Leveraging Neural Machine Translation for Word Alignment

TL;DR

This paper explores extracting word alignments from neural machine translation models by comparing attention-based methods and probability-based scores, and proposes a combined approach using a feed-forward network for improved accuracy.

Contribution

It introduces a novel method that aggregates multiple sources of alignment scores from NMT models to enhance word-alignment quality.

Findings

Combined alignment extractors outperform individual methods.

Probability-based scores can effectively infer word alignments.

Aggregation improves alignment accuracy over single-source methods.

Abstract

The most common tools for word-alignment rely on a large amount of parallel sentences, which are then usually processed according to one of the IBM model algorithms. The training data is, however, the same as for machine translation (MT) systems, especially for neural MT (NMT), which itself is able to produce word-alignments using the trained attention heads. This is convenient because word-alignment is theoretically a viable byproduct of any attention-based NMT, which is also able to provide decoder scores for a translated sentence pair. We summarize different approaches on how word-alignment can be extracted from alignment scores and then explore ways in which scores can be extracted from NMT, focusing on inferring the word-alignment scores based on output sentence and token probabilities. We compare this to the extraction of alignment scores from attention. We conclude with aggregating all of the sources of alignment scores into a simple feed-forward network which achieves the best results when combined alignment extractors are used.