Semi-Autoregressive Training Improves Mask-Predict Decoding

Abstract

The recently proposed mask-predict decoding algorithm has narrowed the performance gap between semi-autoregressive machine translation models and the traditional left-to-right approach. We introduce a new training method for conditional masked language models, SMART, which mimics the semi-autoregressive behavior of mask-predict, producing training examples that contain model predictions as part of their inputs. Models trained with SMART produce higher-quality translations when using mask-predict decoding, effectively closing the remaining performance gap with fully autoregressive models.

References in corpus (2)

• Pay Less Attention with Lightweight and Dynamic Convolutions
• Insertion Transformer: Flexible Sequence Generation via Insertion Operations

Cited by in corpus (1)

• Neural Language Generation: Formulation, Methods, and Evaluation