Output details
11 - Computer Science and Informatics
University of Edinburgh
A Discriminative Latent Variable Model for Statistical Machine Translation
<22> Originality: This paper introduces the idea of maximum translation decoding (accounting for competing translations) and shows how statistical machine translation can be made to scale to millions of features. It also shows for the first time how regularisation methods can improve translation.
Significance: Almost all machine translation research uses heuristic methods that are hard to improve. Our approach is based upon a clean machine learning approach which yields a clear agenda for future developments.
Rigour: Uses optimisation on a training set and evaluation on a test set. Paper appeared in the top NLP venue (acceptance rate 25%).