METHOD

Materials

Six pairs of objects were used in the study. Each pair consisted of one novel object and one familiar object. The pairs are listed in Table 1. Additionally, an infant's sock and a small plastic horse, tree and bed were used in the warm-up procedure.

TABLE 1.

Novel–familiar pairs used in Studies 1 and 2

RESULTS AND DISCUSSION

Consistent with the results from Jaswal & Hansen's (Reference Jaswal and Hansen2006) study with older children, toddlers were more likely to select the familiar object following a request for it than a request involving a novel word. In the gaze-baseline and point-baseline conditions, they selected the familiar object on 3·38 (SD=1·59) and 3·00 (SD=1·64) of 6 trials, respectively, and in the gaze-word and point-word conditions, they selected the familiar object on just 1·33 (SD=1·54) and 2·12 (SD=2·34) of 6 trials, respectively. A 2×2 (pragmatic cue×request type) analysis of variance (ANOVA) on these data showed a significant effect of request type (F(1, 60)=10·27, p=0·002, η_p²=0·15). It did not matter whether the pragmatic cue involved gaze or pointing (F(1, 60)<1), nor was there an interaction between cue and request type (F(1, 60)<1·62).

We next analyzed how often the hypothesis-blind coder indicated that children followed the speaker's gaze or point to the familiar object. This was of interest because we wanted to ensure that children in the word conditions had attended to the speaker's pragmatic cue as much as those in the baseline conditions. Children followed the researcher's gaze on an average of 3·60 (SD=1·64) and 3·50 (SD=1·21) of 6 trials in the gaze-word and gaze-baseline conditions, respectively. They followed the researcher's point on an average of 4·35 (SD=1·50) and 3·56 (SD=1·21) of 6 trials in the point-word and point-baseline conditions, respectively. A 2×2 (pragmatic cue×request type) ANOVA on these data showed no significant effects and no interaction (Fs<1·60). This is important because it indicates that, although children did not always attend to and follow the researcher's pragmatic cue, they were equally likely to do so in all four conditions.

In the next set of analyses, we compared how likely children were to select the familiar object when they followed the researcher's gaze or point with how likely they were to do so when they did not, as a function of the type of request they heard. Figure 1 shows results for the gaze and pointing conditions separately. A 2×2×2 (pragmatic cue×request type×cue-follow) ANOVA on these data showed that there was no effect, nor any interactions, involving the type of pragmatic cue involved (Fs<1): children responded similarly in the pointing and gaze conditions. The other two main effects, however, were significant: children were more likely to select the familiar object when they followed the researcher's gaze or point than when they did not (F(1, 54)=55·76, p<0·0001, η_p²=0·51), and they were more likely to select the familiar object when they heard a request for it than a request involving a novel word (F(1, 54)=18·30, p<0·0001, η_p²=0·25). In addition, there was a significant interaction between these two main effects (F(1, 54)=11·26, p=0·002, η_p²=0·17).

Fig. 1.

Average percentage of selections of the familiar object in Study 1, as a function of whether children followed the researcher's gaze (top) or point (bottom), and whether they heard a baseline request or a novel word request. Error bars show SEM.

We investigated this interaction using simple main effects analyses. When children followed the researcher's cue to the familiar object, they were more likely to select that object than when they did not follow, regardless of request type (Word: F(1, 54)=7·77, p=0·007, η_p²=0·12; Baseline: F(1, 54)=65·92, p<0·0001, η_p²=0·54). Thus, children in the word conditions were not immune to the effects of the speaker's gaze or point; it had a measurable influence on their responses. Importantly, however, when children in the baseline conditions followed the experimenter's cue to the familiar object, they were significantly more likely to select that object than when children in the word conditions followed the experimenter's cue (F(1, 54)=23·44, p<0·0001, η_p²=0·30). When children in the word and baseline conditions did not follow the experimenter's gaze or point to the familiar object, they were equally likely to select the familiar object (F(1, 54)<1·64).

The discounting of the speaker's pragmatic cue in the word conditions was also evident at the individual level, as evaluated by chi-squared tests. Whereas 12 of the 16 children (75%) in the gaze-baseline condition selected the familiar object on 75% or more of the trials where they followed the researcher's gaze to that object, only 2 of 15 (13%) in the gaze-word condition did so (χ² (1, N=31)=11·89, p<0·0001). Similarly, whereas 11 of the 16 children (69%) in the point-baseline condition selected the familiar object on 75% or more of the trials where they followed the researcher's point to that object, only 4 of the 17 children in the point-word condition did so (χ² (1, N=33)=6·80, p=0·009).

