Automatic Extraction of Turkish Hypernym-Hyponym Pairs From Large Corpus

In this paper, we propose a fully automatic system for acquisition of hypernym/hyponymy relations from large corpus in Turkish Language. The method relies on both lexico-syntactic pattern and semantic similarity. Once the model has extracted the seeds by using patterns, it applies similarity based expansion in order to increase recall. For the expansion, several scoring functions within a bootstrapping algorithm are applied and compared. We show that a model based on a particular lexico-syntactic pattern for Turkish Language can successfully retrieve many hypernym/hyponym relations with high precision. We further demonstrate that the model can statistically expand the hyponym list to go beyond the limitations of lexico-syntactic patterns and get better recall. During the expansion phase, the hypernym/hyponym pairs are automatically and incrementally extracted depending on their statistics by employing various association measures and graph-based scoring. In brief, the fully automatic model mines only a large corpus and produces is-a relations with promising precision and recall. To achieve this goal, several methods and approaches were designed, implemented, compared and evaluated.