Steven.Minton@CMU-CS-A.ARPA (04/11/84)
The following references might prove helpful if you're interested in AI and metaphor comprehension: Carbonell, J.G. and Minton, S. "Metaphor and Common-Sense Reasoning" CMU tech report CMU-CS-83-110, March 83 Carbonell, J.G. "Metaphor: An Inescapable Phenomenon in Natural Language Processing", in Strategies for Natural Language Processing, W. Lehnert and M. Ringle (eds.), Erlbaum 1982 Carbonell, J.G. "Invariance Heirarchies in Metaphor Interpretation" Proceedings of the 3rd Meeting of the Cognitive Science Society, 1981 There's a large body of literature on analogical reasoning and other aspects of metaphor comprehension. Much of the relevant research has been done within psychology and linguistics. I'd suggest looking at these for an overview: Ortony, A. (Ed.) "Metaphor and Thought" Cambridge Univ. Press 1979 Lakoff, G. and Johnson, M. "Metaphors We Live By" Chicago Univ. Press 1980 Gentner D. "Structure-Mapping: A Theoretical Framework for Analogy" in Cognitive Science, Vol. 7, No.2 1983 Winston P. "Learning by Creating and Justifying Transfer Frames" in Arificial Intelligence, Vol. 10, No. 2, 1978 I don't know of any natural language system which can handle a wide range of novel metaphors, and I don't expect to see one soon. Any such system would have to contain an enormous amount of knowledge. Unlike most present-day NL systems, a robust metaphor comprehension system would have to be able to understand many different domains. In spite of this difficulty, metaphor comprehension remains a fertile area for AI research. I've spent some time examining how people understand sentences like: "The US/Russian arms negations are a high stakes poker game". When you get right down to it, its amazing that people can figure out exactly what the mapping between "arms negotiatations" and "poker games" is. What's most amazing is that using and understanding metaphors APPEARS to take so little effort. (In fact, they are often the easiest way to to rapidly communicate complex technical information. The next time you are at a talk, try counting the analogies and metaphors used.) -- Steve Minton, CMU