Optimal Transport for Unsupervised Hallucination Detection in Neural Machine Translation

TL;DR

This paper introduces an unsupervised optimal transport-based method for detecting hallucinations in neural machine translation, leveraging attention patterns to distinguish pathological outputs without requiring labeled data.

Contribution

It proposes a novel, fully unsupervised detector using optimal transport to identify hallucinations in NMT, outperforming previous model-based approaches.

Findings

Outperforms all previous model-based hallucination detectors

Competitive with large-model detectors trained on millions of samples

Effective across different attention-based NMT models

Abstract

Neural machine translation (NMT) has become the de-facto standard in real-world machine translation applications. However, NMT models can unpredictably produce severely pathological translations, known as hallucinations, that seriously undermine user trust. It becomes thus crucial to implement effective preventive strategies to guarantee their proper functioning. In this paper, we address the problem of hallucination detection in NMT by following a simple intuition: as hallucinations are detached from the source content, they exhibit encoder-decoder attention patterns that are statistically different from those of good quality translations. We frame this problem with an optimal transport formulation and propose a fully unsupervised, plug-in detector that can be used with any attention-based NMT model. Experimental results show that our detector not only outperforms all previous model-based detectors, but is also competitive with detectors that employ large models trained on millions of samples.