Importance of language for children’s theory of mind: Comparison of verbal and nonverbal theory of mind tasks

The ability to attribute mental states to oneself and others, known as the theory of mind (ToM), has been widely researched over the past 40 years, along with its relation to language comprehension. However, a majority of the research on the relation between ToM and language used only verbal tasks assessing false belief understanding as a measure of ToM. Therefore, this study aimed to analyze the relation between language and ToM, using a larger battery of ToM measures, with different language demands. A total of 203 typically developing children between 46 and 68 months of age, with average nonverbal cognitive skills, were assessed using language comprehension and ToM tasks. The language aspect was assessed using the Reynell Developmental Language Scales (Language Comprehension scale A). To assess ToM, verbal and non-verbal tasks were taken from the ToM subtest of the NEPSY?II. Results indicated a significant correlation between language comprehension and verbal and non-verbal ToM measures. Hierarchical regression showed that language comprehension was a significant predictor for children?s performance on both verbal and non-verbal ToM tasks. Specifically, language comprehension affected ToM, regardless of the language demands of the ToM tasks. However, language comprehension was a stronger predictor for verbal than non-verbal ToM tasks. The results of this study contribute to the view that the relation between language and ToM is fundamental and exceeds the features of specific tasks.


3
Theory of mind and language For many years, the essential indicator that one had developed ToM was understanding the concept of false beliefs (i.e., ToM was defined as successful completion of the false belief task; Wellman, Cross, & Watson, 2001). According to Wellman et al. (2001), a child's understanding of false beliefs (i.e., beliefs contradictory to reality) provided compelling evidence that he or she could appreciate the difference between the mind and the real world, thereby showing understanding of mental states as interpretations of reality. In other words, to successfully pass the false belief task, a child has to understand that one's belief is a mental state representing something and can be different from reality (Perner, 1991). Lately, however, researchers increasingly argue that ToM is a wider concept related to general knowledge of the mind, not only understanding false beliefs, and that its development is not an "all-or-nothing" event, but a more complex process with various competencies developing gradually (Astington, 2001;Doherty, 2009). For example, Astington (2001) emphasized that the concept of ToM also includes understanding desires and intentions, arguing that in everyday social interactions, people take more explicit consideration of other's motivational states. Similarly, other researchers (e.g., Altschuler et al., 2018;Baron-Cohen, Joliffe, Mortimore, & Robertson, 1997) included affective aspects of mental state attribution to the ToM construct, assessing, for example, one's ability to interpret emotional clues in pictures of eyes, recognition of others' affective mental states, or the ability to reason about others' feelings. Due to the need to assess different aspects of ToM, different ToM tasks were created in recent years, including tasks for assessing understanding of different mental states (e.g., desires, beliefs, knowledge access, hidden emotions, etc.; Korkman, Kirk, & Kemp, 2007;Wellman & Liu, 2004). Theory of mind and language Theory of mind is often associated with language. Particularly, children with advanced language abilities often achieve better results on ToM tasks (Farrant, Maybery, & Fletcher, 2012;Jenkins & Astington, 1996). Further, the relation between ToM and language has been found in various groups, such as children with an autism spectrum disorder (Tager-Flusberg & Joseph, 2005), a specific language impairment (Miller, 2001), or a hearing impairment (de Villiers, 2005;Peterson & Siegal, 2000). Specifically, previous research (e.g., Astington & Jenkins, 1999;Cutting & Dunn, 1999;Spanoudis, 2016) has shown that children's ToM is associated with their semantic, grammatical, and pragmatic language competencies.

