Our final section contains chapters that address broad topics of theoretical import. Each outlines a major issue that carries considerable implications for all natural-language processing research. As such, the issues discussed by these last four authors point to areas of research that should expand and strengthen our current strategies for natural-language processing.
In the first chapter Jaime Carbonell discusses the problem of metaphor, and argues that this area deserves more attention from the AI community. It is certainly the case that metaphor has come to be a popular topic among linguists and psychologists in recent years, and yet very few AI researchers have chosen to concentrate on it. Carbonell maintains that a robust representational system must come to grips with the problem of metaphoric expressions very early on; it will not pay to factor this considerable portion of language use out of our initial target domains. Metaphor is pervasive in natural-anguage use, but thus far, strategies of sentence analysis have been largely insensitive to the challenges of metaphor recognition. Many researchers have dismissed it as a "special case" of semantic processing, but Carbonell suggests that it is very central to all semanticprocessing techniques.
The second chapter, by Eugene Charniak, focuses on context as a key to memory indices. It has long been ac