Young children find desires easier to understand than beliefs and may understand desires with little or no understanding of beliefs. In this chapter, I review evidence and explanations for that lag. I conclude by offering a novel explanation, arguing that children’s conception of other people undergoes an important shift when they begin to engage in conversation. Initially young children conceive of other people primarily as agents with goals, but around three years of age they start to construe people as epistemic subjects capable of exchanging information for the formation and updating of beliefs. Hence, they shift from a desire psychology to a belief-desire psychology.
MUCH has changed in social cognition research since Paul Harris opened his contribution to Theories of Theories of Mind with this passage, back in 1996. However, as I will argue in this chapter, nothing in the last twenty years suggests that Harris was wrong to believe that children learn to reason about beliefs in conversation. This is why I will conclude that Theory of Mind has a lot to learn from experimental pragmatics.
Reasoning about other people’s beliefs lies at the heart of human Theory of Mind: our capacity to interpret other people’s behaviour by reference to their desires, intentions, and other mental states. The central role of belief reasoning in Theory of Mind research is most evident in the experimental record, where the large majority of studies in the last thirty years have investigated children’s and infants’ performance in false-belief tasks. Standard false-belief tasks were originally designed for preschoolers and elicit a response by asking the child a direct test question. For example, in the Sally-Anne task (Baron-Cohen et al., 1985), a character named Sally puts a marble in a box before leaving the scene, and in her absence, a second character named Anne moves the marble to a basket. Children are then asked ‘When Sally comes back, where will she look for her marble?’ If children understand that Sally holds a false belief, they should anticipate that she will look for the marble in the container where she had left it, and not where Anne put it. Unlike standard false-belief tasks, gaze-monitoring versions adapted for younger children and infants (e.g. Clements & Perner, 1994; Onishi & Baillargeon, 2005) rely on indirect measures of belief understanding (e.g. anticipatory looking to the location where Sally will look for her marble).
The results of hundreds of Theory of Mind studies have revealed that children under the age of 4 tend to fail false-belief tasks by predicting that the agent will look for the object in its current location (for a metaanalysis, see Wellman et al., 2001). By contrast, infants under 2 years of age are able to pass indirect versions of these tasks by anticipating that the agent will return to the original container (for a recent review, see Scott & Baillargeon, 2017). These two sets of findings have been the main focus of Theory of Mind research in the last three decades. In this chapter, I will argue that this increasingly narrow focus on false-belief tasks has limited the scope of this research field to laboratory tasks, while disregarding belief reasoning in everyday situations.
Given their intense focus on false-belief tasks, it is perhaps surprising that Theory of Mind studies have not addressed the question of when children engage in belief reasoning in their daily lives. The design of false-belief tasks would suggest that they engage in belief reasoning whenever they have to predict the behaviour of a mistaken person. This ability may be important for everyday life, but does not seem so frequent as to deserve the disproportionate attention that false-belief tasks have received in Theory of Mind research. Here I will defend the idea that belief reasoning does indeed play a key role in our daily lives, but not in the kind of scenarios recreated in false-belief tasks. Instead, on an everyday basis, we engage in belief reasoning whenever we engage in communication (Geurts & Rubio-Fernández, 2015).
The received view on the use of Theory of Mind in communication is that it is necessary for rather sophisticated pragmatic phenomena. As Peterson puts it, ‘throughout childhood, an understanding of Theory of Mind is fundamental to skilled communication and social interaction, especially in complex situations including reminiscence, sarcasm, deception, humor, interpersonal conflict and persuasive negotiation’ (2009: 475). However, the role of Theory of Mind in communication is not only limited to figurative language, as in irony comprehension (Spotorno & Noveck, Chapter 17 in this volume) or the recognition of the speaker’s referential intention in metonymy (Schumacher, Chapter 19 in this volume); Theory of Mind use extends to literal language production and comprehension.
If we take communication to serve the primary function of exchanging information among interlocutors, any informative contribution to a conversation must take into account what the interlocutors know and do not know (for a discussion of epistemic awareness in the derivation of scalar implicatures, see Breheny, Chapter 4 in this volume). This means speakers must take into account their common ground with their listeners in order to successfully communicate. If we did not engage in this form of belief reasoning, conversations would be plagued with redundancies and repetitions, as well as ambiguities—just to mention a few pragmatic issues related to common ground (for a discussion of the role of common ground in referential communication, see Rohde and Davies & Arnold, Chapters 27 and 28, respectively, in this volume).
