Understanding and Improving Lexical Choice in Non-Autoregressive Translation
This work addresses the problem of lexical choice errors in non-autoregressive translation models, specifically for low-frequency words, which is an incremental improvement for researchers and practitioners in machine translation.
This paper investigates lexical choice errors in non-autoregressive translation (NAT) models, finding that knowledge distillation propagates errors on low-frequency words from teacher models. By exposing raw data to NAT models and introducing a Kullback-Leibler divergence term, the authors achieved state-of-the-art NAT performance, reaching 27.8 BLEU on WMT14 English-German and 33.8 BLEU on WMT16 Romanian-English.
Knowledge distillation (KD) is essential for training non-autoregressive translation (NAT) models by reducing the complexity of the raw data with an autoregressive teacher model. In this study, we empirically show that as a side effect of this training, the lexical choice errors on low-frequency words are propagated to the NAT model from the teacher model. To alleviate this problem, we propose to expose the raw data to NAT models to restore the useful information of low-frequency words, which are missed in the distilled data. To this end, we introduce an extra Kullback-Leibler divergence term derived by comparing the lexical choice of NAT model and that embedded in the raw data. Experimental results across language pairs and model architectures demonstrate the effectiveness and universality of the proposed approach. Extensive analyses confirm our claim that our approach improves performance by reducing the lexical choice errors on low-frequency words. Encouragingly, our approach pushes the SOTA NAT performance on the WMT14 English-German and WMT16 Romanian-English datasets up to 27.8 and 33.8 BLEU points, respectively. The source code will be released.