This paper has been written with Lance Slade and Mele Taumoepeau.

There is now abundant evidence that false belief understanding in children is linked to their language ability. Although there is clearly more to a theory of mind than false belief alone, false belief is typically regarded as the aci test. Some have claimed that it is a particular aspect of syntax, sentential complements, that relates to false belief understanding (deVilliers & Pyers, 2002; Hale & Tager-Flusberg, 2003). Yet there are both logical grounds (Astington & Jenkins, 1999; Ruffman, Slade, Rowlandson, Rumsey, & Garnham., 2003; Slade & Ruffman, in press), and empirical grounds (Perner, Sprung, Zauner, & Haider, 2003) for doubting such a claim. Others have claimed that it is general syntax that assists false belief understanding (Astington & Jenkins, 1999). Yet direct tests of semantics (e.g., receptive vocabulary) versus syntax (e.g., word order) have established that general language ability (e.g., syntax + semantics) is important rather than syntax or semantics per se. This has been established both within a single time point (Ruffman et al., 2003), and longitudinally (Slade & Ruffman, in press).

One of the important questions is why language is linked to theory of mind. We have argued that semantics and syntax both relate to theory of mind because, in normal development, syntax is a clue to semantics and semantics is a clue to syntax. These ideas have been referred to, respectively, as syntactic bootstrapping (e.g., Naigles, 1996; 1990; Naigles & Hoffginsberg, 1995; Pinker, 1984), and semantic bootstrapping (e.g., Rondal & Cession, 1990). The idea is that both syntax and semantics develop rapidly in childhood and assist theory of mind by helping the child to reflect on and refine implicit knowledge about mind. That is, we have proposed the following. In normal development children’s initial understanding of mind is likely implicit and manifest in their behavior rather than insights that they can verbalize. This understanding could be innate, but learning likely plays a large role given children’s interest in the human face (which acts as a window into the mind), or their observations of the social world which allow them to utilize their capacity for statistical learning (e.g., Saffran, Aslin, & Newport, 1996) to pattern match (i.e., connect various social outcomes with certain preconditions). Over time, and as their language develops, children develop a consciously mediated and verbally based theory on the basis of these implicit intuitions.

We have also argued that children with autism understand the world in a different way (Ruffman, Garnham, & Rideout, 2001). They might lack the initial implicit understanding, and as a result, the social understanding that the gifted group of autistic children eventually develop might be based on processes of self-motivated induction or teaching from others rather than reflecting on implicit intuitions to create explicit theories. One consistent finding is that individuals with autism do not look to the eyes in social situations (Klin, Jones, Schultz, & Volkmar, 2003). This might reflect an absence of implicit knowledge or might be the reason such individuals do not develop implicit social knowledge. If this is correct, then both children with autism and normally developing children develop some sort of an explicit, verbally-based understanding of mind, but children with autism cannot base their understanding on a rich set of implicit intuitions. Further, we have argued that language assists in working out explicit theories because it provides the terminology for thinking explicitly about a mental state (Ruffman et al., 2003). Language allows the child to think explicitly about a person pretending “x” versus thinking “x” (for making fine distinctions between different propositional attitudes and contents), and it facilitates understanding of the causal origins and implications (e.g., subsequent actions) of mental states, enabling explicit predictions. There are a number of research strands that come together to support the above ideas. These are discussed below.

Implicit understanding?

We have obtained evidence that children’s initial understanding of false belief is implicit (i.e., not conscious). Children were given a standard false belief task in which the character placed an object in a left-hand location and then went away (Ruffman, Garnham, Import, & Connolly. 2001). While absent, the object was moved to the right-hand location. Children were given three measures: (a) they were asked the standard verbal question regarding where the character would look for the hidden object, (b) their eye gaze was monitored as to where they looked when anticipating the story character’s return (as in Clements & Perner, 1994), and (c) they were asked to “bet” 10 plastic counters in any configuration on the location(s) they thought the story character would return. Like Clements and Perner, we found that young children often looked to the correct (left-hand) location when anticipating the character’s return, but explicitly claimed he would go to the right-hand location. The betting measure allowed us to determine whether eye gaze indexed low confidence conscious knowledge (e.g., children understood the character might go to the left-hand location but weren’t sure about this so said he would go the right-hand location), or genuinely unconscious knowledge. In the youngest children who showed correct eye gaze but incorrect verbal answers, their eye gaze indexed genuinely implicit knowledge. Despite looking to the left-hand location, these children bet all of their counters on the right-hand location (where they said the character would go). They showed no awareness of the knowledge manifest in their eye gaze. Further, a control condition showed that betting was sensitive to small variations in children’s certainty. For instance, they bet with much less confidence when guessing whether an object would be red or green when a bag contained 9 red objects and 1 green (9-1), in comparison to when it contained 10 red objects and 0 green (10-0).