Consider two simple examples: if you could not remember your best friend’s birthday, you would probably call a mutual friend, but you would not ask a random passer-by on the street (who is unlikely to know your best friend). On the other hand, if you were looking for an underground station in an unfamiliar city, you may ask a passer-by for directions (rather than calling one of your friends). The fact that we try to address our questions to knowledgeable interlocutors (or at least to interlocutors whom we hope will know the answer to our questions) is evidence that communication requires reasoning about other people’s epistemic states.
These examples of belief reasoning in communication seem closer to our everyday experience of interacting with others than the kind of false-belief scenarios used in Theory of Mind tasks. In this respect, experimental pragmatics has a lot to offer to Theory of Mind research. More specifically, the more varied and complex evidence of belief attribution observed in experimental pragmatics studies (relative to the pass-or-fail results of false-belief tasks) has the potential to reveal a much more nuanced picture of the development of children’s epistemological theories.
Children’s performance in false-belief tasks has been shown to correlate with their language development (for a metaanalysis, see Milligan et al., 2007). However, these correlational studies have focused on syntax, semantics, or general language ability (which combines syntax and vocabulary measures) but have not included pragmatic measures of Theory of Mind development (e.g. children’s ability to distinguish between ignorant and knowledgeable interlocutors). Children’s pragmatic abilities have been extensively investigated in developmental studies looking at different communicative settings and speech acts (e.g. Matthews et al., 2006; Rakoczy & Tomasello, 2009; Grosse et al., 2010; Grosse & Tomasello, 2012), but the results of most of these studies have not been examined in connection with children’s performance in Theory of Mind tasks (cf. Dunn et al., 1991; Bernard & Deleau, 2007).
Since Onishi & Baillargeon (2005) published their seminal false-belief study with infants, the hottest debate in Theory of Mind research has revolved around the question of whether infants possess a Theory of Mind, which has unfortunately relegated the results of experimental pragmatics studies even further. There is, however, no consensus as to what abilities infants and toddlers display in looking studies, with some researchers arguing that infants only reveal a very rudimentary or implicit Theory of Mind (Butterfill & Apperly, 2013; Wellman, 2014; Rakoczy, 2015), while others go further in their scepticism and affirm that infants’ success in false-belief tasks does not require a Theory of Mind (Perner & Ruffman, 2005; Heyes, 2014; Ruffman, 2014).
On the issue of nativism, an interesting parallel can be drawn between Theory of Mind and language acquisition research. In linguistics, there is still an ongoing debate about the extent to which language is innately specified (Chomsky, 1965; Goldberg, 2003; Lieven, 2016). However, even if researchers disagree on whether humans are endowed with a universal grammar from birth, there seems to be a general agreement that a full theory of language must explain not only its inception but also its development (Tomasello, 2003; Pinker, 2009). That is, even if language were largely innate, the question remains as to what changes from the time a child is born to the point when they become able to communicate fluently.
By contrast, experimental studies in Theory of Mind have investigated, almost exclusively in recent years, the question of whether the concept of belief is innate or early emerging, rather than looking at how that concept develops in the first years of life. While early studies in the 1980s and 1990s addressed a variety of complex research questions in Theory of Mind development (including the acquisition of mental state verbs or the development of perspective taking in referential communication), the last decade has seen an increasingly narrow interest in infants’ performance in false-belief tasks, at the expense of addressing other interesting developmental questions that span the preschool and early school years.
In favour of adopting a broader developmental approach in Theory of Mind research, in this chapter, I will briefly review a number of experimental studies with a special focus on communication as a way to highlight the important contribution that experimental pragmatics could make to the study of children’s early epistemological theories.
Harris (1996, 1999) argued that conversation has a tutorial role in the development of belief reasoning because young children come to appreciate that interlocutors have different information at their disposal, which they exchange and update in communication. This view led Harris to predict that early references to think and know would have a conversational function (e.g. ‘I don’t know’), rather than referring to mental states as such (e.g. ‘How do you know that?’). Early studies of naturalistic conversations between young children and adults supported Harris’s prediction (Shatz et al., 1983; Bartsch & Wellman, 1995). However, in view of the positive results of infant false-belief studies, Harris and colleagues have recently come to challenge the conclusions of these early studies—and indirectly, their original prediction.
