About Counterfactual Belief

8 February 1999

About + Belief + Counterfactual

Josef Perner

University of Salzburg

Draft chapter to appear in P. Mitchell and K. J. Riggs (Eds.). Children's reasoning and the mind. Hove, East Sussex: Psychology Press.

Most of the contributions to this volume reflect the fact that between 3 and 5 years various intellectual changes take place that are related to children's mastery of the false belief task. This reinforces my impression that the false belief task does index some important intellectual acquisition. Needless to say, just what that intellectual achievement is, is hotly contested. I will try to formulate my position on this issue and, thereby, relate to as many of the other chapters as possible, although, in some cases this will not be possible.

One clear exception to the predominant theme is the chapter by Deanna Kuhn (ch. _K_). She argues that the understanding of the relationship between theory and evidence is not achieved with understanding false beliefs. As someone who has taught research methods to psychology students for many years I am strongly inclined to side with her on that. There are also the voices—the no interesting development camp—that deny any marked, interesting intellectual change. They attribute the mastery of the false belief task to the continuous increase in processing capacity (German & Leslie, ch. _GL_) or the decrease of some cognitive bias (P. Mitchell, ch. _PM_). By concluding that the observed development must reflect some deeper intellectual change I present my own position integrating understanding of aboutness, representation and mental causation. The latter of these factors is also central to the position by Wimmer and Gschaider (ch. _WG_). Subbotsky (ch. _S_), too, raises the issue of mental causation. I then discuss the several camps that report specific developmental correlations with other abilities and who tend to interpret these in terms of their favourite theory. The language camp, Jill and Peter deVilliers (ch. _dV_) argue that it is the grammatical apparatus dealing with complement structures that enables children to mentally represent false beliefs. Doug Frye (ch. _DF_) and Phil Zelazo (ch. _Z_), the CCC camp—named after their theory "cognitive complexity and control"—found correlations between the false belief task and tasks assessing children's ability to reason with embedded conditionals that also requires executive control. Then there is the largest, the counterfactual camp led by Riggs and Peterson (ch. _RP_) supported by data from Robinson and Beck (ch. _RB_), followed by interesting extensions by Harris & Leevers (ch. _HL_) and by Amsel and Smalley (ch. _AS_). Why precisely should understanding counterfactuality underlie an understanding of false belief? Riggs and Peterson see the link via simulation. There are others who display sympathy for the simulation camp: Harris & Leevers (ch. _HL_), Fabricius and Imbens-Bailey (ch. _FIB_), and R. Mitchell (ch. _RM_).

I will try and take issue with all these camps. This, however, leaves one very important issue untouched. It is the question addressed by Freeman (ch. _NF_), Hobson (ch. _H_) and R. Mitchell (ch. _RM_) of what is mental representation and how it is acquired. With much regret I had to leave this important issue aside.

Camp 1: No interesting development.

There has been a veritable onslaught on the finding that children below a certain age of about 4 years do not understand false belief (Wimmer & Perner, 1983; Perner, Leekam & Wimmer, 1987). There were many claims that changes in the testing procedure show earlier competence than when the original, "standard" test is used. I am not denying that some of these procedural changes are effective. However, one could characterise the results of this kind of research in the following points.

1. Replicability.

Some published demonstrations of earlier competence are simply not replicable, e.g., the demonstration by Lewis & Osborne (1990) that a more explicit temporal marker makes the deceptive content task easier. Or the alleged finding by Chandler, Fritz and Hala (1989) and Hala, Chandler and Fritz (1991) that could not be replicated with the necessary control conditions by two quite independently working research groups (Sodian, Taylor, Harris, & Perner, 1991).

2. Volatility.

Some of these findings proved to be rather volatile. Although they have been replicated in some studies, others showed a null result. For instance, the claim by Siegal and Beattie (1990) that by asking "Where will Maxi look first for his chocolate?" yields earlier competence was originally demonstrated for one of the story vignettes used by Wellman and Bartsch (1988). It was replicated by Surian & Leslie (1995) for the standard false belief story. They found a hefty difference of 50% but Clements and Perner (1994) found only a 2.5% advantage. Now, I hasten to add, that we in fact did anticipate a large effect because we intended to show that the ”first” cue only works within Vygotsky's zone of proximal development, whose onset might be indicated by an earlier implicit understanding of false belief (as argued by Goldin-Meadow, Alibali & Church, 1993, for a different domain of knowledge) that we had found in children's visual orienting responses. I emphasise this, since I am aware of the existence of the well known experimenter effects (Rosenthal 1986) that one tends to find what one wants to find.

3. Alternative Interpretations.

To the degree that changes in procedure succeed in demonstrating allegedly earlier competence these changes open the results to alternative interpretations. There are many. I have reviewed several in my response (Perner, 1995) to Fodor (1992). I concentrate here on some of the more interesting ones.