The Relation between Theory of Mind and Language
Although there is some consensus regarding the strong relation between ToM and language, various authors have interpreted it differently. One view is that language does not have a significant role in ToM development, which is mostly supported among modularity theory scholars (Fodor, 1992). The significant relation found between ToM tasks and language comprehension in different studies is believed to be due to the fact most ToM assessment tasks are linguistic tasks, with high language demands (Bloom & German, 2000). However, the most established notion is that of the fundamental relation between language and ToM. Within this framework, different hypotheses have been considered. The first is that ToM influences language development; therefore, difficulties related to ToM could affect language development (Baron-Cohen, Baldwin, & Crowson, 1997). The second hypothesis is that language influences ToM development. This posits that general language ability and/or some specific aspects of language (e.g., syntax) play an important role in the development of ToM, but not vice-versa (Astington & Jenkins, 1999;Farrar, Lee, Cho, Tamargo, & Seung, 2013;Rakhlin et al., 2011). Important evidence supporting that language plays an important role in ToM development comes from longitudinal studies that found children's language comprehension to be an important predictor 5 Theory of mind and language of ToM development (e.g., Astington & Jenkins, 1999;Brooks & Meltzoff, 2015;Ebert, 2015;Farrar & Maag, 2002;Hughes, 1998;Ruffman, Slade, Rowlandson, Rumsey, & Garnham, 2003).
In addition to general language ability, more specific subcomponents of language have also been highlighted as potentially influencing ToM development. For example, some authors proposed that mental state terms, or verbs representing internal mental states (e.g., believe, know, think, feel, etc.), play the most important role in the relation between language and ToM.
More precisely, the syntax and semantics of mental state terms may assist children with processing relations between an individual and his or her beliefs (San Juan & Astington, 2012. However, other authors identified various aspects of syntax as important in the relation between ToM and language, from basic aspects related to word order to advanced aspects encompassing acquisition of relative clauses (Smith, Apperly, & White, 2003) or object complements (de Villiers & de Villiers, 2000). Yet, it is important to acknowledge that most studies on the relation between syntax and ToM have been conducted in English-speaking countries. English has some specific features that do not apply to other languages (e.g., the importance of word order), while highly inflective languages (e.g., Croatian) include various factors not only on the syntactical level but also on the morphological level. Consequently, one should be cautious when generalizing results obtained in one language to other languages. Also, some other studies on Mandarin and Cantonese languages show that the presence of thinkfalsely verbs in Chinese languages does not enhance Chinese children's overall false-belief (Liu, Wellman,Tardif, & Sabbagh, 2008).
The third hypothesis regarding the relation between ToM and language is a combination of the first two hypotheses, which sees the relation between language and ToM as bi-directional (de Villers, 2007;Milligan, Astington, & Dack, 2007). In a meta-analysis, Milligan et al. (2007) 6 Theory of mind and language found that language has a moderate to high effect on children's understanding of false beliefs.
At the same time, earlier false belief task performance predicted later performance on language ability measures. Furthermore, it is important to note that there was a significant difference in effect sizes, as the direction of effect was stronger from language ability to false belief understanding than the reverse.
Nevertheless, ToM and language consist of various components; thus, it is possible that they relate to one another in numerous ways, and that the direction and the effect size of their relation could change over the course of development. The main shortcoming in assessing the relation between ToM and language is that, as mentioned earlier, the majority of ToM tasks are verbal. In these tasks, a child has to follow some narrative or instructions to infer others' mental states. Success in this kind of linguistic task depends not only on the developed ToM but also on memory, attention, understanding the precise meaning of test questions, and a certain degree of linguistic resources (e.g., understanding, processing, production; Bloom & German, 2000;Burnel, Perrone-Bertolotti, Reboul, Baciu, & Durrleman, 2017).
The need for non-verbal ToM tasks has existed for quite some time and consequently some of these tasks have been created or researched (e.g., Call & Tomasello, 1999;Onishi & Baillargeon, 2005;Peterson, Wellman, & Liu, 2005;Schick, de Villiers, de Villiers, & Hoffmeister, 2007). Some of them are studies which use Wellman and Liu's (2004) battery of tasks for assessing ToM (e.g., Brooks & Meltzoff, 2015, Burnel et al., 2017; however, even the tasks in this battery include minimal verbal demands, and children still have to follow some form of verbal narrative. Consequently, performance on these tasks can still be influenced by a child's language comprehension (Burnel et al., 2017).

