the capsicum that the elephant did not eat. (neither) (and > not, the strong reading)

Interpretation (11b), on which the elephant ate neither vegetable, is the strong subset reading, and asymmetrically entails interpretation (11a). The two meanings may, initially, be difficult to see for English speakers, because adult English speakers tend to prefer the weak reading (11a) of sentences like (8). However, looking across languages, the other possible meaning becomes apparent. For example, adult Mandarin speakers prefer the strong reading (11b) of Mandarin sentences like (9) or (10).

The four critical trials were interspersed with four filler trials. On two filler trials, the animal in question had eaten both vegetables. These were “Gold Medal Fillers.” On the other two filler trials, the animal in question had eaten neither vegetable. These were “Black Cross Fillers.” The puppet uttered sentences like (12) and (13) in response to one Gold Medal Filler and one Black Cross Filler trial each.

1. (12) The panda didn't eat anything.

2. (13) The mouse ate everything.

Sentences like (12) were false for Gold Medal Filler trials, but true for Black Cross Filler trials, while sentences like (13) were true for Gold Medal Filler trials and false for Black Cross Filler trials. This meant there were a balanced number of true and false filler trials across the testing session. The fillers were included to check that the children were concentrating and could answer both “yes” and “no” correctly, and to obscure the purpose of the experiment. Children who failed more than one filler were excluded from the data set.

Before commencing a testing session, each subject was introduced to our puppet and given two practice items. On one practice item, our puppet made an obviously true statement about a story, and on the other, he made an obviously false statement about the same story. This was so that the subjects would know that the puppet could say something wrong and to familiarize them with the task. The full list and ordering of practice, test, and filler sentences is given in Appendix A.

Results

All of the English-speaking and Mandarin-speaking children responded correctly to at least three out of four filler trials (two English-speaking children failed to respond correctly in one filler trial each). Therefore, the responses to the test sentences by all of the children were included in the final data set. We present the results grouped by language.

English results

The 21 English-speaking children rejected the test sentences on critical trials 95% of the time (80/84 trials). A typical justification for these rejections from a child aged 4;4 is given in (15).

1. (15) Puppet: The lion didn't eat both the carrot and the capsicum.

   Child: No, he ate one thing.

On two of the remaining four trials, two different children rejected the test sentences but justified their rejections by stating that the animal in question had eaten both vegetables. Because this was not the case (the animal had only eaten one vegetable), these rejections were omitted from the rejection count. On the remaining two trials, 1 child aged 3;10 accepted the test sentences.

The 15 English-speaking adults only rejected the test sentences on critical trials 28% of the time (17/60 trials). A Mann–Whitney test revealed a highly significant difference between the acceptance rates of English-speaking children versus adults (Z = 4.63, p < .001).

Mandarin results

The first group of 20 Mandarin-speaking children heard critical trials without dou. The children rejected the test sentences on these trials 98% of the time (78/80 trials). A typical justification for these rejections from a child aged 4;4 is given in (16). As the example indicates, Mandarin-speaking children rejected the test sentences on the same grounds as English-speaking children.

1. (16)

   

The group of 20 Mandarin-speaking adults who heard test sentences without dou performed similarly to the children. They rejected the test sentences on these trials 95% of the time (76/80 trials). One adult consistently accepted the target sentences. A Mann–Whitney test showed no significant difference between Mandarin-speaking children's and adults’ acceptance rates to critical trials without dou (Z = 0.036, p = .989).

The second group of 20 Mandarin-speaking children heard critical trials with dou. The children rejected the test sentences on these trials 99% of the time (79/80 trials). The group of 20 Mandarin-speaking adults who heard test sentences with dou rejected these sentences on critical trials 100% of the time (80/80 trials). A Mann–Whitney test showed no significant difference between Mandarin-speaking children's and adults’ acceptance rates of critical trials with dou (Z = 1.00, p = .799). Moreover, the presence or absence of dou (and the resulting different word orders) had no effect on children's or adults’ responses. The two groups performed similarly in response to sentences of both types. Mann–Whitney tests confirmed there were no significant differences between the two groups of children's responses (Z = 0.036, p = .989) or the two groups of adults’ responses (Z = 1.00, p = .799) in the two conditions. The English and Mandarin results are summarized in Table 1.

Table 1.

English and Mandarin acceptance rates in Experiment 1

EXPERIMENT 2

Results

English results

The 11 child subjects included in the final English data set responded correctly to at least three out of four filler trials (1 child failed to respond correctly in one noncrucial filler trial). The children rejected the test sentences on critical trials 89% of the time (39/44 trials). A typical justification for these rejections from a child aged 4;3 is given in (21).

1. (21)

   • Experimenter: Did the rabbit eat both the carrot and the capsicum?

   • Puppet: The rabbit didn't eat both the carrot and the capsicum.

   • Child: He ate the carrot.

   • Experimenter: So was the puppet right or wrong?

   • Child: Wrong.

On one of the remaining five trials, 1 child rejected the test sentence, but justified her rejection by stating that the animal in question had eaten both vegetables. Because this was not the case (the animal had only eaten one vegetable), this rejection was omitted from the rejection count. On the remaining four trials, 1 child aged 3;7 accepted the test sentences.

The seven English-speaking adults only rejected the test sentences on critical trials 7% of the time (2/28 trials). A Mann–Whitney test revealed a significant difference between the acceptance rates of English-speaking children versus adults (Z = 3.454, p < .001).

Mandarin results

The 12 child subjects included in the final Mandarin data set responded correctly to all filler trials. The children rejected the test sentences on critical trials 96% of the time (46/48 trials). A typical justification for these rejections from a child aged 4;8 is given in (22).

1. (22)

   

The 10 Mandarin-speaking adults rejected the test sentences on critical trials 100% of the time. A Mann–Whitney test confirmed there were no significant differences between the acceptance rates of Mandarin-speaking children versus adults (Z = 0.913, p = 1). These results are summarized in Table 2.

Table 2.

English and Mandarin Acceptance Rates in Experiment 2