Harris, Ronfard, & Bartz (2017) and Harris, Yang, and Cui (2017) investigated 2-year-olds’ understanding of knowledge by looking at various pragmatic measures, such as their use of the mental state verb know, their expressions of ignorance, their formulation of information-seeking questions and their empirical corroboration of statements. Harris et al.’s interest in investigating Theory of Mind development in communication is driven by their wanting to find the ‘missing developmental link’ between infants’ successful performance in indirect false-belief tasks and 4-year-olds’ success in standard false-belief tasks. Thus, Harris, Ronfard, & Bartz (2017) and Harris, Yang, and Cui (2017) hypothesized that early studies on the acquisition of mental state verbs may have underestimated 2-year-olds’ abilities and that less conservative measures may offer a more accurate picture of their understanding of knowledge.
While I commend Harris, Ronfard, & Bartz (2017) and Harris, Yang, & Cui (2017) for looking for evidence of Theory of Mind development beyond false-belief tasks, I disagree with Harris and colleagues’ motivation for two reasons. First, as discussed earlier, the interpretation of infants’ success in false-belief tasks is highly controversial (e.g. Heyes, 2014; Ruffman, 2014; Rakoczy, 2015). Second, and most importantly, even if infants were able to attribute false beliefs to others, the mindreading abilities demonstrated in looking studies need not be the same as those required in pragmatic tests of belief reasoning in communication (including the use of mental state verbs). Therefore, in my opinion, the positive results of infant false-belief studies should not fuel rich interpretations of early pragmatics studies. In the following, I will discuss some of these rich interpretations of young children’s behaviour.
Shatz et al. (1983) and Bartsch & Wellman (1995) carried out a corpus analysis of the spontaneous utterances of a number of children between the ages of 2 and 4 years with a focus on their use of mental state verbs. In order to reliably evaluate young children’s abilities to talk about the mind, Shatz et al. and Bartsch & Wellman distinguished between conversational functions of mental state verbs (e.g. opening a conversation with the formulaic phrase ‘You know what?’) and genuine references to mental states (e.g. ‘Mommy can’t sing it. She doesn’t know it’). This distinction is important because, as Shatz and colleagues point out, ‘the mere occurrence of a word in a child’s speech does not necessarily indicate that its producer understands its semantic presuppositions’ (1983: 302).
The results of these studies suggest that children start using mental state verbs in the months after their second birthday, but these early uses have conversational functions (i.e. turns of phrase, repetitions, and idiomatic phrases). Genuine references to mental states appear around the third birthday, including expressions of contrast between the contents of mental states and the world (e.g. ‘I was teasing you; I was pretending, ’cept you didn’t know that’). The early appearance of contrastives at age 3 suggests that children may start revealing false-belief reasoning in naturalistic conversation before they are able to pass standard false-belief tasks. If so, these data would support the view that standard false-belief tasks are unnecessarily hard for young children (e.g. because these tasks draw children’s attention to the target object with the test question, increasing the salience of the wrong response; Rubio-Fernández & Geurts, 2013, 2016).
Harris, Ronfard, & Bartz (2017) and Harris, Yang, & Cui (2017) challenged Shatz et al. (1983) and Bartsch & Wellman (1995) for discarding children’s uses of the phrase ‘I don’t know’ as potentially formulaic, arguing that
if we examine this phrase in the context of the preceding and subsequent utterances that were produced by the child’s interlocutor, (…) we can ask if it was used appropriately, notably as a plausible expression of ignorance in response to a range of questions, or, alternatively, if it was used in a more inflexible fashion—for example, as a way to demur or withdraw from a conversation. (Harris, Yang, & Cui, 2017: 287)
Harris, Ronfard, & Bartz (2017) and Harris, Yang, & Cui (2017) reanalysed the corpora used by Shatz et al. (1983) and Bartsch & Wellman (1995) and found that at age 2, the most frequent function of the verb know was to deny their own knowledge; that is, 2-year-olds typically used this verb in the phrase ‘I don’t know’.
