RUM And The Perspectival Relativity Of Sortals
1
Draft: 6 July 1999
RUM, PUM, and the Perspectival Relativity of Sortals.
Josef Perner
University of Salzburg
To appear in J. Astington (Ed.). Minds in the making: Essays in honor of David R. Olson. Blackwell.
When I was a student in the Department of Psychology at the University of Toronto there was little Cognitive Development on offer for me besides Barney Gilmore's one term Piaget course. So I sauntered over to OISE to attend the weekly colloquia of David Olson’s research group and became a fairly regular attendant. At the time I was interested in logical reasoning, in particular, Piaget’s class inclusion task on which Bill Ford was working under David’s supervision. Their approach sensitized me to the pragmatic aspects of the language used with children and influenced by David’s 1970 Psychological Review paper (Olson, 1970) I developed a long lasting interest in pragmatics. This interest resurfaced with vigor later when trying to relate children’s problems in Piagetian tests to their developing theory of mind. I not only had a theory like Margaret Donaldson’s (1978) of why they failed Piaget and Inhelder’s conservation test - because the questions asked are not genuine questions since the experimenter knows the answer. I also could explain why the older children passed - because they are able to understand second order epistemic interest: the experimenter wants to know whether I know. Unfortunately my paper ”The Insincerity of Conservation Questions” (Perner, Leekam & Wimmer, 1984) made it only to two conferences (APA in Toronto, August 1984; and BPS Developmental Section, Lancaster, September 1984) because reality did not cooperate sufficiently for turning it into a proper publication. Our initial data indicated that children’s problems with the conservation question were alleviated when the test question was turned into a genuine looking question asked by a seemingly ignorant newcomer. Unfortunately, this could not be replicated (Arthur, 1985).
David’s paper with Elisabeth Robinson (Robinson, Goelman & Olson, 1983) had another devastatingly strong influence on me. It led me into a spate of activity trying, with the help of Nicola Yuill, to find a link between children’s understanding of knowledge formation and their appreciation of the ambiguity of referential expressions. My memories are vague, but I remember as many as ten studies none of which yielded coherent, interpretable, publishable results. However, the experience was valuable in forcing me to think about the relation between children’s metarepresentional and metalinguistic abilities.
Less directly I could also blame David for my efforts trying to prove that under certain conditions mutual knowledge was logically justified. Fortunately for David, this enterprise led to some success. Although we basically rediscovered the original solutions proposed by David Lewis (1969) and Steve Schiffer (1972) (but ignored by Herb Clark: Clark & Marshall, 1981) we were able to add to them and get it published in the Journal of Semantics (Garnham & Perner, 1990; Perner & Garnham, 1988) and use it for a commentary on Sperber and Wilson (1986).
Clearer success resulted from David roping me into being the external supervisor of Deborrah Howe’s M.A. thesis. Deborrah and David had some neat experimental ideas, which Deborrah and I worked into the still clearest evidence against ”introspectionist” simulation theory (Perner & Howes, 1992). By the time of this collaboration David’s influence on me had taken on a quite different form—as a great and effective sponsor of ”theory of mind” research. It began with an invitation to his theory of mind symposium at the Society for Philosophy and Psychology (SPP) meeting in Toronto in 1985. Then the following year David, with the help of Janet Astington, Lynd Forguson and Alison Gopnik, beat Paul Harris by one month to holding the first ever theory of mind conference. The joint proceedings of both conferences appeared in the well known opus by Astington, Harris and Olson (1988) which gave a tremendous (some would claim, overly effective) boost to theory of mind research.
Legitimately blended with his self interest in this area David continued his moral and financial support of theory of mind research. Whenever I happened to be in North America there was a reason and some money for a stop-over in Toronto and stimulating explorations of meta-this-and-meta-that in David’s office. The latest occasion was David’s workshop "Literacy and Conceptions of Language," in April 1999. He invited me to this workshop because he had been very interested in our report (Doherty & Perner, 1998) of a close relationship between understanding false belief and metalinguistic abilities. David’s excitement about this finding came just in time to rescue me from interpreting it the wrong way due to some recent incommensurable data. Here is the story and the correct interpretation (so I hope).
The Original Finding
Martin Doherty and I (Doherty, 1994; Doherty and Perner, 1998) discovered a surprisingly strong correlation between children’s mastery of the traditional false belief test (Wimmer & Perner, 1983) and our ”synonyms task”.
In the false belief task a story character ,Maxi, puts his chocolate bar into one location (A) and leaves. In his absence the chocolate is unexpectedly transferred to another location (B). On his return hungry for his chocolate, children are asked where he will look. Typically young 3-year-olds answer incorrectly that he will look in location B where the chocolate bar is. After 4 years most children give the correct answer, at about the age that they are able to understand our "synonyms" task: Children are first tested for their knowledge of word pairs that come close to being synonyms, e.g., bunny-rabbit, lady-woman, television-TV, etc. On different occasions children were shown the same object among 3 distractors and asked, "Which one is the bunny?", and later, "Which one is the rabbit?" Then they were given training on three training pairs. It was pointed out that the item could be called a bunny and a rabbit. So when a talking puppet (operated by the experimenter) says that it's a bunny, then the child's task is to call it a rabbit and the other way around. Then five test pairs followed. As one would expect, the younger children made more errors. They tended to repeat what the puppet had just said, while most older children correctly used the alternative expression. What was the pleasant surprise was that this ability did, indeed, as we had predicted, correlate strongly with mastery of the false belief task (r = .71 in Experiment 3 and r =.65 in Experiment 4) even when children's age and some measure of verbal intelligence (BPVS) were partialled out (r = .66 and r =.60, respectively). Because children had to produce a suitable alternative expression, we called this the production version of the synonyms task.
