In a classic New Yorker magazine cartoon, a drowning boy is depicted crying to his dog, “Lassie, get help!”, as the dog observes from the edge of the water. In the next panel, we find Lassie lying on a psychiatrist’s couch. In the present paper, I argue that word learning without theory of mind is useless in two senses. First, because there seem to be important differences between words acquired via a theory of mind mechanism and words acquired without such a mechanism. In fact, if one of the main functions of words is to express a speaker’s disposition towards and about the world and by doing so impact a listener’s disposition towards and about the world, then the latter kind of words might be of little use. This is the sense of “useless” illustrated in the cartoon. And second, because typically developing children – who have at their disposal a theory of mind – will favor it when learning words over any other acquisition mechanism. I will first review evidence in support of these two contentions. I will then discuss how theory of mind gets recruited to the job of word learning, arguing that the mapping is not trivial. This discussion will carry implications for the debate on the coevolution of theory of mind and language capacities.

Before I get into the evidence, let me clarify briefly what I mean by “word learning” and by “theory of mind” (ToM). Word learning is the process by which a cognitive system comes to know the linguistic symbols intentionally associated with concepts. As linguistic symbols, words are arbitrary and conventional forms that stand for particular concepts. That is, words are not simple associations between sounds and percepts. Rather, words have the potential to evoke abstract, absent, and generalizable ideas or actions. Last but not least, words are intentional inasmuch as they manifest a particular disposition of the speaker towards the referents of words.

As for ToM, I do not mean the full-blown mindreading, representational, and metacognitive capacities attributable to adults and perhaps even to 5-year-olds. I do not mean false-belief understanding. Rather, I use ToM to refer to all sorts of attributions a cognitive system makes regarding the mental computations performed by a different behaving cognitive system. For infants, this might be limited to a sensitivity to other people’s intentions, desires, and emotions.

Word learning without ToM

            It is a fact: organisms without ToM can acquire words. Not only Lassie the dog, but also Alex the parrot (Pepperberg & Wilcox, 2000), Kanzi the bonobo (Savage-Rumbaugh, Murphy, Sevcik, Brakke, Williams, & Rumbaugh, 1993), and various other animal species have been shown to understand, and some even produce, word-like symbols. While I agree that these are striking discoveries regarding the cognitive capacities of these animals, it is highly debatable whether these “words” have the paradigmatic characteristics of words described above. One likely obvious shortcoming of this type of words is their capacity to refer to or describe abstract or absent concepts. A further shortcoming more central to the present argument regards the pragmatic characteristics of such words. I will not extend on this matter because others in this forum have already done so in excellent fashion (see Fitch). Reiterating his point, most of the empirical data available suggests that when communicating, animals do not take into account a listener’s state of knowledge or disposition towards the world. Their use of word-like symbols serves only to express what they perceive.

            Word learning without ToM is not exclusive to nonhuman animals. Certain human children – namely young infants and children with autism – might do so too. Recent work using looking time paradigms shows that under tightly controlled experimental situations, children as young as 14-months of age can learn word-to-object pairings (Schafer & Plunkett, 1998; Werker, Cohen, Lloyd, Casasola, & Stager, 1998). For instance, in Werker et al.’s (1998) study, after habituating to a specific word-object pairing, infants looked longer when either word or object changed. But as Werker et al. carefully conclude, before we can grant that these infants truly learned words, we need first to show that these associative capacities are somehow unique to words as opposed to general to any sound-visual pattern pairing. Second, there would have to be evidence that the infants generalize the words to other similar objects or events. And third, I would add, there needs to be evidence that these words are used to express various intents.

            But perhaps the most striking example of word learning without ToM, and at the same time the clearest case of the consequent pragmatic uselessness of words acquired in such a way, comes from studies on children with autism-spectrum disorders. On the one hand, these children, especially those with Asperger syndrome, can acquire a sizeable vocabulary, yet arguably without help from ToM. For instance, Baron-Cohen, Baldwin, & Crowson (1997) showed that children with autism have difficulty associating a novel label to an object, when the label’s referent is not the object they are attending to but rather an object the speaker is attending to. More recently, Preissler (2003) reported that the word learning difficulties exhibited by children with autism in a task similar to Baron-Cohen et al.’s (1997), were related to their difficulties in a non-verbal intentional understanding task. These latter difficulties notwithstanding, Preissler found that these children did avoid assigning two labels to an object, presumably contradicting the claim that this response results from inferences about speakers’ referential intents (Clark, 1990; Diesendruck & Markson, 2001). On the other hand, a central characteristic of these syndromes is the impairment in communication: language is commonly not used to express the range of possible dispositions towards the world, and to influence others’ dispositions towards the world.

