Empirical acquisition of conceptual distinctions via dictionary definitions

This thesis discusses the automatic acquisition of conceptual distinctions using empirical methods, with an emphasis on semantic relations. The goal is to improve semantic lexicons for computational linguistics, but the work can be applied to general-purpose knowledge bases as well. The approach is to analyze dictionary definitions to extract the distinguishing information (i.e., differentia) for concepts relative to their sibling concepts. A two-step process is employed to decouple the definition parsing from the disambiguation of the syntactic relations into the underlying semantic ones. Previous approaches tend to combine these steps through pattern matching geared to particular types of relations. In contrast, here a broad-coverage parser is first used to determine the syntactic relationships, and then statistical classification techniques are used to disambiguate the relationships into their underlying semantics. There are several contributions of this thesis. First, it introduces an empirical methodology for the extraction and disambiguation of semantic relations from dictionary definitions. Second, it introduces a statistical representation for these semantic relations using Bayesian networks, which are popular in artificial intelligence for representing probabilistic dependencies. Third, it shows how improvements in word-sense disambiguation can be achieved by augmenting a standard statistical classifier approach with a probabilistic spreading-activation system using the semantic information extracted using this process.