Early Theory of Mind Competencies: Do Infants Understand Others’ Beliefs?

Authors


should be sent to Birgit Träuble, Department of Psychology, University of Heidelberg, Hauptstrasse 47-51, D-69117 Heidelberg, Germany. E-mail: birgit.traeuble@psychologie.uni-heidelberg.de

Abstract

Recent studies suggest that even infants attend to others’ beliefs in order to make sense of their behavior. To warrant the assumption of early belief understanding, corresponding competences need to be demonstrated in a variety of different belief-inducing situations. The present study provides corresponding evidence, using a completely nonverbal object-transfer task based on the general violation-of-expectation paradigm. A total of = 36 infants (15-month-olds) participated in one of three conditions. Infants saw an actor who either observed an object’s location change, did not observe it, or performed the location change manually without seeing it (i.e., variations in the actor’s information access). Results are in accordance with the assumption that 15-month-old infants master different belief-inducing situations in a highly flexible way, accepting visual as well as manual information access as a proper basis for belief induction.

For more than 25 years it has been assumed that children develop an understanding of others’ beliefs around 4 years of age and that younger children do not understand that mental states, such as beliefs, are not direct reflections of the reality but representations that may be true or false (e.g., for a meta-analysis see Astington & Gopnik, 1991; Flavell, Green, & Flavell, 1990; Perner, 1991; Wellman, Cross, & Watson, 2001). To attribute a false belief to a person, one has to discriminate false and true propositions, be aware of causal efficacy of mental states, and realize that a person believes that something false is true (Perner, Leekam, & Wimmer, 1987). Empirical findings suggest a synchrony in preschoolers passing a large number of verbal false-belief (FB) tasks including different belief-inducing situations (e.g., unseen object transfer, or misinformation about an object’s location) as well as different response demands (answer questions, or going to get the object). These tasks are assumed to test “explicit” (e.g., Clements & Perner, 1994) belief understanding by usually measuring “elicited” (Song, Onishi, Baillargeon, & Fisher, 2008) responses. In these tasks children are usually asked direct questions about an actor’s belief.

Using different kinds of tasks, assumed to test “implicit” (e.g., Clements & Perner, 1994) belief understanding, more recent studies suggest an earlier emergence of belief understanding by measuring children’s “spontaneous” (Song et al., 2008) responses: By the use of anticipatory looking tasks, Ruffman, Garnham, Import, and Connolly (2001) as well as Clements and Perner (1994) have shown that 3-year-olds show correct anticipatory looking behavior (i.e., a looking pattern consistent with the protagonist’s belief) using children’s eye gaze in response to a verbal anticipation prompt (“I wonder where the actor will look for the object”) as dependent measure.

Results obtained with completely nonverbal anticipatory looking paradigms indicate that 25-months-olds correctly anticipate an actor’s behavior that accords with her false belief (Southgate, Senju, & Csibra, 2007; for similar results with 18-month-olds, see Neumann, Thoermer, & Sodian, 2008). In addition to anticipatory looking tasks, results obtained with violation-of-expectation tasks also support the assumption that infants attribute false beliefs to agents: In their widely debated study Onishi and Baillargeon (2005) provided evidence suggesting that even 15-month-old infants do understand false beliefs in a nonverbal, unexpected transfer FB task: Infants saw an actor placing an object in one of two boxes. The actor then either left the scene (FB condition), or remained watching (true-belief (TB) condition), while the object changed its location by moving from the first to the second box. During the test trial, infants saw the actor reaching either into the empty box, or into the box containing the object. Infants participating in the TB condition showed longer looking times when the actor reached for the object in the empty box, whereas infants participating in the FB condition looked longer when the actor searched for the object in its current location. According to Onishi and Baillargeon, infants appreciate false beliefs as mental representations that guide others’ behavior. Similarly, Surian, Caldi, and Sperber (2007) have shown that within a violation-of-expectation task, 13-month-olds attribute false beliefs also to nonhuman agents, namely to a computer-animated caterpillar.

