This paper presents a new method for definition extraction from Slovene domain-specific corpora, based on a model for definition classification learned using machine-learning methods on examples from Slovene Wikipedia. In the first step we extract definition candidates using a Slovene semantic lexicon, automatic terminology recognition and lexico-syntactic patterns. Next, we use the learned classification model to select ŽtrueŽ definitions from the set of definition candidates. The method was tested on a natural science domain corpus from which we extracted more than a thousand definition candidates and achieved up to 70% classification accuracy with the learned classification model.
|
