Consensus Training for Consensus Decoding in Machine Translation

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1418–1427, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Consensus Training for Consensus Decoding in Machine Translation Adam Pauls, John DeNero and Dan Klein Computer Science Division University of California at Berkeley {adpauls,denero,klein}@cs.berkeley.edu Abstract 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 scores when using standard Viterbi decoding. 1 Introduction Increasing evidence suggests that machine trans- lation decoders should not search for a single top scoring Viterbi derivation, but should instead choose a translation that is sensitive to the model’s entire predictive distribution. Several recent con- sensus decoding methods leverage compact repre- sentations of this distribution by choosing transla- tions according to n-gram posteriors and expected counts (Tromble et al., 2008; DeNero et al., 2009; Li et al., 2009; Kumar et al., 2009). This change in decoding objective suggests a complementary change in tuning objective, to one that optimizes expected n-gram counts directly. The ubiquitous minimum error rate training (MERT) approach op- timizes Viterbi predictions, but does not explicitly boost the aggregated posterior probability of de- sirable n-grams (Och, 2003). We therefore propose an alternative objective function for parameter tuning, which we call con- sensus BLEU or CoBLEU, that is designed to maximize the expected counts of the n-grams that appear in reference translations. To maintain con- sistency across the translation pipeline, we for- mulate CoBLEU to share the functional form of BLEU used for evaluation. As a result, CoBLEU optimizes exactly the quantities that drive efficient consensus decoding techniques and precisely mir- rors the objective used for fast consensus decoding in DeNero et al. (2009). CoBLEU is a continuous and (mostly) differ- entiable function that we optimize using gradient ascent. We show that this function and its gradient are efficiently computable over packed forests of translations generated by machine translation sys- tems. The gradient includes expectations of prod- ucts of features and n-gram counts, a quantity that has not appeared in previous work. We present a new dynamic program which allows the efficient computation of these quantities over translation forests. The resulting gradient ascent procedure does not require any k-best approximations. Op- timizing over translation forests gives similar sta- bility benefits to recent work on lattice-based min- imum error rate training (Macherey et al., 2008) and large-margin training (Chiang et al., 2008). We developed CoBLEU primarily to …

Content truncated to 500 words