Machine Translation: public talk by Dr. Jeff Ondich

May 4, 2010

The Carleton Linguistics Program is very pleased to announce the third colloquium of the Carleton Linguistics Colloquium Series.

Machine Translation
Dr. Jeff Ondich
Professor of Mathematics and Computer Science, Carleton College
Chief Technology Officer, Ultralingua
Friday, 7 May 2010, Leighton 304, 4:00 pm

The way people talk about translation software, it can seem as though Babelfish and Google Translate are used primarily for party games where you translate sentences to other languages and back to see what hilarity ensues. Here, for example, is Google Translate going from English to Hindi to Hebrew and back to English again:

O Carleton, our Alma Mater,
We hail the maize and blue;
Thy name is ever dearest
Thy children ever true.

Hey, Carlton, our Alma Mater,
We hail blue corn;
Your name is not cute
Your kids do not.

(Even Google wants to take our "e" away!)

To be sure, modern machine translation often produces results that are awkward, inelegant, or just plain wrong. On the other hand, even a mediocre translation can help you read the news from the perspective of Le Monde or Al Jazeera. And a professional translator of, say, business documents might be able to work more quickly from a bad first draft than from the original text. Sometimes low-quality communication is better than no communication at all.

We know that translation is hard, even for skilled human beings, so it's not surprising that machine translation is imperfect. In this talk, I will introduce you to the main techniques used by modern machine translation systems, with the goal of helping you to understand the sources of translation software's imperfections.

All are invited to attend a reception following the talk.