Verbal-story elicited-response (VS-ER) tasks are widely used to assess false belief understanding (Wellman et al., 2001). In these tasks, a situation is presented where two agents possess different perspectives. For example, in the well-known change-of-location task, children are introduced to a story about two girls, Sally and Anne (Baron-Cohen et al., 1985). Sally places a marble in a basket and then leaves the scene. In her absence, Anne moves the marble from the basket to a box. The child is then asked where Sally will search for the marble upon her return. Passing the task requires understanding that Sally will search for the marble in the empty basket, demonstrating an understanding of her false belief. VS-ER tasks not only assess the child’s ability to attribute false beliefs but also tap into other developmental areas, such as executive functions and verbal ability.

Language development, including pragmatic, lexical, and grammar skills, has been found to be positively related to performance on these tasks (Milligan et al., 2007). Additionally, a child’s conversational environment, particularly the use of mental state language by mothers in conversations with their child, has been linked to the child’s performance on VS-ER tasks (Meins et al., 2002; Ruffman, 2023). This suggests that children’s performance on VS-ER tasks may be influenced by both their own language understanding and production as well as the conversational environment in which they grow up.

False-belief tasks administered to pre-verbal infants employ non-verbal methodologies and assess children’s spontaneous looking patterns as dependent measures (Onishi and Baillargeon, 2005; Song and Baillargeon, 2008; Senju et al., 2011; Surian and Geraci, 2012). These tasks involve presenting a story non-verbally, either through a computer screen or live, without requiring the child to comprehend or employ verbal language to pass the test. For example, in a study by Onishi and Baillargeon (2005), 15-month-old infants were shown an agent reaching for a watermelon in one of two boxes. Subsequently, the watermelon was moved to the second box while the agent either did or did not observe the relocation. The infants’ looking patterns indicated an expectation that the agent would reach for the watermelon in the location consistent with the agent’s false belief, rather than its actual location. These findings, along with numerous other studies, suggest an early capacity in infants to attribute beliefs to agents (Scott and Baillargeon, 2009).

Deaf children born into hearing families experience delays in the development of belief attribution skills when assessed using conventional verbal tasks (Peterson and Siegal, 1995; Meristo et al., 2007). The majority of deaf children are born to parents who do not use sign language and, as a result, these children lack access to language from birth. Growing up in a hearing environment limits their exposure to language-based communication and discussions about mental states, in contrast to their hearing peers. This limited conversational environment is believed to be the primary factor contributing to the delayed development of social perspective-taking skills (Woolfe et al., 2002). Research has shown that deaf children from deaf families, who are exposed to sign language from birth, follow a typical developmental trajectory, emphasizing the significance of environmental factors in social understanding (Woolfe et al., 2002; Meristo and Strid, 2020).

Currently, most deaf children receive cochlear implants (CIs) at an early age. However, studies examining the development of social perspective-taking abilities among children with CIs have produced mixed findings. Some studies indicate that these children demonstrate a disadvantage in VS-ER tasks compared to typically developing hearing children (Peterson, 2004; Walker et al., 2017; Akkaya and Doğan, 2023), while others suggest comparable performance between the two groups (Remmel and Peters, 2009) (also see Lundy, 2002; Macaulay and Ford, 2006; Moeller and Schick, 2006, for studies involving diverse samples of deaf children with and without CIs). In a recent study conducted by Meristo et al. (2016), preschool children with CIs were evaluated using both VS-ER and NV-SR tasks. The results indicated that they exhibited lower proficiency in the NV-SR task compared to hearing children, while performing equally well on the VS-ER. Conversely, deaf children from deaf families achieved scores comparable to those of their hearing peers (Meristo and Strid, 2020).

Successful performance on VS-ER tasks requires the utilization of both verbal and non-verbal cognitive abilities, including advanced pragmatic and executive functioning skills. However, these abilities are not fully developed in infants before 4 years of age, which could explain their difficulty with such tasks. Interestingly, individuals with cochlear implants (Meristo et al., 2016) and those with Asperger syndrome (Senju et al., 2009) appear to find VS-ER tasks easier to solve compared to NV-SR tasks. One possible explanation for this difference is that traditional VS-ER tasks provide additional cues that facilitate task-solving (Bowler, 1992; Happé, 1995; Senju et al., 2009; Schneider et al., 2013). These cues might involve framing the verbal story in a mentalistic manner, explicitly directing attention to the agents’ mental states, or making the connection between mental states and actions more salient.

It has been suggested that high verbal ability could compensate for a lack of intuitive understanding of false belief problems in individuals with autism (Happé, 1995). For example, individuals with Asperger syndrome correctly answered false belief questions in Bowler’s (1992) study, but when asked to explain their solutions, they did so using logical terms, suggesting that traditional tasks can be solved at a non-mental level. More recent research indicates that adults with Asperger syndrome can often provide correct answers to traditional VS-ER tasks as well (Senju et al., 2009), despite experiencing difficulties in everyday social interactions. Thus, while typically developing infants may struggle with VS-ER tasks due to high verbal demands, linguistically advanced adults with Asperger syndrome may instead rely on embedded language cues to solve these tasks.

