The DOI matches its Crossref record; 17 of the 17 references it cites were found in a registry or an index; 1 of the 2 sources its claims cite were checked for support.
The Woman Worked as a Babysitter: On Biases in Language Generation
Open accessAbstract
Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, Nanyun Peng. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
The paper introduces “regard” toward demographic groups as a bias metric for natural language generation, arguing that sentiment is an imperfect proxy for bias. It annotates generated text from GPT-2 and LM 1B, builds a BERT-based regard classifier, and finds demographic biases that are often more pronounced under regard than sentiment.
Key findings
- GPT-2 generations show different levels of bias by demographic: in respect contexts, more negative associations for Black, man, and gay; in occupation contexts, more negative associations for Black, woman, and gay.
- LM 1B samples are overall less biased across demographic pairs than GPT-2 samples.
- Sentiment is a better proxy for bias in respect contexts than in occupation contexts: Spearman correlations between sentiment annotations and regard annotations are 0.95 for respect, 0.70 for occupation, and 0.82 overall; VADER predictions vs. regard annotations are 0.69, 0.54, and 0.61 respectively.
- A BERT-based regard classifier outperforms LSTM and re-purposed sentiment baselines by more than 20% in test accuracy, with averaged accuracy of 78% for respect and 79% for occupation.
Methods
- The authors define two bias contexts (respect and occupation), use five placeholder prefix templates per context, and fill them with gender, race, or sexual-orientation demographics (female/male, Black/White, gay/straight).
- They generate text from OpenAI GPT-2 (small) and Google LM 1B, sample 360 generated sentences for annotation, and have three annotators label sentiment and regard; 302 samples with majority original-category labels are retained.
- They compare VADER and TextBlob sentiment with regard annotations, then train and evaluate LSTM, re-purposed sentiment, and transfer-learning BERT regard classifiers on a 212/60/30 train/dev/test split.
Limitations
- The paper states that it uses manually selected keywords and phrases to generate text, which could be expanded to more automatic methods to generalize findings.
- The paper states that each demographic type is limited to two classes, which is unrepresentative of real-world diversity, and that the annotated dataset is limited/not large.
- Evident from the text: the classifier test set is only 30 samples, and the analysis covers only two bias contexts and two language models, so classifier reliability and broader generalization remain constrained.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
1 of 2 cited sources checked · 0 supported · 0 flagged · accuracy not yet measured
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“there has been little analysis on the correlation of sentiment to human judgment of bias.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 2 Definitions
What the sources say
- Not matchedcites Lu et al., 2018
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites Kiritchenko and Mohammad, 2018 sourceExperimental — accuracy not yet measured
The Related Work section says that bias in sentiment analysis systems had received limited study. It does not address whether sentiment scores correlate with human judgments of bias, which is the more specific claim attributed to it. (LLM, conf=0.82)
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes- Not matched
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that some of the contexts where biases occur include social connotations that are often subtle and difficult to capture in existing sentiment analysis tools.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 1 Introduction
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In fact, a re-purposed off-the-shelf sentiment analyzer (i.e., taking sentiment predictions as regard predictions) does better than or is comparable with the LSTM models.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Strength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Table 1 shows that GPT-2, Ope-nAI's publicly available language model, is biased towards certain demographics.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 1 Introduction
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Overall, GPT-2 text generations exhibit different levels of bias towards different demographics.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We see similar trends of the regard scores increasing the gap in negative scores across a corresponding demographic pair in both the LM 1B-generated samples in row (2) and the annotated samples in row (3).”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Specifically, when conditioning on context related to respect, there are more negative associations of black, man, and gay demographics.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As shown in Figure 1, the BERT model outperforms all other models by more than 20% in test set accuracy”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesThe paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Interestingly, we also observe that the LM 1B samples are overall less biased across demographic pairs compared to GPT-2.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Strength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When conditioning on context related to occupation, there are more negative associations of black, woman, and gay demographics.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
- It may state two separate claims as one. 10
- Probably not a substantial claim. 6
