If language is necessary for ToM, it follows that preverbal infants do not understand ToM. Several recent studies in infants showed that it is not the case. It has been shown that precursor ability (e.g., joint attention, imitation and pointing gestures) of ToM is present early in infancy (e.g., [27-29]). A series of recent studies that used looking-preference paradigms found that even 13-15 month-old infants have the capacity to understand ToM [29-31]. Even though these results and methods used in the false-belief experiments on infants are still debated (see for example, Perner and Ruffman [32] for the criticism of Onishi and Baillargeon’s [29] study), evidence from studies on precursor abilities (see above refs.) of ToM seems to be convincing.

The above results imply that nonverbal ToM precedes verbal ToM developmentally and if so, nonverbal false-belief tasks are easier than verbal ToM tasks. However, mixed results were obtained on this account. In several studies that compared children’s performance in verbal and nonverbal false-belief tasks, children performed better [33, 34], the same [35, 36] or worse [37, 38] in the nonverbal false-belief tasks. For instance, in Call and Tomasello [36] the helper watched the experimenter hide a sticker in one location. When the helper left the room the experimenter moved the sticker from that location to another. Children were asked which box the helper might indicate as containing the sticker. In this experiment, both verbal and nonverbal conditions were acted out with no verbal description, but only the test question in the verbal version was verbal. Children’s performances in the two versions were comparable. Later, Lunn [33] tested children with the same task as Call and Tomasello and this time children performed better in the nonverbal task than the verbal one. Given these results, evidence in support of the linguistic determinism hypothesis is equivocal at best. As several ToM researchers suggest [16, 39], language may be important for ToM not because language is a prerequisite for ToM but because it is used as “scaffolding” for symbolic representations.

While implicit nonverbal ToM may not involve any language, there is relatively little doubt that language is involved in explicit verbal ToM, especially in the false-belief reasoning [40]. However, the extent of the involvement and what aspects of language are the most involved are much in debate. As I mentioned in the introduction, the strong linguistic determinism of ToM posits that language is prerequisite for ToM (e.g., [17]). De Villiers and her colleagues argue that children’s mastery of syntax, especially the sentential complement (e.g., “Sally says that…”) is necessary for ToM development. In this view, other aspects of language (i.e., lexicon, semantics, and pragmatics) are marginalized. More recently, P. de Villiers [41] has elaborated this hypothesis and argued that the mastery of sentential false-complements with communicative verbs plays a central role for explicit ToM reasoning. However, most of the evidence that supports this claim came from their own groups’ research (e.g., [17, 37, 41]). For instance, in their studies of oral deaf children and ASL-learning deaf children, children’s performance of the complement comprehension task predicted their performance on both standard and low-verbal false-belief tests (40). However, a close examination of the results of this study shows that children’s mastery of vocabulary (p = .005), and not the false complement clauses (p = .039) is a stronger independent predictor of their false-belief performance. Given these results, it seems unreasonable to maintain that the mastery of false-complement is the best predictor of false-belief reasoning. Moreover, in Lohmann and Tomasello [42], training children with discourse without sentential complements improved their false-belief task performance. Furthermore, a recent meta-analysis of over a hundred of studies that explored the relationship between language and ToM have demonstrated that various aspects of language (i.e., semantics, syntax, receptive vocabulary memory for complements, and general language) all predict ToM reasoning [18].

The evidence described above seems to increasingly support that the strong linguistic determinism is untenable. However, fortunately there is another moderate version of linguistic determinism that emphasizes the influence of the socio-linguistic effects on ToM [43-46]. It follows Vygotsky’s theory [47] and posit that it is the social-cognitive ability embedded in culture-specific symbolic systems (i.e., social languages) that enables children’s ToM. Its emphasis is not on constitutive aspects (e.g., syntax and semantics) but socio-communicative or pragmatic aspects of language [43, 46]. Bruner regards culture-specific meaning-making process as a central force for development of ToM or folk psychology [43]. Similarly, Nelson posits that ToM is best described as a Community of Minds that is enabled through internalization of language [46]. The Community of Minds is not universal and each person develops his / her own Community of Minds depending upon his / her background, personality, relationships, and history. Intuitively, the pragmatic aspects of language are strongly associated with ToM because pragmatics assumes that both verbal and nonverbal communications are the expression of context-specific intentions [48, 49]. In support of this hypothesis, it has been found that individuals with ASD are most impaired in the pragmatic aspect of the language [19,50]. It has also been suggested that the way in which older children / adolescents with ASD approach false-belief tasks is different from typically developing children in that they rely on syntax and semantics more than pragmatics which seems to be lacking in people with ASD [51]. The lack of pragmatics, and therefore, lack of ToM is evident in children with autism in their learning style of novel words. Baron-Cohen and his colleagues found that children with ASD learned new words whenever they themselves attended to the focal object whether or not the adult also attended to it, but the normally developing preschool children did not learn the new words unless the object was jointly attended to by another person [52].

