Why children aren't universally successful with quantification *.

Musolino, Julien ; Lidz, Jeffrey

Abstract

This paper explores preschooler's knowledge of the linguistic principles (syntactic, semantic and pragmatic) regulating the interaction between universally quantified noun phrases and negation. Previous work has shown that 5-year-olds differ systematically from adults in the way they interpret sentences containing these elements (Musolino 1998; Musolino et al. 2000). On the basis of these results, Musolino et al. conclude that the grammar of 5-year-olds generates only a subset of the interpretations available in the adult grammar. We present here a series of experiments whose results challenge this conclusion. First, we demonstrate that under certain contextual conditions, 5-year-olds can be shown to access the same range of interpretations which characterize the adult system. Second, we show that while 5-year-olds do not differ from adults grammatically, their command of the pragmatic principles associated with the use of quantified statements is much more fragile than that of adults. We further suggest that children's immature pragmatic abilities is what gave rise to Musolino et al. 's findings in the first place. In doing so, we tie together research on grammatical development with a growing body of work on the development of pragmatic abilities and sentence processing in young children (Noveck 2001; Papafragou and Musolino 2003; Trueswell et al. 1999; Hurewitz et al. 2001). Finally, we show that our results have interesting methodological implications for the study of grammatical development.

1. Introduction

In recent years, children's comprehension of sentences containing quantified expressions has received considerable attention from investigators in the field of psycholinguistics (Philip and Aurelio 1991; Drozd and Philip 1993; Philip 1995, 1996; Philip and Lynch 2000; Brooks and Braine 1996; Crain et al. 1996; Crain 2000; Drozd 2000; Musolino et al. 2000; Kramer 2000; Lidz and Musolino 2002; Papafragou and Musolino 2003; among many others). Because quantificational phenomena involve the interplay of multiple levels of linguistic analysis (i.e., syntactic, semantic and pragmatic), the grammar of quantification represents a paradigm particularly well-suited for investigating the development of a broad range of linguistic principles.

A central question emerging from this line of research concerns the nature of the linguistic representations underlying preschoolers' quantificational competence and the extent to which these representations differ from those of adults. This question arises from the observation that preschoolers often differ from adults in the way they interpret sentences containing quantified expressions (ibid). While the validity of this observation is now beyond dispute, the factors giving rise to the phenomenon have been the subject of vigorous debate (Philip 1995; Crain et al. 1996; Drozd and van Loosbroek 1999; Philip 1996; Philip and Lynch 2000; Crain 2000).

To be sure, determining why children differ from adults in the way they interpret quantified sentences involves the delicate task of teasing apart the effect of a large number of variables, ranging from the contribution of the various linguistic factors giving rise to quantificational phenomena to the influence of particular experimental techniques on the child's performance. In this paper we take up this task--if only within a much narrower domain--by exploring preschooler's knowledge of the linguistic principles (syntactic, semantic and pragmatic) regulating the interaction between universally quantified NPs and negation. Previous work has shown that 5-year-olds differ systematically from adults in the way they interpret sentences containing these elements (Musolino 1998; Musolino et al. 2000). On the basis of these results, Musolino et al. have come to the conclusion that the grammar of 5-year-olds generates only a subset of the interpretations available in the adult grammar. We present here a series of experiments whose results challenge this conclusion. First, we demonstrate that under certain contextual conditions, 5-year-olds can be shown to access the full range of interpretations characteristic of the adult grammar. Second, we show that while 5-year-olds do not differ from adults grammatically, their command of the pragmatic principles associated with the use of certain quantificational expressions is much more fragile than that of adults. We further suggest that children's immature pragmatic abilities is what gave rise to Musolino et al.'s findings in the first place. In doing so, we tie together research on grammatical development with a growing body of work on the development of pragmatic abilities and sentence processing in young children (Noveck 2001; Papafragou and Musolino 2003; Trueswell et al. 1999; Hurewitz et al. 2001). We also show that our results have methodological implications for the study of child language and in particular that they call for a refinement of some of the assumptions associated with the use of the Truth Value Judgment Task of Crain and Thornton (1998). We conclude, with Crain and Thornton (1998), that the study of grammatical development cannot proceed without a careful consideration of children's growing language processing and pragmatic abilities.

The discussion is organized as follows: We begin by introducing the phenomenon we shall be concerned with (Section 2.1) as well as previous studies on the acquisition of this phenomenon (Section 2.2). Section 3 and 4 are the main experimental sections of the paper. Section 5 summarizes our results and presents an integrated account of our findings followed by a discussion of their implications for the development of quantificational competence and, more generally, for the study of grammatical development.

2. Preliminaries

2.1. Theoretical background

A well-known property of quantificational expressions is their ability to interact with one another to create scope ambiguity (Horn 1989; Jackendoff 1972; Lasnik 1972; May 1977, among many others). Consider for example the interaction between the universal quantifier and negation in the sentence in (1).

(1) Every student can't afford a new car.

a. [for all] x [student (x) [right arrow] [logical not] can afford a new car (x)]

b. [logical not] [for all] x [student (x) [right arrow] can afford a new car (x)]

On one reading, (1) can be paraphrased as Every student is such that s/he cannot afford a new car. In this case, the universally quantified subject is interpreted outside the scope of negation (abbreviated every > not), as indicated by the representation in (1a). On another reading, (1) can be paraphrased as Not every student can afford a new car. Here, every student is interpreted within the scope of negation (abbreviated not > every), as shown in (1b).

This example illustrates a core property of language that we will refer to as 'covert displacement', namely the fact that, in the absence of overt syntactic movement, certain expressions are interpreted in a position different from the one in which they occur (Rodman 1976; May 1977, 1985; Hornstein 1995; Reinhart 1997). In our example above, notice that the phrase every student c-commands the negative clitic n't in the surface syntax. This configuration naturally leads to the interpretation given in (1a) via a straightforward, isomorphic mapping between (surface) syntactic structure and semantic structure. However, in (1b), this configuration is reversed--negation now takes scope over the universal quantifier--thereby illustrating a mismatch between surface syntactic structure and semantic interpretation. This is the phenomenon of covert displacement in evidence. In sum, while the quantificational expression every student c-commands negation in the surface syntax, either scope relation between these two elements is possible in the semantics.

Interestingly, not all sentences containing a universally quantified NP and negation appear to give rise to scope ambiguity. This can be seen by comparing (1), which finds the universally quantified NP in subject position and is perceived to be ambiguous, and (2), which finds the universally quantified NP in object position and is not perceived to be ambiguous. To be sure, the most natural interpretation of (2) is one on which the professor talked to some of the students but not to others. In other words every student is interpreted within the scope of negation, as shown in (2a).

(2) The professor didn't talk to every student.

a. [logical not] [for all] x [student (x) [right arrow] talked to (professor, x)]

At this point, it is relevant to observe that while the most natural interpretation of (2) is one of which the professor talked to only some of the students, but not all of them (i.e., not > every), this sentence would also be true in a situation in which the professor had talked to none of the students (i.e., every > not) (1). This is so because every situation which satisfies the truth conditions of the every > not (i.e., none) reading also satisfies the truth condition of the not > every (i.e., not all) reading. In other words, if none of the students are such that the professor talked to them, it follows that not all the students are such that the professor talked to them; but not vice-versa. These entailment patterns are shown in (3).

(3) a. [for all] x [[logical not]P(x)] [right arrow] [logical not] [[for all]x[P(x)]] none [right arrow] not all

b. [logical not] [[for all]x[P(x)]] [right arrow] # [for all] x [[logical not]P(x)] not all [right arrow] # none

What is interesting about such entailment patterns is that they give rise to a well-known class of conversational inferences called scalar implicatures (Horn 1972, 1989 among many others). Scalar implicatures typically arise in examples like (4) where the speaker's use of some indicates that s/he had reasons not to use a stronger (i.e., more informative) term, e.g., all. The use of some in (4) thus gives rise to the inference or implicature in (5).

(4) Some students can afford a new car.

(5) Not all students can afford a new car.

What defines informational strength here is precisely the presence of entailment relations. So for example, all the students can afford a new car entails that some students can afford a new car, but not vice-versa. In other words, the proposition that all the students can afford a new car is true in a subset of the set of circumstances which verify the truth conditions of the proposition that some students can afford a new car. This situation is represented schematically in (6).

(6) [for all]x[P(x)] [??] [there exists]x[P(x)]

Assuming that the speaker is trying to be co-operative and will say as much as s/he truthfully can that is relevant to the exchange (Grice 1957, 1989), the fact that s/he chose the weaker term (i.e., some) gives the listener reason to think that s/he is not in a position to offer an informationally stronger statement (i.e., All students can afford a new car). This leads to the inference that the stronger statement does not hold, i.e., to (5). The inference in (2) is typically referred to as a scalar implicature (Horn 1972, 1989).

We can now return to our example in (3), repeated here as (7), and see that the entailment relation between the 'none' (i.e., every > not) and the 'not all' reading (i.e., not > every) gives rise to the same kind of phenomenon as the one described above. In other words, the use of not every in (7) gives rise to the inference that the stronger term on the scale, i.e., 'none' does not hold, i.e., (7a).

