Hallucinations in Neural Machine Translation

By Katherine Lee, Orhan Firat, Ashish Agarwal, Clara Fannjiang, David Sussillo. 2018

183 citations recorded by the source corpus.

Subfields: Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

Semantic Noise Matters for Neural Natural Language Generation (2019, background): “…reports on hallucinations (i.e. adding information not grounded in the input), as observed for image captioning (Rohrbach et al., 2018), sports report generation (Wiseman et al., 2017), machine translation (Koehn and Knowles, 2017; Lee et al., 2019), and question answering (Feng et al., 2018).”
WT5?! Training Text-to-Text Models to Explain their Predictions (2020, background): “, 2019) or diagnosing why they sometimes generate nonsensical text (Lee et al., 2018; Belinkov & Bisk, 2017).”
Survey of Hallucination in Natural Language Generation (2022, methodology): “Finally, abrupt jumps to the end of the sequence and outputs that remain mostly in the source language are also examples of hallucinatory content [95].”
mSLAM: Massively multilingual joint pre-training for speech and text (2022, background): “This is reminiscent of oscillatory hallucinations caused by unexpected inputs (Lee et al., 2018).”
ToTTo: A Controlled Table-To-Text Generation Dataset (2020, background): “…to be prone to hallucination , i.e., generating text that is ﬂuent but not faithful to the source (Vinyals and Le, 2015; Koehn and Knowles, 2017; Lee et al., 2018; Tian et al., 2019) and it is often easier to assess faithfulness of the generated text when the source content is structured…”
Handling Divergent Reference Texts when Evaluating Table-to-Text Generation (2019, background): “Hallucination (Rohrbach et al., 2018; Lee et al., 2018) refers to when an NLG system generates text which mentions extra information than what is present in the source from which it is generated.”
Unsolved Problems in ML Safety (2021, background): “Others could also build tools to detect when models are hallucinating information [117].”
On Extractive and Abstractive Neural Document Summarization with Transformer Language Models (2020, background): “Transformers are also known to hallucinate (Lee et al., 2019) but we notice that including extracted sentences, TLM + Intro + Extract, improve relevance by 3% over TLM + Intro, bringing relevance closer to extractive methods.”
Non-Parametric Adaptation for Neural Machine Translation (2019, background): “…translation quality NMT has shown some glaring weaknesses, including idiom processing, and rare word or phrase translation (Koehn and Knowles, 2017; Isabelle et al., 2017; Lee et al., 2018) - tasks that should be easy if the model could retain learned information from individual training examples.”
Q^{2}: Evaluating Factual Consistency in Knowledge-Grounded Dialogues via Question Generation and Question Answering (2021, background): “Amongst other tasks, this phenomenon has been observed in neural machine translation (Lee et al., 2019) mainly when considering out-of-domain scenarios (Koehn and Knowles,”
Improved Natural Language Generation via Loss Truncation (2020, background): “For example, neural language models often produce ﬂuent text that is unfaithful to the source (Tian et al., 2019; Wiseman et al., 2017; Lee et al., 2018).”
Neural Path Hunter: Reducing Hallucination in Dialogue Systems via Path Grounding (2021, background): “chine translation (Koehn and Knowles, 2017; Lee et al., 2019), image captioning (Rohrbach et al.”
Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, unknown): “In contrast, when the contradiction arises between the generated content and task input, the hallucination is in line with the conventional definition in specific NLG tasks, such as machine translation (Lee et al., 2019) and summarization (Maynez et al., 2020; Pu et al., 2023).”
Domain Robustness in Neural Machine Translation (2019, background): “to produce translations that are fluent, but unrelated to the content of the source sentence (Lee et al., 2018).”
Aligning Large Multimodal Models with Factually Augmented RLHF (2023, background): “…by (Zhang et al., 2023) into: 1) Input-conflicting hallucination, which veers away from user-given input, exemplified in machine translation (Lee et al., 2018; Zhou et al., 2020); 2) Context-conflicting hallucination where output contradicts prior LLM-generated information (Shi et al.,…”
Multi-Fact Correction in Abstractive Text Summarization (2020, background): “…in abstractive summarization systems are inherently difﬁcult to control and prone to hal-lucination (Vinyals and Le, 2015; Koehn and Knowles, 2017; Lee et al., 2018), and often leads to factual inconsistency: the system-generated summary is ﬂuent but unfaithful to the source (Cao et al., 2018).”
Data-QuestEval: A Referenceless Metric for Data-to-Text Semantic Evaluation (2021, background): “Neural NLG systems are known to be ﬂuent, but prone to hallucinations (Lee et al., 2018), i.e. they tend to include nonfactual information.”
Sticking to the Facts: Confident Decoding for Faithful Data-to-Text Generation (2019, background): “However, encoder-decoder architectures are inherently difﬁcult to control, and have been shown to be prone to hallucination , i.e., generating text that is ﬂuent but unfaithful to the source (Vinyals & Le, 2015; Koehn & Knowles, 2017; Wiseman et al., 2017; Lee et al., 2018).”
Unleashing the potential of prompt engineering in Large Language Models: a comprehensive review (2023, methodology): “When using AIGC tools such as GPT-4, it is common to face a problem called “hallucinations”, which refer to the presence of unreal or inaccurate information in the model’s generated output [17, 67].”