Unfortunately, Milligan et al.’s meta-analysis [18] excluded those studies that examined the relationship between pragmatic aspects of language and those that tested non-English speaking children. An earlier meta-analysis that examined over a hundred ToM studies (including studies that tested non-English speakers) found that the basic course of development is the same across different language systems [4]. Children across different language systems passed false-belief tasks between 2.5 and 5 years of age. However, since many of the studies included in the meta-analysis did not examine the relationship between language and ToM and since most of these studies tested children in English speaking countries, the meta-analysis is relatively uninformative in terms of the possible effects of the linguistic variation on ToM. More recently, Collaghan et al. [53] tested children in Canada, India, Peru, Thai, and Samoa with a single false-belief task and found no difference among these children in the passing age of the task. However, the same limitation as the above Wellman et al.’s study [4] applies here. Since this study did not use any specific language measures pitted against ToM measure, little information is provided on the relationship between language and ToM development.

Fortunately, there are a number of studies that tested non-English speaking children with specific linguistic measures / hypotheses. Several of these studies found effects of language on ToM development (e.g., [54-56]). For instance, Junin Quechua children performed poorly on ToM tasks relative to American / European (A/E) children presumably because of their lack of mental state verbs [54]. In Lee et al.’s study with Mandarin-speaking children, even though the children’s performance for the false-belief task was comparable overall to A/E children’s performance, their performance was influenced by the choice of verbs (i.e., three verbs that all mean “think”) used in the false-belief task [55]. Mandarin-speaking children performed significantly better when yiwei and dang, which connote that the belief referred to may be false, were used than when xiang (the more neutral verb) was used. In a study in Turkish and Puerto Rican (PR) Spanish-speaking children, advancement in ToM performance was found [56]. In this study, the Turkish or PR Spanish-speaking children, who have either a specific verb (Turkish) or a case marker (PR Spanish) available to make the false-belief mental state more explicit, performed better than Brazilian Portuguese or English-speaking children who do not have those lexicons.

As I mentioned earlier, linguistic determinism theory posits that children’s successful mastery of complements is necessary for successful performance in the false-belief task [17]. However, it has been found that English-speaking children pass the false-desire task significantly earlier than they pass false-belief tasks (e.g., [57, 58]). To this, P. De Villiers explains that the false-desire task is easier for children because it takes the to+complement (e.g., “X want to…”), while false-belief task is harder because it takes the that+complement (e.g., “X believe that…”) [41]. If this claim is true, German-speaking children should perform equally well in the false-desire and false-belief tasks because in German the desire verb, wollen (want) also takes the that+complement form. However, a study found that it is not the case [59]. German-speaking children still performed better in false-desire (using “want” verb) task than the false-belief (using “believe” verb) task. More importantly, in this experiment, no correlation was found between children’s use of that+complement and their performance in the false-belief task after verbal IQ and age were accounted for. Similarly, Cheung et al. [60] examined whether or not Cantonese children’s mastery of syntactic aspects of language (more specifically, false-complements) uniquely predicts the children’s false-belief understanding. Contrary to this hypothesis, neither syntax nor verb semantics but children’s general language ability (measured by TELD-3 [61]) was the best predictor of ToM performance.

More recently, my dissertation research compared 8-12 year-old Japanese bilingual children’s understanding of the second-order false-belief (i.e., “Sally thinks that Anne thinks that …”) presented in the first language (L1, Japanese) and the second language (L2, English) as a part of the functional brain imaging study ([25], see also [62]). Most linguists agree that in general, center-embedded sentences (those that have relative clauses placed in the middle) are harder to parse [63, 64]. The center-embedding problem is the most relevant to Japanese false-belief sentence processing, because a Japanese false-belief sentence normally has a center-embedded structure. For example, a typical English false-belief sentence is written as:

Anne thinks [that the cake tastes good]

SVCP

But the same sentence in Japanese is written as:

