Negative Lexical Constraints in Neural Machine Translation

TL;DR

This paper investigates methods for enforcing negative lexical constraints in neural machine translation from English to Czech, focusing on preventing specific words or expressions in the output and addressing the challenge of surface form variations.

Contribution

It introduces a training approach using stemmed negative constraints to reduce constraint bypassing in neural machine translation.

Findings

Training with stemmed constraints improves constraint adherence

Methods vary in effectiveness depending on the task

Complete mitigation of bypassing remains challenging

Abstract

This paper explores negative lexical constraining in English to Czech neural machine translation. Negative lexical constraining is used to prohibit certain words or expressions in the translation produced by the neural translation model. We compared various methods based on modifying either the decoding process or the training data. The comparison was performed on two tasks: paraphrasing and feedback-based translation refinement. We also studied to which extent these methods "evade" the constraints presented to the model (usually in the dictionary form) by generating a different surface form of a given constraint.We propose a way to mitigate the issue through training with stemmed negative constraints to counter the model's ability to induce a variety of the surface forms of a word that can result in bypassing the constraint. We demonstrate that our method improves the constraining, although the problem still persists in many cases.