Tense, aspect, and modality are known to present difficult problems in machine translation. In traditional approaches, tense, aspect, and modality have been translated by using manually constructed heuristic rules. Recently, however, such corpus-based approach as the k-nearest neighborhood method have also been applied. This paper is a report on experiments we carried out on the application of a variety of machine-learning methods, including the k-nearest neighborhood, to the translation of tense, aspect, and modality. One experimental result was that support vector machine obtained the highest precisions of the methods we applied. In the previous work, applying the k-nearest neighborhood method, only those strings at the ends of sentences were used for the translation of tense, aspect, and modality. In contrast, our method used all morphemes of the whole sentences as information and the support vector machine thus obtained a higher precision than it did by using the ends of sentences. We therefore found that using all of the morphemes of a whole sentence is effective in the translation of tense, aspect, and modality.
展开▼
