Code-Switching for Enhancing NMT with Pre-Specified Translation

TL;DR

This paper introduces a data augmentation technique for neural machine translation that uses code-switched training data to improve translation of constrained words, outperforming existing methods without modifying the model or decoding process.

Contribution

The proposed method creates code-switched training data by replacing source phrases with target translations, enhancing NMT performance on constrained words without affecting overall translation quality.

Findings

Consistent improvements over existing constrained translation methods.

Enhanced translation of constrained words without degrading unconstrained word translation.

Method does not require changes to the NMT model or decoding algorithm.

Abstract

Leveraging user-provided translation to constrain NMT has practical significance. Existing methods can be classified into two main categories, namely the use of placeholder tags for lexicon words and the use of hard constraints during decoding. Both methods can hurt translation fidelity for various reasons. We investigate a data augmentation method, making code-switched training data by replacing source phrases with their target translations. Our method does not change the MNT model or decoding algorithm, allowing the model to learn lexicon translations by copying source-side target words. Extensive experiments show that our method achieves consistent improvements over existing approaches, improving translation of constrained words without hurting unconstrained words.