ABSTRACT
This chapter concerns the general problem of characterizing structured knowledge in terms of rules. It introduces the notion of a rule set and identifies the major criteria involved in evaluating different rule sets. The chapter discusses some alternative ways of characterizing structured knowledge, and evaluates each with respect to power, parsimony, and kind of performance assumptions necessary for psychological viability. It also discusses some relationships between present goals and those of linguistics. The goal of any rule-based theory of knowledge is to provide an account of a given class of input-output pairs. The fundamental task of linguistic research is to propose and evaluate theories of linguistic competence, just as ours are to evaluate theories of competence generally. In generative linguistics, the theorist is given a finite corpus, say, of sentences-and is required to come up with a characterization which accounts for these and other sentences of the language.
