Looking for a Needle in a Haystack: A Comprehensive Study of Hallucinations in Neural Machine Translation

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 1059–1075 May 2-6, 2023 ©2023 Association for Computational Linguistics Looking for a Needle in a Haystack: A Comprehensive Study of Hallucinations in Neural Machine Translation Nuno M. Guerreiro1,2 Elena Voita4 André F. T. Martins1,2,3 1Instituto de Telecomunicações, Lisbon, Portugal 2Instituto Superior Técnico & LUMLIS (Lisbon ELLIS Unit), Lisbon, Portugal 3Unbabel, Lisbon, Portugal 4University of Edinburgh, Scotland {nuno.s.guerreiro, andre.t.martins}@tecnico.ulisboa.pt lena-voita@hotmail.com Abstract Although the problem of hallucinations in neu- ral machine translation (NMT) has received some attention, research on this highly patho- logical phenomenon lacks solid ground. Pre- vious work has been limited in several ways: it often resorts to artificial settings where the problem is amplified, it disregards some (com- mon) types of hallucinations, and it does not validate adequacy of detection heuristics. In this paper, we set foundations for the study of NMT hallucinations. First, we work in a natural setting, i.e., in-domain data without artificial noise neither in training nor in in- ference. Next, we annotate a dataset of over 3.4k sentences indicating different kinds of crit- ical errors and hallucinations. Then, we turn to detection methods and both revisit methods used previously and propose using glass-box uncertainty-based detectors. Overall, we show that for preventive settings, (i) previously used methods are largely inadequate, (ii) sequence log-probability works best and performs on par with reference-based methods. Finally, we pro- pose DEHALLUCINATOR, a simple method for alleviating hallucinations at test time which sig- nificantly reduces the hallucinatory rate. 1 Introduction Neural machine translation (NMT) is becom- ing increasingly accurate (Vaswani et al., 2017; Akhbardeh et al., 2021), particularly in high re- source language pairs where parallel data is abun- dant. However, even the best systems available today may generate hallucinations. These are ex- tremely pathological translations that contain con- tent that is unfaithful to the source sequence. Criti- cally, a tiny fraction of these mistakes is all it takes to compromise user trust or safe deployment of NMT models in production. Unfortunately, although the problem of hallu- cinations received some attention, research on this highly pathological phenomenon lacks solid ground. First, previous work used multiple and of- ten overlapping definitions and categories of hallu- cinations which makes it hard to draw connections between observations made in different works (Lee et al., 2018; Raunak et al., 2021; Zhou et al., 2021). Next, since hallucinations are extremely rare, pre- vious work focused on settings in which the phe- nomenon is amplified, e.g. perturbing data either in training or at inference, or evaluating under do- main shift (Lee et al., 2018; Raunak et al., 2021; Müller et al., 2020; Wang and Sennrich, 2020; Voita et al., 2021; Müller and Sennrich, 2021; Zhou et al., 2021). Critically, the analysis on these works mostly relied on the adequacy of the automatic hal- lucination detection methods they proposed. How- ever, it is not immediate whether these methods translate well to unperturbed settings. In this work, we set foundations for the study …

Content truncated to 500 words