Paper: Consensus Training for Consensus Decoding in Machine Translation

ACL ID D09-1147
Title Consensus Training for Consensus Decoding in Machine Translation
Venue Conference on Empirical Methods in Natural Language Processing
Session Main Conference
Year 2009
Authors

We propose a novel objective function for dis- criminatively tuning log-linear machine trans- lation models. Our objective explicitly op- timizes the BLEU score of expected n-gram counts, the same quantities that arise in forest- based consensus and minimum Bayes risk de- coding methods. Our continuous objective can be optimized using simple gradient as- cent. However, computing critical quantities in the gradient necessitates a novel dynamic program, which we also present here. As- suming BLEU as an evaluation measure, our objective function has two principle advan- tages over standard max BLEU tuning. First, it specifically optimizes model weights for downstream consensus decoding procedures. An unexpected second benefit is that it reduces overfitting, which can improve test set BLEU sco...