Semantic Noise Matters for Neural Natural Language Generation
Open accessThe DOI matches its Crossref record; 28 of the 28 references it cites were found in a registry or an index.
Abstract
Neural natural language generation (NNLG) systems are known for their pathological outputs, i.e. generating text which is unrelated to the input specification.In this paper, we show the impact of semantic noise on state-of-theart NNLG models which implement different semantic control mechanisms.We find that cleaned data can improve semantic correctness by up to 97%, while maintaining fluency.We also find that the most common error is omitting information, rather than hallucination.
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.
Integrity signals
Registry
Cites
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.
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
BLEU: a method for automatic evaluation of machine translation
Papineni, K., Roukos, S., Ward, T., & Zhu, W.-J. (2001) · Proceedings of the 40th Annual Meeting on Association for Computational Linguistics - ACL '02 · doi:10.3115/1073083.1073135 · Cited by 21,344 · cited 1×
“We use freely available word-overlap-based evaluation metrics (WOM) scripts that come with the E2E data (Duˇsek et al., 2019), 9 supporting BLEU (Papineni et al., 2002), NIST (Doddington, 2002), ROUGE-L (Lin, 2004), METEOR (Lavie and Agarwal, 2007) and CIDEr (Vedantam et al., 2015).”
MethodNeural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213 · cited 1×
“6 TGen is in essence a vanilla sequence-to-sequence (seq2seq) model with attention (Bahdanau et al., 2015) us-ing LSTM cells where input MRs are encoded as sequences of triples in the form (dialogue act, slot, value).”
Background(2004) · Cited by 9,718
CIDEr: Consensus-based image description evaluation
Vedantam, R., Zitnick, C. L., & Parikh, D. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7299087 · Cited by 4,845 · cited 1×
“We use freely available word-overlap-based evaluation metrics (WOM) scripts that come with the E2E data (Duˇsek et al., 2019), 9 supporting BLEU (Papineni et al., 2002), NIST (Doddington, 2002), ROUGE-L (Lin, 2004), METEOR (Lavie and Agarwal, 2007) and CIDEr (Vedantam et al., 2015).”
MethodAutomatic evaluation of machine translation quality using n-gram co-occurrence statistics
Doddington, G. (2002) · Proceedings of the second international conference on Human Language Technology Research - · doi:10.3115/1289189.1289273 · Cited by 1,584 · cited 3×
“, 2002), NIST (Doddington, 2002), ROUGE-L (Lin, 2004), METEOR (Lavie and Agarwal, 2007) and CIDEr (Vedantam et al.”
Meteor: an automatic metric for MT evaluation with high levels of correlation with human judgments
Lavie, A., & Agarwal, A. (2007) · Proceedings of the Second Workshop on Statistical Machine Translation - StatMT '07 · doi:10.3115/1626355.1626389 · Cited by 1,047 · cited 1×
“We use freely available word-overlap-based evaluation metrics (WOM) scripts that come with the E2E data (Duˇsek et al., 2019), 9 supporting BLEU (Papineni et al., 2002), NIST (Doddington, 2002), ROUGE-L (Lin, 2004), METEOR (Lavie and Agarwal, 2007) and CIDEr (Vedantam et al., 2015).”
MethodSemantically Conditioned LSTM-based Natural Language Generation for Spoken Dialogue Systems
Wen, T.-H., Gasic, M., Mrkšić, N., Su, P.-H., Vandyke, D., & Young, S. (2015) · Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d15-1199 · Cited by 874 · cited 11×
“The differences against the original MRs allow us to compute the semantic/slot error rate (SER; Wen et al., 2015; Reed et al., 2018; Dušek et al., 2019):”
BackgroundMethodInfluentialChallenges in Data-to-Document Generation
Wiseman, S., Shieber, S., & Rush, A. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1239 · Cited by 524 · cited 1×
“…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).”
BackgroundObject Hallucination in Image Captioning
Rohrbach, A., Hendricks, L. A., Burns, K., Darrell, T., & Saenko, K. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1437 · Cited by 420 · cited 3×
“…work on related tasks, which mostly 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…”
BackgroundResultEric, M., Goel, R., Paul, S., Kumar, A., Sethi, A. R., Ku, P., Goyal, A. K., Agarwal, S., Gao, S., & Hakkani-Tür, D. Z. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1907.01669 · Cited by 337 · cited 2×
“These findings suggest that efforts to create more accurate training data—whether through stricter crowd-sourcing protocols, conducting follow-up annotations (cf. Eric et al., 2019), or automated cleanup heuristics like we report here—are likely necessary in the NLG and dialogue systems communities.”