Anne wa [sono ke-ki ga oishii-to] omou

(Anne[that the cake tastes good] thinks)

SCPV

As can be noticed easily, in the Japanese false-belief sentence, the complement phrase (CP) is placed between the noun and the verb of the sentence. In the English false-belief sentence, the truth value of the whole sentence can be inferred from the verb (V) placed right after the subject (S), but in the Japanese false-belief sentence, to determine whether or not the whole sentence is true, one has to wait until the end of the sentence. In the second-order false-belief sentence, the problem becomes worse because of the double center-embedding. For examine, when the above sentence is doubly-embedded it becomes:

Anne wa [Sally ga{sono ke-ki ga oishii-to} omotteiru to] omotteiru

(Anne [Sally {that the cake tastes good} thinks that] thinks)

S1S2CP-that1V2CP-that2V1

Thus, my prediction was that if the false-complement structure influences the false-belief understanding Japanese children (who were all balanced bilinguals) would perform worse in the same second-order false-belief task in Japanese (L1) than in English (L2). Contrary to this prediction, there was no effect of language. Japanese children’s performance on the Japanese false-belief task was comparable to their performance on the English false-belief task. These results are consistent with the most recent study that tested Japanese children with the second-order false-belief task [65]. In this study, despite the significant developmental delay in the first-order false-belief task (see also [66, 67]), Japanese children’s onset of the second-order false-belief task was about the same as the A / E children’s (i.e., between 6 and 9 years old [68, 69]). These results clearly show that mastery of that+complement is not necessary for ToM understanding at least in several non-English speaking cultures.

In sum, the above evidence suggests that although it is undeniable that language ability is necessary for developing explicit ToM because it provides scaffolding to think and interpret other people’s intentions and actions, language may not be sufficient for the expression of ToM. Moreover, it may not be the syntactic or semantic aspects of language that is the most important for ToM. Contrary to the strong linguistic determinism, the currently available data suggest that it is pragmatic aspects of language that is the most essential for ToM development. Thus, evidence so far supports the second moderate version of linguistic determinism.

Compared to behavioral or developmental evidence, neurological evidence to support or negate the developmental relationship between language and ToM is still scarce. However, evidence from developmental ToM studies described above suggests that pragmatic aspects of language are the greatest importance for ToM expression. It has been established that in human adults syntactic and semantic aspects of language are specialized in the left hemisphere of the brain (e.g., [70, 71]), while pragmatic aspects of language are specialized in the right hemisphere of the brain (e.g., [72, 73]). For instance, in Winner and Gardner’s study, RHD patients have been observed to choose a literal translation for a metaphorical sentence more frequently than healthy controls [74]. Several studies clearly showed that right hemisphere damaged (RHD) patients have problems with understanding non-literal language, metaphors, and jokes while they have no trouble in understanding literal language (e.g., [72, 75]). It has been found that left hemisphere damage (LHD) patients often perform as well as healthy controls in ToM tasks [21, 76-78], while RHD patients perform worse than LHD patients [79]. For example, Siegal et al. [79] tested both RHD and LHD patients with the standard Sally-Anne style false-belief task. RHD patients failed, whereas LHD patients passed the test. Similarly, in a lesion study, a LHD patient, PH, performed almost perfectly on nonverbal second-order false belief task, while he failed grammatical language tests [80]. Most recently, Champagne-Lavau and Joanette [81] observed a co-occurrence of ToM impairment and deficit in pragmatic language in a group of 15 RHD patients. These results from lesion studies are consistent with the findings from behavioral studies and suggest that constitutive aspects of language or at least grammar are less important for ToM than pragmatic aspects of language.

Lesion studies are, although informative, less specific about the brain areas involved in the aspects of language and their relationship to ToM. In order to determine precisely which brain regions are important for explicit ToM and language, researchers most commonly rely on data from brain imaging studies. The classical language regions encompass Broca’s area or Brodmann area (BA) 44/45 in the inferior frontal gyrus (IFG), Wernicke’s area in the superior temporal gyrus (STG), the angular gyrus, and supramarginal gyrus (SMG) [82]. There is now a broad consensus that syntactic processes are subserved by the left STG and the IFG (specifically, BA 44 and frontal operculum) in adults [70, 71, 83]. Semantic processes, in contrast, are found to be supported by the left middle temporal gyrus, the SMG, and BA 45/47 in the IFG in adults [84, 85].

