Candidate Soups: Fusing Candidate Results Improves Translation Quality for Non-Autoregressive Translation
This addresses the trade-off between speed and quality in machine translation for users needing fast, high-quality translations, offering an incremental improvement over existing NAT methods.
The paper tackles the translation quality degradation of non-autoregressive translation (NAT) models compared to autoregressive ones by proposing Candidate Soups, a method that fuses candidate results to improve quality while maintaining fast inference. It significantly boosts performance on benchmarks like WMT'14 EN-DE and WMT'16 EN-RO, with a best variant outperforming autoregressive models on three tasks with a 7.6 times speedup.
Non-autoregressive translation (NAT) model achieves a much faster inference speed than the autoregressive translation (AT) model because it can simultaneously predict all tokens during inference. However, its translation quality suffers from degradation compared to AT. And existing NAT methods only focus on improving the NAT model's performance but do not fully utilize it. In this paper, we propose a simple but effective method called "Candidate Soups," which can obtain high-quality translations while maintaining the inference speed of NAT models. Unlike previous approaches that pick the individual result and discard the remainders, Candidate Soups (CDS) can fully use the valuable information in the different candidate translations through model uncertainty. Extensive experiments on two benchmarks (WMT'14 EN-DE and WMT'16 EN-RO) demonstrate the effectiveness and generality of our proposed method, which can significantly improve the translation quality of various base models. More notably, our best variant outperforms the AT model on three translation tasks with 7.6 times speedup.