Clustering Lexical Patterns Obtained from a Text Corpus

A system tbr lexical acquisition is presented where word meanings are represented by clusters of phrase patterns obtained from analysis of a text corpus. A sample of cases, in the form of a concordance of phrases in which a particular word occurs in the text, is used Ibr the basic analysis. Clustering techniques are used to group together cases having similar grammar and/or meaning. This view is that words obtain their meaning from the category describing this clustering of cases. This category is theorybm~ed in that it contains a model to represent the word meaning at an abstract level, whereas the cases provide empirical evidence which confirm or disprove the model. A complex category evolves as more cases are encountered. Each new case matches to an existing categoD’, or m~’ dynamically alter existing categories as needed to account for the new case. An experimental system is presented which includes syntactic and semantic analysis of phrases obtained from text. It uses a hand-built lexicon and grammar to bootstrap a learning process. The ability to dynamically alter category structure through interpretation of new cases is shown as a way to build lexical structure semi-automatically.