BackgroundResultInfluentialPathologies of Neural Models Make Interpretations Difficult
Feng, S., Wallace, E., Grissom II, A., Iyyer, M., Rodriguez, P., & Boyd-Graber, J. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1407 · Cited by 245 · cited 1×
“…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).”
BackgroundWhat to talk about and how? Selective Generation using LSTMs with Coarse-to-Fine Alignment
Mei, H., Bansal, M., & Walter, M. R. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1086 · Cited by 237 · cited 1×
“Neural Natural Language Generation (NNLG) is promising for generating text from Meaning Rep-resentations (MRs) in an ‘end-to-end’ fashion, i.e. without needing alignments (Wen et al., 2015, 2016; Duˇsek and Jurˇc´ıˇcek, 2016; Mei et al., 2016).”
BackgroundEvaluating the state-of-the-art of End-to-End Natural Language Generation: The E2E NLG challenge
Dušek, O., Novikova, J., & Rieser, V. (2020) · Computer Speech & Language · doi:10.1016/j.csl.2019.06.009 · Cited by 192
Self-citationMulti-domain Neural Network Language Generation for Spoken Dialogue Systems
Wen, T.-H., Gašić, M., Mrkšić, N., Rojas-Barahona, L. M., Su, P.-H., Vandyke, D., & Young, S. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1015 · Cited by 167
MultiWOZ 2.1: Multi-Domain Dialogue State Corrections and State Tracking Baselines
Eric, M., Goel, R., Paul, S., Sethi, A. R., Agarwal, S., Gao, S., & Hakkani-Tür, D. Z. (2019) · Cited by 124
Phrase-Based Statistical Language Generation Using Graphical Models and Active Learning
Mairesse, F., Gašić, M., Jurčíček, F., Keizer, S., Thomson, B., Yu, K., & Young, S. J. (2010) · Cited by 118
(2018) · Cited by 115
Findings of the E2E NLG Challenge
Dušek, O., Novikova, J., & Rieser, V. (2018) · Proceedings of the 11th International Conference on Natural Language Generation · doi:10.18653/v1/w18-6539 · Cited by 90 · cited 1×
“This phenomenon is also observed by Duˇsek et al. (2018, 2019) for systems in the E2E NLG challenge, but stands in contrast to work on related tasks, which mostly reports on hallucinations (i.e. adding information not grounded in the input), as observed for image captioning (Rohrbach et al., 2018),…”
ResultSelf-citationA Deep Ensemble Model with Slot Alignment for Sequence-to-Sequence Natural Language Generation
Juraska, J., Karagiannis, P., Bowden, K., & Walker, M. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1014 · Cited by 75 · cited 2×
“We plan to further improve our slot matching script based on errors found during the manual evaluation and comparison to Juraska et al. (2018).”
MethodInfluentialA Simple Recipe towards Reducing Hallucination in Neural Surface Realisation
Nie, F., Yao, J.-G., Wang, J., Pan, R., & Lin, C.-Y. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1256 · Cited by 68 · cited 1×
“Contemporaneous with our work is the effort of Nie et al. (2019), who focus on automatic data cleaning using a NLU iteratively bootstrapped from the noisy data.”
MethodSequence-to-Sequence Generation for Spoken Dialogue via Deep Syntax Trees and Strings
Dušek, O., & Jurcicek, F. (2016) · Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) · doi:10.18653/v1/p16-2008 · Cited by 66 · cited 5×
“In this paper, we examine the impact of this type of semantic noise on two state-of-the-art NNLG models with different semantic control mechanisms: TGen (Dušek and Jurčı́ček, 2016) and SCLSTM (Wen et al.”
BackgroundMethodInfluentialSelf-citationChallenges in Data-to-Document GenerationPreprint
Wiseman, S., Shieber, S. M., & Rush, A. M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.08052 · Cited by 53 · cited 1×
“…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).”
BackgroundCrowd-sourcing NLG Data: Pictures Elicit Better Data.
Novikova, J., Lemon, O., & Rieser, V. (2016) · Proceedings of the 9th International Natural Language Generation conference · doi:10.18653/v1/w16-6644 · Cited by 47 · cited 1×
“However, NNLG requires large volumes of in-domain data, which is typically crowdsourced (e.g. Mairesse et al., 2010; Novikova et al., 2016; Wen et al., 2015, 2016; Howcroft et al., 2017), introducing noise.”
BackgroundSelf-citationMulti-domain Neural Network Language Generation for Spoken Dialogue SystemsPreprint
Wen, T.-H., Gašić, M., Mrkšić, N., Rojas-Barahona, L. M., Su, P.-H., Vandyke, D., & Young, S. J. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1603.01232 · Cited by 41
Can Neural Generators for Dialogue Learn Sentence Planning and Discourse Structuring?
