Towards a comprehensive framework for the specification, representation and unification of complex feature structures : the lexical objects theory

Abstract

This paper describes the main characteristics of the Lexical Objects Theory: a framework that comprehends the formal, linguistic and computational aspects involved in the study of unification and features structures in Computational Linguistics. From a functional point of view, our proposal is based on a layered architecture which distinguish between three main levels. Specification level introduces new expressive tools: optionality and special values (null, complete and incomplete), and defines the inference rules that obtain the canonical form of any lexical object. Representation level permits three models: InRange, DataDefined and EndLess, and concentrates on computational techniques aimed to improve the management of lexical objects. Finally, Unification Level defines formally Low Level Logic-based,Weak and Constructive Unification strategies.