Continuity from an Implicit to an Explicit Understanding of False Belief from Infancy to Preschool Age

An implicit understanding of false belief indicated by anticipatory looking has been shown to be significantly correlated with performance on explicit false‐belief tasks in 3‐ and 4‐year‐old children (Low, 2010). Recent evidence from infant research indicates, however, that implicit false‐belief understanding guides infants' expectations about goal‐directed actions even in the second year of life. The present study presents data from a sample of N= 70 infants who were tested longitudinally at 15, 18, 30, 36 and 48 months with implicit and explicit Theory of Mind measures, as well as an assessment of verbal IQ. Belief‐based anticipatory looking in the false‐belief task at 18 months significantly predicted verbal false‐belief reasoning at 48 months, after controlling for verbal IQ. These findings indicate developmental continuity and conceptual specificity in belief reasoning from infancy to preschool age. They are discussed with respect to competing accounts of infants' understanding of the mind.

An explicit understanding of false belief develops around the age of 4 years ([37]). Until recently, there was evidence for an implicit understanding of belief emerging in children's anticipatory gaze patterns around the third birthday, but not earlier ([9]; [11]). However, in the last 7 years, infant research has made substantial progress in demonstrating the ability to take an agent's false belief into account when reasoning about goal‐directed action in infants in their second year of life (for reviews, see [7]; [19]; [29]; [34]). Converging evidence for belief‐based reasoning about intentional action comes from studies using violation‐of‐expectation, anticipatory looking, and interactive paradigms. Looking‐time studies have found evidence for infants' belief‐based action expectations in an unexpected transfer paradigm ([25]; [39]), as well as for their ability to reason about agents' false beliefs about identity, contents, property, and number (see [3], for a review). Similarly, [4] showed in an interactive task that 18‐month‐old infants interpreted an adult's behaviour differently depending on whether or not the adult had witnessed an unexpected location change of an object, supporting the conclusion that they took the adult's false belief into account when inferring his action goals.

While both looking‐time and interactive tasks assess infants'reactions to an agent's psychologically plausible or implausible actions, studies using eye‐tracking technology can provide information about infants' ability to predict agents' actions based on their false beliefs. [37] familiarized 2‐year olds to a video sequence in which a human agent (the 'searcher') standing behind a shoulder‐high wall with two windows in it watched a 'hider' (a hand puppet) placing a target object in one of two boxes. After this, a short action delay was introduced during which infants' fixations at the left and right hand window in anticipation of the searcher's behaviour were recorded. Finally, the searcher reached through the window next to the box and retrieved the object from the box. In test trials, after having placed the object in the box, while the searcher was distracted, the hider took the object out of the scene. Subsequently, the searcher turned back and infants' anticipatory fixations were recorded. Results showed that 85% of the 25‐month‐old infants included in this study first fixated at the window through which the searcher should be expected to reach based on her false belief, and that, as a group, infants spent significantly more time looking at the correct window than at the other one during this anticipatory interval.

Using a simplified task in which a self‐propelled car either moved from one location to another or turned back to its original location, [24] found belief‐based action anticipation in 18‐month olds. Thus, infants in the second year of life can already take another person's false belief into account when anticipating her action. What is the developmental relation between implicit belief‐based action anticipation in infancy and explicit understanding of false belief in preschool age? Is an implicit understanding of false belief a precursor to an explicit understanding?

To date, the relation between implicit and explicit belief understanding has only been studied in 3‐ to 5‐year‐old children. [11] presented 3‐year‐old children with a task adapted from [9] and elicited both anticipatory gaze and verbal responses. In this early study, comparing concurrent implicit and explicit false‐belief performance, the authors found that while the majority of children showed belief‐based anticipatory gaze patterns, only few were able to verbally solve the false‐belief task correctly. The same group ([30]) followed up on these results by asking 3‐ to 5‐year‐old children to bet on a location in which the story character would search for the target object based on a false belief, while also assessing their explicit answers and anticipatory looking. As expected, the majority of participants showed correct anticipatory looking while failing the explicit false‐belief task. Furthermore, the interpretation of anticipatory looking as an indicator of implicit understanding was supported by the finding that children bet highly on their (incorrect) explicit answers, indicating that they were not aware of their (correct) gaze behaviour.