The disambiguation task was a two-alternative forced-choice, so the proportion of selections of the familiar and unfamiliar objects always summed to 1. Because of this problem of dependence, it is not possible to compare directly the proportion of selections of familiar objects to the proportion of selections of unfamiliar ones. However, one can investigate whether one response is more frequent than the other by comparing the proportion of selections of the familiar object to chance of 50%. If children select the familiar object more frequently than expected by chance, this indicates that they selected the familiar object more often than the unfamiliar one. If they select the familiar object less frequently than expected by chance, this indicates that they selected the unfamiliar object more often than the familiar one. Given that none of the analyses thus far have identified any differences between the two types of pragmatic cues (gaze and point), we collapsed across cue type in order to increase power for these chance analyses.

On those trials where they did not follow the researcher's cue, children had a preference for the unfamiliar object regardless of the type of request they heard: they selected the familiar object just 11% (SD=28%) of the time when they heard a request for a new word and 24% (SD=33%) of the time when they heard a request for it, both less often than expected by chance (ts>4·5, ps<0·0001, ds>0·80). When children did follow the researcher's cue, however, as the earlier analyses have shown, the type of request they heard had an important influence. When they heard a request for it and followed the speaker's cue, they selected the familiar object 78% (SD=28%) of the time, more than would be expected by chance (t(31)=5·76, p<0·0001, d=1·02). In contrast, when they heard a request for a novel word and followed the experimenter's cue, they selected the familiar object 36% (SD=38%) of the time, less often than expected by chance (t(31)=2·14, p=0·04, d=0·38) (meaning that they selected the novel object more often than expected by chance).

In summary, children in both the word and baseline conditions were influenced by the speaker's behavioral pragmatic cue indicating the familiar object: in both cases, they were more likely to select the familiar object when they followed the cue than when they did not. However, when children in the baseline conditions followed the speaker's pragmatic cue to the familiar object, they were much more likely to be influenced by that cue than when children in the word conditions followed the cue. Indeed, chance analyses showed that when children in the baseline conditions followed the experimenter's cue, they selected the familiar object more frequently than expected by chance; when children in the word conditions followed the experimenter's cue, they nevertheless selected the novel object more frequently than expected by chance. Thus, as in Jaswal & Hansen's (Reference Jaswal and Hansen2006) study with older children, many toddlers seem to discount some interpretable, behavioral pragmatic cues that conflict with mutual exclusivity.

Given that children can and do learn that a single object can be referred to by multiple labels (e.g. poodle–dog–animal), one question concerns what behavioral pragmatic cues might be more effective in leading them to violate mutual exclusivity. Jaswal & Hansen (Reference Jaswal and Hansen2006) suggested that there were likely to be pragmatic cues that would lead children to violate mutual exclusivity, but they did not investigate this question empirically. In Study 2, we combined the two cues the speaker provided separately in Study 1: the speaker both looked at and pointed toward the familiar object when making a request for the referent of a new label. We reasoned that although neither of those cues alone was sufficient to lead children in the word conditions to infer at above-chance levels that the label referred to the familiar object, the two together might be. Although pragmatic cues are sometimes considered to have all-or-none rather than graded effects, there is some reason to think that a combination of cues could be more powerful than a single cue on its own.

For example, in Hollich et al. (Reference Hollich, Hirsh-Pasek, Golinkoff, Brand, Brown and Chung2000, Study 8), one-year-old infants saw an experimenter look at one of two novel objects and label it repeatedly. During this training phase, infants spent more time looking at the object being labeled than at the other one, showing that they were sensitive to the referential intent of the speaker. However, on a subsequent comprehension test, they did not show evidence of having mapped the new label to the target. In another study (Study 7), however, infants saw the experimenter look at one of two novel objects and handle it while labeling it repeatedly. On the subsequent comprehension test, infants did show evidence of having mapped the label to the target. Hollich et al. argued that infants were better able to learn a new word given multiple, overlapping social cues. Although the children in the present studies are a year and one-half older than those in Hollich et al., it is possible that they, too, will find multiple social cues more persuasive than a single one.

We did not include a baseline condition in Study 2 because, as will be described below, children in a word condition nearly always selected the familiar object as the referent for a new name when they saw the speaker both look and point at that object.