            Taken together, the main point to be taken from this brief review is that when words are acquired without ToM, they are relatively useless for the sake of effective communication. A secondary implication of this review, however, is that it leaves open the question of what mechanisms animals, infants, and children with autism do use to acquire words. The word learning literature offers two candidate mechanisms: associative learning and lexical constraints.

            Associative learning seems to be a plausible candidate account for the word learning found in all these different populations. The small size of vocabulary acquired (except for children with Asperger), the intensive amount of training required (in all cases), the specificity of word-object links (in young infants), and the limited reliance on inferential processes (in all cases), are all consistent with an associationist mechanism.

As for lexical constraints, studies on parrots (Pepperberg & Wilcox, 2000), bonobo monkeys (Savage-Rumbaugh et al., 1993), and on children with autism (Preissler, 2003), imply that a form of mutual exclusivity might be present in these different populations. Questionable in this regard is to what extent these reported mutual-exclusivity-like phenomena – namely, the avoidance of two names for a single referent – result from the intensive training in labeling that participants in these studies underwent. For instance, in Preissler’s (2003) studies on children with autism, participants performed 16 baseline/training trials before the actual testing trials in which they eventually avoided two names for an object. For the sake of comparison, people with Williams Syndrome show a similar response pattern after only two familiarization trials (Stevens & Karmiloff-Smith, 1997). These differences are consistent with the possibility that while the mutual-exclusivity-like phenomena found in parrots, bonobos, and children with autism might result from associative learning, those found in other populations might have a different source – arguably ToM.

Word learning when ToM is available

            A current dominant view in the field is that children rely on multiple cues when acquiring words.For instance, it has been argued that children rely on lexical-specific constraints (Markman, 1989), syntactic information (Hall & Graham, 1999), and mechanisms of attention and memory (Smith, Jones, & Landau, 1996). At the same time, a number of researchers emphasize how cues to speakers’ communicative intents influence this acquisition process (Akhtar & Tomasello, 2000; Baldwin & Moses, 2001; L. Bloom, 1998; P. Bloom, 2000). An important question that arises from these theoretical perspectives regards the relative power of these different cues. A direct way to address this question is to evaluate how children respond when faced with conflicting cues.

            ToM vs. constraints. Mutual exclusivity is one of the lexical constraints presumed to guide children’s word learning (Markman, 1989). Mutual exclusivity stipulates that children avoid assigning two names for an object. In a series of studies, Diesendruck & Markson (2001) investigated an alternative explanation for this type of response. Following Clark’s (1990) Principle of Contrast, we hypothesized that this response might result from children reasoning that when speakers use two different names – or any other referential symbols – it is because they likely have different referential intents in mind, and consequently the names likely refer to different objects. In support of this hypothesis, we found that this response was not unique to the case of names, but also appeared when children were asked to assign facts to objects. That is, the response was not lexical specific but instead more general to referential contexts. Furthermore, we found that children’s avoidance of assigning two facts to an object disappeared when one of the speakers asking children for an object associated with a fact, did not know that the other object presented to children already had a fact associated with it. That is, children took into consideration a speaker’s state of knowledge in order to infer his referential intent.

            A recent study extended this conclusion also to the case of object names (Diesendruck, 2004). In this study, the state of knowledge of a speaker was manipulated by varying whether the speaker was bilingual or monolingual. Specifically, a bilingual speaker taught bilingual children a novel English name for one of two novel objects. In two critical conditions, a different speaker, who was present during the introduction of the English name, then asked children for the referent of a novel Hebrew name. The only difference between the conditions was whether this second speaker was himself bilingual or monolingual. We found that when the second speaker was bilingual, children tended to select the as-yet unnamed object as the referent of the Hebrew name. When he was monolingual, however, children selected randomly between the objects. Children seem to have reasoned that given the fact that the second bilingual speaker understood the English name provided by the first speaker to one of the objects, then his use of a different name – even if in a different language – probably reflected a different referential intent. When the second speaker was monolingual, however, children assumed he did not understand the English name used by the first speaker, and thus could not draw a clear inference about his referential intent.

            These studies indicate that at least in these limited contexts, children’s responses seem to be guided by inferences regarding speakers’ intents rather than by a priori lexical specific constraints.

            ToM vs. syntactic cues. An important source of information for children about the meanings of new words is the syntactic or morphological frame in which words are embedded (Brown, 1957). For instance, young children interpret novel words differently if the words are introduced as a count noun (e.g., “This one is a zav”), a proper noun (e.g., “This one is named zavy”), an adjective (e.g., “This one is very zavy”), or a verb (e.g., “This one is zaving”). In fact, Hall & Graham (1999) found that children rely on such syntactic cues when inferring the relation between labels. Specifically, they found that under certain circumstances, children accept a proper name and an adjective for the same entity, implying that perhaps children allow two words for the same object as long as they are from different form classes.