Whether these results can be taken as evidence that children in their second year of life are able to attribute beliefs to others is still under heated debate and several alternative interpretations have been proposed. In their critical commentary on Onishi and Baillargeon’s (2005) study, Ruffman and Perner (2005) argue that the ascription of FB understanding is only warranted if positive results can be demonstrated in a variety of FB tasks including different belief-inducing situations (see also Perner, 2008). This argument is embedded in a more general discussion concerning the need for mental understanding in order to pass behavioral FB tasks. Referring to Povinelli and Vonk (2004) it was argued that children’s responses might be based on behavior rules relating specific situations to specific actions without taking mental processes into consideration. For example, with regard to infants’ looking behavior in the study of Onishi and Baillargeon, Perner and Ruffman (2005) argued that infants’ looking behavior might be based on a behavioral rule whereupon “…people look for an object where they last saw it…” (p. 215). According to the authors, such a rule might capture something implicit of the mind because it applies as a result of the mind mediating between seeing and acting. Nevertheless, infants could simply deploy the rule without conceiving the mind as the mediator. While this argument also holds with regard to the classic FB tasks used for testing older children, it was argued that a flexible mastering of different belief-inducing situations makes the learning and use of isolated behavior rules implausible (Perner, 2008). In this spirit, some recent studies examined whether infants not only attribute to others false beliefs about the location of an object but also about identity, contents, properties, and number (e.g., He & Baillargeon, 2007; Scott & Baillargeon, 2009; Scott, Song, Baillargeon, & Leslie, 2007; Song & Baillargeon, 2008).

Instead of varying different contents of false beliefs (e.g., number and properties), the present study examines infants’ attention to different types of belief-determining information access. Recently, Song et al. (2008) showed that, when an actor falsely believed that an object was hidden in location A as opposed to location B, 18-month-old infants were not surprised when the actor searched in location A, unless she was informed by words or by pointing gestures that the object was actually in location B. Given that slightly younger infants passed the FB task in the original study of Onishi and Baillargeon (2005), the aim of the present study was to test 15-month-olds’ attention to different types of information access. In addition to the usual TB- and FB-inducing conditions in which infants see an actor either observe (TB condition) or not observe (FB condition) an object’s location change, a third belief-inducing condition was introduced in which the actor manually controls the object’s location change without seeing it. Hence, the belief induction in the third condition does not depend upon the actor’s visual access. The goals of the present study are twofold: (1) we test whether earlier findings obtained by Onishi and Baillargeon with 15-month-olds can be replicated in a different experimental setting; and (2) going one step further, we vary the type of another person’s information access in order to test whether 15-month-olds flexibly attend to relevant factors for belief determination.

Method

Similar to Onishi and Baillargeon (2005), we used a nonverbal violation-of-expectation task.1 In three different experimental conditions, infants saw an actor who: (1) observed the transfer of an object from one location to another (true belief, visual access); (2) did not have any access to information about the location change (false belief, no access); or (3) did not observe the location change but caused it manually (true belief, manual access). Infants’ looking time in response to two test scenes showing the actor searching for the object in each location served as dependent measure. Differing from Onishi and Baillargeon as well as most other infant studies on FB understanding, we thus applied a within-subject design.

Participants

Thirty-six 15-month-old infants (20 girls and 16 boys) participated (M age = 15 months 11 days, range = 15 months 0 days to 15 months 29 days). Seven additional infants were tested but excluded from the final sample due to fussiness (= 3), experimenter error (= 2), and technical problems (= 2).

Material and procedure

The child was sitting in front of a small stage, which could be covered from the infant’s view by a curtain. The open stage revealed a table with a balance beam and an actor sitting behind. The arms of the balance beam consisted of two parallel metal rods (length: 1 m). Attached to each end was a box (14 cm × 11 cm × 11 cm, one green one yellow) opened toward the fulcrum, such that a foam ball (2.7 cm in diameter) could enter them. The ball could easily roll from one end to the other without making any noise. The actor could operate the balance beam by lifting one side manually, thus causing the ball to roll from the upper to the lower end and disappear in the corresponding box. Each box was also open on top such that the actors’ hand could reach into it and place or remove the ball. The child was sitting on her mothers’ lap and could not reach the beam.

Infants received two familiarization trials. In the first trial, infants saw the actor placing the foam ball in one of the two boxes (position counterbalanced across infants), while the beam was in a horizontal position. The actor then lifted one arm of the beam near the fulcrum, thus causing the ball to roll into the opposite box. No auditory cue indicated the location change. The actor followed the movement of the ball with her head. In the next step the balance beam returned to its horizontal starting position. It was moved by a second experimenter, hidden from the view of the infant. Next, the actor moved the balance beam one more time. She followed the ball rolling back into the first box before the balance beam again returned to its horizontal position. At the end of the trial the scene was covered by the curtain. The second familiarization trial was identical to the first one, except that the actor placed the ball into the opposite box at the start of the trial. Hence, in each familiarization trial the ball changed its position twice. The familiarization phase was followed by the belief induction phase. Infants were randomly assigned to one of three belief induction conditions (= 12 infants in each condition).

