CoChat

FaithDial: A Faithful Benchmark for Information-Seeking Dialogue

Open access
Dziri, Nouha, Kamalloo, Ehsan, Milton, Sivan, Zaiane, Osmar, Yu, Mo, Ponti, Edoardo M., Reddy, Siva · 2022 · Transactions of the Association for Computational Linguistics
Cited by 63

The DOI matches its Crossref record; 42 of the 42 references it cites were found in a registry or an index.

Abstract

The goal of information-seeking dialogue is to respond to seeker queries with natural language utterances that are grounded on knowledge sources. However, dialogue systems often produce unsupported utterances, a phenomenon known as hallucination. To mitigate this behavior, we adopt a data-centric solution and create FaithDial, a new benchmark for hallucination-free dialogues, by editing hallucinated responses in the Wizard of Wikipedia (WoW) benchmark. We observe that FaithDial is more faithful than WoW while also maintaining engaging conversations. We show that FaithDial can serve as training signal for: i) a hallucination critic, which discriminates whether an utterance is faithful or not, and boosts the performance by 12.8 F1 score on the BEGIN benchmark compared to existing datasets for dialogue coherence; ii) high-quality dialogue generation. We benchmark a series of state-of-the-art models and propose an auxiliary contrastive objective that achieves the highest level of faithfulness and abstractiveness based on several automated metrics. Further, we find that the benefits of FaithDial generalize to zero-shot transfer on other datasets, such as CMU-Dog and TopicalChat. Finally, human evaluation reveals that responses generated by models trained on FaithDial are perceived as more interpretable, cooperative, and engaging.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1162/tacl_a_00529

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 22

  1. Cao, 2021, CLIFF: Contrastive learning for improving faithfulness and factuality in abstractive summarization, 6633, 10.18653/v1/2021.emnlp-main.532 · DOI · Paper page
  2. DeStefano, 2007, Cognitive load in hypertext reading: A review, Computers in Human Behavior, 23, 1616, 10.1016/j.chb.2005.08.012 · DOI · Paper page
  3. Dinan, 2019, Wizard of Wikipedia: Knowledge- powered conversational agents
  4. Dziri, 2019, Evaluating coherence in dialogue systems using entailment, 3806, 10.18653/v1/N19-1381 · DOI · Paper page
  5. Grusky, 2018, Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies, 708, 10.18653/v1/N18-1065 · DOI · Paper page
  6. Kang, 2020, Improved natural language generation via loss truncation, 718, 10.18653/v1/2020.acl-main.66 · DOI · Paper page
  7. Richard Landis, 1977, The measurement of observer agreement for categorical data, biometrics, 159, 10.2307/2529310 · DOI · Paper page
  8. Li, 2021, Addressing semantic drift in generative question answering with auxiliary extraction, 942, 10.18653/v1/2021.acl-short.118 · DOI
  9. Lin, 2004, ROUGE: A package for automatic evaluation of summaries, 74
  10. Liu, 2019, Roberta: A robustly optimized BERT pretraining approach, CoRR
  11. Lux, 2020, Truth or error? Towards systematic analysis of factual errors in abstractive summaries, 1, 10.18653/v1/2020.eval4nlp-1.1 · DOI · Paper page
  12. Nie, 2021, I like fish, especially dolphins: Addressing contradictions in dialogue modeling, 1699, 10.18653/v1/2021.acl-long.134 · DOI · Paper page
  13. van den Oord, 2018, Representation learning with contrastive predictive coding, CoRR
  14. Papineni, 2002, BLEU: A method for automatic evaluation of machine translation, 311, 10.3115/1073083.1073135 · DOI · Paper page
  15. Prabhumoye, 2021, Focused attention improves document-grounded generation, 4274, 10.18653/v1/2021.naacl-main.338 · DOI
  16. Radford, 2019, Language models are unsupervised multitask learners, OpenAI Blog, 1, 9
  17. Sipos, 2012, Large-margin learning of submodular summarization models, 224
  18. Stiles, 1992, Describing Talk: A Taxonomy of Verbal Response Modes
  19. Welleck, 2019, Dialogue natural language inference, 3731, 10.18653/v1/P19-1363 · DOI · Paper page
  20. Zhang, 2021, On the inductive bias of masked language modeling: From statistical to syntactic dependencies, 5131, 10.18653/v1/2021.naacl-main.404 · DOI
  21. Zhao, 2020, Knowledge-grounded dialogue generation with pre-trained language models, 3377, 10.18653/v1/2020.emnlp-main.272 · DOI · Paper page
  22. Zhou, 2018, A dataset for document grounded conversations, 708, 10.18653/v1/D18-1076 · DOI · Paper page

Cited by

See all 63

Metadata

Authors
Dziri, N., Kamalloo, E., Milton, S., Zaiane, O., Yu, M., Ponti, E. M., & Reddy, S.
Venue
Transactions of the Association for Computational Linguistics
Publisher
MIT Press
Published
2022
Volume
10
Pages
1473-1490
ISSN
2307-387X (ISSN-L)
OpenAlex
W4313459318
Cited by
63
References
42
FWCI
3.96
Type
journal article
Language
en
Open access
diamond

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.