Learning Coupled Policies for Simultaneous Machine Translation using Imitation Learning

Philip Arthur, Trevor Cohn, Gholamreza Haffari


Abstract
We present a novel approach to efficiently learn a simultaneous translation model with coupled programmer-interpreter policies. First, we present an algorithmic oracle to produce oracle READ/WRITE actions for training bilingual sentence-pairs using the notion of word alignments. This oracle actions are designed to capture enough information from the partial input before writing the output. Next, we perform a coupled scheduled sampling to effectively mitigate the exposure bias when learning both policies jointly with imitation learning. Experiments on six language-pairs show our method outperforms strong baselines in terms of translation quality quality while keeping the delay low.
Anthology ID:
2021.eacl-main.233
Volume:
Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume
Month:
April
Year:
2021
Address:
Online
Editors:
Paola Merlo, Jorg Tiedemann, Reut Tsarfaty
Venue:
EACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
2709–2719
Language:
URL:
https://aclanthology.org/2021.eacl-main.233
DOI:
10.18653/v1/2021.eacl-main.233
Bibkey:
Cite (ACL):
Philip Arthur, Trevor Cohn, and Gholamreza Haffari. 2021. Learning Coupled Policies for Simultaneous Machine Translation using Imitation Learning. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 2709–2719, Online. Association for Computational Linguistics.
Cite (Informal):
Learning Coupled Policies for Simultaneous Machine Translation using Imitation Learning (Arthur et al., EACL 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.eacl-main.233.pdf