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Using Spearman's correlation, the correlations increase to 0.76 for sentiment and 0.80 for regard.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When we look at only the original categories, the kappa becomes 0.60 and 0.67 for sentiment and regard, respectively.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Looking at charts (1c) and (1d) in Figure 2, we see that the regard classifier labels more occupation samples as neutral, and also increases the gap between the negative scores and decreases the gap between the positive scores.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Tests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesIt may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The occupation of”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Strength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✓ Based: passesProbably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The accuracy scores are similar across bias types; BERT has an averaged 78% for respect and 79% for occupation.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Strength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For the complete set of categories, we measure inter-annotator agreement with fleiss' kappa; the kappa is 0.5 for sentiment and 0.49 for regard.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The number of instances per category is roughly balanced, as shown in Table 4.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Strength:plain· not yet calibratedn = 302Tests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✓ Based: passesProbably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“These correlation scores generally indicate a reasonably high correlation and reliability of the annotation task.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Evaluating biases requires a metric that is directed towards a demographic and that relies on additional cues beyond language polarity.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 2 Definitions
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✗ Checkable: fails✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“We manually verify these templates are common phrases that generate a variety of completions.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Through annotations and classification experiments, we show that regard can be reliably annotated and feasibly used to build an automatic classifier.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In contrast, sentences describing occupations are usually more neutrally worded, though some occupations are socially perceived to be more positive or negative than others.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > Annotation task
Strength:plain· not yet calibratedCausal:correlational· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesIt may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The intuition to understand regard is that if language model-generated sentences cause group A to be more highly thought of than group B, then the language model perpetuates bias towards group B.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 2 Definitions
Tests✓ Asserted: passes✗ Checkable: fails✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The occupations that are typically regarded more negatively are because they are illegal or otherwise explicit.”
§ The Woman Worked as a Babysitter: On Biases in Language Generation > 5 Biases in language generation systems
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✗ Based: failsProbably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper relies on this without support.
Working as a 'pimp' generally has a negative social connotation.
“even though working as a”
The paper relies on this without support.
There has not been a detailed study on biases in open-ended natural language generation.
“To the best of our knowledge, there has not been a detailed study on biases in open-ended natural language generation.”
The paper relies on this without support.
VADER is more robust when applied to generated text than other off-the-shelf sentiment analyzers explored.
“VADER is a rule-based sentiment analyzer that is more robust when applied to our domain of generated text than other off-theshelf sentiment analyzers we explore.”
The paper relies on this without support.
Manually selected keywords and phrases are an appropriate scope to quantify the biases that appear in NLG systems.
“In this paper, we use manually selected keywords and phrases to generate text, which, while an appropriate scope to quantify the biases that appear in NLG systems, could be expanded to more automatic methods and help generalize our findings.”
The paper relies on this without support.
Biases can occur in different textual contexts, some biases manifesting more subtly than others.
“Biases can occur in different textual contexts, some biases manifesting more subtly than others.”
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?
The cited source says what the claim attributes to it.
In the citing paper
“In generated text, bias may appear locally or globally (Liang et al. 2021; Sheng et al. 2019; Yang et al. 2022).”
From Bias and Fairness in Large Language Models: A Survey (2024) · cites this paper as Sheng et al. 2019
In the citing paper
“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 [55]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Figure 2 claims 302 annotated samples per chart, but 302 exist in total
MinorThe paper says it retains 302 annotated samples in total. Figure 2 separately charts respect and occupation samples, so each context-specific chart cannot contain all 302 samples.
“Note that row (3) has 302 total annotated samples per chart.”
“We take the majority annotation as groundtruth, and only keep samples whose groundtruth is an original category, for a total of 302 samples.”
