Statistical Machine Translation with an Eye on Linguistics


Abstract
 
Machine translation is more relevant than ever, especially in a global economy with many different languages. For instance, the European Union has 23 official languages. What will happen to languages such as Dutch? Will it survive as a language of commerce, or will it be abandoned in favour of English? By lowering translation costs, we would hope to sustain the viability of different languages in a globalized world.

Statistical machine translation holds the promise of instant machine translation. Given open source tools such the Moses decoder, just add a parallel corpus and you have a machine translation system.

This talk will present some problems where the standard phrase-based approach fails, and where attention to the specifics of the languages involved is required. I will present methods that deal with different word order, morphology and agglutinative compounding.



Back to Schedule