We also turned the paradigm around, in order to check to what degree the difficulty of this task has to do with finding an alternative expression. The child names the item first (e.g., "a rabbit") and the puppet then has to produce the alternative expression. The child's task was to judge whether the puppet had conformed to the instructions in each of three trials presented in a random sequence (judgement version). The puppet either [1] correctly used the alternative description ("a bunny", correct judgement: "yes, puppet said the right thing"), or [2] repeated the child's label ("a rabbit", correct judgement: "no"), or [3] said something wrong ("an elephant", correct judgement: "no"). Again correctness of judgements was strongly correlated with passing the false belief task: r = .83 in Experiment 1 and r =.82 in Experiment 2, and stayed so even when the effects of a structurally similar control task and verbal intelligence in Experiment 2 (r = .70) were partialled out.
These correlations between the false belief task and the judgement version of the synonyms task have since been replicated by Martin Doherty (1998): r = .71 and r = .55 after BPVS was partialled out. In a study by Wendy Clements children were taught 4 German color words and then had to use the German word if puppet had used the English word and the other way around. Again the correlation with the false belief task was high: r = .75 and r = .73 after memory for the German color words was partialled out (Clements, Heider, Brooks & Garnham, 1998). A mixed result was obtained in one study with German speaking children in Austria by Waldh?r (1996). In one experiment there was a clear correlation (r = .55 and r = .46 with age and the K-ABC, a verbal intelligence test partialled out), but in the second experiment the correlation was lower and not statistically significant: r = .23 and r = .10 after age and verbal intelligence had been partialled out. This, however appears to be one deviant sample, because Sandra Stummer (Perner & Stummer, 1998) found very strong correlations with Austrian children (see Table 1 below).
Our Original Explanation: Representational Understanding of Mind (RUM)
At face value the false belief task and the "synonyms" task have little in common. Their strong correlation is, therefore, highly surprising. The only explanation available was the one that led us to conceive of this study. This was the claim that mastery of the false belief task indexes the onset of a Representational Understanding of Mind (or RUM, an acronym coined by Martin Doherty). I have argued for some time (Perner, 1988; 1991) that understanding false belief is tied to an understanding of representation. I will retrace this argument below. Martin and I then reasoned that the "synonyms" task also requires an understanding of words as representation. To understand words as representations one needs to be aware of them as a representational vehicle and be aware of their representational content (meaning). The synonyms task requires both. The child has to monitor the meaning of the word used (it has to describe the item correctly; "elephant" won't do) and monitor the particular means (representational vehicle) used to express this meaning, i.e., to use a synonym. The data provided a better confirmation of this theory than we had originally hoped for. Too good to be true.
One nagging problem has always been that most linguists claim that there are no true synonyms. Certainly the word pairs that we have used are not synonymous, at best they are pseudo-synonyms. Nevertheless, we argued that they were synonymous for such young children.
Troublesome Data
Using the production version only we started to use words other than "synonyms" (Perner & Stummer, 1998). All these tasks can be characterized as Say-Something-Different tasks. In one condition we used basic and superordinate categories, e.g., dog-animal. Children had a natural tendency either always to use the basic term ("it's a dog") or to repeat the puppet’s expression. To use "animal" when the puppet had said "dog" was as difficult as the synonyms task and the false belief task and it correlated highly with the false belief task. Essentially the same results came from another categories task using basic and subordinate categories, e.g., dog–poodle. Children again had a tendency to use the basic term (dog) or repeat the term used by puppet. The ability to use the subordinate term after the puppet had used the basic term was as difficult as and correlated highly with the false belief task (See Experiments 1 and 2 in Table 1: The top panel shows the percentage of children passing each task for each experiment, e.g., the false belief task, synonyms task and categories task. The lower panel shows the correlations between the various tasks and the false belief task).
-----------------------------
Insert Table 1 about here
-----------------------------
These data are an embarrassment to the RUM theory because children can obey task instructions purely at the level of the meaning of words: say something true but different about the item. There is no need to look for differences at the level of the representational vehicle (different word form with identical meaning) and yet the task is mastered at the same time as the false belief task.
RUM received even greater embarrassment in the two-colors condition. When puppet named one of the colors of an item with two colors then the child was instructed to name the other color of the item. Again the young children tended to repeat the color named by puppet and the ability to name the other color emerged with the false belief task (see Table 1).
To complete the story I should also mention the color–name condition with items of a single color. If the puppet gave the name of the object ("It's a bird" then the child had to name its color ("It's yellow") or the other way around. This task was much easier than any of the others (see Table 1). This fact suggests that the difficulty with the other tasks is not one of monitoring truth, i.e., that children who fail the false belief task find it difficult to follow task instructions in terms of "say something true but different about the item." If that were their problem then the color–name task should be as difficult.
A Near Miss Explanation: Executive Inhibition
There is an explanation from a totally different theoretical background that would fit these data quite neatly. As we have argued (Perner & Stummer, 1998), the correlation with our say-something-different tasks and the false belief task could be explained by the fact that the say-something-different tasks require executive inhibition of initial response tendencies and we know that such executive inhibition emerges with the mastery of the false belief task (for review see Perner & Lang, in press; Perner, Stummer & Lang, 1999). The say-something-different tasks require executive inhibition on the following grounds: When one is asked a question (or asks oneself a question due to prior instruction) then an automatic answering routine is triggered which produces the first suitable answer that comes to mind. If several answers are possible, of which a particular one should be chosen on the basis of further criteria, then executive control is needed to monitor the automatic answering routine, in particular to inhibit the first predominant answer that comes to mind automatically. This is exactly the case in all our say-something-different tasks except the color–name task.