(7) The professor didn't talk to every student

(7)a. It is not the case that the professor didn't talk to any students

This explains why, even though (7) is true in a context in which none of the students are such that the professor talked to them, it is most naturally interpreted on a 'not all' reading (for similar considerations regarding the interpretation of sentences like (1), see Section 5 and the extensive discussion offered by Horn 1989). Thus, while (7) is semantically compatible with both 'none' and a 'not all' situations, it is pragmatically compatible only with the latter. The diagram below illustrates this situation (2).

Finally, an important property of scalar implicatures is that they are cancelable, as shown by (8).

(8) a. Some of my students got an A. In fact, they all did.

b. Not all of my students got an A. In fact, none of them did.

Moreover, there are specific contexts in which scalar implicatures fail to arise. Such contexts include situations of uncertainty such as predictions or bets. Imagine for example that someone were to make a bet that some of the students in her class will get an A. Suppose now that all of the students end up getting an A. In this case, it seems obvious that the bet is won. In other words, the use of a weaker term in the statement of the bet (i.e., some students will get an A) appears to be consistent with a situation in which all of the students get an A.

To summarize, we have seen that the interpretation of sentences containing universally quantified expressions and negation involves the interplay of syntactic (i.e., c-command relations), semantic (i.e., entailment relations) and pragmatic (i.e., conversational inferences) factors. Given the complexity and abstract nature of the facts described above, explaining when--and more importantly how--learners become aware of these facts raises important psychological questions (Chomsky 1965; Crain 1991; Pinker 1989 among many others). The need to elucidate these questions and to evaluate the claims from previous studies is what led us to the research presented here. Accordingly, we begin by reviewing earlier work investigating children's comprehension of the interaction between negation and other quantificational expressions.

2.2. Previous studies

Several studies on children's comprehension of the interaction between quantifiers and negation can be found in the literature on language acquisition (Roeper and Matthei 1975; Boysson-Bardies 1977; Drozd and Philip 1993, O'Leary and Crain 1994; Thornton 1995; Kramer 2000, Lidz and Musolino 2002; Gualmini 2004). One study designed to investigate the acquisition of scopal interactions between quantificational NPs and negation is Musolino et al. (2000) (based on Musolino 1998). In a series of experiments using the Truth Value Judgment Task (Crain and Thornton 1998), Musolino et al. (2000) tested children's and adults' interpretation of sentences like (9) and (10).

(9) Every horse didn't jump over the fence

a. [for all]x [horse (x) [right arrow] [logical not] jump over the fence (x)]

b. [logical not]x [horse (x) [right arrow] jump over the fence (x)]

(10) The Smurf didn't buy every orange

a. [logical not][for all]x [Smurf (x) [right arrow] buy (x, orange)]

As discussed in Section 2.1, sentences like (9) are ambiguous between a 'none' and a 'not all' reading (9a and 9b respectively). Using Musolino et al.'s terminology, (9a) corresponds to an isomorphic interpretation (because of the isomorphic mapping between syntactic and semantic scope) while (9b) corresponds to a non-isomorphic interpretation. By contrast, (10) seems only to allow an isomorphic, i.e., 'not all' interpretation (but see discussion in Section 2.1). The main finding from Musolino et al.'s study is that while adults can easily access the non-isomorphic interpretation of sentences like (9) and the isomorphic interpretation of sentences like (10), 5-year-olds systematically assign (9-10) an isomorphic interpretation only. In other words, 5-year-olds rarely assign sentences like (9)a non-isomorphic interpretation. This is what Musolino et al. call the 'Observation of Isomorphism'.

In order to account for the observation of isomorphism--as it pertains to universally quantified NPs and negation--Musolino et al. observe that in Chinese, the equivalent of a sentence like (10) allows only an isomorphic, i.e., 'none' interpretation. This situation creates the following, well-known, learnability problem (Berwick 1985; Pinker 1989; Crain 1991; Crain and Thornton 1998; Wexler and Manzini 1987 among others). Suppose a learner of Chinese were to begin acquisition with the (incorrect) hypothesis that his/her grammar allows both the isomorphic and the non-isomorphic interpretation. Such a learner might never be able to recover from this hypothesis because all of the data s/he encounters will be consistent with it. Consequently, the subset argument leads to the conclusion that the initial state must contain the Chinese hypothesis. This hypothesis would be correct for children learning languages that allow only the isomorphic reading and easily disconfirmed for children learning languages that allow both the isomorphic and the non-isomorphic readings. Thus, Musolino et al. explain the observation of isomorphism by claiming that when children fail to access the nonisomorphic reading, it is because they are in a stage of development in which their grammar of quantification is essentially Chinese.

The observation of isomorphism and the account offered by Musolino et al. raise a number of interesting questions. First, one may wonder whether the difference between 5-year-olds and adults observed by these authors does indeed have a grammatical basis. This question is prompted by the observation that the ambiguity shown in (1) arises from the operation of core grammatical mechanisms which many would assume are in place by the time children reach their fifth birthday (3).

An alternative possibility would be that children and adults have similar grammars but that they display different interpretive preferences regarding sentences like Every horse didn't jump over the fence. In other words, children may be aware of the fact that such sentences are ambiguous but they may find the non-isomorphic interpretation more difficult to access than the isomorphic one.

It is interesting to observe in this respect that recent studies on the acquisition of universal quantification report that children's performance improves dramatically when certain features of the context in which universally quantified statements occur are manipulated (e.g., Crain et al. 1996; Drozd and van Loosbroek 1999).

This observation provides an interesting test for Musolino et al.'s grammatical account: to the extent that children's apparent inability to access non-isomorphic interpretations reflects a grammatical difference between children and adults, presenting children with sentences like Every horse didn't jump over the fence in a context which would familiarize children with the intended domain of quantification should yield no significant improvement in performance. Conversely, if performance improves in such contexts, we would then have evidence against a competence explanation and in favor of a performance account (4). In the next section, we present an experiment designed to test these predictions.

Regarding sentences like (10), Musolino et al.'s study shows that 5-year-olds can easily access the 'not all' interpretation which, in this case, corresponds to an isomorphic interpretation. What Musolino et al. didn't test however, is whether children also know that sentences like (10) are true in a 'none' context (see discussion in Section 2.1). This question is worth asking though, for the following reason: since sentences like (10) are true in 'none' contexts by virtue of the entailment relations between 'none' and 'not all' (see discussion in Section 2.1), showing that children accept such sentences in 'none' contexts would provide evidence that they have knowledge of the relevant entailment relations. Put another way, since sentences like (10) are typically used in 'not all' contexts (for pragmatics reasons, as discussed in Section 2.1), it is hard to see how children would know that they are also true in 'none' contexts if they didn't have knowledge of the relevant entailment relations between 'none' and 'not all'. Moreover, testing children (and adults') interpretation of sentences like (10) would give us the opportunity to directly test for the pragmatic bias giving rise to the preference for the 'not all' reading discussed in Section 2.1. In Section 4, we present an experiment designed to address these issues.

3. Experiment 1

Our first experiment was designed to test the competence account of Musolino et al. (2000). In order to do so, we tested children in two conditions. The first condition was designed to replicate Musolino et al.'s (2000) findings. Children were therefore tested using the same method and similar materials as the ones used by Musolino et al. (2000), i.e., sentences like (11) below. In the second condition, children we presented with the same sentences as in condition 1 but, in order to familiarize children with the intended domain of quantification (i.e., the set of three horses present in the story), these sentences were all preceded by statements mentioning the relevant set of horses, i.e., Every horse VP, as in (12).

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

(12) Every horse jumped over the log but every horse didn't jump over the fence.

3.1. Method

3.1.1. Participants. We tested 20 English-speaking children (8 boys and 12 girls) between the ages of 5;0 and 5;11 (mean 5;4). One child was replaced in the design because she could not provide justifications for her answers. We chose 5-year-olds because previous studies, in particular Musolino et al. (2000), showed that children in that age range gave nonadult responses to ambiguous sentences containing a universally quantified subject and negation. The children were selected from daycare centers in the Philadelphia and in the Bloomington (Indiana) areas. In addition to the children, we also tested a control group of 20 adult native speakers of English. The adult participants were all undergraduate students at the University of Pennsylvania and Indiana University.

3.1.2. Procedure. As in Musolino et al.'s study, we tested children using the Truth Value Judgment Task methodology (TVJT) (Crain and Thornton 1998). The TVJT typically involves two experimenters. The first experimenter acts out short stories in front of the participants using small toys and props. The second experimenter plays the role of a puppet who watches the stories alongside the participants. At the end of the story, the puppet makes a statement about what he thinks happened in the story. The participants' role is to decide whether the puppet's statement is 'right' or 'wrong'. Finally, participants are asked to justify their answers by explaining why they think that the puppet was right or wrong. For a more detailed description of the TVJT, see Crain and Thornton (1998).

The children were first introduced to the task as a group and then tested individually in a quiet room away from the class. Each child received two pretest stories and if the child could answer those appropriately (including appropriate justifications for their answers), they would then hear seven more stories: four test stories and three control stories, administered in a pseudo-random order.

Adult participants were shown a videotaped version of the stories witnessed by the children, including the pretest stories. They were given a score sheet and were instructed to indicate, for each story, whether the puppet was right or wrong. They were also asked to provide justifications for their answers.

3.1.3. Materials. In condition 1, whose purpose was to replicate Musolino et al.'s findings, children were tested on their interpretation of sentences like (11), repeated here as (13).