            A second study reported in Diesendruck (2004), investigated the effect of speakers’ knowledge state on children’s inferences of the meanings of words from different form classes. In that study, an experimenter taught children either a novel proper noun or a novel count noun for one of two unfamiliar animate creatures. A second speaker, who was either present or absent during the experimenter’s naming, then asked children for the referent of a different name. We found that when the second speaker was present, then it did not matter the lexical form of the noun taught by the experimenter: in both cases children selected the as-yet unnamed creature in response to the second speaker’s request. The crucial finding came when the second speaker was absent when the experimenter named one of the creatures. In these circumstances, the lexical form of the name used by the experimenter mattered. Specifically, when the name was a count noun, children again selected the as-yet unnamed creature in response to the second speaker’s request. However, when the name was a proper noun, children selected randomly. Children seem to have reasoned that count nouns are conventional linguistic forms known by all speakers of a language, and thus even an absent speaker is presumed to know them (Clark, 1990; Diesendruck & Markson, 2001). In contrast, proper nouns are only known by individuals familiar with the referents, and thus an absent speaker new to the situation might not know the nouns (Birch & Bloom, 2002). This is an initial piece of evidence that inferences about speakers’ state of knowledge and consequent referential intents, modulate children’s reliance on syntactic cues to the meanings of words.

            ToM vs. attentional mechanisms. Given the presumed centrality of associationist and attentional mechanisms for word learning in ToM-less organisms noted above, this conflict is of particular theoretical relevance. Indeed, this conflict has received considerable empirical attention, most of which supporting the claim that children privilege cues to speakers’ intents over automatically attention-grabbing associations.

Baldwin’s (1991, 1993) studies with 18-month-olds showed that temporal contiguity was not necessary for a name-object link to be formed, and that a link was only formed when an interacting person presented the name. These studies further revealed that children associate a novel name not with the object they are attentive to but rather with the object the speaker is attentive to. In fact, 2-year-olds rely on social cues (e.g., eye gaze) for linking names to objects, even when these cues direct children away from salient attention-grabbing aspects of the situation (Hollich, Hirsh-Pasek, & Golinkoff, 2000; Moore, Angelopoulos, & Bennett, 1999).

In contraposition to these studies, Samuelson & Smith (1998) offered an attentional learning account of phenomena previously proclaimed as resulting from intentional inferences. Specifically, Akhtar, Carpenter, & Tomasello (1996) reported that 2-year-olds relied on a speaker’s state of knowledge to infer that a novel name referred to the one object, out of four, that the speaker had not seen. Samuelson & Smith argued that this response was due to the novelty of the context in which the target object was presented, which made it memorable and attention-grabbing. In a recent study, Diesendruck, Markson, Akhtar, & Reudor (2004) demonstrated that the response was not simply due to a change in context. Specifically, 2-year-olds associated a novel name with the target object only when the change in its presentation context resulted from an intentional rather than an accidental action. Moreover, even when the change was intentional, children made the association only when the speaker doing the naming was the one who produced the change. The association was not formed when a different speaker provided the name.

Summarizing this section, in most of the studies in which ToM was directly pitted against other cues, ToM seemed to come out as the dominant mechanism used by children to learn words.

How is the mapping of ToM to word learning done? 

            Having argued for the centrality of ToM in word learning, what is left to discuss is how children recruit ToM for this task. It seems to me that this mapping involves at least three steps. First, children need to have a sufficiently developed ToM prior to substantial vocabulary acquisition. Second, children need to adequately employ their ToM for the sake of understanding human actions. And third, children need to realize that words are a kind of intentional human “product”, which they should consequently deal with through their ToM capacities.

            Do young children have the ToM it takes? The short answer is, “yes.” There is evidence that 9-month-olds encode the goals of agents (Woodward, 1998), that 10-month-olds parse dynamic actions in terms of intentional structure (Baldwin, Baird, Saylor, & Clark, 2001), that 14-month-olds distinguish between accidental and intentional actions (Carpenter, Akhtar, & Tomasello, 1998), and that 18-month-olds attribute different intentions, emotions, and desires to different people (Meltzoff, 1995; Moses, Baldwin, Rosicky, & Tidball, 2001;Repacholi & Gopnik, 1997).

