yndigo

A clear future for machine translation

Machine translation (MT) knows its limits and is targeting internet and technical texts according to Mike Dillinger, President of the Association for Machine Translation in the Americas and Adjunct Professor, Department of Psychology, San José State University.

In yesterday’s article in innovations report, Dillinger goes on to explain where Machine Translation has matured — industrial and military applications — and where it hasn’t scratched the surface — legal and literary texts. And in contrast to the worries expressed by many translators that machines will be taking their jobs, Dillinger’s assessment is that machines are alleviating humans from total drudgery. “It would be cruel even to have people translate e-mails, chats, SMS messages and random web pages.”

People new to machine translation should know that many MT systems are already far more advanced than the free internet translation tools we all mock such as babelfish, so advances in statistical machine translation by the Association for Machine Translation and developers of proprietary tools are serious and are already tackling millions of words with some success. Customized software that is fed human translations and is fine-tuned to a specific context seems to be making even further strides. And where it falls short, a human post-editing (PE) component is often added.

Despite Dillinger’s assertion that Machine Translation’s future is in these limited applications, competition in the language industry is fueling more ambitious goals in my opinion. Fear of being left behind in an era with seemingly endless breakthroughs and the prospect that the competition is adopting the new model is quite palpable. Competition inspires innovation but we need to ensure correct information is being shared both within the language industry and with translation consumers.

One of the main goals expressed by Dillinger in the article seems in fact to be education, i.e., what machine translation can and cannot achieve. “A lot of people think that “translation” is being able to tell what the author means, even if he or she has not expressed himself or herself clearly and correctly. Therefore, many have great expectations about what a translation system will be able to do. This is why they are always disappointed.” While Dillinger appears to blame source documents for MT shortcomings here, I agree with the thrust of his argument, which is the general public needs clearer information. I think those within the translation industry need to educate themselves, too.

The dream is for private industry to join the Association for Machine Translation in its goal, which is to advance machine translation technology while disseminating information about its strengths and limitations; were this to happen, more human translators would surely share Dillinger’s optimistic outlook rather than putting up a fight.