Machine Translation and the treatment of (semi-)fixed expressions

Abstract:

Machine translation (MT) faces various challenges, one of them being the treatment of fixed or semi-fixed expressions, such as idioms ("he kicked the bucket") and collocations ("entertain-hope", "practice-profession", "school of fish"). In this talk, I will present the Its-2 translation prototype (English - French) developed at LATL and show how idioms and collocations are handled in this system. I will suggest that such expressions are best identified on the basis of the parse tree rather than on the basis of the tokenized surface string as is current practice in commercial MT systems.

Bio:

Eric Wehrli is professor of linguistics and computer science at the University of Geneva. He is currently visiting the Language Technology group at Macquarie University.

Enquiries: sals@mri.mq.edu.au