Abstract

Machine readable dictionaries constitute an important source of information in many fields of Natural Language Processing. It has been argued that the richness and quality of dictionaries greatly impacts the performance of the tasks in which they are employed. In recent years, monolingual approaches to the disambiguation of dictionary definitions have been investigated. However, little attention has been paid to the disambiguation of bilingual dictionaries.

In this talk, we present a novel algorithm for the disambiguation of monolingual and bilingual dictionary glosses. The dictionary is represented as a graph and cyclic patterns are sought in the graph to assign the most appropriate sense to each gloss word. We first report on the algorithm’s performance on monolingual dictionaries and then focus on a two-fold experimental evaluation of the approach on the Ragazzini-Biagi English-Italian dictionary: first, we show that disambiguation performance achieves very high accuracy, thus providing a high-quality sense assignment for most ambiguous translations in the dictionary; second, we will see how the algorithm’s output can be used to enhance the dictionary. Different automatic suggestions can be produced which aim to reduce the degree of lexical and semantic inconsistency within the dictionary entries, thus improving their quality.