Non-autoregressive Streaming Transformer for Simultaneous Translation

作     者：Ma, Zhengrui Zhang, Shaolei Guo, Shoutao Shao, Chenze Zhang, Min Feng, Yang 

作者机构：Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China School of Future Science and Engineering Soochow University China 

出 版 物：《arXiv》 (arXiv)

年 卷 期：2023年

核心收录：

主　　题：Machine translation 

摘      要：Simultaneous machine translation (SiMT) models are trained to strike a balance between latency and translation quality. However, training these models to achieve high quality while maintaining low latency often leads to a tendency for aggressive anticipation. We argue that such issue stems from the autoregressive architecture upon which most existing SiMT models are built. To address those issues, we propose non-autoregressive streaming Transformer (NAST) which comprises a unidirectional encoder and a non-autoregressive decoder with intra-chunk parallelism. We enable NAST to generate the blank token or repetitive tokens to adjust its READ/WRITE strategy flexibly, and train it to maximize the non-monotonic latent alignment with an alignment-based latency loss. Experiments on various SiMT benchmarks demonstrate that NAST outperforms previous strong autoregressive SiMT baselines. Source code is publicly available at https://***/ictnlp/NAST. Copyright © 2023, The Authors. All rights reserved.