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

In the story corresponding to (13), for example, three horses decided to test their skills by jumping over various obstacles. First, the horses considered jumping over a barn but they realized that it was too tall and therefore that it might be dangerous. Next, the horses considered jumping over a fence. The first two horses cleared the fence but the third one, who had hurt its leg the day before, decided not to jump. At the end of the story, the puppet described the situation using the sentence in (13), see picture 1.

Notice that in the context of the story just described, (13) is false on an isomorphic (i.e., 'none') reading since it is not the case that none of the horses jumped over the fence. So if participants access the isomorphic interpretation of (13), we expect them to reject the puppet's statement. By contrast, (13) is true on a non-isomorphic interpretation (i.e., not all) since it is true that not all the horses jumped over the fence, i.e., two did and one didn't. So if participants access the non-isomorphic interpretation, we expect them to accept the puppet's statement.

In condition 2, children were presented with the same sentences as those used in condition 1, but these sentences were preceded by affirmative statements, as shown in (14).

(14) Every horse jumped over the log but every horse didn't jump over the fence.

The stories corresponding to sentences like (14) were very similar to the ones used with sentences like (13). In the case of (14), three horses decided to test their skills by jumping over various obstacles. First, the horses all jumped over a log and then only two of them jumped over the fence. In the end, therefore, the situation is fully parallel to the one described in condition 1: the isomorphic reading is false in this context since it is not the case that none of the horses jumped over the fence whereas the nonisomorphic reading is true since not all the horses jumped over the fence (see picture 2).

[FIGURE 2 OMITTED]

The complete sets of test sentences used in conditions 1 and 2 are given in Table 1 and Table 2 below.

In addition to the four test stories, children also heard three control stories. Unlike the test items, the statements made by the puppet on the control stories were not ambiguous. The experimenter holding the puppet had a choice between two different statements for each of the control stories. One statement was true in the context of the story and the other was false. If the child had answered YES to a given test story, the experimenter holding the puppet was instructed to pick the statement for the following control story corresponding to a NO answer, and vice-versa. This precaution was taken to ensure that children could provide either kind of answer, i.e., YES or NO, when appropriate (5).

[FIGURE 3 OMITTED]

The list of statements made by the puppet on control stories is given in Table 3.

Finally, the 40 participants (20 children and 20 adults) were randomly assigned to each of the two conditions, yielding a 2 by 2 design with age (children vs. adults) and condition (condition 1 vs. condition 2) as between subject factors and 10 participants per cell. The group of children assigned to condition 1 ranged in age between 5;2 to 5;11 (mean 5;7) and the children assigned to condition 2 ranged in age between 5;0 to 5;9 (mean 5;2).

3.2. Results

In the analyses below, we used as our dependent measure the proportion of YES responses to the puppet's statements. Beginning with test items, we found that in condition 1 (our replication condition), children almost never accepted the puppet's statements, i.e., 15% of the time, whereas adults almost always did so, i.e., 92.5% of the time (t(18) = 6.83, p < .0001). By contrast, we found that children in condition 2 accepted the puppet's statement reliably more often, i.e., 60% of the time (6) (compared to 15% in condition 1, t(18) = 2.3, p < .05) (7). Adults in condition 2 always accepted the puppet's statements, i.e., 100% of the time.

The proportions of YES responses to the test items were entered into an analysis of variance (ANOVA) with two factors: age (children vs. adults) and condition (condition 1 vs. condition 2). The analysis revealed a significant main effect of age (F(1,36) = 34.94, p < .0001), a significant main effect of condition (F(1,36) = 6.97, p = .01) and a marginally significant interaction between age and condition (F(1,36) = 3.55, p = .06), see graph 1.

[FIGURE 6 OMITTED]

On control items, children gave correct responses 100% of the time in condition 1 and 96% of the time in condition 2. Adult participants always gave correct answers, i.e., 100% of the time in both conditions.

In condition 1, children rejected the puppet's statements 85% of the time. When asked why the puppet was wrong, children typically invoked the fact that two of the horses did jump over the fence. By contrast, adults--who almost always accepted the puppet's statements (i.e., 92.5% of the time)--typically said that the puppet was right because not all the horses jumped over the fence, i.e., two did but one didn't. In condition 2, the children who rejected the puppet statements invoked the same reasons as the children in condition 1, namely the fact that two horses did jump over the fence. By contrast, the children who accepted the puppet's statements justified their answers in the same way that adults did, namely by invoking the fact that while two horses had made it over the fence, one didn't.

[FIGURE 4 OMITTED]

3.3. Discussion

First, it is important to observe that our results replicate the findings of Musolino et al. (2000) in the case of sentences like Every N didn't VP. That is, we found, as did Musolino et al., that while adults can easily access the non-isomorphic (i.e., not all) interpretation of such sentences (i.e., 92.5% of the time, including appropriate justifications), 5-year-olds almost never do so (i.e., 15% of the time (8)). More importantly however, our results show that children's ability to access the non-isomorphic interpretation increases dramatically in the case of sentences like Every N VP but every N didn't VP (i.e., 15% vs. 60% acceptance rate, p < .05) (9). That children are indeed accessing non-isomorphic interpretations--rather than simply saying 'Yes' more often because they may be confused by the more complex statements--is clearly demonstrated by the justifications they provide. Recall that children who accept the puppet statements in condition 2 typically explain that the puppet is right because two of the horses jumped over the fence but the third one didn't; in other words, not all the horses jumped over the fence.

Showing that children's ability to access the not all, non-isomorphic interpretation of sentences like Every N didn't VP improves under certain contextual manipulations casts serious doubts on the grammatical account proposed by Musolino et al. (2000) (10). Such a 'competence' account makes a strong, falsifiable prediction: since children's non-adult interpretations reflect a lack of grammatical knowledge, contextual support should have no effect on the child's ability to access the nonisomorphic interpretation of sentences like Every N didn't VP (11). Clearly, the facts presented here indicate otherwise and so the grammatical account must be abandoned.

As we have seen, the data presented so far seem to indicate that 5-year-old children do have the grammatical competence to generate the nonisomorphic interpretation of sentences like Every horse didn't jump over the fence. However, one possible explanation of the findings that we have reported is that Musolino et al. were right to conclude that children fail to generate the non-isomorphic interpretation but that some other factor explains their apparent abilities in our contrast condition. In particular, it is possible that children believe that "but" contains a negation operator in its semantic representation. If this were so, then the nonisomorphic reading would not arise from a movement operation inverting the scopes of "every" and "n't" but rather from the fact that there is a negative operator that c-commands the universal quantifier on the surface, namely the one inside "but". On this view, the "not every" interpretation would be isomorphic since the relevant negation would be inside "but".

In order to test this hypothesis, we tested a group of 10 children between the ages of 4;6 to 6; 1 (mean 5;4) on their interpretation of sentences like (15) where but was replaced by and (the mean ages of the children in the 'but' and in the 'and' condition, i.e., 5;2 and 5;4 respectively, do not differ significantly, i.e., t(18) = -1.09, p = .28). The procedure and materials used in this condition were otherwise identical to the ones used in condition 2.

(15) Every horse jumped over the log and every horse didn't jump over the fence.

What we found is that in response to sentences like (15) children accessed the non-isomorphic interpretation (i.e., accepted the puppet's statements and provided appropriate justifications) 55% of the time, compared to 60% of the time in response to sentences containing but. No reliable differences were found between these two acceptance rates (t(18) = 0.2, p = 0.8). We therefore conclude that it is familiarization with the intended domain of quantification and not a negative operator contained in "but" that gives rise to the effect we have observed. In other words, the nonisomorphic reading IS generated by children's grammars, contra Musolino et al. (2000), but this reading requires contextual support in order to be detected in children.

Another possibility has to do with the potential effect of prosody on the resolution of ambiguous sentences like Every horse didn't jump over the fence. It is sometimes claimed in the linguistics literature (e.g., references cited in Horn 1989) that the 'none' and the 'not all' readings may be associated (at least in English) with different intonational contours (12). On the assumption that this is the case, one could speculate that preschoolers, unlike adults, are insensitive to the prosodic contour associated with the 'not all', non-isomorphic interpretation and therefore that they so often fail to access such interpretations. Interestingly, the claim that young children often fail to use prosodic cues to resolve syntactic ambiguity has recently received experimental support (Snedeker and Trueswell 2001). So far so good. Severe complications arise however, when one tries to make sense of children's improved performance in response to sentences of the form Every N VP but/and every N didn't VP. One could hypothesize that once embedded within such larger structures, sentences like Every N didn't VP receive a different prosodic contour which may facilitate children's access to the 'not all' interpretation. However such an account leads to the following paradox: in order to account for children's difficulty with the 'not all' interpretation in the first place, i.e., in the case of sentences like Every horse didn't jump over the fence, one needs to assume that children CANNOT use prosodic cues to resolve such ambiguities. However, in order to account for the fact that children are better with sentences like Every horse jumped over the log but every horse didn't jump over the fence, one now needs to assume that children CAN use prosodic information to resolve such ambiguities--hence the paradox. Thus an account on which prosody is the sole variable affecting children's differential rates of success with the 'not all' interpretation of sentences like Every N didn't VP does not appear to be very promising (13). Moreover, notice that even if such an account turned out to be correct, it would not affect our claim that children's difficulty with the 'not all' interpretation cannot be due to the fact that children are semantically restricted to the 'none' reading; as argued by Musolino et al. (2000).

