Context-Sensitive Grammar and Natural Language Syntax

Abstract

A theory of grammar, such as transformational grammar, context-free grammar, categorial grammar, or any of the many descendants of these grammatical formalisms, serves at least two functions. The formal model should be rich enough to allow descriptions for the full range of data observed for natural language syntax (or at least a good approximation to the full range). Additionally, the formalism should embody a description of the nature, and hence limits, of natural language syntax. The importance of the first function is obvious: if a theory is inconsistent with the data it cannot be correct. The importance of the second function, or even what it is, may not be as clear. Certainly, any theory must explain and must ultimately aid understanding, but the meaning of these terms can be illusive when applied to a formal model for natural language. This is especially true when the topic is weak generative capacity since that context strips languages of all but their surface strings leaving them no structure other than that of a mathematical set. In this context, the facts to be explained are why natural languages produce the strings sets that they do, and not some larger, or smaller, or incomparable collection of string sets. The most obvious way that a theory can explain the phenomena of these natural language string sets is to (weakly) generate exactly these sets.