On Exposure Bias, Hallucination and Domain Shift in Neural Machine Translation
arXiv:2005.03642
Abstract
The standard training algorithm in neural machine translation (NMT) suffers from exposure bias, and alternative algorithms have been proposed to mitigate this. However, the practical impact of exposure bias is under debate. In this paper, we link exposure bias to another well-known problem in NMT, namely the tendency to generate hallucinations under domain shift. In experiments on three datasets with multiple test domains, we show that exposure bias is partially to blame for hallucinations, and that training with Minimum Risk Training, which avoids exposure bias, can mitigate this. Our analysis explains why exposure bias is more problematic under domain shift, and also links exposure bias to the beam search problem, i.e. performance deterioration with increasing beam size. Our results provide a new justification for methods that reduce exposure bias: even if they do not increase performance on in-domain test sets, they can increase model robustness to domain shift.
arXiv commentACL 2020
References in corpus (6)
- Sequence to Sequence Learning with Neural Networks
- Assessing agreement on classification tasks: the kappa statistic
- Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks
- Pay Less Attention with Lightweight and Dynamic Convolutions
- Mutual Information and Diverse Decoding Improve Neural Machine Translation
- Domain Robustness in Neural Machine Translation
Cited by in corpus (4)
- Understanding and Improving Encoder Layer Fusion in Sequence-to-Sequence Learning
- Sketch and Refine: Towards Faithful and Informative Table-to-Text Generation
- Hallucinations in LLMs: A Lifecycle-Based Survey of Causes, Detection, Mitigation, and Prevention
- Exposure Bias versus Self-Recovery: Are Distortions Really Incremental for Autoregressive Text Generation?