In a more recent series of studies examining concurrent relations between implicit and explicit belief understanding, [20] assessed 3‐ and 4‐year‐olds' competence in an implicit anticipatory looking task, and a battery of false‐belief tasks, as well as verbal and nonverbal ability, sentence complement mastery, and cognitive flexibility. His findings showed not only that implicit and explicit ability to reason about false‐beliefs was correlated, but also that they showed distinct patterns of correlations with the other measures: While implicit false‐belief‐based action anticipation was not related to verbal ability, complement structure, and cognitive flexibility, these variables had unique contributions in explaining the variance in explicit false‐belief reasoning, even when implicit false‐belief competence was accounted for.

While these concurrent studies indicate that implicit and explicit understanding of false belief are linked, but not equivalent, to date, no longitudinal research has addressed the relation between the competencies observed in infancy, and false‐belief understanding in preschool age.

The idea that there may be long‐term developmental continuity from infancy to preschool age in the social domain is derived from domain‐specific accounts of conceptual development such as the core cognition theory (e.g., [5]). Recent longitudinal studies indicate significant predictive relations between infants' encoding of goal‐directed actions in habituation tasks and preschoolers' Theory of Mind (ToM) performance, independent of general language ability, executive function, and IQ (e.g., [2]; [39]; [40]). This developmental link has been shown to be specific to the social domain ([41]).

The findings on false‐belief representation in infancy have sparked an ongoing debate about the depth of infants' insight into the mind ([1]). Proponents of a rich interpretation argue that infants possess a representational ToM (e.g., [21]; [36]), whilst advocates of a lean interpretation argue that the findings can be explained at the level of behavioural rules (e.g., [27]; [31]). The rich interpretation predicts developmental continuity from infancy to preschool age and would lead us to expect developmental relations between implicit and explicit false‐belief tasks and, further, links among a wide range of ToM tasks. If infants possess a ToM in the sense of a rich conceptual framework for predicting and explaining human action, then infants' ability to take another's beliefs into account when inferring his goals, should be related to both belief and desire reasoning at a later age.

A lean account of infant false‐belief reasoning in terms of behaviour rules would also lead to the prediction of developmental relations between implicit and explicit false‐belief competencies, since an analysis of intentional action in terms of behaviour leads to the same predictions as an analysis in terms of mental states ([26]). However, on this account, situation‐specific relations with later ToM reasoning, rather than a broad generalization over a wide range of situations of mentalistic action explanation should be expected. Similarly, the accounts by [1] and [20] propose a limited and inflexible system of mentalistic action explanation in infancy that over development becomes interdependent with language and executive function and thus enables an explicit representation of false belief. Such accounts would predict task‐ and situation‐specific continuity from infant psychological reasoning to later explicit reasoning about false beliefs, rather than a theory‐like coherence of belief reasoning that remains continuous from infancy to preschool age.

The present longitudinal study is the first to address the developmental relation between implicit false‐belief reasoning in infancy and explicit false‐belief understanding in preschool age. Task specificity is addressed by including two explicit false‐belief tasks, a false‐belief‐about‐location task, and a content false‐belief task. Concept specificity and conceptual coherence are addressed by assessing both belief reasoning and desire reasoning, and by assessing visual perspective‐taking abilities at different ages. Finally, by controlling for verbal IQ, it is determined that the developmental link cannot be accounted for by general cognitive abilities.

It should be emphasized that it will not be possible to dismiss a rich or a lean theoretical account of infant ToM competencies based on a single set of longitudinal correlational findings, since several sources of shared variance contribute to such correlations, and the developmental relations are likely to be complex. It is important to begin to gather such data in order to better understand developmental relations from infancy to preschool age.

In sum, the present longitudinal study addressed three issues. First, we examined interrelations between conceptually related abilities in infancy by assessing Level 1 perspective‐taking ability in a looking‐time task at the age of 15 months and false‐belief understanding in an unexpected‐transfer anticipatory looking task at 18 months. Second, we examined developmental links between infant competencies and ToM competencies in the third year of life (an explicit perspective‐taking task at 30 months and an explicit desire‐understanding task at 36 months). Finally, we examined developmental links between infant belief‐based action anticipation and explicit false‐belief understanding in matching tasks (location) and non‐matching tasks (contents) at the age of 48 months. To control for the specificity of the presumed relations, verbal intelligence (WPPSI‐III; German version [28]) was assessed at 48 months.