METHOD

RESULTS AND DISCUSSION

Toddlers overwhelmingly selected the familiar object rather than the unfamiliar one as the referent for the new name. They selected the familiar object on an average of 5·19 (SD=1·60) of 6 trials (87% of the time), more often than expected by chance of 3 (t(15)=5·47, p<0·0001, d=1·36). In fact, 11 of the 16 children (69%) selected the familiar object on each of the 6 trials. It is possible that had we included a condition in which children observed an experimenter both point and look at the familiar object when making a neutral request (i.e. a baseline condition), children would have selected the familiar object more often than 87% of the time. However, it seems unlikely that such a manipulation with children aged 2 ; 6 would produce responses much closer to 100%.

Recall that in the word conditions of Study 1, children followed the researcher's gaze or point to the familiar object 60% (3·60 out of 6 trials) and 73% (4·35 out of 6 trials) of the time, respectively. In contrast, in Study 2, children followed the researcher's gaze and point to the familiar object 97% (5·82 out of 6 trials) of the time. Indeed, 13 of the 16 participants in Study 2 followed the researcher's gaze and point on all six trials; the other three children followed on five of the six trials. This suggests that the combination of the two cues was more salient than either of the cues on its own.

Although Studies 1 and 2 were conducted separately, the objects, counterbalancing, novel words, and point and gaze gestures were the same. Thus, we next compared how often children in the two word conditions of Study 1 selected the familiar object with how often they did so in Study 2. In the gaze-word and point-word conditions of Study 1, when they followed the researcher's cue to the familiar object, they selected that object, on average, just 31% (SD=36%) and 39% (SD=39%) of the time, respectively. In the gaze+point condition of Study 2, they selected the familiar object, on average, 87% (SD=27%) of the time. A one-way ANOVA on these data confirmed that children in Study 2 were more likely to select the familiar object as the referent for the new name than children in either condition of Study 1 (F(2, 45)=11·71, p<0·0001, η_p²=0·34). Thus, seeing the researcher both point and look at the familiar object when requesting a new word was more likely to lead children to select that object than just seeing the researcher point or look toward it.

General discussion

Toddlers in Study 1 weighted a speaker's overt pragmatic cues differently, depending on whether those cues accompanied a neutral request or a request involving a new word. Children who saw the speaker look at or point toward a familiar object in a novel–familiar pair were more likely to select the familiar object when the speaker requested it than when she requested e.g. the blicket. Hearing a new word seemed to lead children to discount an otherwise interpretable behavioral cue suggesting that the speaker intended to refer to the familiar object. Study 2 demonstrated one circumstance in which toddlers did consistently violate mutual exclusivity on the basis of behavioral pragmatic cues: when the speaker both looked at and pointed toward the familiar object, children overwhelmingly selected that object as the referent for a new word.

These findings are noteworthy for several reasons. First, Study 1 replicates Jaswal & Hansen's (Reference Jaswal and Hansen2006) results and extends them to an age group an average of 1.5 years younger. Like the preschoolers in Jaswal & Hansen, the toddlers in Study 1 were less likely to select the familiar object (and therefore more likely to select the novel one) in response to a request involving a new label than to a neutral request – even though overt behavioral cues in both conditions were available to suggest that the speaker intended to refer to the familiar object.

Second, we directly compared the likelihood that toddlers would discount gaze and pointing gestures that conflicted with mutual exclusivity. Some researchers have pointed out that one's eyes often wander to locations absent-mindedly or without any intention to communicate (Johnson et al., Reference Johnson, Ok and Luo2007). In contrast, it is unlikely that a hand would take the configuration of a point unless it was intended to convey referential information. As a result, gaze cues may be less readily interpreted as intentional than pointing gestures. However, analyses in Study 1 failed to reveal any differences between the gaze and pointing conditions: in all cases, children were less likely to select the familiar object of a novel–familiar pair when they heard the speaker make a request involving a new word than when they heard her make a neutral request.

The pattern of results obtained in Study 1 is consistent with the explanation of the disambiguation effect derived from mutual exclusivity, the proposed default assumption children have that category labels are mutually exclusive (e.g. Behrend, Reference Behrend1990; Markman, Reference Markman1989). The pattern of results is less consistent with the social–pragmatic account: if children had been reasoning about the speaker's communicative intent, those in the word conditions should have been as likely as those in the baseline conditions to select the familiar object, which the speaker was looking at or pointing toward.

