IBM Research has an incredible translation software product dubbed MASTOR (The Multilingual Automatic Speech-to-Speech Translator).  The service has seen rave reviews for both accuracy and ease of use.  At present, it works in both Arabic and Mandarin but the translation engine is very interesting.  Rather than attempting a word by word translation, the engine translates phrases and concepts, known as multi-task and unconstrained dialog translation.  This helps get it past the translation vs. interpreting barrier imposed on all other machine translation.  In brief tests it seems to get the idea across accurately.  The Business 2.0 reporter tested it, saying:

“I ate some bad chicken and my stomach feels bad”

It translated the second “bad” as “that,” but otherwise got it right. Perhaps I was mumbling so I started again, and this time it went without a hitch

Tools like Dragon’s NaturallySpeaking can’t even get the first language right with great accuracy, much less put the words in another language.  This is in spite of the fact that MASTOR requires no tedious and time-consuming training, reading line after line of canned text to help the software learn your voice.  I’m impressed that IBM has been able to pull this off with any sort of accuracy.

As a final note, this project was funded by the Defense Advanced Research Projects Agency (DARPA) CAST program so I’m not sure what if any future windfall from products sales and licensing would belong to IBM.  C-net’s article on MASTOR is entirely focused on the commercial side of the research, but I’m not sure they considered the funding source.

Tags:No Tags