TransAlign: Machine Translation Encoders are Strong Word Aligners, Too

TL;DR

TransAlign leverages multilingual machine translation encoders to produce highly accurate word alignments, significantly improving label projection in cross-lingual transfer tasks compared to existing methods.

Contribution

This paper introduces TransAlign, a novel word aligner based on MT model encoders, demonstrating superior performance over traditional aligners and non-WA label projection methods.

Findings

TransAlign outperforms popular word aligners in accuracy.

TransAlign improves label projection quality in cross-lingual token classification.

MT-based alignments with TransAlign surpass cross-attention methods in encoder-decoder models.

Abstract

In the absence of sizable training data for most world languages and NLP tasks, translation-based strategies such as translate-test -- evaluating on noisy source language data translated from the target language -- and translate-train -- training on noisy target language data translated from the source language -- have been established as competitive approaches for cross-lingual transfer (XLT). For token classification tasks, these strategies require label projection: mapping the labels from each token in the original sentence to its counterpart(s) in the translation. To this end, it is common to leverage multilingual word aligners (WAs) derived from encoder language models such as mBERT or LaBSE. Despite obvious associations between machine translation (MT) and WA, research on extracting alignments with MT models is largely limited to exploiting cross-attention in encoder-decoder architectures, yielding poor WA results. In this work, in contrast, we propose TransAlign, a novel word aligner that utilizes the encoder of a massively multilingual MT model. We show that TransAlign not only achieves strong WA performance but substantially outperforms popular WA and state-of-the-art non-WA-based label projection methods in MT-based XLT for token classification.