The final sample consisted of N= 701 (36 girls, 34 boys) healthy, full‐term, monolingual children who participated in the present longitudinal study of social‐cognitive development.

Ages at the measurement points were as follows: 15 months (M= 15 months, 3 days; SD= 10 days), 18 months (M= 18 months, 2 days; SD= 7 days), 30 months (M= 30 months, 4 days; SD= 11 days), 36 months (M= 36 months, 8 days; SD= 8 days), and 48 months (M= 48 months, 16 days; SD= 24 days).

Assessments were carried out in a child‐friendly University laboratory and all children were accompanied by a caretaker. Participation was voluntary. Families received a 5 € travel compensation and a small age‐appropriate gift at each measurement point. Data from additional five children were excluded because they contributed data to less than three measurement points.

Infants' ability for Level 1 visual perspective taking was assessed with a looking‐time procedure adopted from [35]. Infants were seated in a high chair facing a puppet‐stage‐like opening. When the stage curtain opened, they saw a live female agent sitting at a table on which two equally sized plastic toys (a blue fish and a red duck) were placed about 60 cm apart towards the left and right side of the stage. An opaque screen was placed on the stage next to the two objects, covering about 40% of the stage, but not hiding the toys from either the experimenter's or child's view during the familiarization phase. Infants were first presented with a total of six familiarization trials in which the female agent asked: 'Where is my toy?' while looking across the table and finally grasping one of two objects (henceforth referred to as the goal object) while announcing 'Here!' Each of the objects was established as the goal object during familiarization trials for approximately half the sample. During the six test trials following familiarization, the goal object was either placed behind the opaque screen so that the experimenter could no longer see it while the observing child still could or it was positioned next to the screen such that both experimenter and child could see it. The non‐goal object was placed in the position at which the goal object had been during familiarization trials. As during familiarization, the experimenter looked for her toy and subsequently reached for the object that had not been her goal object during familiarization trials. Rational (original object hidden by screen) and irrational (original goal object placed next to the screen) test trials were presented alternately and children were assigned randomly to test order conditions. For further procedural details, see [35]. Looking time was recorded by two independent observers blind to the experimental condition who pressed an electronic button whenever they judged the child to be looking at the display. A given trial ended when the main observer had judged the child not to have been looking at the display (i.e., had not pressed the electronic button) for 2 s. Looking times and observer agreement were calculated using Xhab. Only infants for whom observers had achieved an observer agreement of K >.8 were included in the final sample. For correlational analyses, a weighted test preference score was calculated by dividing the time looked at the inconsistent trial by the sum of looking at both test trials. This score was multiplied by 100 to obtain a percentage score. Thus, a score of >50% indicates that infants looked more at the inconsistent trial (and thus were considered 'passers', as depicted in Table 1).

1 Descriptive statistics of performance on all measures

This task was adapted from [24]. In order to assess belief‐based action understanding, infants watched animated coloured movies on a 17‐inch screen with an integrated eye‐tracking system (Tobii T60) from a distance of about 60 cm. The movies showed a female agent watching a yellow toy car moving from one garage to another (see Figure 1). During familiarization (two trials for each child, each lasting 32 s), once the actor had seen the car arrive at the second box, she disappeared behind a screen. Subsequently, two small doors located right above each of the two boxes were illuminated (accompanied by a simultaneous chime). This was followed by a freeze frame of 3 s (the anticipatory period) during which anticipatory fixations at the two doors (areas of interest; AOI) were assessed. Then the agent's face appeared at one of the doors (the one above the box where she had seen the car disappear) and she reached through this door in order to retrieve the car from the box. These familiarization trials served to establish that toddlers understood the story line. Ten infants failed to show correct anticipatory fixations in at least one of the familiarization trials and were therefore excluded from further analyses. Infants subsequently received one test trial. Test trials (lasting 41 s) differed from familiarization in that a phone ring distracted the actor temporarily from observing the car during test trials while the car continued to move. Different from familiarization trials, after reaching the second garage, it went backwards to the first garage and then went on driving through the garage, thus disappearing from the screen. Once the car exited the screen, the doors illuminated with an accompanying chime and infants' fixations were recorded over a 3‐s anticipatory period. Figure 1 shows the timeline for the test event of this belief‐based anticipatory looking task. Throughout the experimental procedure, participants were filmed by a scene camera. Only participants for whom the recording system indicated sufficient eye‐tracking quality and whom the reviewer judged to have paid consistent attention to the screen were included in the final sample. For correlational analyses, a proportion score of looking at the belief‐based (correct) door AOI (that is, the door in the wall at which the agent would be expected to search based on his belief about the car's location) was calculated by dividing the time spent looking at this AOI by the sum of looking at this and the incorrect door AOI and multiplying this quotient by 100 to obtain a percentage score. Infants who received a score of >50% were considered 'passers'.

