Search-Aware Tuning for Machine Translation

Parameter tuning is an important problem in statistical machine translation, but surprisingly, most existing methods such as MERT, MIRA and PRO are agnostic about search, while search errors could severely degrade translation quality. We propose a searchaware framework to promote promising partial translations, preventing them from being pruned. To do so we develop two metrics to evaluate partial derivations. Our technique can be applied to all of the three above-mentioned tuning methods, and extensive experiments on Chinese-to-English and English-to-Chinese translation show up to +2.6 BLEU gains over search-agnostic baselines.