Elamur Research⌕
☰
home / papers / 2022 / Evaluating and Mitigating Inherent Linguistic Bias of African American English through Inference

Evaluating and Mitigating Inherent Linguistic Bias of African American English through Inference

Paper · 2022 · Jamell Dacon, Haochen Liu and Jiliang Tang

Summary

NLP models trained mostly on Standard American English (SAE) often misjudge text in African American English (AAE). This paper studies that "inherent linguistic bias" in natural language inference (NLI). The authors built CodeSwitch, a greedy, one-directional, rule-based translator that rewrites SAE sentences into AAE. Its rules come from sociolinguistic literature and from AAE versions of SNLI and MNLI sentence pairs produced by trained annotators. The rules are collected in a protocol called TMsP. With CodeSwitch, they made AAE versions of NLI data in which meaning and gold labels stay the same, so model behaviour on SAE and AAE can be compared directly. They then tested two debiasing approaches with BERT-style models:

  • CDA: augmenting training data with AAE examples.
  • LSD: adversarial training that removes dialect-style information from sentence representations so that only semantic content remains.

Both methods shrink the SAE–AAE performance gap. SAE accuracy is kept or even improved, and LSD generally beats CDA. The work matters because dialect speakers should not get worse results from language technology.

Key points

  • The source data come from SNLI (about 570K pairs, built from Flickr30k captions) and MNLI (about 433k pairs across 10 genres). The authors sampled 5,000 pairs from each corpus, 10,000 in total, and every sampled sentence had at least 8 words.
  • Six AMT workers who are trained sociolinguists took part (HIT approval ≥ 96%, more than 10,000 completed HITs, located in the US). Three of them translated the SAE pairs into AAE, and the others helped with human evaluation of contextual accuracy.
  • CodeSwitch works in a few steps. It lowercases the text and tokenizes it. It then replaces any token covered by the TMsP rules, for example "-ing" becomes "-in", "the" becomes "da", and "they" becomes "dey". It also drops some auxiliaries.
  • LSD generally beats CDA on both SAE and AAE test data. The authors claim LSD should carry over to similar models such as DeBERTa, XLNet and T5.
  • The paper names its own limitations:
    • The tool only works SAE→AAE, so it cannot normalize text back to SAE.
    • Its substitutions are deterministic, so it misses slang and regional variants.
    • It does not model the contexts in which copula deletion is grammatical.

Interesting ideas

  • I1. Turn existing labelled data into a dialect variant whose labels stay valid, then compare scores on the two versions, so the gap measures dialect sensitivity alone.
  • I2. Mine substitution rules from a small set of human-made parallel translations instead of writing them only from theory. This keeps the rules grounded in real usage.
  • I3. Add an adversary that tries to guess the language style from the representation, and train the encoder to defeat it, so the semantic signal is separated from the style signal.
  • I4. Treat the SAE-vs-AAE performance gap as a first-class metric beside plain accuracy, because a model can gain accuracy while staying unfair.
  • I5. The authors admit that contextual copula deletion is the hard case that simple substitution misses. This shows where a rule system needs real syntactic conditions.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Run a CodeSwitch-style perturbation over our evaluation texts as a negative control. Frames and answers should not change, and any change exposes a dialect-sensitive step.
  • I2 (tokens): A small aligned SAE/AAE sample could yield the reverse mapping for lemmas and morphology ("smilin" → smile, "ion" → I don't know). The paper's tool cannot do this reverse direction itself.
  • I3 (tree): gparse could get an auxiliary dialect-discriminator head with gradient reversal, which would push its encoder toward dialect-robust attachments.
  • I4 (answer): Report fable QA and word-problem scores on both original and dialect-perturbed texts, and track the gap across releases.
  • I5 (tree): In UD dialect conversion, handle zero-copula clauses explicitly, with the predicate as head and no cop node, so that AAE sentences still yield correct trees.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksSafety Research · 543 works

🧭 Topics 2Research subjects and languages the work is about.English · 2,263 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Jamell Dacon, Haochen Liu and Jiliang Tang. 2022. Evaluating and Mitigating Inherent Linguistic Bias of African American English through Inference. Proceedings of the 29th International Conference on Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{dacon2022evaluating,
  title = {{Evaluating and Mitigating Inherent Linguistic Bias of African American English through Inference}},
  author = {Jamell Dacon, Haochen Liu and Jiliang Tang},
  booktitle = {Proceedings of the 29th International Conference on Computational Linguistics},
  year = {2022},
  doi = {10.48448/bd66-qt03},
  url = {https://doi.org/10.48448/bd66-qt03}
}