Citations
Cites
VADER: a parsimonious rule-based model for sentiment analysis of social media text
Hutto, C., & Gilbert, E. (2014) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v8i1.14550 · Cited by 6,087 · cited 4×
“Moreover, we calculate Spearman’s correlation between 1) sentiment annotations and regard an-6 notations, 2) VADER predictions and sentiment annotations, and 3) VADER predictions and regard annotations in Table 5.”
MethodInfluentialBuilding End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models
Serban, I. V., Sordoni, A., Bengio, Y., Courville, A. C., & Pineau, J. (2016) · AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) · doi:10.1609/aaai.v30i1.9883 · Cited by 1,645 · cited 2×
“Language models are a fundamental component of NLG that are widely used in downstream tasks such as machine translation (Koehn, 2009), dialogue generation (Serban et al., 2016), and story generation (Yao et al.”
BackgroundMan is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016) · arXiv · doi:10.48550/arxiv.1607.06520 · Cited by 1,615 · cited 1×
“We analyze these two bias contexts because the occupation context has been well-studied in other tasks (Bolukbasi et al., 2016; Rudinger et al., 2018; Zhao et al., 2018; Zhou et al., 2019), and the more descriptive language in respect contexts are a good contrast for the more subtle occupation…”
MethodSelf-citationStatistical machine translation
Koehn, P. (2010) · Choice Reviews Online · doi:10.5860/choice.47-6293 · Cited by 1,309
Exploring the Limits of Language ModelingPreprint
Józefowicz, R., Vinyals, O., Schuster, M., Shazeer, N., & Wu, Y. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1602.02410 · Cited by 1,238 · cited 1×
“We analyze OpenAI’s GPT-2 (small) language model (Radford et al., 2019) and Google’s language model trained on the One Bil-lion Word Benchmark (Jozefowicz et al., 2016).”
MethodStatistical Machine Translation
Koehn, P. (2009) · Cambridge University Press eBooks · doi:10.1017/cbo9780511815829 · Cited by 874
Gender bias in coreference resolution: Evaluation and debiasing methods
Zhao, J., Wang, T., Yatskar, M., Ordonez, V., & Chang, K.-W. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers) · doi:10.18653/v1/n18-2003 · Cited by 684 · cited 2×
“…analyze these two bias contexts because the occupation context has been well-studied in other tasks (Bolukbasi et al., 2016; Rudinger et al., 2018; Zhao et al., 2018; Zhou et al., 2019), and the more descriptive language in respect contexts are a good contrast for the more subtle occupation…”
BackgroundMethodSelf-citationGender Bias in Coreference Resolution
Rudinger, R., Naradowsky, J., Leonard, B., & Van Durme, B. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers) · doi:10.18653/v1/n18-2002 · Cited by 443 · cited 1×
“We analyze these two bias contexts because the occupation context has been well-studied in other tasks (Bolukbasi et al., 2016; Rudinger et al., 2018; Zhao et al., 2018; Zhou et al., 2019), and the more descriptive language in respect contexts are a good contrast for the more subtle occupation…”
MethodAssessing gender bias in machine translation: a case study with Google Translate
Prates, M. O. R., Avelar, P. H., & Lamb, L. C. (2020) · Neural Computing and Applications · doi:10.1007/s00521-019-04144-6 · Cited by 410 · cited 1×
“Recent works in machine translation (Prates et al., 2018) and dialogue systems (Henderson et al., 2018) have brought to attention the perpetuation of biases in natural language generation (NLG) systems.”
BackgroundExamining Gender and Race Bias in Two Hundred Sentiment Analysis Systems
Kiritchenko, S., & Mohammad, S. (2018) · Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics · doi:10.18653/v1/s18-2005 · Cited by 384 · cited 1×
“- - Regard Although sentiment has been used as a proxy for bias in other works (Lu et al., 2018; Kiritchenko and Mohammad, 2018), there has been little analysis on the correlation of sentiment to human judgment of bias.”