In the TB condition infants saw the actor placing the ball in one of the two boxes. The actor moved the balance beam and followed the location change of the ball with her head. The ball was then returned to the first box as the beam was moved by the second (hidden) experimenter while the actor was again watching this transfer. During the test phase which immediately followed the belief induction phase, the actor first put both empty hands on the stage, then reached into one of the boxes and kept this position for 20 sec (freeze scene). Each infant saw two belief induction trials with two different test-outcomes, respectively: The actor reached either into the empty box or into the box containing the ball (henceforth denoted as ball box). The direction of the first reach (ball box or empty box) as well as the position of the ball box (left or right) was counterbalanced across the sample.

The FB condition started in the same way as the TB condition: Following the first location change, a bell was shortly ringing in the background, thus motivating the actor to turn away from stage. While the actor was sitting with her back turned toward the beam, the ball changed its location again, but this time the event was only being observed by the infant. Then, the actor turned around and the test phase began. This phase was identical to the TB condition. Note, that in the TB as well as FB condition the second movement of the balance beam was not caused by the actor. Hence, both conditions were fully comparable with each other with respect to the causation of the movement of the beam (as well as the location change of the ball).

The manual-control (MC) condition was highly similar to the FB condition, but with one important difference: While the visually conveyed information for the actor was the same in both conditions, the actor in the MC condition caused the second location change of the ball by manipulating the beam. More specifically, she reached behind her back and moved the beam while her gaze was turned away. It seems important to note that she moved the beam near the fulcrum. Hence, the true position of the ball was not cued by the direction of her reach. After turning around and facing the beam again, the test phase started. It was identical to the previous conditions. Figure 1 illustrates the experimental design.

Figure 1.

 Illustration of selected frames from the three conditions. In each condition infants were tested twice. After the belief induction infants saw the actor reaching for the ball either in the empty box or in the ball box at test (order of test-outcome was counterbalanced). In the false-belief condition the actor’s turn is preceded by a short ringing in the background to signal a sudden distraction.

If infants appeal to the actor’s belief in the TB condition, they should expect the actor to search at the true location of the object, as the actor observed the object transfer, and therefore holds a true belief. Accordingly, infants should be surprised (i.e., look longer) if the actor reaches into the empty box compared with the test-outcome when the actor reaches into the ball box. If infants’ participating in the FB condition consider the actor’s belief they should expect the actor to reach into the empty box because she did not see the object transfer and therefore holds a false belief. Hence, infants should show longer looking times for the test-outcome when the actor reaches into the ball box compared with the test-outcome when the actor reaches into the empty box. This inverse looking pattern for the TB and FB conditions would replicate the findings by Onishi and Baillargeon (2005). However, the same looking pattern might also be accounted for the deployment of a “search where you last saw” behavior rule (Perner, 2008; Ruffman & Perner, 2005). In the TB condition, infants should expect the actor to reach into the ball box because this was the place where she last saw it, whereas in the FB condition this was the case for the empty box.

In the MC condition the seeing–searching relation was disentangled because the belief induction was not dependent upon the actor’s visual access. This leads to the following outcome predictions: If infants flexibly attend to relevant factors for belief determination (e.g., different ways of information access), they should not be surprised when the actor in the MC condition searches in the ball box. The actor has information considering the ball’s location change independent of visual information access. By contrast, the deployment of a “see where you last saw” behavior rule should lead to longer looking times when the actor reaches for the ball box (independent of the fact that the actor might know the location of the object at test even if she did not see the former location change).

Results

Infants’ looking behavior was video-taped and coded offline for total accumulated duration to each of the two test events (freeze scene only). Twenty-five percent of the cases were coded by a second coder. Mean intercoder correlation for the looking duration on each test event (in sec) was = .90.

A repeated measures mixed model analysis of variance with Condition (TB, FB, MC) and order of test-outcome (ball box first, empty box first) as between-subject factor and test-outcome (search in ball box, search in empty box) as within-subject factor revealed a marginally significant main effect for test-outcome, F(1, 30) = 3.86, < .06, η2 = .11, qualified by a significant test-outcome × Condition interaction, F(2, 30) = 7.52, < .01, η2 = .33. Figure 2 reveals the corresponding means for each condition.

Figure 2.

 Mean looking times (in sec) for the test-outcomes, separated by the three conditions (TB = true belief; FB = false belief; MC = manual control). *< .05. **< .01. Error bars represent ±1 SE.