Graph: 1 Timeline of the familiarization (a) and test (b) events. The solid frame marks the anticipatory scene. (Note that movies were presented in colour; thus, the garages were clearly discriminable).

Hiding skill was assessed with a task adopted from [10] (see also [22]) and consisted of two sub‐tasks, the hiding and the judgement task, administered in this order2.

• 1

• For the hiding task, experimenter and child sat at opposite sides of a sofa table. A cardboard screen (18‐cm high × 23‐cm wide) with an attached wooden base was placed broadside on the table between them. The child was then asked to 'put the teddy (a 10 cm × 5 cm × 3 cm Pooh© toy teddy) on the table so that I don't see him'. This was conducted four times, with varying positions of the experimenter in relation to the child's position: sitting 180° opposite the child, 90° to the child's right, 90° to the child's left, and then next to the child. For each trial, the child was awarded one point for each trial in which the teddy was placed so that the screen was between the experimenter and the teddy. Thus, the score for this perspective hiding task could range from 0 to 4. A random sample of 30% of the valid sample was coded by a second coder blind to the judgement of the first coder, inter‐rater agreement was κ=.95.

• 2

• The judgement was modelled after [22]. Overall, children received two trials with four test questions. On the first trial (180° trial), the experimenter (E) sat opposite of the child at a narrow table (both experimenter and child were seated on chairs). The same props as in the hiding task were used. E then put the cardboard screen broadside to herself, in such a way that the bear was blocked from her view, but in clear sight of the child. The child was asked: 'Can I see the bear, now?' (correct answer = no) and 'Can you see the bear, now?' (correct answer = yes). Then, the experimenter put the cardboard screen broadside to the child, so that the bear was blocked from the child's view, but in clear sight of the experimenter. Again, the child was asked: 'Can I see the bear, now?' (correct answer = yes) and 'Can you see the bear, now?' (correct answer = no). Approximately 30% of the sample was coded by a second coder blind to the judgement of the first coder, which yielded 100% inter‐rater agreement.

Depending on the trial, to be correct, children had to either nod or answer the question with 'yes' or to shake their head or answer the question with 'no', respectively. For each correct answer, children received a score of 1. For each incorrect answer children received a score of 0. Thus, across the two trials, children could receive a maximum score of 4.

Due to experimenter error or lack of cooperation, not all children received all trials. Consequently, only children who had received and answered at least three of four questions of each subtask were included in the analyses and the percentage of correct answers out of all valid questions was used as the dependent variable. For subject‐level analyses, children were scored as passers if they answered at least 66% of the valid questions of a given subtask correctly (i.e., two out of three or three out of four questions).

The easiest task of the German version of the ToM Scale ([14], [15]; [38]; for the full German version with accompanying material, see [12]) was administered at 3 years of age. The subjective desires task presented children with a character (a Playmobile© figure) with a snack preference that differed from their own (e.g., if children stated that they prefer cookies, the character, Mr. Miller is introduced as preferring carrots and disliking cookies). Both kinds of foods were depicted as coloured drawings without context. Children were then asked to predict which kind of food Mr. Miller would eat for snack time if he could only have one of the two. As a control, children were asked to restate which kind of food he preferred.

Two false‐belief tasks from the German version of the ToM Scale ([15]) were administered: (1) The contents false‐belief task asked the child to judge another person's false belief about what was in a container while the child himself knew what was is in the container. The child was shown a prototypical Smarties© container and asked what he or she thought was inside. Then the container was opened and a piglet was found inside, which was labelled by the experimenter. Then the container was closed again and the child was asked the memory control question to state the current content of the container. (If the child gave a wrong answer, the container was opened again.) Then, a Playmobile© figure was introduced as never having looked inside the container. The child was then asked the test question 'What does Lukas (the playmobile figure) think is inside the container?' and the control question 'Has he ever looked inside?'