Reed, L., Oraby, S., & Walker, M. A. (2018) · International Conference on Natural Language Generation · doi:10.18653/v1/w18-6535 · Cited by 37 · cited 2×
“To produce a cleaned version of the E2E data, we used the original human textual references, but paired them with correctly matching MRs. 4 To this end, we reimplemented the slot matching script of Reed et al. (2018), which tags MR slots and values using regular expressions.”
MethodWhat to talk about and how? Selective Generation using LSTMs with Coarse-to-Fine AlignmentPreprint
Mei, H., Bansal, M., & Walter, M. R. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1509.00838 · Cited by 22 · cited 1×
“Neural Natural Language Generation (NNLG) is promising for generating text from Meaning Rep-resentations (MRs) in an ‘end-to-end’ fashion, i.e. without needing alignments (Wen et al., 2015, 2016; Duˇsek and Jurˇc´ıˇcek, 2016; Mei et al., 2016).”
BackgroundCrowd-sourcing NLG Data: Pictures Elicit Better DataPreprint
Novikova, J., Lemon, O., & Rieser, V. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1608.00339 · Cited by 8 · cited 1×
“However, NNLG requires large volumes of in-domain data, which is typically crowdsourced (e.g. Mairesse et al., 2010; Novikova et al., 2016; Wen et al., 2015, 2016; Howcroft et al., 2017), introducing noise.”
BackgroundSelf-citationThe Extended SPaRKy Restaurant Corpus: Designing a Corpus with Variable Information Density
Howcroft, D. M., Klakow, D., & Demberg, V. (2017) · Interspeech (USB) · doi:10.21437/interspeech.2017-1555 · Cited by 6 · cited 2×
“However, NNLG requires large volumes of in-domain data, which is typically crowdsourced (e.g. Mairesse et al., 2010; Novikova et al., 2016; Wen et al., 2015, 2016; Howcroft et al., 2017), introducing noise.”
BackgroundSelf-citation
Research works printed but not linked: 1
- [10]10. Philipp Koehn and Rebecca Knowles. 2017. Six Chal- lenges for Neural Machine Translation. In Proceed- ings of the First Workshop on Neural Machine Trans- lation, pages 28-39, Vancouver, Canada.
Cited by
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025
“Considering this issue, other works [37, 137] correct paired training samples, specifically the input data, according to the references.”
BackgroundPaLM: Scaling Language Modeling with Pathways
Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A., Barham, P., Chung, H. W., Sutton, C., Gehrmann, S., et al. (2022) · arXiv · doi:10.48550/arxiv.2204.02311 · Cited by 2,590
“[cz] • WebNLG 2020 (Gardent et al., 2017; Castro Ferreira et al., 2020) – Verbalize subject-predicate-object triples in one or more sentences in a grammatical and natural way.”
BackgroundInfluentialFinetuned Language Models Are Zero-Shot Learners
Wei, J., Bosma, M., Zhao, V. Y., Guu, K., Yu, A. W., Lester, B., Du, N., Dai, A. M., & Le, Q. V. (2021) · arXiv · doi:10.48550/arxiv.2109.01652 · Cited by 1,101
“E2ENLG (Dušek et al., 2019) 4.”
GLM-130B: An Open Bilingual Pre-trained Model
Zeng, A., Liu, X., Du, Z., Wang, Z., Lai, H., Ding, M., Yang, Z., Xu, Y., Zheng, W., Xia, X., et al. (2022) · arXiv · doi:10.48550/arxiv.2210.02414 · Cited by 281
“, 2020), Clean E2E NLG (Dušek et al., 2019) and WikiLingua (Scialom et al.”
BackgroundMethodInfluentialSPoT: Better Frozen Model Adaptation through Soft Prompt Transfer
Vu, T., Lester, B., Constant, N., Al-Rfou’, R., & Cer, D. (2022) · Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2022.acl-long.346 · Cited by 170
Metadata
- Authors
- Dušek, O., Howcroft, D. M., & Rieser, V.
- Affiliations
- Charles University (CZ)
- Heriot-Watt University (GB)
- Venue
- Proceedings of the 12th International Conference on Natural Language Generation
- Published
- 2019
- Pages
- 421-426
- OpenAlex
- W2987956546
- Cited by
- 85
- References
- 24
- Citations by year
- 2026: 1 · 2025: 3 · 2024: 2 · 2023: 23 · 2022: 19 · 2021: 23 · 2020: 13 · 2019: 1
- FWCI
- 5.36
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- International Conference on Natural Language Generation (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Univerzita Karlova v Praze
- Engineering and Physical Sciences Research Council (EP/N017536/1, EP/M005429/1)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Keywords
- semantic noise, semantic control, hallucinations, information omission, data cleaning, semantic correctness, fluency
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.