Extraction of Semantic Word Relations in Turkish from Dictionary Definitions

Abstract

Many recent studies have been dedicated to the extraction of semantic connections between words. Using such information at semantic level is likely to improve the performance of Natural Language Processing (NLP) systems, such as text categorization, question answering, information extraction, etc. The scarcity of such resources in Turkish, obstructs new improvements. There are many examples of semantic networks for English and other widely-used languages to lead the way for studies in Turkish. In this study, developing a semantic network for Turkish is aimed by using structural and string patterns in a dictionary. The results are promising, so that approximately two relations can be extracted from 3 definitions. The overall accuracy is 86% if we consider the correct sense assignment, 94 % without considering word sense disambiguation. 1