The data suggested that toddlers produced ‘I don’t know’ at appropriate points in the conversation, which led Harris and colleagues to conclude that 2-year-olds are able to make relevant contributions to the ongoing conversation and acknowledge their own ignorance (Harris, Ronfard, & Bartz, 2017: 221; Harris, Yang, & Cui, 2017: 304). However, it must be noted that Harris and colleagues’ reanalysis did not address Shatz and colleagues’ and Bartsch & Wellman’s concerns when discarding such uses as potentially formulaic: unelaborated uses of the phrase ‘I don’t know’ may be a conventional negative response simply meaning ‘I can’t answer’. The fact that young children are able to produce such negative responses at appropriate points in a conversation cannot be taken as evidence that 2-year-olds are admitting their own ignorance in doing so.
Harris, Ronfard, & Bartz (2017) and Harris, Yang, & Cui (2017) also found a clear asymmetry in their data: 2-year-olds would deny their own knowledge (rather than that of their interlocutors) but would ask information-seeking questions to their interlocutors (rather than to themselves). Harris and colleagues propose that such an asymmetry reflects a deep-seated difference between children’s privileged access to their own mental states as opposed to those of their interlocutors—from which it follows that asking self-directed questions, for example, would be otiose (Harris, Ronfard, & Bartz, 2017: 225). This interpretation of young children’s uses of the verb know is again unnecessarily rich: a more parsimonious interpretation of this asymmetry would relate it to the frequency of such uses in child-directed speech. It is an empirical question, of course, but it seems plausible that adults would more often tell young children ‘I don’t know’ than ‘You don’t know that’, and likewise, that children would more often hear adults ask ‘Do you know where my toothbrush is?’, for example, than the self-directed version ‘Do I know where my toothbrush is?’
It is therefore possible that children’s early knowledge denials and know questions simply reflect the frequency of such uses in the adult speech that they hear, rather than an appreciation of the differential accessibility of one’s and others’ knowledge. A similar hypothesis has been corroborated in the investigation of children’s acquisition of wh-questions: whereas traditional accounts had posited that the order in which English speaking children acquire wh-questions is determined by their syntactic and semantic complexity, Rowland et al. (2003) found that the acquisition order of wh-questions could be successfully predicted from the frequency with which particular wh-words and verbs occurred in the children’s input and, moreover, that syntactic and semantic complexity did not reliably predict acquisition once input frequency was taken into account. Therefore, the frequency of different functions of mental state verbs in caregiver speech needs to be investigated before we can derive any conclusions about young children’s understanding of other people’s mental states from their use of those verbs (see Dudley et al., under review).
Some Theory of Mind researchers have interpreted infants’ success in false-belief tasks as evidence of an implicit awareness of mental states (e.g. Clements & Perner, 1994; Apperly & Butterfill, 2009). Contrary to such lean interpretations, Harris, Yang, & Cui (2017) posited that a less conservative analysis of 2-year-olds’ uses of mental state verbs may actually reveal an explicit understanding of belief. However, in order to take 2-year-olds’ uses of mental state verbs as evidence of explicit Theory of Mind (comparable to the evidence gathered through standard false-belief tasks), we need to go beyond the mere production of such forms, and investigate whether young children understand the factivity of mental state verbs. That is, we need to examine whether young children appreciate that know is used in contexts where the complement is taken to be true. For example, if I had just met Sally in the garden, I could claim ‘I know that Sally is outside’, whereas if I was not sure of her whereabouts, I could at most say ‘I think that Sally is outside’. Likewise, if I say ‘Sally doesn’t know that her marble is in the basket’, my statement presupposes that the marble is in the basket.
Early studies suggest that children do not differentiate verbs like know and think until the age of 4 (e.g. Macnamara et al., 1976; Johnson & Maratsos, 1977), or even later (Schulz, 2003; Léger, 2007). However, Dudley and colleagues (2015) have recently argued that the tasks used in those studies were too demanding for young children, for reasons independent of the factivity of the verbs that were tested (e.g. children were asked to make metalinguistic judgements about utterances, which are known to be difficult for young children). Dudley and colleagues tried to better estimate 3-year-olds’ understanding of factivity by using a more appropriate task for that age group.
