On the Off-Target Problem of Zero-Shot Multilingual Neural Machine Translation
This addresses a critical issue in multilingual translation systems, particularly for zero-shot tasks, with incremental improvements in performance.
The paper tackles the off-target problem in zero-shot multilingual neural machine translation, where translations are in the wrong language, by proposing Language Aware Vocabulary Sharing (LAVS) to increase lexical distance between languages, reducing the off-target rate from 29% to 8% and improving BLEU scores by an average of 1.9 points.
While multilingual neural machine translation has achieved great success, it suffers from the off-target issue, where the translation is in the wrong language. This problem is more pronounced on zero-shot translation tasks. In this work, we find that failing in encoding discriminative target language signal will lead to off-target and a closer lexical distance (i.e., KL-divergence) between two languages' vocabularies is related with a higher off-target rate. We also find that solely isolating the vocab of different languages in the decoder can alleviate the problem. Motivated by the findings, we propose Language Aware Vocabulary Sharing (LAVS), a simple and effective algorithm to construct the multilingual vocabulary, that greatly alleviates the off-target problem of the translation model by increasing the KL-divergence between languages. We conduct experiments on a multilingual machine translation benchmark in 11 languages. Experiments show that the off-target rate for 90 translation tasks is reduced from 29\% to 8\%, while the overall BLEU score is improved by an average of 1.9 points without extra training cost or sacrificing the supervised directions' performance. We release the code at https://github.com/PKUnlp-icler/Off-Target-MNMT for reproduction.