We propose a dynamic lexical model based on the Generative Lexicon 1. This model is composed of a lexical database and a rule system. We first built the lexicon editor which allowed users to enter lexical items according to the GL feature structure format. Argument structure and qualia structure in each lexicon are automatically set by the categories of nominal types and event types. We built an algorithm to generate complex words or phrases from limited entries in this editor. Modifiers are added to a headed word to form complex word structures by tiered rule applications. In our model, multi-senses of a word are generated form a single lexical entry, and, when composing a sentence, the most appropriate sense is determined by type constraints which co-occurring words impose on the word. In an evaluation experiment, our dynamic lexical model was also proven to be suitable for resolving ambiguity.
展开▼