Cognitive artificial intelligence is the branch of computer science concerned with the modelling of human cognition. Cognitive artificial intelligence is still in the process of developing paradigms for research, yet there has been a great deal of agreement among workers in the field regarding the question of which structures are useful in the writing of cognitive simulations. In particular, a number of research programs in cognitive artificial intelhgence are founded upon the use of one or another kind of conceptual hierarchy or graph. In this essay, I would hke to criticize the use of conceptual graphs in artificial intelligence models of mind, and to describe some alternative structures that would improve computational models of mind. My strategy will be to present my argument in the context of analyzing the concept of analogy, primarily because analogy has been one of the central notions studied in cognitive artificial intelligence. In the first third of the essay, I show how the use of conceptual graphs has made artificial intelligence accounts of analogy inadequate and unintuitive. In the second third of the paper, I present my own theory of analogy. In the final third of the paper, I show how several notions from situation semantics may figure in the representation structures needed to implement the account of analogy presented in part two of this paper.
I. CONCEPTUAL GRAPH THEORIES OF ANALOGY
The two most extensive computational treatments of analogy, Patrick Winston's and Roger Schank's, have been founded on the notion of matching. 1 Matching, in artificial intelligence programming, usually involves a datum and a template. The datum, for example, could be a sentence, e.g., 'The little kitten was lost.' The template, then, might be 'The ~ was lost.' Typically, a program will, after matching a datum to a template, subsequently associate the blanks in the template with the portions of the datum which