On the Shortcut Learning in Multilingual Neural Machine Translation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Wenxuan, Jiao, Wenxiang, Huang, Jen-tse, Tu, Zhaopeng, Lyu, Michael R.
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910700724551680
author Wang, Wenxuan
Jiao, Wenxiang
Huang, Jen-tse
Tu, Zhaopeng
Lyu, Michael R.
author_facet Wang, Wenxuan
Jiao, Wenxiang
Huang, Jen-tse
Tu, Zhaopeng
Lyu, Michael R.
contents In this study, we revisit the commonly-cited off-target issue in multilingual neural machine translation (MNMT). By carefully designing experiments on different MNMT scenarios and models, we attribute the off-target issue to the overfitting of the shortcuts of (non-centric, centric) language mappings. Specifically, the learned shortcuts biases MNMT to mistakenly translate non-centric languages into the centric language instead of the expected non-centric language for zero-shot translation. Analyses on learning dynamics show that the shortcut learning generally occurs in the later stage of model training, and multilingual pretraining accelerates and aggravates the shortcut learning. Based on these observations, we propose a simple and effective training strategy to eliminate the shortcuts in MNMT models by leveraging the forgetting nature of model training. The only difference from the standard training is that we remove the training instances that may induce the shortcut learning in the later stage of model training. Without introducing any additional data and computational costs, our approach can consistently and significantly improve the zero-shot translation performance by alleviating the shortcut learning for different MNMT models and benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2411_10581
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On the Shortcut Learning in Multilingual Neural Machine Translation
Wang, Wenxuan
Jiao, Wenxiang
Huang, Jen-tse
Tu, Zhaopeng
Lyu, Michael R.
Computation and Language
Artificial Intelligence
In this study, we revisit the commonly-cited off-target issue in multilingual neural machine translation (MNMT). By carefully designing experiments on different MNMT scenarios and models, we attribute the off-target issue to the overfitting of the shortcuts of (non-centric, centric) language mappings. Specifically, the learned shortcuts biases MNMT to mistakenly translate non-centric languages into the centric language instead of the expected non-centric language for zero-shot translation. Analyses on learning dynamics show that the shortcut learning generally occurs in the later stage of model training, and multilingual pretraining accelerates and aggravates the shortcut learning. Based on these observations, we propose a simple and effective training strategy to eliminate the shortcuts in MNMT models by leveraging the forgetting nature of model training. The only difference from the standard training is that we remove the training instances that may induce the shortcut learning in the later stage of model training. Without introducing any additional data and computational costs, our approach can consistently and significantly improve the zero-shot translation performance by alleviating the shortcut learning for different MNMT models and benchmarks.
title On the Shortcut Learning in Multilingual Neural Machine Translation
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2411.10581