Two more factors need to be considered when interpreting the results of experiment 1. First, it is worth pointing out that previous psycholinguistic work suggests that negative statements are more difficult for people to evaluate than their affirmative counterparts (Wason 1965). (14) The following examples, due to Wason, illustrate this point.

(16) Five is an odd number

(17) Five is not an even number

While both sentences are true, it has been found that (17) is more difficult to evaluate than (16). Moreover, it has been found that the difficulty associated with sentences like (17) can be significantly reduced if the sentence is preceded by an affirmative statement, as in (18).

(18) 4 is an even number, but 5 is not an even number.

In a similar vein, de Villiers and Tager Flusberg (1975: 279) point out that a statement like I didn't drive to work "is more plausible, and consequently easier to comprehend, if it is made by someone who normally drives rather than by someone who commutes by train." According to these authors, this property of negative sentences is due to the fact that "negative statements are generally used to point out discrepancies between a listener's presumed expectations and the facts".

Coming back to the results of experiment 1, one can now see that the presence of the initial affirmative statement in sentences like Every horse jumped over the log but every horse didn't jump over the fence may have contributed to the observed effect. In other words, just as adult speakers find it easier to evaluate negative statements like (17) when they are preceded by affirmative statements, as in (18), children may also find it easier to access the non-isomorphic, i.e., not all, interpretation of sentences like Every horse didn't jump over the fence when these negative statements are directly preceded by an affirmative lead-in as in Every horse jumped over the log but/and every horse didn't jump over the fence. Finally, notice that the manipulation described in condition 2 of experiment 1, i.e., first having the three horses all jump over a log, may have created the kind of positive expectation (i.e., all the horses did jump over the log) required to satisfy the felicity conditions on negative statements discussed by de Villiers and Tager Flusberg (1975). Recall that in condition 1 of experiment 1 (based on the original Musolino (1998) experiment) all the horses initially fail to jump over the barn, thereby creating a negative expectation. For an experimental demonstration of the role of such expectations and their effect on children's comprehension of negative statements, see Gualmini (2004).

The data presented here do not of course allow us to tease apart the potential contribution of the factors we have reviewed so far, namely (a) familiarization with the intended domain of quantification, (b) the presence of a preceding affirmative statement and (c) the satisfaction of felicity conditions associated with the use of negative statements. Disentangling the effects of such factors will clearly require further research. However, the contribution of experiment 1 should also be clear: whatever the precise contribution of the factors discussed above, it is no longer possible to maintain, as did Musolino (1998) and Musolino et al. (2000), that children are grammatically limited to only a subset of the interpretative options available in the adult grammar.

Let us take stock. We've been concerned here with children's knowledge of the linguistic principles regulating the interaction between the universal quantifier and negation in sentences like (19) and (20) below:

(19) Every N didn't VP

(20) NP didn't V every N

As discussed in Section 2.1, the interpretation of such sentences involves the interplay of syntactic, semantic and pragmatic factors. So far, we have shown--contra Musolino et al. (2000)--that 5-year-olds can access both interpretations of sentences like (19) (i.e., the 'not all' and the 'none' interpretation). This result demonstrates that preschoolers know that such sentences are scopally ambiguous and therefore that they must have knowledge of covert displacement.

Our next task involves determining whether preschoolers have adult-like knowledge of the set of truth conditions associated with sentences like (20). Since Musolino et al. established that 5-year-olds know that sentences like (20) are true in a 'not all' context, we ask whether children also know that such sentences are true in a 'none' context (see Section 2.1). In addition, we ask whether we can find experimental evidence, in children and in adults, for the pragmatic bias associated with the use of such sentences (see Section 2.1). Our goal therefore, is to determine whether children have knowledge of the syntactic, semantic and pragmatic principles associated with the interpretation of sentences like (19-20).

4. Experiment 2

This experiment was designed to test children and adults' interpretation of sentences like (21). As discussed above, such sentences are TRUE in situations in which either 'none' or 'not all' of the elephants were placed on the table. However, the use of 'not every' in (21) (as opposed to the stronger 'not any') gives rise to the implicated meaning in (22) (see Section 2.1).

(21) The strong guy didn't put every elephant on the table.

(22) It is not the case that the strong guy didn't put any of the elephants on the table.

Recall from our discussion in Section 2.1. however, that there are contexts in which such implicated meanings fail to arise. An interesting consequence of this fact for experimental design is that contexts can be constructed in which the implicated meaning either arises or fails to arise which allows the experimenter to directly test participants' sensitivity to such a pragmatic bias. In the experiment described below, we tested children and adults' interpretation of sentences like (21) in contexts which either licensed or failed to license the implicated meaning in (22) based on a technique discussed in Chierchia et al. (1998). The basic idea is to present statements like (21) either as a description of a set of events that took place in a story (in which case an implicated meaning is expected to arise) or as a prediction (or a bet) regarding what will happen in a story (in which case the implicated meaning is not supposed to arise). For further discussion of this technique, see Chierchia et al. (1998).

4.1. Method

4.1.1. Participants. We tested 20 English speaking children (9 boys and 11 girls) between the ages of 5;0 and 5;10 (mean 5;4). Three children were replaced in the design because they were either too shy or could not provide justifications for their answers. The children were selected from daycare centers in the Philadelphia area. In addition to children, we also tested a control group of 20 adult native speakers of English. The adult participants were all undergraduate students at the University of Pennsylvania.

4.1.2. Procedure. As in experiment 1, we tested participants using the TVJT. Statements like (21) were presented either as a description of the story witnessed by the participants (i.e., description mode), in which case participants were asked whether the puppet 'said it right' or 'said the right thing' (15) or as a prediction made by the puppet about what was going to happen in the story (i.e., prediction mode) (16), in which case participants were asked whether the puppet made a correct prediction. Recall from our discussion in Section (2.3) that an important property of scalar implicatures is that they can be cancelled. Moreover, scalar implicatures are typically cancelled in the context of a prediction or a bet. By using such contexts, we were therefore able to create situations in which the scalar implicature associated with sentences like (21) was not expected to arise. By contrast, a scalar implicature was expected to arise when statements like (21) were used as a description of the stories witnessed by our participants.

Let us illustrate these points by taking a closer look at the story used to test participants' interpretation of sentences like (21). In the description mode, the story begins with a strong guy, three dogs and three elephants. In order to test his strength, the strong guy tries to lift the animals one by one and to put them on a large table behind him. The strong guy begins with the dogs and easily manages to place each of them on the table. He then turns to the elephants and tries to lift the bigger one. Unfortunately, the big elephant is far too heavy for the strong guy who cannot even lift it off the floor. The strong guy then turns to the medium elephant, hoping that it is lighter. Still no luck though--elephants are heavy! Finally, the strong guy tries to pick up the smaller elephant but he still fails to lift it off the ground. In the end therefore, the strong guy was able to put all the dogs on the table but none of the elephants. At this point, the puppet describes what happened by saying that "The strong guy didn't put every elephant on the table". In this case, the puppet's use of a weaker expression (i.e., 'not every') to describe a situation which satisfies the truth conditions of a stronger term on the scale (i.e., 'not any') gives rise to pragmatic infelicity due to the scalar implicature associated with the use of the expression 'not every'. To the extent that participants are sensitive to scalar implicatures, we expect them to find the puppet's statements to be 'bad' descriptions of the stories they witnessed.

In the prediction mode, participants are shown the strong guy, the dogs and the elephants and they are told that the strong guy is going to try to put all the animals on the table behind him. At this point, the puppet remarks that the elephants look heavy and predicts that "The strong guy won't put every elephant on the table". The story then unfolds as described above (17) and at the end, participants are asked whether the puppet's prediction was correct. Here, since the scalar implicature associated with the use of 'not every' is cancelled, we expect participants to find that the puppet's prediction was correct.

[FIGURE 5 OMITTED]

As before, children first heard two pretest stories and then four test stories involving sentences like (21) and three control stories, administered in a pseudorandom order. The complete sets of statements made by the puppet on the critical and control trials are given in Tables 4-6 below.

As before, controls were designed to ensure that children could provide both YES and NO responses: the experimenter holding the puppet had a choice between two different statements for each of the control stories. One statement was true in the context of the story and the other was false. If the child had answered YES to a given test story, the experimenter holding the puppet was instructed to pick the statement for the following control story corresponding to a NO answer, and vice-versa.

Finally the 40 participants (20 children and 20 adults) were randomly assigned to the two conditions (i.e., description vs. prediction) which gave rise to a 2 x 2 design with age and testing mode as between subject factors and 10 participants per cell. The group of children assigned to the description condition ranged in age between 5;0 and 5;10 (mean 5;5) and the one assigned to the prediction condition ranged in age between 5;0 and 5;9 (mean 5;4). These two means did not differ significantly, i.e., t(18) = 1.01, p = .32.