Before continuing, it is important to address the possibility that children discounted the gaze and pointing gestures used in Study 1 because they were somehow unnatural. Our baseline conditions show that, in fact, the gestures were not unnatural; they were readily interpretable cues to the speaker's communicative intent. Children in the baseline conditions who followed the speaker's gaze or point to the familiar object overwhelmingly selected that object. Children in the word conditions were as likely to follow the speaker's gaze or point to the familiar object as those in the baseline conditions. When they followed her gaze or point, they were more likely to select the familiar object than when they did not. However, they were significantly less likely than those in the baseline conditions to subsequently select that object, and they were less likely than chance to select that object (meaning they were more likely than chance to select the novel object). When children in Study 1 did not follow the speaker's gaze or point to the familiar object, the two conditions did not differ: in this case, children in all conditions preferred the novel object.

One could argue that the children in Study 1 were faced with a conflict between two social–pragmatic expectations – an expectation about the speaker's intent based on contrast and conventionality and one based on where she was pointing or looking. Perhaps contrast and conventionality were simply weighted more heavily than ostensive communicative gestures. We cannot rule out this possibility, but it seems unlikely because it would mean that children favored a non-obvious inference about a speaker's intent (i.e. if she had meant the shoe, she would have said shoe) over an obvious cue to her intent (i.e. her gaze or point toward the shoe). Perhaps the pragmatic account could be modified to weight non-obvious inferences about intent more heavily than obvious cues, but we are unaware of any theoretical justification for doing so.

One could also argue that children in the word conditions of Study 1 construed the situation as a game in which the speaker's gaze or point was an attempt to deceive them. This is possible, but unlikely in light of research that shows that children older than those used here have difficulty understanding that communicative gestures can be used deceptively. For example, Couillard & Woodward (Reference Couillard and Woodward1999) found that three-year-olds who were told that an experimenter was going to try to trick them about where an object was hidden nonetheless often searched in the location where that experimenter pointed instead of in the other available location.

Although the expectation that each object has just one category label is so strong that children will discount some pragmatic cues that conflict with it, there are clearly situations in which it will be overridden. Jaswal & Hansen (Reference Jaswal and Hansen2006) had made the argument that in a disambiguation setting, some pragmatic cues (or a combination of cues) could lead children to override mutual exclusivity, but they did not investigate this empirically. Study 2 showed that the combination of a speaker's gaze and point toward the familiar object was sufficient to lead children to override mutual exclusivity. We expect that many other pragmatic and/or linguistic cues would undoubtedly also be sufficient. For example, when children hear a new label introduced in the context of an inclusion relationship (e.g. ‘a poodle is a kind of dog’), they seem to readily accept that the same referent can have two labels (e.g. Callanan & Sabbagh, Reference Callanan and Sabbagh2004; Clark & Grossman, Reference Clark and Grossman1998). Clearly, then, given sufficient pragmatic and/or linguistic evidence, children can learn multiple labels for a familiar object. But, to return to the results from Study 1, a single pragmatic cue (i.e. pointing alone or gaze alone) indicating that the speaker intended to refer to a familiar object did relatively little to influence their expectation that the referent of a new word was a novel object.

These results add to a growing body of work suggesting that the mutual exclusivity assumption explains aspects of early word learning that other theories do not. For example, Golinkoff, Mervis & Hirsh-Pasek (Reference Golinkoff, Mervis and Hirsh-Pasek1994) suggested that children's performance on disambiguation tasks may reflect an expectation children have that new words map onto objects from categories for which they do not yet already have a name (the novel name–nameless category principle) rather than an expectation that each object has just one label. On this account, when children see a new object, they are motivated to fill a gap in their lexicon (see also Merriman & Bowman, Reference Merriman and Bowman1989). Children do seem to be motivated to learn the names of new objects (e.g. Clark, Reference Clark, Gelman and Byrnes1991; but see Kemler Nelson, Egan & Holt, Reference Kemler Nelson, Egan and Holt2004), but a recent study by Markman et al. (Reference Markman, Wasow and Hansen2003, Study 3) casts doubt on lexical gap-filling as an explanation for the disambiguation effect. In Markman et al., infants as young as 1 ; 3 were presented with a single, familiar object (e.g. a spoon). When the experimenter requested a crimp, children often engaged in search behaviors, looking around the room and under the table. Even though there was no new object available as a referent for crimp (and thus, no lexical gap to fill), they were apparently reluctant to assume that the new word referred to a familiar object (see also Liittschwager & Markman, Reference Liittschwager and Markman1994).

