Metaphor Interpretation as Selective Inferencing

Metaphor pervades natural language discourse. This paper describes a computational approach to the interpretation of metaphors. It is based on a natural language processing system that uses the discourse problems posed by a text to select the relevant inferences. The problem of interpreting metaphors can then be translated into the problem of selecting the relevant inferences to draw from the metaphorical expression. Thus, a metaphor is frequently given a correct interpretation as a by-product of the other things a natural language system has to do. Two examples of metaphors are given -- a spatial metaphor schema from computer science, and a novel metaphor -- and it is shown how the interpretation problem for each can be translated into a selective Inferencing problem and solved by the ordinary operations of the system. This framework sheds light on the analogical processes that underlie metaphor and begins to explain the power of metaphor.