Martin Kay - Linguistics, Mathematics, Computer Science and Machine Translation

In this talk, I will attempt a more comprehensive assessment of current approaches to machine translation than is frequently attempted. I will try to develop a more general picture of the overall architecture of existing and possible statistical systems which, if it is successful, would make them more flexible and capable of development in a variety of new ways. Linguistics will play a major roll in systems that are statistical-expert hybrids and which are bound to attract more attention as pressure grows to strive for results of higher quality. Mathematics will help give a picture of the combinatorial problems that some such approaches might encounter. Computer Science provides ways of making at least some of these problems more tractable.