Noverbal understanding in autism.

In a different study (Ruffman, Garnham, & Rideout, 2001), we found that children with autism were unimpaired relative to a group of children with MLD (moderate learning difficulty) on verbal questions tapping social understanding, but they were impaired on a measure of eye gaze. Nevertheless, their eye gaze was unimpaired on the 10-0 and 9-1 tasks (tapping non-social knowledge) described above. Further, we found that in comparison to verbal performance, eye gaze was a much better correlate of teachers’ ratings of how severely autistic children were. This finding, along with our finding that eye gaze on the social tasks differentiated the autistic and MLD groups whereas verbal performance did not, suggests that the core insight is the nonverbal not the verbal insight.

Is the language-theory of mind relation bi- or uni-directional?

There have been three longitudinal studies relevant to this question. Astington and Jenkins (1996) found that language at an early time point predicted later theory of mind at two of three sets of time points, whereas the reverse relation never held. de Villiers & Pyers (2002) found that language predicted later theory of mind on two occasions over their three time points, and that the reverse relation held once. We found that sampling differences (the language items randomly selected from a composite language measure when evaluating language and social understanding) have a large impact on the end result (Slade & Ruffman, in press). In essence, we found that early false belief understanding was as likely to correlate with later language (receptive vocabulary) as the reverse. This finding fits with the idea that children’s theory of mind assists them in learning about word meanings (e.g., Baldwin, 1991).

How language relates to verbal and nonverbal social understanding.

We have found that language tends to correlate more highly with verbal measures than it does with nonverbal measures of social understanding. These studies have been conducted both with normally developing children and with atypical children, and have involved a range of ages and nonverbal measures. In one study, 2- to 4-year-old children were given a task that tapped their ability to understand how someone would react in a situation analogous to the visual cliff (Ruffman, 2000; in preparation). We called this measure the Emotion-Behaviour task and tested 39 children in two experimental tasks. In each task there were two rooms (e.g., red and green) and two associated windows, and children were told one room was safe and the other was not. A boy asked his father whether the red room was safe to enter. In one task the father smiled (a happy emotion expression was placed on his face), and in the other he looked fearful. Having seen this expression, the story character set off for one of the rooms (in the general direction of both). The child had been shown that the story character would eventually appear in the window of the room he entered and we were interested in whether children understood how the father’s emotional expression would lead the story character to a particular room. To elicit nonverbal understanding, the child then heard a prompt narrated on audio tape which was meant to direct the child’s eye gaze towards one of the rooms in expectation of the story character appearing there ("I wonder which window Sam will go to?"), and we videotaped their anticipatory eye movements.  The child was then directly asked a verbal question ("Which window will Sam go to?"). Two-, 3-, and 4-year-olds were all above chance on the nonverbal measure, but only the 4-year-olds were above chance on the verbal measure. Over all children, their composite score on a language measure correlated with verbal performance, r = .54, p < .001, but not with their eye gaze, r = .10, n.s.

Recently, we have examined emotion recognition in 40 7- to 9-year-olds. Children were given three blocks of trials. In the predictable block they viewed an emotion face on a computer monitor. There were twelve different faces: two each of fear, sadness, anger, disgust, surprise, and happiness. Each face was comprised of an emotion morph (e.g., 60% sadness, 40% fear). There were also two possible emotion labels at the bottom of the screen, one for each of the emotions comprising the emotion morph (e.g., “sad” and “fear”). These labels corresponded to two keys on the keyboard. After a few seconds the emotion face disappeared and was replaced by a target emotion word. In the predictable block, the target emotion word always matched the dominant emotion of the emotion face (e.g., “sad”). When the target word appeared, children’s task was to press the key on the keyboard that matched the target word as quickly as possible. In the predictable block, because the emotion face and target emotion word always matched, the face could act as a prime and reduce reaction times (i.e., understanding of the emotion present in the face would lead to the expectation that the target word would be “sad”).

The unpredictable block was identical, and used the identical emotion faces, except that there was no consistent relation between the emotion face and the target word. Half the time they matched, and half the time they mismatched. Reaction times were expected to be longer because the emotion face wouldn’t help predict the target word. The predictable and unpredictable blocks were counter-balanced, and were always followed by the verbal trials in which children were sequentially presented with the 12 emotion faces and two labels (e.g., “sad” and “fear”), and were asked to explicitly label each face.