References

43 references, 25 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. All authors must warrant mentioning that the in- Moustafa Alzantot, Yash Sharma, Ahmed Elgohary, creased performance for underrepresented dialects Bo-Jhang Ho, Mani Srivastava, and Kai-Wei Chang. 1449 2018. Generating natural language adversarial examples. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2890–2896, Brussels, Belgium. Association for Computational Linguistics.
  2. Guy Bailey, John Baughan, Salikoko S. Mufwene, and John R. Rickford. 1998. African-American English: Structure, History and Use (1st ed.). Routledge. find
  3. Linda M. Bland-Stewart. 2005. Difference or deficit in speakers of african american english? https://leader.pubs.asha.org/doi/ 10.1044/leader.FTR1.10062005.6. ↗
  4. Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna M. Wallach. 2020. Language (technology) is power: A critical survey of "bias" in NLP. CoRR, abs/2005.14050. ↗
  5. Su Lin Blodgett, Lisa Green, and Brendan O’Connor. 2016. Demographic dialectal variation in social media: A case study of african-american english. CoRR, abs/1608.08868. ↗
  6. Su Lin Blodgett, Johnny Wei, and Brendan O’Connor. 2018. Twitter Universal Dependency parsing for African-American and mainstream American English. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1415–1425, Melbourne, Australia. Association for Computational Linguistics. ↗
  7. Tolga Bolukbasi, Kai-Wei Chang, James Zou, Venkatesh Saligrama, and Adam Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Proceedings of the 30th International Conference on Neural Information Processing Systems, NIPS’16, page 4356–4364. ↗
  8. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A large annotated corpus for learning natural language inference. CoRR, abs/1508.05326. ↗
  9. Jamell Dacon. 2022. Towards a deep multi-layered dialectal language analysis: A case study of African-American English. In Proceedings of the Second Workshop on Bridging Human–Computer Interaction and Natural Language Processing, pages 55–63, Seattle, Washington. Association for Computational Linguistics.
  10. Jamell Dacon and Jiliang Tang. 2021. What truly matters? using linguistic cues for analyzing the #blacklivesmatter movement and its counter protests: 2013 to 2020. ↗
  11. Thomas Davidson, Debasmita Bhattacharya, and Ingmar Weber. 2019. Racial bias in hate speech and abusive language detection datasets. In Proceedings of the Third Workshop on Abusive Language Online, pages 25–35, Florence, Italy. Association for Computational Linguistics. ↗
  12. Charles E. DeBose. 1992. Codeswitching: Black english and standard english in the african-american linguistic repertoire. Journal of Multilingual and Multicultural Development, 13(1-2):157–167. ↗
  13. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805. ↗
  14. Yanai Elazar and Yoav Goldberg. 2018. Adversarial removal of demographic attributes from text data. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 11–21, Brussels, Belgium. Association for Computational Linguistics.
  15. Anjalie Field, Su Lin Blodgett, Zeerak Waseem, and Yulia Tsvetkov. 2021. A survey of race, racism, and anti-racism in NLP. CoRR, abs/2106.11410. ↗
  16. Penelope Gardner-Chloros et al. 2009. Code-switching. Cambridge university press. find
  17. Lisa J. Green. 2002. African American English: A Linguistic Introduction. Cambridge University Press. ↗
  18. Sophie Groenwold, Lily Ou, Aesha Parekh, Samhita Honnavalli, Sharon Levy, Diba Mirza, and William Yang Wang. 2020. Investigating African-American Vernacular English in transformer-based text generation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 5877–5883, Online. Association for Computational Linguistics. ↗
  19. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2020. Deberta: Decoding-enhanced BERT with disentangled attention. CoRR, abs/2006.03654. ↗
  20. Dirk Hovy and Shannon L. Spruit. 2016. The social impact of natural language processing. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 591–598, Berlin, Germany. Association for Computational Linguistics. ↗
  21. Mandar Joshi, Danqi Chen, Yinhan Liu, Daniel S. Weld, Luke Zettlemoyer, and Omer Levy. 2019. Spanbert: Improving pre-training by representing and predicting spans. CoRR, abs/1907.10529. ↗
  22. Sharese King. 2020. From african american vernacular english to african american language: Rethinking the study of race and language in african americans’ speech. Annual Review of Linguistics, 6(1):285–300. ↗
  23. Haochen Liu, Jamell Dacon, Wenqi Fan, Hui Liu, Zitao Liu, and Jiliang Tang. 2020a. Does gender matter? towards fairness in dialogue systems. In Proceedings of the 28th International Conference on Computational Linguistics, pages 4403–4416, Barcelona, Spain (Online). International Committee on Computational Linguistics.
  24. Haochen Liu, Wei Jin, Hamid Karimi, Zitao Liu, and Jiliang Tang. 2021. The authors matter: Understanding and mitigating implicit bias in deep text classification. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 74–85, Online. Association for Computational Linguistics. find
