Glancing Transformer for Non-Autoregressive Neural Machine Translation
This work addresses the efficiency-quality trade-off in machine translation for applications requiring fast, parallel decoding, representing a strong incremental improvement over prior NAT methods.
The paper tackled the problem of non-autoregressive neural machine translation (NAT) by proposing Glancing Transformer (GLAT), which uses a single-pass parallel decoding method to achieve high-quality translations with 8-15 times speedup, reducing the gap to Transformer to 0.25-0.9 BLEU points on multiple WMT language directions.
Recent work on non-autoregressive neural machine translation (NAT) aims at improving the efficiency by parallel decoding without sacrificing the quality. However, existing NAT methods are either inferior to Transformer or require multiple decoding passes, leading to reduced speedup. We propose the Glancing Language Model (GLM), a method to learn word interdependency for single-pass parallel generation models. With GLM, we develop Glancing Transformer (GLAT) for machine translation. With only single-pass parallel decoding, GLAT is able to generate high-quality translation with 8-15 times speedup. Experiments on multiple WMT language directions show that GLAT outperforms all previous single pass non-autoregressive methods, and is nearly comparable to Transformer, reducing the gap to 0.25-0.9 BLEU points.