4.2. Results

In the analysis below, we used as our dependent measure the proportion of YES responses to the puppet's statements. Beginning with participants' responses to test items, we found that in the prediction mode, both children and adults overwhelmingly accepted the puppet's statements, i.e., 85% and 82.5% of the time, respectively. In the description mode however, we found that while adults' acceptance rate was reliably lower as compared to their acceptance rate in the prediction mode, i.e., 20% acceptance rate vs. 82.5%, respectively (t(18) = -3.67, p = .0017), children's acceptance rate remained high and did not differ significantly across the two conditions, i.e., 85% in the predictive mode vs. 75% in the descriptive mode (t(18) = -0.64, p = .52). The proportion of YES responses were entered into a 2 (children vs. adults) x 2 (predictive vs. descriptive mode) analysis of variance (ANOVA). The analysis revealed a significant main effect of age (F(1, 36) = 6.2, p = .01), a significant main effect of condition (F(1, 36) = 9.9, p = .003) and a significant interaction between age and condition (F(1, 36) = 5.2, p = .02), see graph 2. On control items, children gave correct responses 93% of the time in both conditions. Adults gave correct responses 100% of the time in the descriptive condition and 96% of the time in the predictive condition.

Finally, turning to justifications, we found that adults rejected the puppet's statements 80% of the time in the descriptive condition. When asked why the puppet 'said it wrong', adults typically invoked the fact that the main character didn't V ANY of the Ns, thereby making direct reference to the stronger term on the scale (i.e., not any vs. not every). By contrast, children accepted the puppet's statements 75% of the time. When asked for justifications, children typically answered that the puppet had said 'the right thing' because the main character didn't V the Ns/every N on the table. We also found that children were perfectly capable of repeating the puppet's statements. In the predictive condition, both children and adults overwhelmingly accepted the puppet's statements (i.e., 85 and 82.5% of the time, respectively). When asked for justifications as to why the puppet should win his bet, both groups typically invoked the fact that none of the Ns had been Ved. Here again, children were perfectly capable of repeating the puppet's predictions that the main character won't V every N.

4.3. Discussion

We found that in accordance with the predictions of pragmatic theory, adults overwhelmingly rejected statements like "The strong guys didn't put every elephant on the table" (i.e., 80% of the time) when they were used to describe a situation in which none of the elephants (i.e., 'not any') were placed on the table. Moreover, when asked to justify their negative answers, adults typically explained that the puppet didn't say the right thing "because the strong guy didn't put ANY elephant on the table" thereby directly invoking the stronger term on the scale (i.e., not any). By contrast, adults overwhelming accepted such statements when they were offered as predictions regarding the stories they were about to witness (i.e., 82.5% of the time).

The second, noteworthy observation is the remarkable lack of sensitivity displayed by 5-year-olds vis-a-vis scalar implicatures. While adults in the descriptive condition rejected the puppet's infelicitous statements 80% of the time, children only did so 25% of the time. What is interesting here is that adults' semantic knowledge of the universal quantifier (revealed fully in the predictive condition) is masked by pragmatic factors in the descriptive condition, thereby giving the illusion of 'illogical' behavior (see Noveck 2001 for a similar conclusion). 5-year-olds, on the other hand, seem to not yet be sensitive to the pragmatic bias seen in adults (but see Papafragou and Musolino 2003). Children's 'logical' behavior can therefore be construed as a direct reflection of their semantic knowledge of the universal quantifier (18). Finally, note that these results comport well with other recent findings on the development of pragmatic abilities also showing that preschoolers are remarkably insensitive to scalar implicatures (Noveck 2001; Chierchia et al. 2001; Papafragou and Musolino 2003).

5. General discussion

In this study, we investigated children's knowledge of the linguistic principles governing the interaction between the universal quantifier and negation, in sentences like (23) and (24).

(23) Every N didn't VP

(24) NP didn't V every N

We observed that interpreting such sentences involves (a) a proper calculation of the scope relations between the universal quantifier and negation, based on syntactic principles such as c-command and covert displacement (b) semantic knowledge of the truth conditions associated with such sentences based in part on entailment relations holding between different interpretations and (c) the ability to derive certain conversational inferences, associated with the use of these quantificational elements.

Our results indicate that 5-year-olds have adult-like knowledge of the truth conditions associated with sentences like (23) and (24). In the case of sentences like (23), we have shown--contra Musolino et al. (2000)--that children could access both the isomorphic (i.e., 'none') and non-isomorphic interpretation (i.e., not all)--albeit with different rates of success. This finding is important because it demonstrates, among other things, that children have knowledge of the covert displacement operation that is responsible for the non-isomorphic interpretation. This is an interesting finding because covert displacement, by its very nature, is very abstract and presumably impossible to detect in the surface form of the input.

Regarding sentences like (24), previous results, i.e., Musolino et al. (2000), indicate that children are able to access the isomorphic (i.e., not all) interpretation. Our present study further demonstrates that children also know that such sentences are true in a 'none' context. This too is interesting since sentences like (24) are rarely, if ever, used to express the equivalent of sentences like NP didn't V any N, i.e., the 'none' reading. This finding therefore suggests that children have knowledge of the entailment relations discussed earlier. That is, they know that sentences like (24) are true in a 'none' context--presumably not because they have heard them being used in such contexts but rather because they have implicit knowledge of the fact that 'none' entails 'not all' and therefore that 'none' situations satisfy the truth conditions of sentences like (24). Entailment relations, just like covert displacement, are not the kind of relations that can be easily detected in the input. Their apparent mastery by 5-year-olds should therefore be regarded as an impressive achievement.

What we have shown so far is that differences in the way 5-year-olds and adults interpret sentences containing the universal quantifier and negation do not reflect grammatical differences between the two groups. That is, 5-year-olds have adult-like knowledge of the grammatical mechanisms responsible for the interpretation of such sentences (scope and covert displacement, i.e., syntactic knowledge (19)) and, consequently, adult-like knowledge of the set of truth conditions associated with such sentences (i.e., semantic knowledge). Children do differ from adults in two important respects however: first, 5-year-olds and adults display opposite preferences regarding their interpretation of sentences like Every horse didn't jump over the fence. Adults show a preference for the 'not all', non-isomorphic interpretation whereas children display a preference for the 'none', isomorphic interpretation.

At this point, it is important to emphasize that the preference displayed by adults for the 'not all' interpretation of sentences of the form Every N neg VP goes beyond the data reported in experiment 1. In this regard, it is interesting to consider spontaneous uses of sentences like every/all N neg VP. In a corpus compiled by Musolino, we find that such constructions are frequently used in TV commercials (e.g., "All salad dressings/cars/ shampoos are not the same") or at train stations ("All doors will not open"). However, their use is not the exclusive province of train conductors and marketing directors: they can also be found in magazines and newspapers (Figure 7), fortune cookies (e.g., "Everything is not yet lost", "Everybody doesn't have to be the same") and other texts (Figure 7), and they are spontaneously used by people all over the United States, ranging from distinguished academics (25)-(27) to restaurant waitresses ("All the desserts aren't on the menu"), landlords ("Everybody doesn't pay the same rent") and insurance brokers ("Everybody doesn't keep receipts for the things they buy") (see Horn 1989 for more examples).

[FIGURE 7 OMITTED]

(25) "There are extra copies of the handout on the chair here, in case everybody didn't get one." (Kenneth Wexler, linguistics colloquium, University of Maryland, October 16, 1998.)

(26) "All the birds don't seem to be quite the same" (Lila Gleitman, Psych 135 Lecture, University of Pennsylvannia, February 25, 1999).

(27) "All semantic features are not going to be under direct syntactic control" (Merrill Garrett, IRCS Colloquium talk, February 26, 1999.)

Crucially, one finds that spontaneous examples of sentences of the form Every/all N neg VP are invariably used on a 'not all' interpretation (only one of the dozens of examples from the Musolino corpus was used to convey a 'none' reading (20)). Given the fact that the grammar of English makes available two interpretations for such sentences, i.e., 'not all' and 'none', one may wonder why adult speakers of English display such a strong preference for the 'not all' interpretation, both in production and comprehension. Given the nature of the alternative interpretations of sentences like Every/all N neg VP, the preference for the 'not all' reading has all the attributes of a Gricean phenomenon (see Section 2.1). That is, adult speakers of English prefer to produce/interpret sentences like Every/all N neg VP on a 'not all' interpretation because of the availability of sentences of the form Nobody/no-none VP to express the alternative interpretation, i.e., the 'none' reading (for a more detailed account, along essentially similar lines, see Horn 1989).

An important consequence of the account discussed above regarding the nature of adult preferences--to the extent that it is along the right lines--is that it paves the way to a better understanding of children's difficulty with the 'not all' reading of sentences like Every horse didn't jump over the fence (as witnessed by Musolino et al.'s findings and our replication of their results in experiment 1). The explanation would be that children fail to access the 'not all' reading of such sentences because they are less sensitive to the pragmatic factors which gives rise to the adult preference. What is attractive about this hypothesis is that it is in perfect agreement with the results of experiment 2, which directly show that adults are sensitive to the kind of pragmatic factors described above, but that 5-years aren't. Seen in this light, children's difficulty with sentences like Every horse didn't jump over the fence loses its idiosyncratic character and can now be seen to follow from a much more general--and by now well documented (Noveck 2001; Chierchia et al. 2001; Papafragou and Musolino 2003)--fact about 5-years-olds, namely their lack of sensitivity to the pragmatic effects of entailment relations.

