Focus on the Target's Vocabulary: Masked Label Smoothing for Machine Translation

TL;DR

This paper introduces Masked Label Smoothing (MLS), a novel technique that improves neural machine translation by better integrating label smoothing with vocabulary sharing, leading to enhanced translation quality and calibration.

Contribution

The paper proposes MLS, a new masking mechanism that addresses conflicts between label smoothing and vocabulary sharing in neural machine translation models.

Findings

MLS consistently outperforms original label smoothing across datasets.

MLS improves translation quality and model calibration.

Experiments include bilingual and multilingual translation tasks.

Abstract

Label smoothing and vocabulary sharing are two widely used techniques in neural machine translation models. However, we argue that simply applying both techniques can be conflicting and even leads to sub-optimal performance. When allocating smoothed probability, original label smoothing treats the source-side words that would never appear in the target language equally to the real target-side words, which could bias the translation model. To address this issue, we propose Masked Label Smoothing (MLS), a new mechanism that masks the soft label probability of source-side words to zero. Simple yet effective, MLS manages to better integrate label smoothing with vocabulary sharing. Our extensive experiments show that MLS consistently yields improvement over original label smoothing on different datasets, including bilingual and multilingual translation from both translation quality and model's calibration. Our code is released at https://github.com/PKUnlp-icler/MLS