Secoco: Self-Correcting Encoding for Neural Machine Translation

TL;DR

Secoco introduces a self-correcting encoding framework for neural machine translation that explicitly corrects noisy inputs during translation, leading to improved robustness and interpretability on real-world datasets.

Contribution

The paper proposes Secoco, a novel self-correcting encoding method that handles input noise in NMT by correcting errors during decoding, unlike previous approaches.

Findings

Significant improvements over strong baselines on real-world test sets

Effective correction of noisy inputs during translation process

Good interpretability of the proposed method

Abstract

This paper presents Self-correcting Encoding (Secoco), a framework that effectively deals with input noise for robust neural machine translation by introducing self-correcting predictors. Different from previous robust approaches, Secoco enables NMT to explicitly correct noisy inputs and delete specific errors simultaneously with the translation decoding process. Secoco is able to achieve significant improvements over strong baselines on two real-world test sets and a benchmark WMT dataset with good interpretability. We will make our code and dataset publicly available soon.