As expected, reaction times were significantly faster in the predictable block than the unpredictable block. Thus, the emotion faces were priming anticipation of the target emotion word, thereby speeding reaction time. This indicates that children possessed knowledge on some level of what emotions were expressed in the faces. More interestingly, reaction times seemed to be tapping knowledge that was mainly implicit in that faster reaction times on the predictable block were in no way related to the number of emotion faces children explicitly identified on the verbal trials when asked directly to label each face. Thus, whether children were correct when verbally labeling 4 to 6 faces, 7 to 9 faces, or 10 to 12 faces, their reaction times were quicker on the predictable block by the same margin relative to the unpredictable block. In other words, it seemed to be implicit knowledge of the emotions that primed children’s expectation as to what target word would appear, and sped up reaction times in the predictable block. As above, language (receptive vocabulary) correlated significantly with the percentage of items correct on the verbal measure, r = .33, p < .05, but not with the percentage of items correct on the nonverbal measure (the percentage of items in which reaction time was quicker on the predictable block), r = -.15, n.s.

Finally, we examined language-theory of mind relations in autism and MLD in the study described above (Ruffman, Garnham, & Rideout, 2001). In the children with autism, a composite measure of language correlated with verbal performance on the tasks tapping social understanding, r = .42, p < .05, but not with the measure of eye gaze on the social tasks, r = .10, n.s. Only in the children with MLD was the pattern different. In this group, language correlated with both verbal performance, r = .46, p < .01, and with eye gaze, r = .37, p < .05.

To sum up over all studies, language correlated with verbal performance on 4 of 4 occasions, whereas it correlated with nonverbal performance on only 1 of 4 occasions. The question is why language correlates more consistently with verbal performance. Some of our tasks have more obvious verbal demands than others. The emotion-behavior task described above, and social tasks used in the study of autistic and MLD children, have relatively high verbal demands because children must follow a narrative and then answer a verbal question at the end (e.g., “Which window will Sam go to?”). These tasks are like false belief tasks which also correlate reliably with language ability. This leads to a trivial explanation of the language-theory of mind relation. Perhaps language correlates simply because children can’t follow story narratives or can’t parse verbal questions. Yet this explanation does not fully account for the pattern of findings because the emotion labeling task described above has relatively low verbal demands. Children must simply label an emotion photo using one of two labels, yet language correlates with verbal labeling (see also Ruffman et al., 2003), and it does so even at 7 and 9 years of age when the emotion terms are within the children’s vocabulary. Thus, the fact that language correlates with verbal performance on all tasks suggests that it is not simply the verbal complexity of a task that leads to the correlation.

A second possibility is that the language-theory of mind relation is mediated by children’s age. Both things develop over age and it could be age (experience in the world) that is important to theory of mind rather than language per se. Yet we found that the language-theory of mind relation was robust even after accounting for children’s age (Ruffman et al., 2003).

A third possibility is that language is a proxy measure of IQ and that is IQ that correlates with theory of mind. Although this explanation is perhaps partially correct, it is not the full story because in contrast to language, spatial IQ does not correlate consistently with social understanding (Jenkins & Astington, 1996; Ruffman, Garnham, & Rideout, 2001; Tager-Flusberg & Sullivan, 1994).

A fourth possibility is that language is related only to theory of mind insights in the initial stages of development (e.g., amongst preschoolers) when language differences between children are more profound. Yet we found language was related to verbal labeling of emotions amongst relatively experienced language users; 7- to 9-year-olds.

Instead, there seems to be something more fundamental about verbal ability itself. As stated above, we think this is that language is needed to formulate, refine and explicitly reflect on ideas about social concepts. In the experiment described above, children’s faster reaction times on the predictable block showed that they possessed some knowledge of the emotion expressed in different faces even though they were incorrect when explicitly labeling the faces. Recall that reaction times did not relate significantly to the child’s language, whereas verbal performance did. In other words, reaction times were not improving between 7 and 9 years (r = -.11, n.s.), even though language was (r = .40, p < .01), and verbal performance was (r = .42, p < .01). These findings are in keeping with studies demonstrating that implicit learning is as good in 4-year-olds as it is in adults (Vinter & Perruchet, 2000), even though language improves dramatically during this period. That is, it is thought that implicit learning is based on processes of pattern induction or statistical learning (Boucher & Dienes, 2003) rather than language. We have posited the same for eye gaze (and reaction times) in tasks tapping social understanding (Ruffman, 2000).

