Why MT?

Why work on Machine Translation?

A potential applicant to the Linguistics graduate program wrote me the following:
>   Originally I thought that machine translation should be a very useful and 
>   productive subject, because it facilitates the cross-language human 
>   communication. After reading some papers, I'm convinced that it's really 
>   hard to make a breakthrough in this field, especially in literature 
>   translation, because translation is not primarily a linguistic operation, it 
>   draws on philosophy, psychology, social culture knowledge, etc. It seems to 
>   me that the research on machine translation is doomed. I am surprised to 
>   know that some people, including you are still working on this field. I am 
>   curious to know what is your expectation from the research.

Here's what I replied:

Translation of literature is indeed likely to be a very difficult
problem.  However, translation of literal text, such as news wire, has
made significant progress in the last 15-20 years thanks to the
introduction of large scale statistical methods.  Briefly, text that
has been translated by people is an enormous natural resource.  To
consider an analogy, one might have imagined that just creating a
bilingual dictionary would have some of the same problems you mention,
and yet large bilingual dictionaries of fairly high reliability can be
extracted by looking statistically at how words are used when human
beings have created translations from one language into another.  More
to the point, I consider myself at least as much a scientist as an
engineer: for me, the goal is understand how language works, not just
to build things.  I believe we can learn a great deal about language,
and about ourselves as human beings, from working on translation --
even if the problem is not completely solved.

Or perhaps I'm just an optimist. :-)