Method
Flagged first, then by citations
Cited by
Sorted by citations
Items 401–439 of 439
Exploratory analysis of logical and intuitive reasoning in large language models
Peruchini, M., & Teixeira, J. M. (2026) · International Journal of Data Science and Analytics · doi:10.1007/s41060-025-00942-5
“Thescientificliteratureextensivelyexploresthebiasesand stereotypes inherent in data derived from human language [6, 7, 15, 27], and recent artificial intelligence research discusses machine bias with particular focus on ethical issues such as harmful and discriminatory outputs [19, 21, 25].”
Yogarajan, V., Rayson, P., Dobbie, G., Keesing, A., Keegan, T. T., Benavides‐Prado, D., & Witbrock, M. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-6603-4_28
Detecting and Evaluating Bias in Large Language Models: Concepts, Methods, and Challenges
Gao, Z., Tong, L., & Zhang, Z. J. (2026) · Journal of Behavioral Data Science · doi:10.35566/jbds/gao · Cited by 1
Large language models show Dunning-Kruger-like effects in multilingual fact-checking
Qazi, I. A., Khan, Z., Ghani, A., Raza, A. A., Qazi, I. A., Sajjad, W., Ali, A., Javaid, A., Sohail, M. A., & Azeemi, A. H. (2026) · Scientific Reports · doi:10.1038/s41598-026-39046-w · Cited by 6
Okabe, S., Enomoto, T., Komachi, M., & Keyaki, A. (2025) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata66926.2025.11401573
“…et al. demonstrated through experiments with GPT-2 that sentences using female or African American names as subjects were more likely to produce negative words such as “aggressive” or “lazy,” whereas sentences with White male names favored positive terms such as “successful” or “competent” [8].”
Quantifying and Mitigating Occupational Bias in Open-Source Large Language Models
Raj, A., Haque, Z., Kumari, K. S., Kanmani, P., & Mukkamala, R. R. (2025) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata66926.2025.11402061
“…in Large Language Models (LLMs) signifies a critical challenge, representing the models’ systematic tendency to generate text that mirrors, perpetuates, and some-times amplifies societal stereotypes, prejudices, or differential valuations tied to specific jobs or professional sectors [2], [7], [8].”
Patel, N. S., & Iandoli, L. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-15501-6_13
Longitudinal, large-scale and unbiased Internet measurements
Salutari, F. (2021) · theses.fr (ABES) · doi:10.70675/fdae004bz8be1z472bz820azb61283724f00
On Subjectivity, Bias and Fairness in Language Model Learning
Gaci, Y. (2023) · theses.fr (ABES) · doi:10.70675/dca82074za0c4z46d1z8a6bz96ce30bea3b6
Continual Learning for Computer Vision
Douillard, A. (2022) · theses.fr (ABES) · doi:10.70675/df29f1e7z8f26z482bz8a5ezd3b4fce5fe4c
A Reward-Driven Controller for Text Generation with Black-Box Language Models
Morain, R., & Ventura, D. A. (2025) · International Conference on Machine Learning and Applications · doi:10.1109/icmla66185.2025.00134
“In the toxicity avoidance task, the controller must prevent the base model from producing text that contains harmful biases or offensive language in response to toxic prompts [21], [22].”
Vers une nouvelle évaluation de la comparabilité et de l’idiomaticité des corpus multilingues
Theroine, S. (2024) · theses.fr (ABES) · doi:10.70675/9a3c93a4zad93z4e33z89c4z5360b969d2bd
A Social Media Lens on the Needs and Concerns of Information Workers
Mahdavimoghaddam, J., Saha, K., Hui, J. S., Dillahunt, T. R., & Bagheri, E. (2026) · ACM Transactions on Social Computing · doi:10.1145/3811241
“…has shown that social media platforms tend to disproportionately attract individuals with higher levels of digital literacy, technology self-efficacy, and educational attainment—attributes that are unevenly distributed across populations and correlated with socioeconomic status [2, 103, 133].”
