Solving the Polysemy Problem of Persian Words Using Mutual Information Statistics

In recent years, large monolingual, comparable and parallel corpora have played a very crucial role in solving various problems of computational linguistics including machine translation, information retrieval, natural language processing, and the like. This paper tries to solve the problem of polysemy of Persian words while translating them into Persian by the computer. We use Mutual Information statistics obtained from a very large monolingual corpus of Persian. Mutual information values are calculated based on co-occurrence frequencies of words and used to measure the correlation between words. Using mutual information statistics the occurrence or co-occurrence frequencies of different equivalents of an ambiguous word in the target language is calculated and the most probable equivalent for every ambiguous word is selected. When mutual information value is high, the word associations are strong and provide dependable results for translational disambiguation and vice versa. The method discussed in this paper not only can be directly applied in the system of Persian-English machine translation, but also it can certainly increase performance effectiveness of the retrieval tasks, especially in cross-language information retrieval.