Lexical Disambiguation of Polysemeous Adjectives in MT: A Corpus Based Study

Ambiguous nouns and adjectives are words with more than one meaning. Having several meanings creates problems in MT particularly in automatic programs where human plays no role in choosing the equivalent target word and there is no context available to the machine. One of the ways to solve the problem is to provide the Machine with proper and enough contexts; here by context we mean the collocations of a word.. In order to show this we chose English adjectives for our study. We extracted polysemous adjectives from a bilingual English-Persian Middle Dictionary (Hezareh) and checked them all in a bilingual parallel English-Persian Corpus to get the context and collocations for each adjective. Then we recorded all the adjectives with their associate words، context and the meaning provided by the corpus. Then a concordance was created out of these adjectives and their equivalents along with their collocations and a disambiguating program was written for it so that it could choose the best equivalent for the target adjective according to the highest meaning frequency of the given adjective or its collocation. This program’s reliability was tested by using five translators. Their choices were compared through a statistical method (Cronbach alpha) and their covariance was calculated by spss software. The results show that in more than 50% of the cases translators’ choices were the same as the program’s choices. The results of this study could be helpful in MT، bilingual information retrieval، word nets and even in teaching Persian to foreigners.