Another important distinction between VS-ER and NV-SR tasks is the amount of time available for deliberate reflection on others’ mental states. In VS-ER tasks, after the verbal question is presented (e.g., “Where will Sally look for her marble?”), children are not limited by short processing times to arrive at the correct answer. In contrast, NV-SR tasks do not involve traditional testing with verbal responses but require immediate and spontaneous responses. Bowler (1992) suggests that individuals with Asperger syndrome use a slower and more effortful approach to comprehend social situations, resulting in awkward everyday social interactions. Therefore, individuals with Asperger syndrome employ compensatory verbal and non-verbal cognitive abilities when performing on VS-ER tasks. Individuals lacking intuitive or spontaneous access to mental state understanding may rely on different strategies to correctly answer test questions in traditional verbal test settings. In other words, some participants’ success on traditional tasks may be attributed to problem-solving strategies that rely on verbal or attentional cues, rather than an intuitive and spontaneous ability to understand and reason about others’ mental states.

The differences in performance between verbal and non-verbal tasks raise intriguing questions about the specific cognitive processes underlying the understanding of false beliefs in children with CIs. Does their struggle with non-verbal spontaneous-response (NV-SR) tasks imply a dependency on verbal narratives, or does it point to difficulties in spontaneously focusing on important details related to the mental states of individuals? To examine these possibilities, we introduced a novel non-verbal story elicited-response (NVS-ER) task (Pyers and Senghas, 2009). This task eliminates the need for participants to engage with any verbal narratives about the story character or respond to verbal questions. Instead, participants are presented with a series of cards arranged by the experimenter. The only language cue is when the experimenter verbally indicates that a card is missing and poses the question “Which one comes next?.” In this task, minimal language cues are provided, such as indicating a missing card and asking a simple verbal question. However, these cues are intentionally minimal and do not explicitly convey information about the character’s beliefs. They are designed to guide attention and prompt predictions based on the story sequence, rather than relying on language-based strategies. While present, these cues are insufficient to support an alternative strategy solely based on language. If children’s difficulties with the NV-SR task are due to the absence of an explicit question and the resulting need to rely on spontaneous mindreading, they should perform well on this new NVS-ER task. However, if the challenges with the NV-SR task primarily stem from its non-verbal nature, children should perform poorly also on the NVS-ER task.

Twelve Estonian children with CIs participated (8 female, 4 male; mean age = 76 months, range 47–101 months; see Table 1) in our study. Eight children used bilateral and four children unilateral implants. The mean age of first implantation was 20 months (range: 16–33 months) and the children had used implants for the last 56 months on average (range: 28–76). Ninety-five per cent of parents had completed at least high school, and nine children had at least one sibling. None of the children with CIs had any deaf relatives, or native signers, in their immediate family. All children used spoken Estonian as their main way of communication at home. Eight children attended a bilingual school where classes were given either in spoken Estonian or ESL. These children had the freedom to select their preferred mode of communication, allowing them to switch between various subjects and school years. One child attended a deaf school where the mode of education was spoken Estonian, and three children were integrated in mainstream schools. All the children did not experience any language delays or deficiencies, except for the time prior to the implantation and adjustment of CI-s. The children did not have any known language deficiencies.

The comparison group included 18 typically developing hearing Swedish children (nine female, nine male; mean age = 67 months, range 49–107). Ninety-four per cent of parents had completed high school and 16 children had at least one sibling. We did not add an additional comparison group of hearing children from Estonia as the societal contexts for children are very similar to Sweden.

We selected this specific age range because it aligns with the typical developmental stage when children begin to exhibit positive results in the VS-ER false belief tasks. Initially, we had reached out to parents of children with CIs in Sweden, but only two families agreed to take part in the study. Subsequently, we extended our outreach to the two deaf schools in Estonia and the Estonian organization for parents of children with cochlear implants. The current study comprises the families who agreed to participate.

All children were healthy and without known additional disabilities. The parents were informed about the purpose and procedure of the study and gave written consent. The data reported here were part of a larger study, the results of the VS-ER and NV-SR task have been previously reported in Meristo et al. (2016). The Regional Swedish Government Ethical Review Board and Tallinn Medical Research Ethics Committee in Estonia approved the study (Decision numbers: 484 and 1941).

All children were tested individually in a quiet room, at a school in Tallinn, Estonia, or at a research lab in Gothenburg, Sweden. The CI-children were tested in spoken Estonian by a hearing research assistant, except one child who preferred ESL and was tested by a deaf teacher (that is, she only consented to participate in the study when she learned that the deaf teacher, whom she was familiar with, would be the one conducting the tests. As a student at the bilingual school, she was proficient in ESL as well). The hearing children were tested in spoken Swedish by a research assistant. The teacher and research assistants were carefully instructed on how to administer the tasks. All test sessions were video recorded.