  25. Haochen Liu, Wentao Wang, Yiqi Wang, Hui Liu, Zitao Liu, and Jiliang Tang. 2020b. Mitigating gender bias for neural dialogue generation with adversarial learning. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 893–903, Online. Association for Computational Linguistics. find
  26. Chandler May, Alex Wang, Shikha Bordia, Samuel R. Bowman, and Rachel Rudinger. 2019. On measuring social biases in sentence encoders. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 622–628, Minneapolis, Minnesota. Association for Computational Linguistics.
  27. Marcelo O. R. Prates, Pedro H. C. Avelar, and Luís C. Lamb. 2018. Assessing gender bias in machine translation - A case study with google translate. CoRR, abs/1809.02208. ↗
  28. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2019. Exploring the limits of transfer learning with a unified text-to-text transformer. CoRR, abs/1910.10683. ↗
  29. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2018. Semantically equivalent adversarial rules for debugging NLP models. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 856–865, Melbourne, Australia. Association for Computational Linguistics. ↗
  30. Rachel Rudinger, Jason Naradowsky, Brian Leonard, and Benjamin Van Durme. 2018. Gender bias in coreference resolution. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 8–14, New Orleans, Louisiana. Association for Computational Linguistics. ↗
  31. Maarten Sap, Dallas Card, Saadia Gabriel, Yejin Choi, and Noah A. Smith. 2019. The risk of racial bias in hate speech detection. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1668–1678, Florence, Italy. Association for Computational Linguistics. ↗
  32. Roger C Schank. 1972. Conceptual dependency: A theory of natural language understanding. Cognitive psychology, 3(4):552–631. ↗
  33. Shanya Sharma, Manan Dey, and Koustuv Sinha. 2021. Evaluating gender bias in natural language inference. CoRR, abs/2105.05541. ↗
  34. Ian Stewart. 2014. Now we stronger than ever: African-American English syntax in Twitter. In Proceedings of the Student Research Workshop at the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 31–37, Gothenburg, Sweden. Association for Computational Linguistics. Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, and Rob Fergus. Intriguing properties of neural networks.
  35. Samson Tan, Shafiq Joty, Min-Yen Kan, and Richard Socher. 2020a. It’s morphin’ time! Combating linguistic discrimination with inflectional perturbations. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2920– 2935, Online. Association for Computational Linguistics.
  36. Samson Tan, Shafiq Joty, Lav Varshney, and Min-Yen Kan. 2020b. Mind your inflections! Improving NLP for non-standard Englishes with Base-Inflection Encoding. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 5647–5663, Online. Association for Computational Linguistics. find
  37. Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. A broad-coverage challenge corpus for sentence understanding through inference. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1112–1122. Association for Computational Linguistics. ↗
  38. Zhilin Yang, Zihang Dai, Yiming Yang, Jaime G. Carbonell, Ruslan Salakhutdinov, and Quoc V. Le. 2019. Xlnet: Generalized autoregressive pretraining for language understanding. CoRR, abs/1906.08237. ↗
  39. Peter Young, Alice Lai, Micah Hodosh, and Julia Hockenmaier. 2014. From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions. Transactions of the Association for Computational Linguistics, 2:67–78. ↗
  40. Vershawn Ashanti Young. 2009. "nah, we straight": An argument against code switching. JAC, 29(1/2):49– 76. find
  41. Vershawn Ashanti Young and Rusty Barrett. 2018. Other people’s English: Code-meshing, code-switching, and African American literacy. Parlor Press LLC. find
  42. Jieyu Zhao, Yichao Zhou, Zeyu Li, Wei Wang, and Kai-Wei Chang. 2018. Learning gender-neutral word embeddings. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, October 31 - November 4, 2018, pages 4847–4853.
  43. Xiang Zhou and Mohit Bansal. 2020. Towards robustifying NLI models against lexical dataset biases. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8759– 8771, Online. Association for Computational Linguistics. note that these are only a few examples of the most commonly used morphological linguistic AAE features (which we adapt from AAE literature). Our deterministic translative morpho-syntax protocol (TMsP) and its cases are as follows: Ran Zmigrod, Sabrina J. Mielke, Hanna Wallach, and Ryan Cotterell. 2019. Counterfactual data augmentation for mitigating gender stereotypes in languages with rich morphology. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1651–1661, Florence, Italy. Association for Computational Linguistics. 1. Consonant (‘t’) deletion (Special case) : e.g. “just” → “jus”; “must” → “mus”

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/dacon-2022-evaluating-mitigating-inherent-linguistic-bias/