Approach to Recognition and Understanding of the Time Constituents in the Spoken Chinese Language Translation

Abstract

In the spoken Chinese language, the time constituents occur frequently, especially in the domain of appointment schedule, ticket booking and hotel reservation etc. However, in the current Chinese-to-English Machine Translation (MT) systems, it is still a problem to deal with the time constituents. According to our test results of some commercial Chinese-to- English MT systems, about 57.1% Chinese time constituents are wrongly translated, and 58.3% of the errors are caused by false recognition and misunderstanding of the time constituents. In the paper, we present a new approach to recognition and understanding of the time constituents in the Chinese language, which is integrated by a shallow level analyzer and a deep level analyzer. The shallow level analysis is realized by a Finite State Transition Network (FSTN). The time constituents are first recognized by the FSTN, and the results are divided into three types. Two types of the results are determinate and one is uncertain. Aimed at the uncertain results, the deep level analyzer checks the semantic context of time constituents, performs necessary phrase structure analysis, and finally decides the type of the time constituents. The approach has been employed in our Chinese-to-English spoken language translation system, which is limited in the domain of hotel reservation. However, the approach is domain-independent. The preliminary experiment has proven that the approach is effective and practical.

The research work described in this paper is supported by the National 863 Hi-Tech Program of China under Grant 863-306-ZT03-02-2, the National Key Fundamental Research Program (the 973 Program) of China under Grant G1998030504 and also the National Natural Science Foundation of China under Grant 69835030.