WordGraph - Development of a unified graph-theoretical system for acquiring lexico-semantic phenomena

点击翻译按钮获取中文摘要

The meaning of a word crucially depends on the relationships it has with other words. The goal of the project is to develop intuitive and flexible methods that enable us to model lexical semantics using words and their relationships in diverse linguistic contexts. Our main representational formalism is graph-theoretical which is naturally suited to represent (i) words as nodes and (ii) relationships between words as edges.Many linguistic contexts are easily modeled with graphs. In parallel, vector space models are capable of modeling lexical semantics based on linguistic contexts as well. Different contexts have different properties and are therefore better modeled in one or the other of the two formalisms. We propose a dual model to explore which formalism is best used to cover which phenomena.We continue our work on automatic lexicon acquisition by exploiting lexical knowledge encoded in various linguistic contexts using the appropriate metaphor (graph or vector space model). We employ a new way of ambiguity resolution (incidence graphs) and improved scalability of our algorithms (threshold sieving). We define a new application for the graph model (semantic head identification) and will continue work on automatic lexicon acquisition.