One point was credited for each correct selection of a picture that matched a false belief scenario in each story, with a maximum possible score of three. The dependent variable was the response accuracy, with a correct response coded as the baseline outcome. The binomial test was used to compare the number of correct responses out of three trials to the expected probability of guessing correctly by chance, which was 0.125 (12.5%) given two possible outcomes for each trial. In the TH group, 5 out of 18 children correctly responded to the test questions on all three trials (p = 0.032, one-tailed), whereas in the CI group 3 out of 12 children correctly responded to the test questions (p = 0.096, one-tailed).

We also applied a generalized linear mixed model to examine the effects of group membership and test occasions on the dichotomous outcome variable. The model included random intercepts for subjects to account for within-subject correlation across the three tests. The fixed effects revealed that the main effect of group membership was not statistically significant (β = −0.16, SE = 0.89, p = 0.86), nor were the main effects of the test occasions (Test 2: β = −0.82, SE = 0.76, p = 0.28; Test 3: β = −1.05, SE = 0.75, p = 0.16) or their interactions with group membership (Group × Test 2: β = −0.65, SE = 1.18, p = 0.58; Group × Test 3: β = 0.27, SE = 1.17, p = 0.82). The only significant finding was the intercept (β = 1.28, SE = 0.59, p = 0.03), suggesting a baseline effect. These results suggest that neither group membership nor test occasion significantly predicted the outcome when controlling for subject-level variability.

There was a significant difference in DLS between the groups [t(21) = 2.58, p = 0.017, d = 1.03], with the typical hearing children outperforming the children with CIs. The hearing children (M = 0.72, SD = 0.40), t(14) = 6.87, p < 0.001, η² = 0.77, but not the children with CIs (M = 0.16, SD = 0.65), t(7) = 0.68, p = 0.519, η² = 0.06, scored significantly above zero, that is they looked longer at the correct location. Three children in each group did not make a first anticipatory look toward one of the two relevant locations, that is the first fixation toward either the right or left exit of the tunnel from the moment the cat reemerges in search of the mouse. Thirteen of the 15 hearing children who made a first anticipatory look anticipated the reappearance at the correct FB location (p = 0.007, binomial test), compared to only 3 of 8 in the CI-group (p = 0.727).

In line with the NVS-ER task, a score was assigned for accurately responding to the false belief question in each story, with a maximum score of three. The response accuracy was the dependent variable, and a binomial test was employed to evaluate the number of correct responses from three trials against the probability of correctly guessing by chance, which was set at 0.125 (or 12.5%), considering there were two possible outcomes for each trial. The results showed that both the TH-children (12 out of 18 children, p < 0.001, one-tailed) and the CI-children (6 out of 10 children, p < 0.001, one-tailed) correctly responded to the test questions on all three trials.

We also employed a GLMM repeated measures design to investigate the influence of group (TH-children vs. CI-children) and the three different test trials, each a dichotomous variable. The main effect of group membership was not significant (β = 7.15, SE = 4.20, p = 0.09). The main effect for the third test occasion was significant (β = 5.57, SE = 2.61, p = 0.03), but not the second test occasion (β = 2.43, SE = 1.91, p = 0.20). Notably, the interaction effects between group membership and the second (β = −9.88, SE = 4.45, p < 0.03) and third (β = −13.02, SE = 4.94, p < 0.01) test occasions were both significant, suggesting that the effect of group membership on the outcome varied across the tests. These findings suggest that the likelihood of passing the false-belief tests varied between the typically hearing children and those with cochlear implants, and this effect was not consistent across the tests.

This study is subject to several limitations that should be taken into account when interpreting the results. First, the study’s limited sample size poses a significant constraint on the extent to which the findings can be generalized to the broader population of children with cochlear implants (CIs). With a larger sample size, the study would have greater statistical power, allowing for a more comprehensive analysis of the data and enhancing the reliability of the findings. Second, due to the small sample size, the statistical robustness of the findings is limited. While the results are suggestive, they should be interpreted with caution and confirmed by future research with larger sample sizes. For example, the performance of the TH group on the NVS-ER task was not too different from the CI group given that only 5 out of 18 children in the TH group scored 3 out of 3 and this result was significant only at a one-tailed 0.03 significance level. Third, it is important to emphasize that this paper is exploratory in nature, given the small sample size for children with CIs. The findings should be considered preliminary and in need of replication with larger samples to establish their generalizability to the broader population of children with CIs. Finally, looking time measures can offer valuable insights into cognitive processes, but their interpretation is a subject of controversy, as demonstrated by the different theoretical perspectives presented in the introduction (Heyes, 2014; Wellman, 2014; Tomasello, 2018; Ruffman, 2023). For example, some authors argue that infants’ false belief tasks in looking time studies can be interpreted as their understanding of epistemic states such as knowledge/ignorance rather than false beliefs. There is however considerable evidence about infants’ understanding of true and false beliefs in their second year of life (for review see Baillargeon et al., 2010).