A statistical approach on Persian word sense disambiguation

This article studies different aspect of a new approach for resolving lexical ambiguities using statistical information gained from a monolingual corpus. The proposed approach resolves the problem of target word selection in an machine translation system. This Method is an unsupervised graph-based approach which uses a bilingual dictionary to find all possible translations of each ambiguous word in the source sentence (English) and then chooses the most appropriate alternative regarding the statistical information gathered from target language (Persian) corpora. Also, two new methods to measure the semantic similarity based on source and target language corpora are introduced. The experiments show that the unsupervised graph-based WSD which uses the proposed semantic similarity measures in the dependency graph outperforms all other methods on WSD for translating English to Persian words, significantly.