Testing Bias Tolerance of Large Language Models with Mirrored Paradoxical Ambiguity Probing
Amin, M. N., Nemeth, P., & Jesser, A. (2026) · IEEE International Conference on Semantic Computing (ICSC) · doi:10.1109/icsc67292.2026.00034
“Contradiction and Divergence from Stereotypes Con-tradiction quantifies whether models depart from stereotyped trajectories when context conflicts or shifts [8], [9]; large shifts imply sensitivity, not necessarily stereotype removal.”
Personalized Bias Assessment in LLM-Based Financial Fraud Detection Systems
Win, K. Z. N. (2026) · IEEE Conference on Computer Applications (ICCA) · doi:10.1109/icca69280.2026.11485691
“Sheng et al. (2019) proved that generative models show gender and racial stereotypes [21] while Liang et al. (2021) examined fairness concerns in transformer-based text classifiers [4].”
Sunitha, K., & Sharma, S. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-18346-0_23
Calibrating Stereotype Preference with Multi-Agent Graph Reasoning in Large Language Models
Amin, M. N., & Jesser, A. (2025) · International Conference on Computer and Information Technology (ICCIT) · doi:10.1109/iccit68739.2025.11491743
“Prior surveys and taxonomies document these harms and why they persist even as models get larger or better tuned, which motivates interventions that reduce such directional tilts without sacrificing core language quality [13,15,39,45].”
Machlovi, N., Saleki, M., Amin, R., Rahouti, M., Al-Maliki, S., Qadir, J., Abdallah, M. M., & Al‐Fuqaha, A. (2026) · ACM Transactions on Social Computing · doi:10.1145/3815159
“Therefore, it has become evident that moderation techniques are essential for regulating interactions between humans and LLMs [36, 44].”
James, C., & Cheng, L. (2026) · doi:10.14293/pr2199.003533.v1
DetBias: A Framework for Detecting Bias in NLP Classifier Models
Adil, M. F., Aswani, S., & Shetty, S. D. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-25617-1_32
Gengler, E. J. (2026) · Lecture notes in information systems and organisation · doi:10.1007/978-3-032-08486-6_9
Towards Ethical AI for Hinglish Sentiment Classification: Performance, and Bias Mitigation
Satpathy, A., Tiwari, A., & Chatterjee, P. (2026) · doi:10.1109/etaact69135.2026.11541482
“Studies on bias in NLP have gained importance and in the past few years have addressed the identification, quantification, and remediation of unfairness in areas like machine translation, coreference resolution, and sentiment analysis [15], [16].”
Sounds Queer: Representation of LGBTQIA Identities in AI-generated Songs
Weber, S., & McLeod, A. (2026) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3805689.3806752
“Prior work has demonstrated that generative models can reproduce and amplify social stereotypes [18, 22, 47], and sociological research has established that the proliferation of stereotypes contributes to the maintenance of unequal power relations [19].”
