Measuring Bias in Contextualized Word Representations
Open accessAbstract
Contextual word embeddings such as BERT have achieved state of the art performance in numerous NLP tasks.Since they are optimized to capture the statistical properties of training data, they tend to pick up on and amplify social stereotypes present in the data as well.In this study, we (1) propose a template-based method to quantify bias in BERT; (2) show that this method obtains more consistent results in capturing social biases than the traditional cosine based method; and (3) conduct a case study, evaluating gender bias in a downstream task of Gender Pronoun Resolution.Although our case study focuses on gender bias, the proposed technique is generalizable to unveiling other biases, including in multiclass settings, such as racial and religious biases.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Contextual word embedding bias studies use artificial context that does not reflect natural usage of a word.
“First, the context does not reflect the natural usage of a word.”
From StereoSet: Measuring stereotypical bias in pretrained language models (2021) · cites this paper as Kurita et al., 2019
The Introduction (page unavailable) says the authors create simple template sentences to measure bias, supporting the claim that the study uses constructed contexts. It does not say that those contexts fail to reflect a word’s natural usage; that criticism is not established by this source. (LLM, conf=0.79)
Experimental — accuracy not yet measuredCurrent evaluations focus on measuring bias without considering the language modeling ability of a model.
“current evaluations focus on measuring bias without considering the language modeling ability of a model, which could lead to misleading trust on a model even if it is a poor language model.”
From StereoSet: Measuring stereotypical bias in pretrained language models (2021) · cites this paper as Kurita et al., 2019
The Introduction and §5 Real World Implications discuss existing bias-evaluation methods and say they can give inconsistent or unclear results for contextualized embeddings. But the paper does not say those evaluations neglect a model’s language-modeling ability, or that they could inspire misplaced trust in a poor language model. Its proposed method uses masked-token probabilities to measure bias, not to assess overall language-modeling quality. (LLM, conf=0.78)
Experimental — accuracy not yet measuredExisting bias evaluations mainly evaluate bias of pretrained language models on a small set of artificial sentences.
“they mainly evaluate bias of pretrained language models on a small set of artificial sentences, even though these models are trained on natural data”
From StereoSet: Measuring stereotypical bias in pretrained language models (2021) · cites this paper as Kurita et al., 2019
The Introduction and Table 1 describe evaluating BERT bias with a small set of simple template sentences, and the paper discusses an earlier sentence-embedding bias test. But it does not establish that existing bias evaluations mainly use artificial sentences; it also examines bias on a Wikipedia-derived pronoun-resolution dataset. Page numbers are unavailable in our copy. (LLM, conf=0.78)
Experimental — accuracy not yet measuredContextual word embedding bias studies focus on single word terms and ignore multiword terms like construction worker.
“Third, they focus on single word terms and ignore multiword terms like construction worker.”
From StereoSet: Measuring stereotypical bias in pretrained language models (2021) · cites this paper as Kurita et al., 2019
The Introduction (page unavailable) describes this paper’s bias measure as querying BERT for a particular token and illustrates it with single-word targets and attributes. It does not establish that contextual-embedding bias studies generally ignore multiword terms, or specifically discuss “construction worker.” (LLM, conf=0.78)
Experimental — accuracy not yet measuredContextual word embedding bias studies study bias of a model independent of its language modeling ability.
“Lastly, they study bias of a model independent of its language modeling ability which could lead to undeserved trust in a model if it is a poor language model.”
From StereoSet: Measuring stereotypical bias in pretrained language models (2021) · cites this paper as Kurita et al., 2019
The Introduction describes measuring BERT’s bias using masked-token predictions and normalizing those predictions against a prior probability. It does not establish that the bias measure is independent of language-modeling ability or discuss undeserved trust in a poor language model. (LLM, conf=0.84)
Experimental — accuracy not yet measured
Contextual word embedding bias studies require stereotypical attribute terms to be predefined.
“Second, they require stereotypical attribute terms to be predefined (e.g., pleasant and unpleasant terms).”
From StereoSet: Measuring stereotypical bias in pretrained language models (2021) · cites this paper as Kurita et al., 2019
Language models inherit biases and dominant views from their training corpus.
“Text suggestions based on language models have been known to inherit biases and dominant 'views' from the text corpus used to train the language model [22,39,43,55].”
From Interacting with Next-Phrase Suggestions: How Suggestion Systems Aid and Influence the Cognitive Processes of Writing (2023) · cites this paper as [43]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 31 · Cited by 348
Research works printed but not linked: 3
- Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Cor- rado, and Jeff Dean. 2013. Distributed representa- tions of words and phrases and their compositional- ity. In Proc.of NIPS, pages 3111-3119.
- Matthew E Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep contextualized word rep- resentations. In Proc. of NAACL.
- Ian Tenney, Patrick Xia, Berlin Chen, Alex Wang, Adam Poliak, R. Thomas McCoy, Najoung Kim, Benjamin Van Durme, Samuel R. Bowman, Dipan- jan Das, and Ellie Pavlick. 2019. What do you learn from context? probing for sentence structure in con- textualized word representations. In Proc. of ICLR.
- Authors
- Kurita, K., Vyas, N., Pareek, A., Black, A. W., & Tsvetkov, Y.
- Affiliations
- Carnegie Mellon University (US)
- Venue
- Proceedings of the First Workshop on Gender Bias in Natural Language Processing
- Published
- 2019
- Pages
- 166-172
- OpenAlex
- W2972413484
- Cited by
- 348
- References
- 31
- Citations by year
- 2026: 15 · 2025: 32 · 2024: 45 · 2023: 80 · 2022: 67 · 2021: 75 · 2020: 29 · 2019: 5
- FWCI
- 45.91
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Workshop on Gender Bias in Natural Language Processing (GeBNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation (1812327)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Keywords
- bias measurement, contextual word embeddings, BERT, social stereotypes, gender bias, racial bias, religious bias
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