Planned comparisons between conditions revealed a significant test-outcome × Condition interaction for the TB–FB comparison, F(1, 20) = 8.59, p < .01, η2 = .30, as well as for the FB–MC comparison, F(1, 20) = 14.84, < .001, η2 = .43, while for the TB–MC comparison there was only a significant main effect for test-outcome, F(1, 20) = 11.97, < .01, η2 = .37, with higher looking times for the empty-box outcome (= 18.43, SE = .32) compared with the ball-box outcome (= 16.71, SE = .57) in both groups.

Corresponding analyses within each condition revealed that infants showed longer looking times for the empty-box outcome compared with the ball-box outcome in the TB condition, t(11) = 1.90, < .05, one-tailed, = .46, as well as in the MC condition, t(11) = 3.35, < .01, one-tailed, = 1.11. By contrast, infants participating in the FB condition showed the reversed looking pattern with longer looking times for the ball-box outcome compared with the empty-box outcome, t(11) = −3.00, p < .01, two-tailed, = .58.2

Nonparametric analysis (Wilcoxon signed-rank tests) showed that within each condition significantly more infants showed a looking pattern that was consistent with the hypothesis. In the TB condition nine (of 12) infants showed longer looking times when the actor searched in the empty box compared with the ball box (p < .05). In the FB condition 10 (of 12) infants showed the opposite looking pattern (p < .01), and in the MC condition 10 (of 12) infants showed the same looking pattern as the majority of infants in the TB condition with longer looking times when the actor searched in the empty box (p < .01).

Discussion

Using a novel nonverbal object-transfer task, the present findings replicated earlier results reported by Onishi and Baillargeon (2005): In the TB condition 15-month-old infants showed longer looking times when the actor reached for the object in the empty box, whereas infants in the FB condition showed longer looking times when the actor reached for the object in its current location. To our knowledge this is the first study to test 15-month-olds’ belief understanding using a within-subject design. Most importantly, the design of the present study allowed us to test whether 15-month-old infants attend to different ways of information access when interpreting another person’s actions.

As earlier studies with infants of the tested age range realized different belief inductions by manipulating the actor’s visual access to an object transfer, the deployment of the “search where you last saw” rule can not be ruled out in corresponding procedures. The design of the present study allows us to disentangle the usual seeing–searching relation by contrasting the TB condition with an FB condition, as well as with an MC condition. That is, the belief induction in the MC condition and in the TB condition leads to the same (true) belief about the final location of the object but based on different types of information access (i.e., visually or manually conveyed information). By contrast, the belief induction in the MC as well as FB conditions leads to different beliefs concerning the object’s final location with the visually conveyed information being the same.

Based on a comparison of results in all three conditions (TB, FB, MC), we conclude that 15-month-olds attend to relevant factors for belief determination in a rather flexible way. In the MC condition, the actor had no visual, but only manual information access. Still, infants looked longer when the actor searched for the ball in the wrong location (i.e., the empty box). This observation is inconsistent with the assumption that infants applied a simple “search where you last saw” rule in the present context. It could be, however, that infants use a different behavioral rule in events when an actor is manually engaged.

The present report does not allow any definite conclusions regarding the rule-based approach. For instance, infants might well use a rule whereupon people search for objects according to their perceptual access in a more general sense, including various forms like visual, auditory, and tactile access. From that point of view future research might focus on the question to what type of perceptual information access infants attend to at what time in development. However, it has also been stated that a flexible mastering of different belief-inducing situations provides indirect support for the assumption of a mentalistic belief understanding in young children. Our findings demonstrate that at 15 months of age infants attend to different forms of an actor’s information access suggesting at least some flexibility in their use of relevant factors for belief determination. The reported findings fit well into a gradually growing number of studies suggesting that even infants master different belief-inducing situations in a highly flexible way.

Footnotes

  • 1

    It has been argued that results obtained within a violation-of-expectation procedure leaves open the possibility that infants attribute ignorance rather than false beliefs to an agent. Studies with older children suggest that children expect an ignorant person to err rather than perform at chance (Ruffman, 1996). However, recent research suggests that infants do not expect ignorant agents to err (He & Baillargeon, 2007; Scott & Baillargeon, 2009; Scott et al., 2007).

  • 2

    For the MC condition, a two-tailed test was chosen because the mentalist assumption and the rule-based assumption would predict opposite looking patterns for the test-outcomes. In the TB as well as FB conditions, both assumptions would predict the same looking pattern with longer looking times for the empty-box outcome in the TB condition and longer looking times for the ball-box outcome in the FB condition. Therefore, a one-tailed test was used in these two conditions.

Ancillary