(2) The Location False‐Belief task required the child to judge where someone would search based on a mistaken belief. The child was told a story about a Playmobile© figure (e.g., 'Paul"), who was looking for his mittens that could be either in the closet or in his backpack (both of which were presented as decontextualized coloured drawings on A4 format). The child was told that really the mittens were in his backpack, but that he believed they were in the closet. The child was then asked the test question: 'Where will Paul look for his mittens?' and the control question: 'Where are they really?'

For each task, the child was awarded one point if test and control question was answered correctly. Hence, the two false‐belief tasks could yield a score between 0 and 2. (For further procedural details, see [15]; for test materials, [12]).

The verbal IQ at 48 months was assessed by administering two subtests of the German version of WPPSI ([28]) verbal IQ scale, Similarities and Information. Both tasks were administered according to the testing manual. The subtest Similarities measures verbal reasoning and concept formation. The child was read an incomplete sentence containing two concepts that share a common characteristic, such as 'Dogs and cats, both are...?'. The child's task was to complete the sentences by stating a shared property or superordinate concept (e.g., '...four‐legged' or '...pets'.).

The subtest Information involved the recall of facts addressing a broad range of general knowledge topics. Items were presented verbally, for example, 'How many legs does a bird have?' and the child was required to answer verbally or by showing.

Following the standard scoring procedures ([28]) based on the raw values, normalized scores for the given age group for each subtest were assigned. In order to arrive at a normalized score estimate for the verbal scale with only two rather than three subtests, the sum of the normalized values of the two administered subtests was divided by two and multiplied by three. Based on this, estimated Verbal IQ scores were assigned.

A brief descriptive overview of the results for individual tasks will be provided initially, followed by correlational analyses. Descriptives for all tasks are provided in Table 1 and Table 2 shows zero‐order correlations between tasks.

2 Zero‐order correlations among measures

1 Note. (*)p <.1; *p <.05, **p <.01, all two‐tailed. As a measure of association between continuous (1. −4., 8.) and dichotomous (5.–7.) variables, Spearman's rank correlations were calculated, associations between dichotomous variables are expressed by phi‐coefficients.

The implicit Level 1 perspective‐taking task administered at 15 months of age presented infants with test events sequentially, and looking time at the entire display in reaction to the agent's action was assessed. No significant differences between girls' and boys' absolute looking times and weighted test preference scores were found.

Table 1 shows the mean proportion of time infants spent looking at the belief‐based door. Fifty‐five percent of the participants received a score of >50% and thus were considered 'passers'. Five children failed to show overt fixations towards the agent during the distracting event in the information phase. However, as exclusion of these children did not affect the results, these children were retained in the sample. No significant difference in girls' and boys' fixation patterns was found.

As can be seen in Table 1, toddlers performed better in the judgement (M= 74% correct trials) than the hiding (M= 42% correct trials) task (t (60) = 7.80; p <.001). Girls (hiding: M= 49%[SD= 32] correct trials; judgement: M= 84%[SD= 16]) outperformed boys (hiding: M= 35%[SD= 23]; judgement: M= 65%[SD= 25]) in both tasks (hiding: t (60) = 2.04; p <.04; judgement: t (493) = 3.45; p <.01).

Of the 63 children who completed the desire task at 36 months of age, 38 (60%) passed the task, while 25 (40%) failed. Boys (55% passers) and girls (65%) performed equally well.

Each of the tasks was passed by 20 (39%) of the 51 children who had completed both tasks. Task performance was associated (φ (51) =.32; p <.05): 23 (45%) failed both tasks, while 12 (24%) passed both. An equal number of children (30%; eight children) passed the contents, but not the location task or showed the reverse pattern.

Girls outperformed boys in both tasks (Contents: Girls: 54% passers, boys: 22% passers; χ2(1, 51) = 5.37; p <.05; Location: Girls: 57% passers, boys: 17% passers; χ2(1, 51) = 8.37; p <.05).4

Children reached a mean verbal IQ estimate of 107 points, with no difference between girls and boys.

