Accurate Word Alignment Induction from Neural Machine Translation
This addresses the issue of word alignment accuracy for machine translation researchers, offering incremental improvements over existing methods.
The paper tackles the problem that Transformer attention weights capture poor word alignments in neural machine translation, showing they can be accurate and proposing two methods, Shift-Att and Shift-AET, which outperform neural baselines and GIZA++ by 1.4-4.8 AER points.
Despite its original goal to jointly learn to align and translate, prior researches suggest that Transformer captures poor word alignments through its attention mechanism. In this paper, we show that attention weights DO capture accurate word alignments and propose two novel word alignment induction methods Shift-Att and Shift-AET. The main idea is to induce alignments at the step when the to-be-aligned target token is the decoder input rather than the decoder output as in previous work. Shift-Att is an interpretation method that induces alignments from the attention weights of Transformer and does not require parameter update or architecture change. Shift-AET extracts alignments from an additional alignment module which is tightly integrated into Transformer and trained in isolation with supervision from symmetrized Shift-Att alignments. Experiments on three publicly available datasets demonstrate that both methods perform better than their corresponding neural baselines and Shift-AET significantly outperforms GIZA++ by 1.4-4.8 AER points.