Wordnet Semantics From Dictionaries : Semi-automatic Extraction of Semantic Relations Between Verbs From a Dictionary

Abstract

The motivation for this master thesis originates from Lars Nygaard s cand. philol. thesis from 2006 (Nygaard, 2006), from which a prototype for a Norwegian wordnet was generated by analyzing definitions for nouns in Bokmålsordboka (BOB)1. BOB is a dictionary for Norwegian Bokmål which is available both in book format and as an on-line resource. This thesis will propose a set of components for a method that is in some respects an extension of Nygaard s method, in other respects a different approach altogether. An investigation will be done for every stage of the process, including the extraction and analysis of the dictionary data, the generation of a semantic network and the evaluation phase of such a semantic network. The proposed method will be tested on a dictionary for Norwegian Bokmål, and a thorough annotation study will be presented in order to clarify the terms for evaluating a semi-automatically generated semantic network. The main focus of study is on the task of generating wordnet data from verb definitions. This is an area not covered by Nygaard s method, and is generally a subject of study not as frequently covered as nouns. Verbs exhibit different behavior than nouns in many respects, therefore they should present a somewhat different set of challenges. The goal is to investigate to which extent a method for the described purpose can be automated, and to which extent it can be said to be general. The observations made throughout the thesis will be analyzed in the attempt to gain insights into lexical semantics as well as verbs, both with regard to semantic properties and the challenges one encounters when modeling verbs in a wordnet.