Understanding and Detecting Hallucinations in Neural Machine Translation via Model Introspection

• Published in: Transactions of the Association for Computational Linguistics Vol. 11; pp. 546 - 564
• Main Authors: Xu, Weijia, Agrawal, Sweta, Briakou, Eleftheria, Martindale, Marianna J., Carpuat, Marine
• Format: Journal Article
• Language: English
• Published: One Broadway, 12th Floor, Cambridge, Massachusetts 02142, USA MIT Press 12.06.2023; MIT Press Journals, The; The MIT Press
• Subjects: Chinese languages; Elections; Employment; English language; German language; Hallucinations; Hypotheses; Linguistics; Machine translation; Translations
• ISSN: 2307-387X; 2307-387X
• DOI: 10.1162/tacl_a_00563

Neural sequence generation models are known to “hallucinate”, by producing outputs that are unrelated to the source text. These hallucinations are potentially harmful, yet it remains unclear in what conditions they arise and how to mitigate their impact. In this work, we first identify internal model symptoms of hallucinations by analyzing the relative token contributions to the generation in contrastive hallucinated vs. non-hallucinated outputs generated via source perturbations. We then show that these symptoms are reliable indicators of natural hallucinations, by using them to design a lightweight hallucination detector which outperforms both model-free baselines and strong classifiers based on quality estimation or large pre-trained models on manually annotated English-Chinese and German-English translation test beds.