In first step, relations between ToM competencies were inspected. Zero‐order correlations (see Table 2) indicated a marginally significant link between performance on both early implicit measures, the Level 1 perspective‐taking looking‐time task at 15 months and belief‐based anticipatory looking at 18 months (r (32) =.34; p=.05). When this correlation was controlled for verbal IQ (which was marginally significantly correlated with anticipatory looking: r (35) =.28; p <.1), it still reached significance (rp (26) =.39; p <.05).5

Neither explicit perspective‐taking abilities at 30 months nor desire reasoning at 36 months were positively related to earlier implicit perspective‐taking or false‐belief performance, nor to explicit false‐belief performance 1 year later. There was a negative correlation (ρ (40) =.34; p <.05) between implicit perspective taking at 15 months and desire understanding at 36 months; however, this finding was not confirmed by non‐parametric analyses. Of the 40 infants for whom data from both tasks were available, 42% performed equally well across tasks (12% failed both and 30% passed both), 30% passed the implicit perspective‐taking task at 15 months but not the explicit desire task at 36 months, and 28% showed the reverse pattern (χ2 (1, 40) = 1.38; ns).

Finally, belief‐based anticipatory looking at 18 months showed a significant zero‐order correlation with the sum score of explicit false‐belief reasoning at four years (ρ (31) =.37; p <.05). When relations to the explicit tasks were inspected separately, it proved to be strongly related to location false‐belief competence at 4 years (ρ (31) =.5; p <.05), but not to performance in the unexpected contents task (ρ (31) =.16; ns).

This finding was confirmed by subject‐based non‐parametric analyses: While none of the 13 children who had not succeeded in the belief‐based anticipatory looking task at 18 months (i.e., had looked 50% or less at the belief‐based AOI) mastered both false‐belief tasks at 48 months, eight of the 18 children who had looked longer at the belief‐based AOI at 18 months answered both false‐belief tasks correctly (χ2 (1, 31) = 7.79; p <.005). Separate analyses for the two false‐belief tasks showed that this pattern held only for the location false‐belief task: only two children who had not looked longer at the belief‐based AOI at 18 months of age passed the location false‐belief task at 48 months (χ2(1, 31) = 5.13, p <.05), while no significant association was found for the contents false‐belief task (χ2(1, 31) =.41; ns).

When the correlation between belief‐based anticipatory looking at 18 months and explicit false‐belief reasoning at 48 months was controlled for gender and verbal IQ, it proved to be stable (composite score: rp (26) =.52; p <.01; location false belief: rp (26) =.63; p <.01). Hence, after controlling for verbal ability, the only stable predictor of false‐belief competence at 48 months turned out to be belief‐based anticipatory looking 30 months earlier.

To substantiate the above correlational findings, three separate stepwise conditional binary logistic regressions predicting explicit desire understanding at 36 months, and explicit location and content false‐belief understanding at 48 months were conducted, including subject gender and verbal IQ estimate as a first step, and then precursor measures in chronological order (implicit perspective taking at 15 months, belief‐based action anticipation at 18 months, perspective hiding and perspective judgement skill at 30 months). For explicit desire understanding at 36 months and content false‐belief competence at 48 months, no significant predictors could be identified at any step. However, explicit location false‐belief performance at 48 months was predicted significantly by gender and implicit false‐belief performance at 18 months (see Table 3).

3 Stepwise conditional Binary logistic regression for false‐belief location understanding with subject sex entered first and implicit perspective taking, implicit false‐belief understanding, hiding skills, and verbal IQ estimate entered blockwise (ordered by age). Contribution of individual variables and summary estimation of model at different steps to predict explicit location false‐belief performance.

2 Note. Only variables included in the equation are shown. (*)p <.1; *p <.05; **p <.001).

The present study found a significant longitudinal predictive relationship between individual differences in an anticipatory looking task, administered at the age of 18 months, testing an implicit understanding of false belief, and an explicit understanding of false belief at the age of 48 months, independently of verbal IQ. This finding is consistent with previous studies showing a concurrent correlation between implicit and explicit false‐belief understanding in 3‐ to 4‐year‐old children ([30]; [20]), and is the first evidence for developmental continuity of belief understanding over an age range of two and a half years, from infancy to preschool age.

The pattern of correlations indicated task specificity, the relation being due to a high correlation between an implicit and explicit location‐false‐belief task that did not generalize to a content‐false‐belief task. Furthermore, implicit false‐belief understanding predicted explicit false‐belief understanding, but no other ToM concepts, such as desire understanding. In infancy, an implicit visual perspective‐taking task administered at 15 months was marginally significantly related to performance on the false‐belief task at 18 months; however, the implicit visual perspective‐taking task did not correlate with explicit visual perspective‐taking competencies assessed at 30 months.