Once again, the idea is that social knowledge typically develops in an implicit form and only later becomes explicit. The reason it is implicit initially is that it develops through relatively slow processes of induction in which many different bits of social information are gradually pieced together through statistical learning processes to arrive at initial implicit insights. That is, it seems less common for social learning to occur through effortful processes of deliberate theorizing or explicit teaching (although the exception as argued above, is in autism). Indeed, we have found that false belief understanding in 3- to 4-year-olds correlates with children’s ability to detect statistical patterns in nonsense syllables (Ruffman & Taumoepeau, in preparation). Children were exposed to a string of nonsense syllables based on Saffran, Aslin, and Newport(1996) while they colored (e.g., …bidakutupirogolabutupiro…). Some strings of syllables repeated to form “words” (e.g., “tupiro”) such that the probability of “pi” given “tu” was 1.00, whereas the probability of “go” given “tu” was 0. This statistical information was the only clue to what was a “word”. Intonation and spacing were held constant between syllables. Children were later asked to identify words such as “tupi” or “tugo” as ‘old’ or ‘new’. Although the false belief task was not an implicit task, performance on the statistical learning task correlated significantly with false belief even after controlling for language ability, pr = .26, p < .05. Again, we hypothesize that this is because explicit false belief is initially based on implicit insights and implicit learning is statistical learning. And to repeat the ideas discussed above, once implicit understanding is in place, the first children to develop explicit understanding are those with better language skills because language provides the terminology to reflect on and refine implicit intuitions.

References

Astington, J. W., & Jenkins, J. M. (1999). A longitudinal study of the relation between language and theory-of-mind development. Developmental Psychology, 35, 1311-1320.

Baldwin, D. (1991). Infants’ contribution to the achievement of joint reference. Child Development, 62, 875-890.

Boucher, L., & Dienes, Z. (2003). Two ways of learning associations. Cognitive Science, 27, 807-842.

Clements, W. A., & Perner, J. (1994). Implicit understanding of belief. Cognitive Development, 9, 377-395.

de Villiers, J. G., & Pyers, J. E. (2002). Complements to cognition: A longitudinal study of the relationship between complex syntax and false-belief-understanding. Cognitive Development, 17, 1037-1060. 

Hale,  M., & Tager-Flusberg, H. (2003). The influence of language on theory of mind: A training study. Developmental Science, 6. 346.

Jenkins, J. M., & Astington, J. W. (1996). Cognitive factors and family structure associated with theory of mind development in young children. Developmental Psychology, 32, 70-78.

Klin, A., Jones, W., Schultz, R., & Volkmar, F. (2003). The enactive mind, or from actions to cognition: Lessons from autism. Philosophical Transactions of the Royal Society of London: B, 358, 345-360.

Naigles, L. (1990). Children use syntax to learn verb meaning. Journal of child language, 17, 357-374.

Naigles, L. R. (1996). The use of multiple frames in verb learning via syntactic bootstrapping. Cognition, 58, 221-251.

Naigles, L. R. & Hoffginsberg, E. (1995). Input to verb learning: Evidence for the plausibility of syntactic bootstrapping. Developmental Psychology, 31, 827-837.

Perner, J., Sprung, M., Zauner, P., & Haider, H. (2003). Want thatis understood well before say that, think that, and false belief: A test of de Villier’s linguistic determinism on german-speaking children. Child Development, 74, 179-188.

Pinker, S. (1987).  The bootstrapping problem in language acquisition. In B. MacWhinney (Ed.), Mechanisms of language acquisition. Hillsale, NJ: LawrenceErlbaum Associates.

Rondal, J. A., & Cession, A. (1990). Input evidence regarding the semantic bootstrapping hypothesis. Journal of Child Language, 17, 711-717.

Ruffman, T. (2000). Nonverbal theory of mind: Is it important, is it implicit, is it simulation, is it relevant to autism? In J. W. Astington (Ed.), Minds in the making: Essays in honor of David R. Olson (pp.250-266). Oxford: Blackwell.

Ruffman, T., Garnham, W., Import, A., & Connolly, D. (2001). Does eye direction indicate implicit sensitivity to false belief?: Charting transitions in knowledge. Journal of Experimental Child Psychology. 80, 201-224.

Ruffman, T., Garnham, W., & Rideout, P. (2001). Social understanding in autism: Eye gaze as a measure of core insights. Journal of Child Psychology and Psychiatry, 42, 1083-1094.

Ruffman, T., Slade, L., Rowlandson, K., Rumsey, C., & Garnham, A. (2003). How language relates to theory of mind. Cognitive Development, 18, 139-158.

Ruffman, T., & Taumoepeau, M. (in preparation). False belief understanding is assisted by statistical learning.

Saffran, J. R., Aslin, R.N., & Newport, E.L. (1996). Statistical learning by 8-month-old infants. Science, 274, 1926-1928.

Slade, L., & Ruffman, T. (in press). How language does (and does not) relate to theory-of-mind: A longitudinal study of syntax, semantics, working memory and false belief. British Journal of Developmental Psychology.

Tager-Flusberg, H., & Sullivan, K. (1994). Predicting and explaining behavior: A comparison of autistic, mentally retarded and normal children. Journal of Child Psychology and Psychiatry, 35, 1059-1075.

Vinter, A., & Perruchet, P. (1999). Implicit learning in children is not related to age: Evidence from drawing behavior. Child Development, 71, 1223-1240.