A Practical Approach to Multiple Default Inheritance for Unification-Based Lexicons

This paper describes a unification-based lexicon system for NLP applications that incorporates mechanisms for multiple default inheritance. Such systems are intractable in the general case---the approach adopted here places a number of restrictions on the inheritance hierarchy in order to remove some of the sources of complexity while retaining more desirable properties. Implications of the design choices are discussed, comparisons are drawn with related work in computational linguistics and AI, and illustrative examples from the lexicons of German and English are given.