Identifying Useful Human Correction Feedback from an On-line Machine Translation Service
Omega-S208 Campus Nord - UPC
Mon Apr 22, 2013
from 14:30h to 15:00h
Post-editing feedback provided by users of online translation services offers a good opportunity for the improvement of statistical machine translation systems. However, feedback provided by casual users is noisy, and must be filtered in order to identify potentially useful instances. In this talk I present our recent study on automatic feedback filtering from a real weblog. I will discuss on the features we have considered to decide whether a user's translation proposal is indeed better than the one automatically generated. Regardless of the inherent difficulty of the task (even for humans!), I will show how the selected instances allow to improve the performance of a phrase-based translation system.