Brain imaging studies that examined relationship between language and ToM are still scarce, but an fMRI study found that adults employ medial prefrontal cortex (mPFC), which has been the most consistently implicated in ToM brain imaging studies across studies [23], for processing pragmatically coherent stories [22]. More recently, my colleagues and I tested Japanese-English bilinguals with a second-order false-belief task (see above), using fMRI ([24], see also [62] for example stimuli and methods). During L1 (Japanese) compared to L2 (English) ToM story condition, more activity in the bilateral dorsolateral prefrontal cortex (DLPFC) was seen. During L2 (English) compared to L1 (Japanese) ToM story condition, more activity in the left precentral gyrus and caudate nucleus was found. Since these two story conditions were matched in semantics and syntactic difficulty (see above), these differences in brain activity are most likely accounted for by pragmatic difference between Japanese and English. Difference in brain activity was also seen while Japanese and American participants performed the same nonverbal cartoon ToM task which was also in the second-order false-belief format. During the cartoon task, compared to Japanese adults, American adults activated the left postcentral gyrus, right mPFC, posterior cingulate cortex, and anterior cingulate cortex. Compared to American adults, Japanese adults activated the right putamen more [see [62] for more detailed results].

Compared to those in adults, brain imaging studies of language in children are still too scarce to reach a broad agreement on the specialization of those linguistic processes in the different areas of the brain. But increasing evidence suggests that language is less lateralized and involves broader regions in children than in adults [86]. In contrast to what Saxe [87] described as “exploded” ToM brain imaging studies in adults those in children are still numbered, but several studies examined neural basis of ToM or related social cognitive functions in children [25,88-93]. However, in many of the studies, ToM (or related social cognition) task specific activity in the aforementioned language regions has been observed (for detailed discussion see [94]). For example, significant activity in the IFG was found while children were engaged in facial imitation [89], story and cartoon-based irony [90, 91] and ToM tasks [25, 93]. So far, our study is the only brain imaging study that has compared adults and children in false-belief task related neural activity and we found difference in false-belief specific activity between the age groups in some of the language areas (see above) including the right IFG (BA 47) and left STG (BA 62). Children activated these regions more than adults [94]. Furthermore, we found similar differences in ToM-specific brain activity between 8-12 year-old Japanese and American children [25] to those between Japanese and American adult [24]. In this experiment [25], American children activated the right TPJ more than Japanese children who activated the left STS and right IFG more while they performed the same cartoon based ToM task, although the significance levels were less (p-values were more) than what are normally expected in fMRI experiments in general. What pragmatic aspects between the two groups account for the differential brain activity remains to be tested, but it has been observed that Japanese use indirect expression more than Americans to inform everyday communicative intentions [95, 96]. Also, a recent study found that Japanese children’s justifications of protagonists’ behaviors in ToM stories are less agent-centered and more interpersonal situation-centered than American children’s [67]. Thus, any of these or other pragmatic / communicative differences might account for the above difference in the brain activity between the two cultural groups.

Through a further analysis, we found the most convergence of ToM-specific activity (measured by ToM vs non-ToM conditions) between the Japanese and American adult and child groups in the right DLPFC (besides precuneus) for both verbal (story) and nonverbal (cartoon) versions of the task (Fig. 1; see also [62]), although in this analysis the bilateral tempro-parietal junction (TPJ) was also active, with the right TPJ being slightly more active than the left TPJ. Besides the mPFC, the TPJ has been implicated also most consistently across ToM brain imaging studies (e.g., [97]). These results, together with Saxe and Wexler’s results [97] which implicate the right TPJ specifically in mental state reasoning, indicate that the right-hemisphere pragmatic region of the brain plays a more important role for ToM than the left-hemisphere constitutive language region.

Fig. (1). Convergence of ToM-specific brain activity between American and Japanese adult and child (ages between 8 and 12 years old) groups. Convergent activity was found in the precuneus (x y z Talairach coordinate [98]: -4 -60 38), right dorsolateral prefrontal cortex (DLPFC: 34 34 20), and bilateral tempro-parietal junction (TPJ: 53 -67 11 and -51 -53 19) through ToM minus non-ToM contrast (p < .0005 uncorrected: see also [62]).

In sum, evidence from both brain lesion studies and neuroimaging studies of ToM has begun to converge to support the right hemisphere-pragmatics hypothesis of ToM. These results are again consistent with the account that pragmatic (not constitutive) aspects of language are necessary for the explicit ToM understanding.