Another way to view the problem is in terms of current models of sentence processing. Over the past 15 years or so, advances in psycholinguistic research on human sentence processing have led investigators to conclude that the recovery procedure whereby a comprehender extracts meaning from the speech signal involves the rapid integration of multiple, probabilistic sources of information (linguistic and contextual) and that this procedure is best captured by a constraint-satisfaction model (MacDonald et al. 1994; Tanenhaus and Trueswell 1994). In the case at hand, i.e., the resolution of ambiguous sentences like Every horse didn't jump over the fence, syntactic factors (i.e., the surface hierarchical relations between the quantified subject and negation) and pragmatic factors (as discussed above) exert antagonistic influences: the former favors a 'none', isomorphic interpretation while the latter favors a 'not all', non-isomorphic interpretation. In adults comprehenders, we find that the resolution of this conflict leads to a situation where pragmatic factors are able to override syntactic factors--hence the ultimate bias towards the 'not all' reading. In children however, pragmatic factors do not yet exert such a potent influence and syntactic factors have the last word, ultimately leading to the selection of the 'none' interpretation.

Interestingly, this situation comports well with recent findings on the development of processing abilities in preschool children (Trueswell et al. 1999). These authors have found that when resolving syntactic ambiguities (e.g., PP attachment ambiguities like Put the frog on the napkin in the box) "adults showed sensitivity to ... discourse constraints at the earliest possible stages of processing, and were capable of revising incorrect parsing commitments." (Trueswell et al. 1999: 89) but that "Five Year Olds did not take into account relevant discourse/pragmatic principles [our italics] when resolving temporary syntactic ambiguities and showed little or no ability to revise initial parsing commitments." (1999: 89). In other words, Trueswell et al. found, as we did, that children relied more heavily on linguistic factors (in this case lexical biases) than they did on discourse/pragmatic factors and that this led to "systematic differences in how children and adults process spoken language ..." (1999: 89).

What remains to be explained, of course, is why children's difficulty with the 'not all' reading of sentences like Every horse didn't jump over the fence greatly diminishes when children are familiarized with the intended domain of quantification, as in Every horse jumped over the log but every horse didn't jump over the fence. Given the account developed so far, another way to put the question is to ask why such familiarization boosts children's sensitivity to the relevant pragmatic properties and thus eventually leads them to access the 'not all' interpretation. First, it is important to note that contextual manipulations akin to the one discussed above have been shown to increase children's sensitivity to pragmatic phenomena such as scalar implicatures (Papafragou and Musolino in press). In fact, this conclusion agrees with a large body of work which emphasizes the role of context and verbal instructions in children's performance on a variety of linguistic and non-linguistic tasks (Bever et al. 1968, Gelman and Greeno 1989, Greeno, Riley and Gelman 1984, Shipley 1979, Markman and Seibert 1976, Rose and Blank 1974, Samuel and Bryant 1984, McGarrigle and Donaldson 1974).

As to the precise mechanics of the account, we can at this point only offer some speculative remarks. As discussed above sentence processing is now viewed as involving the rapid integration of probabilistic cues from multiple sources of imformation. While adults can carry out this task very efficiently, we have discussed evidence that preschoolers are not yet as savvy as adults in this domain, especially when it comes to the integration of certain contextual/pragmatic factors (e.g., Trueswel et al. 1999); perhaps because of certain limitations on working memory. A case in point might be the on-line calculation of certain pragmatic inferences such as scalar implicatures. To the extent that familiarization with the intended domain of quantification (21) renders sentences containing the universal quantifier easier to process, the corresponding reduction in processing load might free up some of the computational resources deployed by children during language comprehension and thus permit the child to now attend to the pragmatic factors leading to the relevant preference. In the absence of such contextual support, children are found to rely heavily on a strong factor pulling in the opposite direction, namely the surface syntactic structure of the sentence, leading to over-isomorphic interpretations (Lidz and Musolino 2002).

As we mentioned earlier, we believe that the results presented here have interesting implications for experimental methodology. First, it has long been argued that children's non-adult behavior on tasks designed to tap grammatical knowledge are often due to violations of the felicity conditions associated with the use of certain linguistic constructions (e.g., Hamburger and Crain 1982 in the case of relative clauses and, more recently, Crain et al. 1996; Crain 2000; Meroni et al. 2001 in the case of sentences containing the universal quantifier). To quote Meroni et al. (2001):

If this is correct, then children's non-adult behavior is simply due to the difficulty they experience in responding to sentences in infelicitous contexts. This means that there is just one difference in the linguistic skills of children and those of adults--adults are apparently better able to ignore infelicities than children are. (Meroni et al. 2001: 7)

What is interesting about the results of experiment 2 (as well as the results of other recent studies of scalar implicature, e.g., Noveck 2001; Chierchia et al. 2001; Papafragou and Musolino 2003) is that they suggest just the opposite, namely that adults are highly sensitive to 'infelicities' associated with under-informative statements whereas children blatantly ignore them. These two findings need not be contradictory though. In fact, they raise interesting questions for future research on the development of pragmatic abilities. It is no longer sufficient to say that children are or are not sensitive to 'felicity' conditions. We now need to ask why children are sensitive to certain felicity conditions (e.g., those discussed by Crain and colleagues and the ones discussed in experiment 1) but not to others (e.g., those arising from certain conversational inferences).

In this regard, our results, along with other recent results (i.e., Trueswell et al. 1999; Hurewitz et al. 2001; Papafragou and Musolino 2003), indicate that children's sensitivity to pragmatic factors in language comprehension tasks should not be regarded as an all or nothing affair. In other words, the relevant question should not be whether children are sensitive to pragmatics or not but rather when, and in what circumstances, do pragmatic factors affect children's (and adults') comprehension. The interesting challenge for future research will be to specify when, how and under what circumstances children's (and adults') comprehension of spoken language is affected by pragmatic/contextual factors (for a specific proposal, see Papafragou and Musolino 2003). In the present case, such an account would amount to providing a principled explanation of children's differential sensitivity to the pragmatic effects seen in adults in the case of sentences like Every N didn't VP and NP didn't V every N on the one hand (where children are not sensitive to the effect) and Every N VP but/and every N didn't VP on the other (where children display increased sensitivity to the effect). Until further research illuminates these issues, it is fair to conclude from the present data that (a) preschooler's pragmatic abilities are more fragile that that of adults and (b) that children's sensitivity to pragmatic factors can be boosted under certain contextual manipulations.

With these considerations in mind, let us now turn to the implications of our results for experimental methodology. While there is no doubt that Crain and Thornton's (1998) TVJT has been enormously successful and has enabled researchers to learn a great deal about children's developing linguistic knowledge, certain features of the technique should nevertheless not be overestimated. In particular, two assumptions often associated with the use of the TVJT may need to be refined in light of the results presented here. The first one has to do with the kinds of inferences regarding children's underlying linguistic competence that can be made based on behavioral data collected using the TVJT. To quote Crain and Thornton (1998):

A second reason for discussing these two tasks in particular is that, when used properly, they are relatively free from the influence of performance factors that have been found to mask children's linguistic knowledge in other tasks. Properly designed studies adopting the truth value judgment task and the elicited production task have resulted in extremely consistent and reliable performance by children--indeed, performance that is often on a par with that of adults. Even when children's behavior differs from that of adults, the pattern of children's responses is so consistent that the responses can be confidently attributed to linguistic knowledge and not to performance factors. (Crain and Thornton 1998: 4)

As the present study demonstrates however, use of the TVJT does not necessarily guarantee adult-like performance, even when children do not differ from adults grammatically. In other words, the TVJT, like other tasks, is not impervious to the influence of performance factors. An important consequence of this fact is that differences between children and adults arising from the use of the TVJT should not necessarily lead investigators to conclude that such discrepancies reflect differences in linguistic competence between children and adults; a conclusion that, it turns out, was erroneously embraced by Musolino et al. (2000) regarding children's interpretation of sentences like Every horse didn't jump over the fence. In any given case, it therefore remains an empirical matter whether children's non-adult behavior reflects a difference in grammatical competence or the influence of performance factors (22).

A second, related assumption often associated with the use of the TVJT regards the biases that children are claimed to exhibit when faced with the task of interpreting ambiguous sentences. To quote Crain and Thornton (1998) again:

In our experience with children, we have been able to identify several factors that conspire to determine which reading of an ambiguous sentence is selected. First, children pick the reading that makes an ambiguous sentence true in the context, assuming that both readings are consistent with the context. (Crain and Thornton 1998: 111)

Suppose that the child has a preference for one interpretation over another and that the preferred interpretation corresponds to the "false" reading in the context that has been presented. It has been found that many children nevertheless give a high proportion of positive responses. The assumption is that children want the puppet to say things that are true. That is, the child prefers to say "Yes" if possible [our italics]. This bias to say "Yes" is apparently enough to boost the dispreferred interpretation of an ambiguous sentence in the child's mind, thereby making it easier for the child to generate that interpretation ... (Crain and Thornton 1998:211)

Here again, our results indicate that the 'Yes' bias described by Crain and Thornton should be treated with caution. Recall from experiment 1 that children were presented with contexts in which either a 'Yes' or a 'No' response to sentences like Every horse didn't jump over the fence was in principle possible. However, in this case, children did not prefer to say 'Yes', even though this was clearly an option in the context of the experiment and, more importantly, an option that is allowed by the child's grammar (as we have learned from experiment 2). This fact suggests that there may not be any hard and fast rules determining how children will behave when interpreting ambiguous sentences. Claims that children will tend to select a 'Yes' answer whenever one is made available by their grammar should therefore be treated with caution as they could lead, in the absence of disconfirming empirical evidence, to erroneous inferences regarding children's grammatical knowledge (i.e., concluding that children's grammar does not generate the reading corresponding to a Yes answer if children persist in selecting the 'No' answer of an ambiguous sentence, as in the case of Musolino et al. 2000).

