Preparing content for machine translation with spelling/grammar tools

Last week I wrote a post on integrating grammar tools in the MT post-editing process. The post generated some interesting comments. One of the comments was made by Dr. Frans Wijma of Shufra, a technical documentation company in Singapore.

Grammar-checking on the output of MT will only expose flaws in the MT software. It makes much more sense to check (and edit) the input to MT so the MT results are more accurate. We do this using Simplified Technical English, and it dramatically improves the quality and accuracy of both the source and any translations.

This is a good idea: prepping content that you are about to feed into machine translation with spelling and grammar tools. For it is a well known fact that entering text with spelling mistakes can be disastrous to the outcome of the machine translation; the MT does not have any way of knowing that it is a mistake and usually just translates the word as it sees it. There are also some types of sentences that MT does not process well; long sentences should be shortened into a few sentences for example. Sentences with ambiguous meanings tend to come out poorly in MT. Running the text through grammar checking tools can help make the text more palatable to the MT. Misplaced or omitted punctuation can also adversely affect the MT quality.

Having said this, however, we should realize that the job of a translator is to accurately convert the thoughts of the writer in another language. The job of a translator is not to second-guess the writer and change the meaning of the words. That would be betray our jobs are translators. So the prepping work should be limited to fixing spelling mistakes and crude grammar and punctuation errors.

The MT vendors are working on new technology that would correct input errors automatically. This includes natural language disambiguation, context sensitive spelling error correction and other neat stuff. In the meantime, use of available spell and grammar checkers can help improve MT quality.