In their paradigm (adapted from Moore & Davidge, 1989; and Moore et al., 1989), children had to find a toy hidden in one of two boxes using clues in the form of attitude reports containing think and know. Crucially, the experimenter (who knew where the object had been hidden) echoed such clues on behalf of a puppet (Lambchop) who whispered into her ear. Thus, children had to figure out whether the object was in the red or the blue box by using clues such as ‘Lambchop knows/thinks that it’s in the blue box’. Unlike previous research, the results of Dudley et al.’s study revealed that about half of their 3-year-olds’ were able to distinguish think and know, and appreciated that know presupposes the truth of its complement, while think does not.
As in the early studies by Shatz et al. (1983) and Bartsch & Wellman (1995), Dudley et al.’s (2015) investigation of young children’s understanding of knowledge in a communicative task seems to reveal better performance than is usually observed in traditional Theory of Mind tasks. Thus, like Harris, Yang, & Cui (2017), Dudley et al. (2015) concluded that their results support the view that there is continuity between infants’ success in indirect false-belief tasks and preschoolers’ performance in standard false-belief tasks because 3-year-olds can be sensitive to their interlocutors’ epistemic states in a linguistic task.
However, the individual performances observed by Dudley et al. (2015) do not offer strong support to their conclusions about the continuity view of Theory of Mind development. In the critical condition, the children in their study had to understand that the clue ‘Lambchop doesn’t know that it’s in the blue box’ presupposed (or at least conveyed) that the object was actually in the blue box. Dudley and colleagues report that, of the twenty-eight children in their study, 43 per cent selected the correct box (corresponding with a factive interpretation of know). However, that figure results from grouping together those children who gave two or three correct responses (out of three trials). The most striking result when looking at the individual performances is that the preferred response was actually the non-factive one:1 when looking only at those children who gave the same response in the 3 trials, 68 per cent systematically selected the non-factive response, whereas only 32 per cent systematically selected the correct answer. Therefore, the results of Dudley et al. (2015) are closer to the findings of previous studies than their statistical analyses suggest.
Future studies should therefore investigate why the semantics of know might be initially non-factive, especially if one assumes that infants already possess an explicit Theory of Mind. However, young children’s use and comprehension of mental state verbs is not the only measure of their understanding of knowledge that can be gathered through communicative tasks. Other naturalistic situations where children need to engage in belief reasoning include updating their own knowledge when receiving reliable information in conversation, and relying on their common ground with their interlocutor in referential communication.
In a pioneering study, O’Neill (1996) asked whether 2-year-olds would adapt their referential strategies depending on their parents’ knowledge of the location of a toy. Her results showed that toddlers were more likely to refer to the toy’s new location via pointing gestures and vocalizations when the parents were absent at the time when the toy was placed out of reach on a high shelf than when the parents had witnessed the transfer. Older and younger 2-year-olds (mean age: 31 months in Experiment 1 and 27 months in Experiment 2) revealed comparable sensitivity to their parents’ knowledge states. However, O’Neill defended a non-mentalistic interpretation of her data, according to which toddlers would only need to appreciate, in a general sense, that the parents who missed the transfer of the object had disengaged from the critical events (rather than understanding that they were ignorant of the toy’s location). Then, in wanting to retrieve the object, 2-year-olds would be motivated to update their parents regarding the critical events.
The results of a follow-up study by Dunham and colleagues (2000) replicated O’Neill’s original findings and supported her lean interpretation of toddlers’ pragmatic abilities as a disengagement and updating heuristic. Dunham and colleagues hypothesized that one reason why the younger toddlers in O’Neill’s study might have pointed more towards the location of the toy in the Ignorant condition is that the parents had covered their eyes (rather than leaving the room), potentially suggesting that they were engaging in a hiding game. If the toddlers in that experiment had experience with such games, they might have simply produced more pointing gestures because that is what they would normally do in such situations.
Confirming Dunham and colleagues’ hypothesis, the 27-month-olds in their study pointed as much in the Ignorant condition as they did in a Sham-Ignorant condition in which the parents first covered their eyes but then looked at the experimenter moving the object. By contrast, the 33-month-olds in their sample pointed similarly in the Sham-Ignorant condition and in the Knowledgeable condition, showing an appreciation of the parents’ perceptual access to the critical event. However, Dunham et al. (2000) encouraged conservative interpretations of their findings, arguing that older 2-year-olds may simply be able to make more accurate use of the same disengagement and updating heuristic by relying on more specific cues to the parents’ perceptual access to the events.
