STACL: Simultaneous Translation with Implicit Anticipation and Controllable Latency using Prefix-to-Prefix Framework

TL;DR

This paper introduces a prefix-to-prefix framework for simultaneous translation that enables models to anticipate words and control latency, achieving low delay and good quality across multiple language pairs.

Contribution

It proposes a novel prefix-to-prefix translation framework with an effective wait-k policy for low-latency, high-quality simultaneous translation.

Findings

Achieves low latency and reasonable translation quality

Effective anticipation in a single model

Works well across multiple language pairs

Abstract

Simultaneous translation, which translates sentences before they are finished, is useful in many scenarios but is notoriously difficult due to word-order differences. While the conventional seq-to-seq framework is only suitable for full-sentence translation, we propose a novel prefix-to-prefix framework for simultaneous translation that implicitly learns to anticipate in a single translation model. Within this framework, we present a very simple yet surprisingly effective wait-k policy trained to generate the target sentence concurrently with the source sentence, but always k words behind. Experiments show our strategy achieves low latency and reasonable quality (compared to full-sentence translation) on 4 directions: zh<->en and de<->en.