Improving Anaphora Resolution in Neural Machine Translation Using Curriculum Learning

Dario Stojanovski, Alexander Fraser

MT Summit 2019

Modeling anaphora resolution is critical for proper pronoun
translation in neural machine translation. Recently it has been
addressed by context-aware models with varying success. In this work,
we propose a carefully designed training curriculum that facilitates
better anaphora resolution in context-aware NMT. As a baseline, we
train context-aware models as was done in previous work. We leverage
oracle information specific to anaphora resolution during training.
Following the intuition behind curriculum learning, we are able to
train context-aware models which are improved with respect to
coreference resolution, even though both the baseline and the improved
system have access to exactly the same information at test time. We
test our approach using two pronoun-specific evaluation metrics for
MT.