Immature concepts. Bartsch and Wellman (1989) claimed that children can use false belief to explain mistaken actions before they can predict mistaken actions on the basis of false belief as required in the standard false belief task. Moses and Flavell (1990) found this difficult to replicate. Eventually, they did succeed in getting more children to answer the questions "Why did Sam (in search of bandaids) go to the empty bandaid box? What did he think?" correctly with "bandaids there." However, they found that this improvement was a rather questionable sign of genuine understanding of belief because these children started to make errors on a subsequent question. When asked what Sam, after finding out that the box was empty, now thought was in the box, they still answered "bandaids". I attributed this to an immature concept of "prelief" (Perner, Baker & Hutton, 1994) where children do not differentiate between similar states like believing and pretending. Going to an empty Band-Aids box for bandaids indicates that Sam is believing/pretending that the bandaids are in there. If the "think" in the test question is assimilated to this undifferentiated understanding of pretending/believing (i.e., "prelieve") then it makes sense that Sam after looking inside the empty box can still be prelieving that it has bandaids in it.

Baseline Differences. In their very elegant "identical twin" version of contrasting explanation with prediction Robinson and Mitchell (1995) were able to provide some support for Bartsch and Wellman's contention. They showed that more children gave the correct answer to "Which twin (one going to the old the other to the object's new location) was in the room when the object was moved?" than were able to predict where the twin who was absent during the transfer would look for the object. However, the difference can be almost completely accounted for by a baseline difference in "correct" responses by those children who understand nothing (Perner, 1995, p. 253). These children tend to have a naive theory that people look for an object where the object is. Hence, they tend to give 100% wrong answers in the prediction task. They have no such wrong theory about why people go to wrong places. The worst they can do is guess, which means they are on the average 50% correct. Once one is aware of this baseline difference it is a very obvious mistake to be avoided. However, I have to admit, it took me years to realise that some of our own data were afflicted: The alleged developmental lag between understanding knowledge vs. ignorance and understanding false belief (Hogrefe, Wimmer & Perner 1986; Perner & Wimmer, 1988) can be entirely reduced to this baseline effect.

Pragmatic Pressures. German and Leslie (this volume, ch. _GL_) cite the finding that false belief based action predictions improve greatly when the object has disappeared. One problem with this finding are pragmatic pressures. From the young child's point of view the question where someone will look for an object that has gone out of existence may be absurd and require the answer "nowhere". Yet the question puts some pressure on the listener to answer with a definite location. In the absence of any clear understanding they use that location that happens to come to mind first. Now we know from the research by Clements & Perner that an implicit understanding of belief emerges fairly sharply at the age of about 2 years and 11 months. Since this implicit knowledge of "he'll look in the original location" exists that may determine children's explicit responses in the disappear condition due to a clear alternative. It would now be interesting whether the explicit response in the disappear condition is due to pragmatic pressure to come up with a definite location or whether it reflects some earlier explicit understanding (in the absence of tempting alternative theories) of belief. This could be tested with the knowledge control condition, i.e., where Max sees that his chocolate disappears. Where will he look for the chocolate? German and Leslie argue that they didn't run this control since it was pragmatically awkward. Yes, but for a young child who does not understand belief, this question is equally awkward in the false belief condition. Pragmatically awkward or not, children's answers in this control condition would shed light on the theoretical alternatives. If they still answer with the original location then the reported finding is difficult to interpret as an indication of understanding false belief. If they say "nowhere" then the reported finding is a clear sign of early explicit understanding of false belief. If they now answer with any of the possible locations then this attests to the presence of pragmatic pressure to pick a definite location and suggests that their choice of the correct location in the false belief condition may be governed by their implicit understanding of false belief.

Content, Attitude and Spatio-Referential Confusion. Peter Mitchell (ch. _PM_) capitalises heavily on the findings by Saltmarsh, Mitchell and Robinson (1995) using the deceptive container paradigm (Hogrefe, Wimmer & Perner, 1986; Perner, Leekam & Wimmer, 1987): children are shown a Smarties box, they say there are Smarties in there. They are shown that they were wrong; there is a pencil in it. Closed up again, with pencil inside, children are then asked either what another naive person would think is in that box, or what they themselves had earlier thought was inside (Gopnik & Astington, 1988).

Although it is nice to know that children find this task about as difficult as the traditional (unexpected transfer) task and it correlates reasonably with it (e.g., Gopnik & Astington, 1988; Holmes, Black & Miller, 1996) I have had my problems knowing what to conclude purely on the basis of performance on this task. To see my problems we need to draw a clear distinction between the content and the attitude of belief (see Perner, et al., 1994). The content is the proposition "There are Smarties in the box". The attitude is given by what role this proposition plays in my mind, i.e., whether I hold it as a belief, as a mere thought, as pretence or as a desire, etc. For our interest in when children understand belief, it is critical to be able to infer from test results that children understand the attitude of belief. The Smarties task, however, relies heavily on getting the content right. The problem, thus, is that children might render the content correctly without fully understanding that the person holds this content as a belief (i.e., has the attitude of belief towards this content). Unlike in the traditional (unexpected transfer) task, in the Smarties task the content of the belief is highlighted and brought to attention by the looks of the box. Had we found that this task was much easier for children, we would have immediately concluded that it was artificially easy: they just say Smarties because they are looking at a Smarties box. As it turned out, it isn't easier, but one needs to always keep that possibility in mind when investigating new variants of this task.

