Page 111 - Encyclopedia of Philosophy of Language

5. Representations
The term 'representation' has been used throughout this article without giving a precise account of what a representation is. Behaviorist psychologists have often criticized cognitive psychologists for using such hypo- thetical constructs without precision, and hence with- out explanatory power. The problem is not easy, and it is not made any easier by the need to consider how much of a subject's performance is to be ascribed to mental representations per se, and how much can be accounted for by the mental processes used to access these representations.
Occasionally psychologists have produced theories where the representations have been specified very precisely. For example, Johnson-Laird's theory of syl- logistic reasoning proposes that propositions such as All A are B or Some B are not C are encoded by subjects into a 'mental model,' and the form of en- coding for each proposition, and how the subject combines propositions, are fully specified. The theory is very successful at predicting the relative difficulty of different syllogisms.
A rather different approach to representations is to ask subjects to make judgments about some domain (perceptual, such as the similarity of rectangles of different heights and widths; or conceptual, like the similarity of various animals) and then use math- ematical and statistical techniques to infer the struc- ture of the subjects' underlying representations (for an authoritative but difficult review, see Suppes, et al. 1989: ch. 14). A general summary of this work is that perceptual judgments can often be successfully char- acterized by representations with spatial properties (with percepts corresponding to points in a space, and the dissimilarity of two percepts corresponding to distance apart of the corresponding points in the space). However, categories with verbal labels, such as animals or countries, are often better described by tree structures or (equivalently) by collections of features. Here, dissimilarity can be characterized by distance apart in a tree, where each concept cor- responds to a terminal node of the tree (end of a branch) and distance is measured by distance between the branches.
Features have the advantage of providing a richer description of the relation between two concepts: in particular, the features can be parceled into three sets (what features concept A has that concept B does not, what features concept B has that concept A does not, and what features they have in common). With such a characterization, judgmental asymmetries and the effects of context can be neatly handled. For example, it has been shown that subjects' judgments of the similarity of North Korea to China (concentrating on features that North Korea has but China does not) exceeded their judgments of the similarity of China to North Korea (concentrating on features that China has but North Korea does not). Also, when asked to
make judgments about similarity, subjects put greater emphasis on common features than when asked to make judgments of dissimilarity: this explains why one group of subjects rated the former West and East Germany as more similar than Sri Lanka and Nepal, but a different group of subjects also rated the former West and East Germany as more dissimilar than Sri Lanka and Nepal.
The implication of this work for the relation between language and thought is that concepts, at least those that can be readily labeled, seem to be characterized largely with features, and features also play an important role in linguists' characterization of language. Thus, one important common ground between thought representations and linguistic descriptions is at the feature level. Note, however, that such a statement stops well short of Whorf's claim that linguistic features determine conceptual features.
6. Inner Speech
The final area of interaction to be considered in this survey concerns the supervisory role of language in monitoring complex tasks. Vygotsky is prominent among psychologists who have suggested that talking to ourselves is an important aspect of problem solving. In the course of cognitive development overt use of speech when thinking gives way to what Vygotsky called 'inner speech,' but such activity retains all the grammatical and semantic properties of overt speech. By talking to ourselves we can bring together in work- ing memory strands of ideas which might otherwise be kept separate in different modules of our cognitive system. This is much the same function as has been suggested for consciousness itself (see Oatley 1988 for a discussion of 'Vygotskyan consciousness') though consciousness would embrace more than inner speech.
Linking speech with consciousness should suggest that speech is not always advantageous for efficient cognitive functioning. It can readily be shown that on occasion conscious processes interfere with an activity: describing what we are doing while we are tying a shoelace or riding a bicycle disrupts perform- ance. Formal demonstration of this point is provided, for example, by Hayes and Broadbent (1988). They asked subjects to interact with a computer so as to control the computer output. The output was deter- mined by one of two equations linking the subject's input and the computer's present or previous output. Hayes and Broadbent discriminated two forms of learning: S-mode (selective) learning, which is explicit and reportable; and U-mode (unselective) learning, which is implicit and not readily reportable. They were able to show that one of the equations in the computer-control task led to most subjects using S- mode learning, while the other led to U-mode learn- ing. (This is based on the amount of material that needs to be held in working memory: if the capacity of working memory is exceeded, only implicit U-mode
Thought and Language
89