Theory of mind and language
However, the majority of research studying the relation between ToM and language used only the understanding of false beliefs as a measure of ToM (e.g., Milligan et al., 2007;San Juan & Astington, 2017). For example, in their meta-analysis, Milligan et al. (2007) included only different standards and experimental tasks that assessed one's understanding of false beliefs. The authors concluded that there is an association between children's language comprehension and their scores on different false beliefs tasks, regardless of the language demands of those tasks. However, this conclusion was limited to children's understanding of false beliefs as only one ability within children's ToM. If any of the three hypotheses regarding the fundamental relation between language and ToM is correct, we would expect to find a significant relation even when assessing different ToM abilities (e.g., understanding of beliefs, intentions, desires, etc.) In one study, Brooks and Melzoff (2015) attempted this by using the Wellman and Liu's (2004) battery of tasks, and they indeed found a significant relation between children's language comprehension and ToM. However, they used only children's vocabulary production/comprehension as a language measure.
The aim of this study was to evaluate the relation between children's language comprehension and ToM, as assessed with both verbal and non-verbal ToM tasks. Specifically, the primary research question was whether language comprehension is a good predictor of ToM because tasks for assessing ToM are linguistic tasks, or whether there is a fundamental connection between language and ToM. We hypothesized a positive correlation between language comprehension and ToM, regardless of whether ToM was tested using verbal or predominantly non-verbal tasks.  Table 1.

Instruments
A list of tests and tasks used in this study is presented in Table 2.
Intellectual Abilities. The Matrix Reasoning subscale of the WPPSI-III (Wechsler, 2002); (Croatian translation done as a part of the project Cognitive and language development in children at neurodevelopmental risk -013-1081870-2627) was used as a measure of nonverbal or general cognitive skills, due to high correlations with nonverbal (r = .82) and full (r = .74) IQ scales (Wechsler, 2002). In this study, it was used as an inclusion criterion. The reliability expressed by the Cronbach's alpha coefficient on our sample is α = .82. For the verbal scale (15 tasks), a child is read various scenarios or was shown pictures, and is then asked questions that require knowledge of another individual's perspective to answer correctly. These tasks assess a child's understanding of mental states (e.g., desires, intentions, beliefs, deceptions, and emotions) and that others can have mental states that differ from his or her own. The reliability expressed by the Cronbach's alpha coefficient on our sample is α = .72.
In the Contextual tasks scale, the child is shown a picture depicting a simple social situation (for example Laura fell off her bike) in which the face of the target individual is not shown. The child is then asked to select the photograph from four options that depicts the appropriate affect (happy, sad, anger, fear) for the target individual in the picture. It assesses 10 Theory of mind and language the child's ability to recognize facial affect and to understand how emotion relates to social context and to recognize the appropriate affect given various social cues.

Procedure
All participants were assessed individually. At the beginning, to ensure a sample of children with typical cognitive development, the nonverbal Matrix Reasoning subtest of the WPPSI-III (Wechsler, 2002) was administered, followed by the principal assessment. The sequence of tests and tasks in the sample was randomized to control the effects of training and tiredness on test scores. Subsequently, participants were assessed using the Reynell Developmental Language Scales (Reynell & Huntley, 1995) and the ToM subtest from the NEPSY-II (Korkman et al., 2007).

Results
Descriptive statistics for each of the measures are shown in Table 2.

First, we calculated correlations between predictor variables and verbal and non-verbal
ToM scales, as shown in Table 3. The correlation between verbal and non-verbal ToM scales was .07 (p > .05). Table 3, there were significant correlations between language comprehension and both the verbal and non-verbal ToM scales. Additionally, children's scores on the verbal ToM scale were significantly correlated with age.

As shown in
Next, hierarchical regression analysis was used to predict children's ToM for each ToM measure. In the first stage, control variables of age, gender, and mother's level of education were entered into the model, and in the second stage, children's language comprehension task scores were also entered into the model. Table 4 shows the results of hierarchical regression analysis on two different criteria for all participants. A regression equation was statistically significant at the first stage for the verbal ToM scale (control variables explained 16% of the variance); however, it was not statistically significant for the non-verbal ToM scale. Introducing language comprehension tasks at the second stage explained an additional 10% of the variance for the verbal ToM scale and 6% of the variance for the non-verbal ToM scale.

