Generalized Example-Based Machine Translation

[This page is under construction -- please check back.]

8. Statistical Dictionaries

A large bilingual dictionary such as the alignment algorithm requires is often hard to obtain for a particular language pair. Fortunately, it is possible to generate one with sufficient accuracy from the translation examples which are required anyway. The method is a simple count-and-filter approach, optionally followed by one or more refinement passes which use the same pruned correspondence table as in the word-level alignment.

[[Work Area]]

For full details, see the papers "Automated Dictionary Extraction" and " Automatically-Extracted Thesauri for Cross-Language IR".

Next: Applications

[LTI Home Page] [EBMT Main Page] [Generalization] [Results] [Alignment]
(Last updated 04-Aug-99)