Diesendruck & Markson (Reference Diesendruck and Markson2001) argued that one strength of the social–pragmatic account is that it predicts the disambiguation effect whenever a speaker expresses contrasting referential intentions, not just when new words are involved. They found that when two novel objects were introduced, and a speaker provided a fact about one of them (e.g. ‘This is the one my uncle gave me’), three-year-olds tended to assume that the other object was the referent for a different fact (e.g. ‘Can you give me the one my dog likes to play with?’). However, Scofield & Behrend (Reference Scofield and Behrend2007) recently reported that children are more likely to show the disambiguation effect with some types of information than others. In their study, the speaker provided a word or fact about one of two novel objects, and then made a request for the referent of a different word or fact. Children at all ages tested, from 2 ; 6 to 4 ; 0, disambiguated when a word was provided for one object and the referent of another word was requested. But the children aged 2 ; 6 did not do so in any of the other possible combinations of words and facts (e.g. word–fact, fact–word, fact–fact), and three- and four-year-olds showed inconsistent patterns in these other combinations as well. Scofield & Behrend (Reference Scofield and Behrend2007: 884) argued that that these results challenge the social–pragmatic explanation for the disambiguation effect because ‘disambiguation as a pragmatic process should presumably be applied broadly to a variety of referential acts’.

The present work has focused on how mutual exclusivity can take precedence over some behavioral cues to a speaker's intent; other research has shown that mutual exclusivity can also take precedence over contextual pragmatic cues. In a study by Haryu (Reference Haryu1991; described by Haryu & Imai, Reference Haryu and Imai1999), children heard a researcher explain, ‘Mary [a puppet] is hungry now. I would like to give Mary the heku.’ They were then presented with an apple (a familiar object) and a lipstick holder (a novel object), and asked to select the heku. Haryu argued that, given that Mary was just said to be hungry, the most pragmatically felicitous response would be to assume that heku referred to the apple. However, children aged 3 ; 6 did not do so; they tended to select the lipstick holder as the referent for heku. Importantly, a control condition showed that children did understand the pragmatics of this situation; they knew that when Mary was hungry she would want the apple rather than the lipstick holder.

Finally, the central tenet of the social–pragmatic account of the disambiguation effect is that children select the novel object rather than the familiar one because of inferences they make about the speaker's intent (e.g. Bloom, Reference Bloom2000; Diesendruck & Markson, Reference Diesendruck and Markson2001). However, children with autism, who have well-known difficulties reading other people's intentions (e.g. Baron-Cohen, Reference Baron-Cohen1995), are as likely to select the unnamed object in a disambiguation task as typically developing children (Preissler & Carey, Reference Preissler and Carey2005). Additionally, some non-human animals, who presumably do not have an understanding of contrast and conventionality, respond to disambiguation situations in a similar way to humans. For example, Kaminski, Call & Fischer (Reference Kaminski, Call and Fischer2004) found that a border collie tended to select an object it did not yet know the name for rather than any of several familiar objects when asked to retrieve the referent of a new word. One could argue that there are two (or more) routes to the disambiguation effect: one that requires the reading of the speaker's communicative intent, and one that does not. However, given the results of Study 1, in which children discounted an overt behavioral cue to the speaker's intent, it is not clear what additional benefit the social–pragmatic route to the disambiguation effect provides.

The controversy we have highlighted is not just a debate about which of two explanations best explains children's performance on a particular task. It represents a fundamental disagreement about the nature of early word learning. Constraint theorists recognize the importance of social–pragmatic cues to a speaker's intent (e.g. Hollich et al., Reference Hollich, Hirsh-Pasek, Golinkoff, Brand, Brown and Chung2000; Woodward & Markman, Reference Woodward, Markman, Kuhn and Siegler1998), but they argue that without additional constraints on the hypotheses children consider, the inductive problem posed by word learning would be insurmountable. Social–pragmatic theorists counter that constraints actually reflect pragmatic reasoning processes (e.g. Bloom, Reference Bloom2000; Clark, Reference Clark1997; Nelson, Reference Nelson1988; Tomasello, Reference Tomasello, Bowerman and Levinson2001). The studies described here suggest that, in fact, mutual exclusivity does not reduce to a pragmatic reasoning process.