Parameterized type expansion in the feature structure formalism TDL

Abstract

Over the last few years, unification-based grammar formalisms have become the predominant paradigm in natural language processing systems because of their monotonicity, declarativeness, and reversibility. From the viewpoint of computer science, typed feature structures can be seen as data structures that allow representation of linguistic knowledge in a uniform fashion. Type expansion is an operation that makes the constraints on a typed feature structure explicit and determines their satisfiability. We describe an efficient expansion algorithm that takes care of recursive type definitions and allows exploration of different expansion strategies through the use of control knowledge. This knowledge is specified in a separate layer, independently of grammatical information. Memoization of the type expansion function drastically reduces the number of unifications. In the second part, nonmonotonic extensions to TDL and the implementation of well-typedness checks are presented. Both are closely related to the type expansion algorithm. The algorithms have been implemented in Common Lisp and are integrated parts of TDL and a large natural language dialog system.