David Zhang wrote:
I challenge the chance of success albeit I know some people are working on it. For example, exposure, expose to, etc. may mean 暴露 in Chinese. But they mean more than that in English and sometimes it's quite a challenge for a Chinese translator to find an equivalent Chinese word. Maybe I can cite more like access, reference, ..., and so on. Also, the word sequence in a Chinese sentence sometimes is likely a big difference from an English equivalent, which may constitute a nightmare for a researcher.
David, for MT about Chinese, see my post in ProZ forums at http://www.proz.com/forum/machine_translation_mt/98-machine_translation_anyone_using_it_successfully-page3.html#1271702
for your example of (exposure, expose to), there are several Rule-based MT systems which allow to easily enter in variant translation equivalents in the dictionary. I have done this for "continuer à" and "continuer de" which both translate the same into English. Also have done it for different words which translate from 1 in source to 2 in target (issue in English can be 1) problème or 2) numéro de publication) in French. Handling this type of ambiguity both at the source language level and at the target language has been implemented in such systems for many years, but only in the paid software and systems, not in the free non-customizable online systems which I mentioned in the panel.
I specifially described these built in features in software product reviews of Reverso Pro 5 and PROMT v6 back in 2001 and in 2004. See the Language Technology Evaluation (LangTechEval) site in my ProZ profile and look up my name for the software reviews on those products (links are provided to those freely available software reviews).
The customizable statistical MT systems also provide for the possibility to train them based on the existing variants found in legacy translated materials, such as in TMs.