Discussion
The majority of ToM research has used assessment tasks that are in some way linguistic.
Due to the language demands of these tasks, it is difficult to draw conclusions regarding the nature of the relation between children's language and ToM. The NEPSY-II (Korkman et al., 2007) tries to overcome these problems with a ToM subtest. The ToM subtest includes a verbal scale, in which a child is read various scenarios or shown pictures and then asked questions that require knowledge of another individual's mental state, and a non-verbal scale, in which a child is shown a picture representing a social context and asked to select a photograph from four options showing the appropriate affect. Thus, compared to the verbal scale, the non-verbal scale does not include any special language demands. This differentiation of the verbal and nonverbal scales is an optimal base for research on the relation between ToM and language, especially for answering questions regarding the predictive value of language for ToM development and determining the nature of this relation.
Furthermore, the majority of research on the relation between ToM and language has used only false belief tasks as a measure of ToM (e.g., Milligan et al., 2007, San Juan & Astington, 2017; thus, one of our aims was to investigate the relation between language and the wider construct of ToM, beyond only false belief understanding. As noted, the verbal scale assesses children's understanding of different mental states. Consequently, children's performance on this scale is an assessment of different ToM abilities, not only false belief understanding. We found a significant relation between language and the verbal ToM scale, and language was a significant predictor for children's scores on this scale. This is in line with a previous study by Brooks and Melzoff (2015), and could indicate there is a relation between children's language and their ToM, beyond false belief understanding, as was proposed by different authors (e.g., Astington & Baird, 2005;de Villiers, 2007).
Additionally, the results of our study indicated that children's language comprehension is a significant predictor for their performance on both verbal and non-verbal ToM scales.
Children's language comprehension seems to have a significant effect on their ToM, regardless of language demands in ToM tasks. Our results indicated the relation between language and ToM remains significant, even when ToM is assessed using non-verbal tasks. This is a strong indicator that the relation between language and ToM is fundamental and goes beyond features of specific tasks. Additionally, it is in line with previous findings (e.g., de Villers, 2007;Milligan et al., 2007;San Juan & Astington, 2012;Brooks & Meltzoff, 2015), emphasizing the importance of language for ToM development. Furthermore, most researchers agree that children's communication with others plays an important role in ToM development (Astington & Baird, 2005). Through communication, children learn that different people know different things. Consequently, children come to understand that sometimes they know something others do not, and sometimes others have information that is unknown to them. It can be said that language gives children access to the mental world (Harris, 1999). Similarly, some authors posit that it is through communication that children adopt mental states (e.g., Astington, 1996;Bartsch & Wellman, 1995;Peterson & Siegal, 2000). Moreover, some authors claim that it is actually language that enables children to think about mental states; thus, language is a tool for 13 Theory of mind and language representation of mental states. As such, language is meant to help children, especially when thinking about more complex mental representations (Low, Apperly, Butterfill, & Rakoczy, 2016). Therefore, we would expect language comprehension to be an important predictor of performance in verbal and non-verbal ToM tasks (i.e., we would expect language and ToM to be fundamentally related because, through language, children acquire concepts of mental states). The results obtained in this study supported this viewpoint. Children's language comprehension was shown to be a significant predictor of ToM, regardless of whether ToM was assessed using verbal or non-verbal tasks.
However, the results of our study indicated language comprehension was a stronger predictor for performance on the verbal ToM scale, compared to the non-verbal scale. This could be an indicator that the relation between language and ToM found in different studies is partly the result of language demands in ToM tasks. Previously, several authors argued that the language demands in ToM tasks are so high they distort researchers' conclusions regarding the relations between language and ToM (e.g., Bloom & German, 2000).
There are, however, other possible reasons for differences in the relation between language comprehension and verbal and non-verbal ToM scales. First, the stronger relation between the verbal ToM scale and language comprehension could be due to verbal tasks, compared to non-verbal tasks, being clearer and more explicit, because the child is already given a description of the situation. For non-verbal tasks, the child has to realize what happened by him or herself, which could demand higher cognitive resources. In this case, children's performances on the non-verbal ToM scale could partially depend on factors such as children's attention, cognitive flexibility, social knowledge, etc. affective ToM (knowledge about others' emotions), which are thought to involve overlapping but dissociable brain systems. If this was the case in our study, and the difference in relations between language comprehension and verbal and non-verbal ToM scales was the effect of the different abilities assessed by each scale, we could still gain some insight into the relation between ToM and language. More specifically, we could hypothesize that language is an important predictor for cognitive and affective aspects of mental state attribution and understanding. However, this is just one possible explanation, and we cannot draw any definite conclusions about it.
Additionally, as mentioned previously, there are still some questions regarding differences in the relations between language and verbal and non-verbal tasks. Further studies could examine these relations to provide additional insight. Furthermore, in the present study, we only assessed children's language comprehension. In the future, it could be interesting to also assess children's expressive language competence, then compare the predictive value for children's ToM.