Infants as young as 12 months of age have been found to be sensitive to an interlocutor’s engagement with an object in a referential communication setting. Tomasello & Haberl (2003) observed that, when infants interacted with two objects with one experimenter, and with a third object with another experimenter, they would reliably interpret the first experimenter’s expression of surprise on her return (‘Wow! Look at that one!’) as referring to the third object that they had not shared. Moll & Tomasello (2007) and Moll et al. (2008) further observed that 14-month-old infants had to become familiar with the objects in joint engagement with the experimenter in order to disambiguate the experimenter’s request for the target object in the test phase. Thus, witnessing the experimenter interact with the first two objects individually did not allow infants to pass this task until age 18 months (Moll & Tomasello, 2007).
Using the same paradigm, Moll et al. (2011) investigated 2-year-olds’ pragmatic abilities in referential communication and observed that their reliance on the engagement heuristic led them to overestimate their interlocutor’s knowledge: when the experimenter remained present in the scene but could not see the third object because of a barrier, toddlers wrongly assumed that she was familiar with all three objects. Thus, 2-year-olds only showed a preference for the third object in the test phase if the experimenter had left the room before the third toy was introduced; otherwise, they wrongly assumed that all three objects were part of their common ground with the first experimenter.
The results of referential communication studies with infants and toddlers suggest that, at a very early age, children rely on their interlocutor’s engagement (or disengagement) in a scene as a proxy for their knowledge (or ignorance) of relevant events. Interestingly, a similar effect has been observed with adults: Gorman et al. (2013) asked participants to instruct another participant to select one of three objects in a series of trials. Critically, the objects were novel and the two participants had to learn their names prior to the instruction-selection phase. However, the participant who gave the instructions had to learn more names than the participant who had to follow the instructions. As expected, participants named the novel objects more often when the other participant knew the name for the novel object than when only they knew the name (in which case they opted for describing the novel object). Interestingly, participants were better able to discriminate shared and private names when they learned the shared names together with the other participant than when they learned them independently (but with mutual knowledge of their learning). Gorman and colleagues concluded that shared experience provides a strong memory cue to the common-ground status of names.
The results of referential communication studies with infants, toddlers, and adults suggest that ordinary memory processes tracking our interlocutor’s engagement in an event may provide access to shared knowledge without necessarily involving metarepresentation of our interlocutor’s epistemic states. This basic ability may be a precursor of belief reasoning in early referential communication, and an efficient mechanism to establish common ground among adult communicators.
In an early study, Povinelli & DeBlois (1992) observed that 4-year-old children consistently followed the pointing gesture of a knowledgeable informant (who had hidden a surprise) vs. an ignorant informant (who had left the scene before the surprise was hidden). In contrast, 3-year-olds were not able to discriminate between the knower and the guesser. Sodian and colleagues (2006) argued that ignoring the pointing gesture of the ignorant adult might have posed high executive control demands on the younger children, and designed a follow-up study that aimed to provide a more accurate assessment of 3-year-olds’ understanding of knowing and guessing.
Children in Sodian and colleagues’ study had to choose between two helpers, only one of whom had witnessed where a sticker had been hidden. Two-year-old children performed at chance in this task, and it was not until the age of 35 months that they started to consistently discriminate between the knowledgeable and the ignorant helper when asked ‘Who can help you?’ Importantly, all the children in the study were at ceiling in a non-epistemic control task, where they had to understand that only the helper who had a key could open a treasure box. In view of their results (and those of Povinelli & DeBlois, 1992), Sodian and colleagues concluded that ‘the seeing-knowing relation is a genuine conceptual problem for children below the age of three years’ (2006: 234).
The results of Sodian et al. (2006) might seem surprising given the results of social learning studies, which have repeatedly shown that young children prefer to learn from accurate speakers than from inaccurate ones (for a review, see Mills, 2013; cf. Heyes, 2017). However, preferring an informant who labels familiar objects correctly to one who does so incorrectly need not require appreciating the epistemic states of such informants: the mere normativity of the accurate speaker (or the other way around: the strange behaviour of the inaccurate speaker) might be enough to enable children to form the right preference (for discussion, see Lucas & Lewis, 2010). Therefore, rather than manipulating speaker accuracy, a more reliable evaluation of children’s understanding of a speaker’s mental states may be established by manipulating the quality of the information provided (i.e. whether it is relevant or sufficient vs. irrelevant or insufficient).
