A RESOLUÇÃO DA AMBIGÜIDADE LEXICAL SEM APOIO DO CONHECIMENTO DE MUNDO

Abstract

One of the most serious problems found in machine translation, which may have interesting implications for language teaching, is the resolution of lexical ambiguity when translating text into another language. One of the most frequent resources, used in this case, has been the use of encyclopedic knowledge. The attempts made so far to endow the computer with a representation of world knowledge, however, have resulted in combinatorial explosions, which turn any system unfeasible. This study investigates word sense ambiguity through selective restrictions, using English as the source language and portuguese as the target language. Initially 10 ambiguous pairs were selected from the source language and analyzed against an English language corpus made up of 10,000,000 words of expository text. Six examples for each pair were randomly picked from the corpus, using a concordancer, and given to bilingual subjects, who should tell which words surrounding the ambiguous pair could be used to disambiguate it. The results showed that every ambiguous pair was preceded or followed by words which had some semantic features that disambiguated the pair. The conclusion is that selective restrictions replace with many advantages the use of encyclopedic knowledge to solve lexical ambiguities.