The predictive relation between implicit and explicit false‐belief understanding is consistent with earlier research indicating continuity from infancy to preschool age in the social domain ([38]; [39]; [2]). While previous research showed a developmental relation between goal encoding and ToM, the present study addresses more specific relations between conceptually equivalent tasks.

As was outlined in the Introduction, continuity in belief reasoning from infancy to preschool age is consistent with both rich and lean interpretations of infant false‐belief understanding; however, the evidence for task specificity emerging from the present study appears to support lean, rather than rich accounts of ToM in infancy, since a rich account of infant ToM would predict generalized developmental relations with conceptually equivalent tasks, that is, with false‐belief tasks of different formats, and would furthermore lead us to expect both concurrent and predictive relations with other mental state attributions, such as the attribution of desires. It should be noted, however, that we can only tentatively conclude that belief reasoning in infancy is task specific since we used only one false‐belief task in infancy, but two different ones in preschool age. It is therefore not possible to test for task‐specific relations between an implicit version of a content false‐belief task in infancy and the 'Smarties' task in preschool age. The present findings indicate a specific relation between an implicit and explicit location false‐belief task, when both tasks were administered in an action prediction format.

Initial action prediction systems may be centred around specific task demands such as predicting goal‐directed action in tasks that require a selection between different locations. A core knowledge system of agency ([5]) may guide infants' encoding of belief‐based action, based on a fast and automatic computation of predictions of goal‐directed action. The implicit action prediction system that responds to such specific task demands appears to provide the underpinnings for an explicit judgement in a false belief about location task at the age of 48 months; it does, however, not serve as a basis for a response to a false‐belief task of a different format. Thus, infants' skills in predicting goal‐directed action may be centred around distinct classes of action goals, with information about agents' beliefs remaining implicit in such context‐specific action predictions, rather than being explicitly accessible as a generalized framework for mentalistic action explanation.

A pre‐conceptual account of infants' reasoning about others' perspectives is also supported by the present findings on relations among visual perspective‐taking tasks. Performance in an implicit visual perspective‐taking task at 15 months predicted performance on the implicit false‐belief task 3 months later, which is consistent with the view that visual perspective understanding (at Level 1) is a precondition for belief‐based action prediction. However, neither the visual perspective‐taking task nor the implicit false‐belief task predicted explicit visual perspective understanding at 30 months and there was no relation between implicit visual perspective taking in infancy and later explicit false‐belief understanding. These findings are consistent with proposals by [32] and [20] that a pre‐conceptual cue‐based ability to reason about others' perspectives may be functional early, and independently of later flexible perspective representation, whereas rich accounts of infants' reasoning about perspectives ([16]; [17]; [21]) would predict conceptual coherence with later explicit perspective‐taking abilities.

In sum, the present study presents first evidence for (task‐specific) continuity in false‐belief reasoning over a large age range from infancy to the preschool period. These findings add to our understanding of domain‐specific developmental continuity between social information processing in infancy and ToM development. As was outlined above, the present findings are consistent with the view that domain‐specific core cognition supports reasoning about agents' goal‐directed (and belief‐based) action in infancy. However, they also support the notion that such an initial system is transcribed into an explicitly accessible and flexible reasoning format over development, in conjunction with the emergence of language and action control systems ([20]).

Future studies should aim at exploring mechanisms of transition from implicit to explicit understanding. One strategy could be to develop and administer strictly parallel implicit and explicit tasks, addressing different classes of belief‐based action, over a wider age range. Consistent with previous cross‐sectional and short‐term longitudinal studies of ToM, language, and executive function ([20]; [6]; [13]; [33]), the present findings strongly suggest that the developmental relations between implicit and explicit false‐belief understanding can only be understood when investigated longitudinally in the context of emerging executive functions and language acquisition.

The research reported here was funded by DFG grant # So 213/27‐1, 2 to Beate Sodian. The authors would like to thank all children and parents who generously donated their time to participate in this longitudinal study. Thanks also to all undergraduate research assistants for help with assessment and data coding, and to Iyad Aldaqre for assistance in eye‐tracking data analysis.