Gillis & Nilsen (2013) conducted such a study and asked children to seek help from one of two speakers: one who had provided enough information to find a sticker (e.g. ‘It’s under the black figure’, when there was only one black figure) and one who had not provided enough information (e.g. ‘It’s under the triangle figure’, when all figures were triangles). Both preschoolers (4-5 years old) and schoolchildren (6-7 years old) preferred to solicit information from the speaker who had provided sufficient information. However, a second experiment confirmed that the preschoolers (but not the schoolchildren) based their preference on their own success in finding the sticker when following the clues of the two speakers: when children were not allowed to look for the sticker after the clue, only the schoolchildren were able to evaluate the reliability of the two speakers (with the preschoolers performing at chance).
Gillis & Nilsen concluded that, compared with speaker accuracy (which children as young as 2 years are able to use in similar paradigms), the ability to use information quality to assess speaker reliability seems to emerge only at a later age. Before they are able to evaluate speaker reliability in terms of information quality, children rely on their own success when following a speaker’s clues. It must be noted that, from a pragmatic perspective, the underinformative speaker in Gillis & Nilsen’s paradigm need not have been ignorant of the location of the sticker (and hence unable to provide more specific clues); alternatively, this speaker might have simply been uncooperative (i.e. unwilling to provide sufficient information). In either case, selecting the sufficiently informative speaker over the underinformative one would require, at a minimum, appreciating that the clues provided by the underinformative speaker were pragmatically infelicitous.
The results of the studies reviewed in this section suggest that children’s epistemological theories continue to develop in the preschool years. An interesting question, which has not been directly addressed in the literature, is the relationship between preschoolers’ understanding of other people’s knowledge and their understanding of the verb know. Robinson et al. (1999) investigated children’s ability to infer speaker reliability from a speaker’s visual access to the contents of a box. Both preschoolers (4 years old) and schoolchildren (5 years old) changed their opinion more often when a better-informed adult disagreed with their guess about the contents of a box, than when they knew better than the adult what was inside the box. However, when the same group of children were asked ‘Who knows better what is inside the box?’, the schoolchildren performed comparably in the two tasks whereas the preschoolers performed significantly worse with the know-better question. Moreover, performance on the two tasks (i.e. changing opinion in disagreement vs. who knows better?) was unrelated in both age groups.
Robinson and colleagues explained the surprising difference between the two tasks as a difference in the salience of the adult’s beliefs (which were verbalized in the disagreement task, but not in the question task). An alternative explanation that was not considered by the authors has to do with preschoolchildren’s understanding of the verb know: preschoolers may have an implicit understanding of other people’s knowledge that allows them to evaluate the reliability of their opinions (and update their own accordingly), but does not allow them to verbally express such an evaluation in terms of ‘knowing better’.
The results of this brief review highlight the relevance of experimental pragmatics studies for Theory of Mind research. More in line with early accounts of the development of belief reasoning in conversation (Harris, 1996, 1999) and contrary to what the current debate on infant Theory of Mind might suggest, children’s epistemological theories undergo a steep development from infancy to the first school years, which need not be reflected in their performance in standard false-belief tasks. Early studies by Shatz et al. (1983) and Bartsch & Wellman (1995) suggest that children start producing contrastives (i.e. verbal expressions of the difference between the contents of mental states and the corresponding reality) at age 3 years, which is earlier than they start passing standard false-belief tasks. Likewise, the results of Dudley et al. (2015) revealed that some 3-year-olds understand that ‘Sally doesn’t know that the marble is in the basket’ presupposes (or at least conveys) that the marble is in the basket—contrary to how children of that age tend to respond to the false-belief question in the Sally-Anne task. As Sodian and colleagues put it, ‘it is possible that an awareness of others’ mental states emerges in communicative exchanges before it becomes available to predictions and explanations of others’ behaviours made from a third-person perspective’ (2006: 235; for a related point, see Harris, Ronfard, & Bartz, 2017; Harris, Yang, & Cui, 2017).