            Do infants restrict ToM to humans? The answer here is, “may be not from the start!” On the one hand, a few of the studies cited earlier indicate that intentional attribution seems to be restricted to humans (Meltzoff, 1995; Woodward, 1998). On the other hand, a number of studies show that young infants attribute goal-directed behavior not only to humans but also to plain circles moving on a computer screen (Csibra, Gergely, Biro, Koos, & Brockbank, 1999), and to other non-human objects (Johnson, Booth, & O’Hearn, 2001). In fact, 1-year-oldseven rely on previously witnessed behavior of non-human agents to interpret these agents’ subsequent actions in a different context (Kuhlmeier, Wynn, & Bloom, 2003). Whatever the reasons for the differences in findings or for infants’ “overextension” of ToM to non-human agents, the studies intimate that during the first 12 to 18 months of their lives, infants’ ToM capacities undergo fine tuning, so as to eventually become consistently and systematically recruited for the processing of human actions.

            When do infants treat words as intentional expressions? Recent studies indicate that early in the word learning process, infants treat a variety of symbols as equally suitable for expressing referential intents. In particular, it has been found 12- to 18-month-olds treat emotional expressions, sounds, and gestures as referential (Campbell & Namy, 2003; Moses et al., 2001; Namy & Waxman, 1998; Woodward & Hoyne, 1999). Importantly, between 20- to 26-months of age, children start narrowing this range of acceptable referential symbols to words (Namy & Waxman, 1998; Woodward & Hoyne, 1999).

            Evidently, then, all three steps involved in the mapping of ToM to word learning are not trivial and take time. Curiously, the three processes converge to reach the minimal level of required sophistication at around the 18 to 24 month age period, perhaps not coincidentally a period in which substantial vocabulary expansion occurs.

Implications for the coevolution of language and ToM

            The argument presented so far has straightforward implications for the coevolution issue. Yes, “words” could be acquired without ToM. But these words probably would have very little similarity to what we standardly consider as words. For this latter kind of words to be acquired, a basic mindreading capacity had to be in place. Moreover, once such a capacity was in place, humans would typically and consistently recruit this capacity for learning words. This conclusion notwithstanding, as the last section of my paper points out, such recruitment might not be done automatically. Rather, the recruitment seems to involve a series of specific mappings. The fact that all the processes involved in the mappings develop independently with unique functional properties, supports the hypothesis that there need not be a special/modular mechanism devoted only for the intentional processing of words (Bloom, 2000; cf. Sperber & Wilson, 2002).

References

Akhtar, N., Carpenter, M., & Tomasello, M. (1996).  The role of discourse novelty in early word learning.  Child Development, 67, 635-645.

Akhtar, N., & Tomasello, M. (2000).  The social nature of words and word learning. In R. M. Golinkoff & K. Hirsh-Pasek (Eds.), Becoming a word learner: A debate on lexical acquisition (pp. 115-135). Oxford: OxfordUniversityPress.

Baldwin, D. A. (1991). Infants’ contribution to the achievement of joint reference. Child Development, 63, 875-890.

Baldwin, D. A. (1993).  Early referential understanding: Young children’s ability to recognize referential acts for what they are.  Developmental Psychology, 29, 832-843.

Baldwin, D. A., & Baird, J. A., Saylor, M. M., & Clark, M. A. (2001).  Infants parse dynamic action. Child Development, 72, 708-717.

Baldwin, D. A., & Moses, L. J. (2001). Links between social understanding and early word learning: Challenges to current accounts. Social Development, 10, 309-329.

Baron-Cohen, S. (1995).  Mindblindness: An essay on autism and theory of mind.  Cambridge, MA: MIT Press.

Baron-Cohen, S., Baldwin, D. A., & Crowson, M. (1997). Do children with autism use the direction of gaze strategy to crack the code of language? Child Development, 68, 48-57.

Birch, S. A. J., & Bloom, P. (2002). Preschoolers are sensitive to speakers’ knowledge when learning proper names.  Child Development, 73, 434-444.

Bloom, L. (1998).  Language acquisition in its developmental context.  In W. Damon (Series Ed.), D. Kuhn & R. S. Siegler (Vol. Eds.), Handbook of child psychology: Vol. 2, Cognition, perception, and language (5^th edition, pp. 309-370).  New York: Wiley.

Bloom, P. (2000).  How children learn the meanings of words.  Cambridge, MA: MIT Press.

Brown, R. (1957). Linguistic determinism and the part of speech. Journal of Abnormal and Social Psychology, 55, 1-5.

Campbell, A. L., & Namy, L. L. (2003) The role of social-referential context in verbal and nonverbal symbol learning. Child Development, 74, 549-563.

