Some Thoughts on Procedural Semantics
Yorick Wilks University of Essex
Does Artificial Intelligence (AI) either have, or need, a theory of meaning? It is hardly necessary to argue the need for a theory of meaning these days, whether we are considering the meaning of the utterances or sentences of a natural language like English, or of the internal representational languages of AI systems, which make use of inventories of predicates like ANIMATE, CAUSE, ACT, STATE, and even some, like TRANS, that are not identical to English words.1
The original question then divides into two subquestions: (1) Does AI have, or need, a theory of meaning for the sentences or utterances of a language like English; and (2) does AI have, or need, a theory of meaning for its internal languages containing predicates like those listed earlier? To put the last question crudely: What do ANIMATE, LOCATION, CAUSE, and so on, really mean?
One response in AI to both questions has been to appeal to external authority: the Tarskian semantics of denotations and truth conditions for some suitably augmented version of the predicate calculus ( Hayes, 1974; McDermott, 1978). This solution is directed immediately to question (2), and derivatively to (1) once the correspondence between sentences of natural language and the representational formalism has been settled. The present chapter is not the place to argue against such a view; it has been much contested in the philosophical____________________