In sum, while the TVJT remains a powerful tool in the arsenal of the developmental psycholinguist, methodological assumptions alone and confidence in one's experimental technique are no substitute for empirical evidence when it comes to disentangling the role of competence and performance factors in studies designed to assess children's linguistic knowledge.

Received 20 October 2003

Revised version received

13 July 2004

Indiana University

Northwestern University

References

Boysson-Bardies, Benedicte de; and Bacri, Nicole (1977). The interpretation of negative sentences. International Journal of Psycholinguistics 4, 73-81.

Berwick, R. C. (1985). The Acquisition of Syntactic Knowledge. Cambridge, MA: MIT Press.

Bever, T. G.; Mehler, J.; and Epstein, J. (1968). What children do in spite of what they know. Science 162, 979-981.

Chierchia, Gennaro; Crain, Stephen; Guasti, Maria Teresa; and Thornton, Rosalind (1998). "Some" and "or": a study on the emergence of logical form. In BUCLD 22. Proceedings of the 22nd Annual Conference on Language Development, Annabel Greenhill, Mary Hughes, Heather Littlefield, and Hugh Walsh (eds.), 97-108. Somerville, MA: Cascadilla Press.

--; Crain, Stephen; Guasti, Maria Teresa; Gualmini, Andrea; and Meroni, Luisa (2001). The acquisition of disjunction: Evidence for a grammatical view of scalar implicatures. In BUCLD 25: Proceedings of the 25th Annual Conference on Language Development, Anna H.-J. Do, Laura Dominguez, and Aimee Johansen (eds.), 157-168. Somerville, MA: Cascadilla Press.

Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: MIT Press.

Crain, Stephen (1991). Language acquisition in the absence of experience. Behavioral and Brain Sciences 14, 597-650.

--(2000). Sense and sense ability in child language. In BUCLD 24: Proceedings of the 24th Annual Conference on Language Development, S. Catherine Howell, Sarah A. Fish, and Thea Keith-Lucas (eds.), 22-44. Somerville, MA: Cascadilla Press.

--; and Thornton, Rosalind (1998). Investigations in Universal Grammar: A Guide to Research on the Acquisition of Syntax and Semantics. Cambridge, MA: MIT Press.

--; Thornton, Rosalind; Boster, Carole; Conway, Laura; Lillo-Martin, Diane; and Woodams, Elaine (1996). Quantification without qualification. Language Acquisition 5(2), 83-153.

Diesing, M. (1992). Indefinites. Cambridge, MA: MIT Press.

Drozd, Kenneth (2000). Children's weak interpretations of universally quantified questions. In Language Acquisition and Conceptual Development, Melissa Bowerman and Stephen Levinson (eds.), 340-376. Language, Culture, and Cognition 3. Cambridge: Cambridge University Press.

--; and Philip, William (1993). Event quantification in preschoolers' comprehension of negation. In Proceedings of the 24th Annual Child Language Research Forum, E. V. Clark (ed.), 72-86. Stanford, CA: Center for the Study of Language and Information.

--; and van Loosbroek, Erik (1999). Weak quantification, plausible dissent, and the development of children's pragmatic competence. In BUCLD 23: Proceedings of the 23rd Annual Conference on Language Development, Annabel Greenhill, Heather Littlefield, and Cheryl Tano (eds.), 184-195. Somerville, MA: Cascadilla Press.

Grice, Paul (1957). Meaning. Philosophical Review 67: 377-388.

--(1989). Studies in the Way of Words. Cambridge, MA: Harvard University Press.

Gualmini, Andrea (2004). Some knowledge children don't lack. Linguistics 42, 957-982.

Hamburger, Henry; and Crain, Stephen (1982). Relative Acquisition. In Language Development, vol. II, S. Kuczaj (ed.), 245-274. Hillsdale, N J: Lawrence Erlbaum.

Horn, Laurence R. (1972). On the semantic properties of the logical operators in English. Doctoral dissertation, UCLA, Indiana University, Bloomington.

--(1989). A Natural History of Negation. Chicago: University of Chicago Press.

Hornstein, Norbert (1995). Logical Form: from GB to Minimalism. Oxford: Blackwell.

Hurewitz, Felicia; Brown-Schmidt, Sarah; Thorpe, Kristen; Gleitman, Lila R.; and Trueswell, John C. (2001). One frog, two frog, red frog, blue frog: factors affecting children's syntactic choices in production and comprehension. Journal of Psycholinguistic Research 29(6), 597-626.

Jackendoff, Ray (1972). Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.

Koopman, H.; and Sportiche, D. (1991). The position of subjects. Lingua 85, 211-258.

Kramer, Irene (2000). Interpreting Indefinites: An Experimental Study of Children's Language Comprehension. Nijmegen: MPI Series in Psycholinguistics.

Ladusaw, W. (1988). Adverbs, negation and QR. In Linguistics in the Morning Calm 2, Selected Papers from SICOL-1986, 481-488. The Linguistic Society of Korea.

Lasnik, Howard (1972). Analyses of negation in English. Unpublished doctoral dissertation, MIT.

Lidz, Jeffrey; and Musolino, Julien (2002). Children's command of quantification. Cognition 84, 113-154.

MacDonald, Maryellen C.; Pearlmutter, Neal J.; and Seidenberg, Mark S. (1994). The lexical nature of syntactic ambiguity resolution. Psychological Review 101, 676-703.

May, Robert (1977). The grammar of quantification. Unpublished doctoral dissertation, MIT.

--(1985). Logical Form. Cambridge: MIT Press.

McCloskey, J. (1996). Subject positions. In Elements of Grammar, L. Haegemann (ed.), 197-235. Dordrecht: Kluwer.

McGarrigle, James; and Donaldson, Margaret (1974). Conservation accidents. Cognition 3, 341-350.

Meroni, Luisa; Gualmini, Andrea; and Crain, Stephen (2001). All for one, and one 'all' for all. Unpublished manuscript, University of Maryland.

Musolino, Julien (1998). Universal Grammar and the acquisition of semantic knowledge: an experimental investigation of quantifier-negation interaction in English. Unpublished doctoral dissertation, University of Maryland.

--; Crain, Stephen; and Thornton, Rosalind (2000). Navigating negative quantificational space. Linguistics 38(1), 1-32.

Noveck, Ira (2001). When children are more logical than adults: experimental investigations of scalar implicature. Cognition 78: 165-188.

O'Leary, Carrie; and Crain, Stephen (1994). Negative polarity items (a positive result) positive polarity items (a negative result)'. Paper presented at the Boston University Conference on Language Development.

Papafragou, Anna; and Musolino, Julien (2003). Scalar implicatures: experiments at the syntax-semantics interface. Cognition 86(3), 253-282.

Philip, William; and Aurelio, Sabina (1991). Quantifier spreading: pilot study of preschooler's "every". In Papers in the Acquisition of WH: Proceedings of the Umass Round-table, May 1990, T. Maxfield and B. Plunkett (eds.), 267-282. Amherst, MA: Umass Graduate Linguistics Students Association.

--(1995). Event quantification in the acquisition of universal quantification. Unpublished doctoral dissertation, University of Massachusetts, Amherst.

--(1996). The event quantificational account of symmetrical interpretation and a denial of implausible infelicity. In Proceedings of the 20th Annual Boston University Conference on Language Development, Andy Stringfellow, Dalia Cahana-Amitay, Elizabeth Hughes and Andrea Zukowski (eds.), 564-575. Somerville, MA: Cascadilla Press.

--; and Lynch, Emily (2000). Felicity, relevance, and acquisition of the grammar of every and only. In BUCLD 24: Proceedings of the 24th Annual Boston University Conference on Language Development, S. Catherine Howell, Sarah A. Fish, and Thea Keith-Lucas (eds.), 583-596. Somerville, MA: Cascadilla Press.

Pinker, Steven (1989). Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA: MIT Press.

Roeper, Thomas; and Matthei, Edward (1975). On the acquisition of all and some. Papers and Reports on Child Language Development 9. Stanford: Stanford University Press.

Samuel Judith; and Bryant, Peter (1984). Asking only one question in the conservation experiment. Journal of Child Psychology and Psychiatry 25, 315-318.

Snedeker, Jesse; and Trueswell, John (2001). Unheeded cues: Prosody and syntactic ambiguity in mother-child communication. Paper presented at the 26th Boston University Conference on Language Development, Boston, MA.

Tanenhaus, Michael K.; and Trueswell, John (1994). Sentence comprehension. In Speech, Language, and Communication, J. L. Miller and P. D. Eimus (eds.), 217-262. San Diego: Academic Press.

Thornton, Rosalind (1995). Children's negative questions: a production/comprehension asymmetry. In Proceedings of ESCOL, J. Fuller, H. Han, and D. Parkinson (eds.). Ithaca, NY: Cornell University.