Carpenter, M., Akhtar, N., & Tomasello, M. (1998).  Fourteen through eighteen month infants differentially imitate intentional and accidental actions.  Infant Behavior and Development, 21, 315-330.

Clark, E. V. (1990). On the pragmatics of contrast. Journal of Child Language, 17, 417-431.

Csibra, G., Gergely, G., Biro, S., Koos, O., & Brockbank, M. (1999). Goal attribution without agency cues: the perception of 'pure reason' in infancy. Cognition, 72, 237-267.

Diesendruck, G. (2004). The principle of conventionality in word learning: Two empirical examinations. Manuscript under review.

Diesendruck, G., & Markson, L. (2001). Children’s avoidance of lexical overlap: A pragmatic account. Developmental Psychology, 37, 630-641.

Diesendruck, G., Markson, L., Akhtar, N., & Reudor, A. (2004). Two-year-olds’ sensitivity to speakers’ intent: An alternative account of Samuelson and Smith. Developmental Science, 7, 33-41.

Hall, D. G., & Graham, S. A. (1999).  Lexical form class information guides word-to-object mappings in preschoolers.  Child Development, 70, 78-91.

Hollich, G. J., Hirsh-Pasek, K., & Golinkoff, R. M. (2000).  Breaking the language barrier: An Emergentist Coalition Model for the origins of word learning.  Monographs of the Society for Research in Child Development, 65 (3), Serial no. 262.

Johnson, S. C., Booth, A., & O’Hearn, K. (2001). Inferring the unseen goals of a non-human agent. Cognitive Development, 16, 637-656.

Kuhlmeier, V., Wynn, K., & Bloom, P. (2003). Attribution of dispositional states by 12-month-olds. Psychological Science, 14, 402-408.

Markman, E. M. (1989).  Categorization and naming in children.  Cambridge, MA: MIT Press.

Meltzoff, A. N. (1995).  Understanding the intentions of others: Re-enactment of intended acts by 18-month-olds children.  Developmental Psychology, 31, 838-850.

Moore, C., Angelopoulos, M., & Bennett, P. (1999).  Word learning in the context of referential and salience cues.  Developmental Psychology, 35, 60-68.

Moses, L. J., Baldwin, D. A., Rosicky, J. G., & Tidball, G. (2001).  Evidence for referential understanding in the emotions domain at 12 and 18 months. Child Development, 72, 718-735.

Namy, L. L., & Waxman, S. R. (1998). Words and gestures: Infants’ interpretation of different forms of symbolic reference. Child Development, 69, 295-308.

Pepperberg, I.M., & Wilcox, S. E. (2000). Evidence for a form of mutual exclusivity during label acquisition by grey parrots. Journal of Comparative Psychology, 114, 219-231.

Preissler, M. A. (2003). Mutual exclusivity as a word learning constraint in children with autism. Poster presented at the Biennial Meeting of the Society for Research in Child Development, Tampa, FL.

Repacholi, B. M., & Gopnik, A. (1997). Early reasoning about desires: Evidence from 14- and 18-month-olds. Developmental Psychology, 33, 12-21.

Samuelson, L. K., & Smith, L. B. (1998).  Memory and attention make smart word learning: An alternative account of Akhtar, Carpenter, and Tomasello.  Child Development, 69, 94-104.

Savage-Rumbaugh, E., Murphy, J., Sevcik, R. A., Brakke, K. E., Williams, S. L., & Rumbaugh, D. M. (1993). Language comprehension in ape and child. Monographs of the Society for Research in Child Development, 58(3-4, Serial No. 233).

Schafer, G., & Plunkett, K. (1998). Rapid word learning by fifteen-month-olds under tightly controlled conditions. Child Development, 69, 309-320.

Smith, L. B., Jones, S. S., & Landau, B. (1996).  Naming in young children: A dumb attentional mechanism?  Cognition, 60, 143-171.

Sperber, D., & Wilson, D. (2002). Pragmatics, modularity, and mind-reading. Mind & Language, 17, 3-23.

Stevens, T., & Karmiloff-Smith, A. (1997). Word learning in a special population: Do individuals with Williams syndrome obey lexical constraints? Journal of Child Language, 24, 737-765.

Werker, J. F., Cohen, L. B., Lloyd, V. L., Casasola, M., & Stager, C. L. (1998). Acquisition of word-object associations by 14-month-olds infants. Developmental Psychology, 34, 1289-1309.

Woodward, A. L. (1998).  Infants selectively encode the goal object of an actor’s reach.  Cognition, 69, 1-34.

Woodward, A. L., & Hoyne, K. L. (1999). Infants’ learning about words and sounds in relation to objects. Child Development, 70, 65-77.