Non-Autoregressive Neural Machine Translation

TL;DR

This paper presents a non-autoregressive neural machine translation model that produces outputs in parallel, significantly reducing inference latency while maintaining competitive translation quality through innovative training strategies.

Contribution

The paper introduces a novel non-autoregressive translation model that leverages knowledge distillation, input token fertilities, and policy gradient fine-tuning to achieve low-latency translation with minimal BLEU score loss.

Findings

Achieves near-state-of-the-art BLEU scores on WMT datasets.

Reduces inference latency by an order of magnitude compared to autoregressive models.

Demonstrates effectiveness of combined training strategies in non-autoregressive translation.

Abstract

Existing approaches to neural machine translation condition each output word on previously generated outputs. We introduce a model that avoids this autoregressive property and produces its outputs in parallel, allowing an order of magnitude lower latency during inference. Through knowledge distillation, the use of input token fertilities as a latent variable, and policy gradient fine-tuning, we achieve this at a cost of as little as 2.0 BLEU points relative to the autoregressive Transformer network used as a teacher. We demonstrate substantial cumulative improvements associated with each of the three aspects of our training strategy, and validate our approach on IWSLT 2016 English-German and two WMT language pairs. By sampling fertilities in parallel at inference time, our non-autoregressive model achieves near-state-of-the-art performance of 29.8 BLEU on WMT 2016 English-Romanian.