Trueswell, John C.; Sekerina, Irina; Hilland, Nicole; Logrip, Marian (1999). The kindergarten-path effect: studying on-line sentence processing in young children. Cognition 73, 89134.

Wason, P. (1965). The contexts of plausible denial. Journal of Verbal Learning and Verbal Behavior 4:7-11.

Wexler, Kenneth; and Manzini, M. Rita (1987). Parameters and learnability in binding theory. In Parameter Setting, T. Roeper and E. Williams (eds.), 41-76. Dordrecht: Reidel.

Notes

* We would like to thank the children and staff of the following daycare centers without whom the research reported here would not have been possible: the Schoolhouse Learning Center in the Philadelphia area and the Prep School in the Bloomington area. Special thanks are due to Wendy Scott, Director of the Schoolhouse learning center and to Lia Ozolins and Amy Blouin, Director and Associate Director of the Prep School for their precious help and support during the testing phase of our study. We would also like to thank the audiences of the 2000 and 2001 Boston University Conference on Language Development and the 2001 Meeting of the Linguistic Society of America as well as audiences at Indiana and Northwestern University, the University of California at Los Angeles, Rutgers University where portions of the material reported here were presented. Special thanks are also due to members of the CHEESE seminar at the University of Pennsylvania as well as to Andrea Gualmini at the University of Maryland for helpful discussion of the issues addressed in this paper. The authors are engaged in a continuing collaboration in which the order of names alternates from one paper to the next. The authors contributed equally to the work reported here. Correspondence address: Julien Musolino, Department of Speech and Hearing Sciences, Indiana University, 200 S. Jordan Avenue, Bloomington, IN 47405-7002, USA. E-mail: musolino@indiana.edu.

(1.) Because the truth conditions of the 'none' reading are a subset of the truth conditions of the 'not all' reading, we are unable to tell whether the grammar generates this reading directly. That is, the fact that the sentence can be true in situations that are consistent with a representation in which the universal takes scope over negation does not indicate that this representation is actually produced.

(2.) The account presented here essentially follows standard neo-Gricean lines (i.e., Horn 1972, 1989 among others). For a different view, see Chierchia et al. (2001).

(3.) The mechanism involved in interpreting a universal subject inside the scope of negation requires the subject to be interpreted internal to VP (Koopman and Sportiche 1991; Ladusaw 1988; McCloskey 1996), whether this is achieved by a lowering operation (cf., Diesing 1992) or by interpreting only the "copy" of the subject which occurs in [spec, VP] and pronouncing only the "copy" in [spec, IP] (cf., Chomsky 1995).

(4.) For similar considerations in the case of sentences of the form NP didn't V some N, see Gualmini (2004).

(5.) There is by now overwhelming evidence that children as young as 3;5--and certainly 4 and 5 year-olds--experience no difficulty whatsoever with the task and are perfectly capable of giving either 'Yes' or 'No' answers when appropriate (Crain and Thornton 1998; Musolino et al. 2000; Lidz and Musolino 2002; Papafragou and Musolino in press).

(6.) In fact, six of the ten children always accepted the puppets statements (i.e., 4 times out of 4) whereas the remaining four children never did (0 times out of 4).

(7.) Also notice that the children in the replication condition, i.e., those who did not perform like adults, were on average reliably older than the children in condition 2, who performed in a more adult-like fashion (mean 5;2 vs. mean 5;7, t(18) = -3.45, p < .01).

(8.) Notice here too that the justifications provided by children in our replication condition match those reported by Musolino et al. (2000). In other words, in both studies, children overwhelmingly rejected the puppet's statements that Every N didn't VP by invoking the fact that two Ns had indeed Ved.

(9.) Notice that in spite of the improvement observed between condition 1 and 2, children in condition 2 still do not assign the non-isomorphic interpretation at an adult-like level (compare 60% vs. 100%). This shows that use of the TVJT does not guarantee adult-like performance.

(10.) For a similar conclusion regarding sentences like NP didn't V some N, see Gualmini (2004).

(11.) In fact, the logic used here, i.e., measuring performance in different contextual conditions, represents the standard approach to the resolution of competence-performance questions in the linguistic and psychological literature (e.g., Crain et al. 1996, Gelman and Meck 1983 among many others).

(12.) See McMahon et al. (2004) for experimental evidence revealing that parents do not produce intonational cues to interpretation when they produce scopally ambiguous sentences in speech to children.

(13.) Incidentally, we are not aware of a single study on the acquisition of universal quantification in which prosodic cues are manipulated.

(14.) We would like to thank Andrea Gualmini for bringing this fact to our attention.

(15.) Since we were interested in pragmatic judgments--and not judgments about truth conditions--we avoided asking participants whether what the puppet had said was TRUE or FALSE or RIGHT or WRONG. Instead, by asking participants whether the puppet 'said it right' or 'said the right thing' we hoped to direct participants' attention to the pragmatic nature of the task. For further discussion of this point, see Papafragou and Musolino (2003).

(16.) The terms 'description mode' and 'prediction mode' are borrowed from Chierchia et al. (1998).

(17.) In fact, after the strong guy fails to put the first two elephants on the table the puppet reiterates his prediction that the strong guy won't be able to put all the elephants on the table. This step was taken to ensure that children would remember what the prediction was.

(18.) As one of the reviewers pointed out, the statement made by the puppet in the predictive mode was repeated twice (see Note 19). The concern here is that repeating the statement might have blurred the distinction between prediction and description and thus led to similar behavior in the two conditions. While this possibility raises an interesting question, notice that this concern does not affect the main conclusion from experiment 2, namely the fact that preschoolers, unlike adults, fail to reject the puppet's statements in the description condition, thereby displaying a lack of sensitivity to scalar implicature.

(19.) For a demonstration that children compute scope relations on the basis of c-command relations, see Lidz and Musolino (2002).

(20.) The example in question was: "This is the case where all the variables are not equal to zero", uttered by a student in a statistics class at the University of Pennsylvania. In the case that the student was referring to, none of the variables was equal to zero.

(21.) Or the presence of a preceding affirmative statement or satisfaction of the felicity conditions on the use of negative statements discussed earlier.

(22.) Of course, one could always try to argue that the guarantees offered by Crain and Thornton (1998) only apply when the TVJT is used 'properly' and that perhaps Musolino et al. (2000) did not use the TVJT 'properly'. This line of argumentation is not likely to lead anywhere, since this study followed Crain and Thornton's prescriptions to the letter (which is unsurprising since Crain and Thornton were coauthors on the study). What Musolino, Crain and Thornton failed to do, however, is use explicit contrast statements like Every horse jumped over the log but every horse didn't jump over the fence. But this failure can hardly count as an instance of improper use of the TVJT for the obvious reason that the necessity of an explicit contrast is not something that could have been predicted in advance of the results. In sum, 'proper' use of the TVJT, i.e., use in accordance with the specific principles proposed by Crain and Thornton, is not sufficient to guarantee adult-like performance or to completely eliminate the effect of performance factors. The methodological conclusion to be drawn here is simply that an experimenter cannot always predict in advance of collecting data what kinds of discourse factors might have an effect on children's performance. Consequently, any result which appears to indicate a lack of grammatical knowledge, even a result using the TVJT methodology, should be held up to closer discourse-level scrutiny. This conclusion is fully consistent with the methodological principles of the TVJT and more generally represents one of the basic strictures of scientific inquiry: one should always distrust one's own results. Table 1. Puppet's statements on test stories in Condition 1 Test story 1 Every horse didn't jump over the fence Test story 2 Every Smurf didn't buy a cat Test story 3 Every bunny didn't play on the merry-go-round Test story 4 Every dinosaur didn't eat a tree Table 2. Puppet's statements on test stories in Condition 2 Test story 1 Every horse jumped over the log but every horse didn't jump over the fence Test story 2 Every Smurf bought a dog but every Smurf didn't buy a cat Test story 3 Every bunny played on the swing but every bunny didn't play on the merry-go-round Test story 4 Every dinosaur ate a bug but every dinosaur didn't eat a tree Table 3. Puppet's statements on control stories in both conditions Control Donald found all of his friends TRUE story 1 Donald only found two of his friends FALSE Control The big guy ate all the pizza FALSE story 2 The big guy only ate two slices of pizza TRUE Control The big guy put all the elephants on the table FALSE story 3 The big guy only put two elephants on the table TRUE Table 4. Puppet's statements on test stories in descriptive and predictive conditions Test story 1 The strong guy didn't/won't put every elephant on the table Test story 2 Joe didn't/won't buy every plane Test story 3 The karate guy didn't/won't break every brick Test story 4 The painter Smurf didn't/won't paint every balloon Table 5. Puppet's statements on control stories in the descriptive condition Control Donald found all of his friends FALSE story 1 Donald only found two of his friends TRUE Control Minnie bought all the rings FALSE story 2 Minnie only bought two of the rings TRUE Control All the girls played on the merry-go-round FALSE story 3 Only two of the girls played on the merry-go-round TRUE Table 6. Puppet's statements on control stories in the predictive condition Control Donald won't find all the rings FALSE story 1 Donald will only find two of the rings TRUE Control Minnie won't buy any of the balloons FALSE story 2 Minnie will only buy two of the balloons TRUE Control The girl won't buy a bird FALSE story 3 The girl will buy a bird and a cat TRUE