Similarly, in Gopnik and Astington's memory version by having the false belief "Smarties in the box" that content registers in children's mind. Now when they answer with this content to the question, "What did you think was in the box", it is not clear that they understand that they held this content as a belief? They might come up with it because it is in their mind and because the word "think" suggests that the answer requires something different than what was really in the box (i.e., the pencil). By emphasising the content of the false belief (Mitchell & Lacohée, 1991) this content may be more prevalent in their minds and this raises the probability that they answer with this content.

This paradigm is fraught with another complication: proliferation of temporal × mental spaces. And this leads to a serious problem of knowing which space the experimenter is talking about. To some degree this is even a problem with temporal spaces alone. In a long time unpublished study (Perner, Leekam, Myers, Davis, & Odgers, 1993) we had two puppets, Blondie and Curly. Blondie printed a star on his tee shirt. Then either Curly printed the same pattern on his tee shirt, or we made a picture of Blondie with the star on his tee shirt. Then Blondie washed off his print and printed a fish. In both cases we have two contexts: t1 and t2. In t1 Blondie had a star on his shirt, in t2 he had a fish on his shirt. Children had no memory problem, provided one used a unique description of the earlier shape: "which shape did Blondie wash off? All children remembered that it was a star. However, if one used an initially ambiguous reference: "Which was the shape on Blondie's tee shirt..." and then disambiguated by marking which context one is referring to: "...before he changed," then children had "memory problems". It seemed that children had difficulty understanding disambiguating context markers like "now", "before", etc. That they have such a problem was also confirmed when we look at the question about the picture of Blondie: "What shape does Blondie have...in the picture?" Again, the initial reference is ambiguous, "the shape on Blondie's tee shirt"—in reality or in the picture? In contrast, when the question was about Curly (who had the same shape on his tee shirt as Blondie in the picture) there was no problem because the referential expression was unique: "What shape does Curly have on his tee shirt?" (he only ever had one shape, the star, on his shirt).

This problem of referential confusion with time spaces becomes more severe if it is compounded by reference to mental spaces. One reason is that the problem is being aggravated simply because with two dimensions (time and mind) spaces multiply as can be seen from the following development of the deceptive contents task. In the traditional version two spaces (or contexts) have to be distinguished: what is really in the box and what is in the box according to the naive person's view. In the memory version two temporal spaces have to be added: what was in the box before the box was opened (at t1) and what is in the box now (at t2). This then interacts with one's belief, creating more different spaces: What I now (at t2) believe was in the box (at t1), what I now (t2) believe is in the box now (t2), what I then (t1) believed was in the box (t1) and what I then (t1) believed would be now (t2) in the box. Although, not all of them are relevant and have to be kept apart, there is a serious problem of knowing which context the test question is referring to. As deVilliers (1995) noted, children of this age have a particular problem keeping apart the time of the mental (or speech) act (e.g., the time of the thinking) and the time of the content event (e.g., the time of the event thought about).

Moreover, this compounding of temporal and mental spaces may become a specially serious and unpredictable problem when a word like "think" is used, because this word can be used with quite different meaning depending on context (e.g., Perner, 1991, ch. 8) ranging from wanting, to entertaining a thought, to believing. And young children have been shown to have problems finding the intended interpretation. For instance, they inappropriately also interpret it as wanting when it was intended as believing (Wellman & Bartsch, 1988, Experiment 1).

Now, I mention all this because in some of the Smarties task variations in the study by Saltmarsh, et al. (1995) contexts proliferate. There are not only the ones listed above but in Experiment 2 there is also Duffy as a second observer and what he thinks now, thought then was/is in the box. The critical finding was the contrast between two conditions. In one condition the child observed how the expected content (Smarties) was removed and replaced by an atypical content (pencil). In the other the child discovered the atypical content in the box. Both conditions then continued with the introduction of Duffy being confronted with the Smarties box and children are asked what Duffy thinks is in the box now. The finding was that children answered more often with "Smarties" when there had been Smarties in the box initially than when there had always been a pencil inside. This finding can be expected if children are not quite certain what the "think" refers to and it makes them answer with any thought about what is or was in the box (content). The likelihood of the thought "Smarties in box" coming to mind is greater in the condition where there were actually Smarties in the box than when there never had been any Smarties in the box. This may account for the 24% higher incident of "Smarties" responses in this condition.