While experimental pragmatics tasks may reveal early Theory of Mind abilities that are not observable in standard false-belief tasks, they may also reveal shortcomings that Theory of Mind tasks do not: infants, toddlers, and young children show limitations in their understanding of other people’s knowledge in communication, which could be taken to contradict rich interpretations of infants’ successful performance in indirect false-belief tasks. Thus, the results of the studies reviewed in this chapter (and many others in the early Theory of Mind literature) have implications for theoretical accounts of Theory of Mind development. The first implication is that children’s epistemological theories are more complex and slow-developing than the results of indirect false-belief tasks might suggest. Therefore, in order to understand children’s belief reasoning capacities, we need to combine developmental data from traditional Theory of Mind tasks with experimental pragmatics studies where children need to use their developing concept of belief in interaction with others.
The second implication is for specific theories of Theory of Mind development, which do not account for the evidence observed in experimental pragmatics studies. For example, Apperly & Butterfill (2009; Butterfill & Apperly, 2013; Low et al., 2016) have recently argued that humans possess two separate systems for belief reasoning: a simpler, evolutionary, and ontogenetically ancient system that operates fast and independently from central cognitive resources, and a more sophisticated system that develops later and operates more slowly, making substantial demands on executive-control processes. Rather than operating on belief representations, the more efficient system operates on registrations, which Butterfill & Apperly (2013; see also Rakoczy, 2015) define as a relation between an agent, an object, and a location. These non-mentalistic representations could in principle explain the patterns of results observed in early referential communication studies (e.g. O’Neill, 1996; Dunham et al., 2000; Tomasello & Haberl, 2003; Moll & Tomasello, 2007; Moll et al., 2008, 2011), in which infants and toddlers seem to keep track of their interlocutor’s engagement (and disengagement) in a communicative situation. Likewise, low-level memory representations or registrations could also explain the improved performance of adults in Gorman et al.’s (2013) investigation of common ground in referential communication. However, like other Theory of Mind researchers, Apperly, Butterfill and colleagues focus on the results of false-belief tasks, but do not offer a satisfactory analysis of the development of young children’s epistemic theories reported in the experimental pragmatics literature.
Theory of Mind researchers defending rich interpretations of infants’ and toddlers’ success in false-belief tasks (e.g. Carruthers, 2013; Helming et al., 2014) also need to explain the limitations that toddlers and preschoolers reveal in experimental pragmatics studies if we are to assume that infants’ concept of belief is as sophisticated and flexible as some of them argue (e.g. Scott & Baillargeon, 2017). In particular, they would need to explain why infants under 2 years of age would be able to infer that when an agent sees an object being placed in a box, she comes to believe that the object is in that box (as it is assumed in false-belief tasks; e.g., Onishi & Baillargeon, 2005) but 3-year-olds cannot apply the same rationale when they have to select a knowledgeable helper (over an ignorant helper) on the basis of his visual access to the contents of a box (Povinelli & DeBlois, 1992; Sodian et al., 2006; see also Robinson et al., 1999).
In conclusion, a comprehensive theory of the development of children’s epistemological theories needs to account not only for infants’ and children’s performance in false-belief tasks, but also for their appreciation of sensory perception as a source of beliefs, and their understanding of their interlocutors’ knowledge in communication, including the use and comprehension of mental state verbs. Social cognition researchers might argue that communication falls outside the scope of Theory of Mind research, and that it should therefore be investigated by psycholinguists and pragmatists. The problem with passing the buck in this fashion is that, in everyday life, children need to assess the epistemic states of their interlocutors (e.g. when asking and responding to questions) more often than they need to predict the behaviour of an agent with a false belief (as in the Sally-Anne task). Consequently, there is a danger that the bulk of Theory of Mind data will continue to fall outside of the scope of the theories that are designed to explain them.
This chapter is based on a paper presented at the Summer School ‘Understanding Communication and Understanding Minds: The Role of Metarepresentation’ (Central European University, Budapest, June-July 2016). I would like to thank all the students and the rest of the faculty at the summer school for a very stimulating discussion of this work, which helped me shape this chapter. This research was supported by a Young Research Talent Grant from the Research Council of Norway (Ref. 230718). I also gratefully acknowledge this funding.
1 The number of children who gave three, two, one, and zero factive responses were the following: six, six, three, and thirteen—with the last group treating know as a non-factive verb in all three trials.