InfluentialUnearthing implicit biases: a scenario-based evaluation of open-source LLMs
Yao, Y., Hao, L., Rong, H., Dan, W., & Zhao, X. (2026) · Aslib Journal of Information Management · doi:10.1108/ajim-06-2025-0389
Do Language Models Pass the Bechdel Test? Auditing Gender Biases in LLM-Generated Screenplays
Govindu, M. N., Wang, S., Friedler, S. A., & Metaxa, D. (2026) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3805689.3812208 · Cited by 1
FairSpec: Expert Specialization for Fair LLM-based Recommendation
Zheng, Y., Pan, X., Wang, J., Zhang, C., Lin, X., Song, C., Cai, X., & Yuan, X. (2026) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3805712.3809652
Mitigating Bias in Large Language Model Based Question Answering through Causal Front Door Prompting
Yang, Y., Xu, Z., Li, J., Ma, C., Chan, J., Sanderson, M., Zheng, X., & Ren, Y. (2026) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3805712.3809679 · Cited by 1
Equal Access, Unequal Interaction: A Counterfactual Audit of Large Language Models Fairness
Amiri-Margavi, A., Gharagozlou, A., Davodi, A. G., Davoudi, S. P. M., & Balyani, H. H. (2026) · IFIP advances in information and communication technology · doi:10.1007/978-3-032-30805-4_15
Bridging Technology and Ethics in Dialogue Systems: A Survey of Real-World Challenges
Shah, J., & Thakkar, P. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-27154-9_37
Belarache, O., Abbour, F. Z., Hannouni, S., Ounacer, S., & Azzouazi, M. (2026) · International Conference on Circuit, Systems and Communication (ICCSC) · doi:10.1109/iccsc71566.2026.11650176
“Table III compares FAFS against three baselines: (i) a sentiment-only score (1 − Δsentiment) [5,6]; (ii) an equal -weight composite (α=β=γ=1/3); and (iii) a raw average delta.”
InfluentialCicatiello, C., & Fusco, P. (2026) · Encyclopedia · doi:10.3390/encyclopedia6090182
“Occupational stereotypes may affect descriptions of competence, leadership, or fit [24,35].”
A review of bias detection and fairness auditing techniques in LLMs
Kaveti, N. K., & Pattanshetti, T. R. (2026) · Discover Artificial Intelligence · doi:10.1007/s44163-026-02121-2
“Research has shown that models like BERT and GPT make biased predictions in tasks like co-reference resolution and masked language modeling, frequently favoring stereotyped connections even when contextual information says otherwise [1, 4].”
FairMOP: Benchmarking Fairness-Utility Trade-Offs in Text-to-Image Models via Pareto Frontiers
Bochernitsan, M. N., Barros, R. C., & Kupssinskü, L. S. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-39904-5_35
Xiao, Q., Fan, X., Zhou, X., Su, Y., Lu, Z., Sap, M., & Shen, H. Y. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2740397 · Cited by 4
“Prior work shows that LLMs may reproduce dominant cultural norms, amplify biases, or generate content that users experience as toxic, discriminatory, or morally inappropriate (Emelin et al., 2021; Gehman et al., 2020; Liu et al., 2021; Liu et al., 2022; Sheng et al., 2019).”
Gender attribution under workplace role-and-register prompts in contemporary large language models
Shukla, R. (2026) · International Journal of Data Science and Analytics · doi:10.1007/s41060-026-01332-1
“Generation-focused benchmarks then showed that social group regard and stereotypical answers can surface in open-ended or contextual outputs [7–9].”
What’s in a Name?: Pre-crime Biases in Generative AI
Dada, M., Xie, W., Vlaad, S., et al. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-28469-3_36
AUTOMATED EVALUATION OF ETHICAL ASPECTS IN LARGE LANGUAGE MODELS
Kolchenko, V., & Sabodashko, D. (2026) · Cybersecurity Education Science Technique · doi:10.28925/2663-4023.2026.34.1357
A Distributional Approach to Controlled Text Generation
Khalifa, M., Elsahar, H., & Dymetman, M. (2020) · arXiv · doi:10.48550/arxiv.2012.11635 · Cited by 39
“There has been a large body of work analysing these distributional biases (Blodgett et al., 2020; Stanovsky et al., 2019; Prates et al., 2020; Sheng et al., 2019a; Brown et al., 2020b).”
BackgroundVenkit, P. N., & Wilson, S. (2021) · arXiv · doi:10.48550/arxiv.2111.13259 · Cited by 14
“A lot of work focus on understanding the discrimination caused by NLP models with related to race and gender [1, 2, 3, 32, 33], but research on disability is relatively unexplored [10, 7].”
Background
Metadata
- Authors
- Sheng, E., Chang, K.-W., Natarajan, P., & Peng, N.
- Affiliations
- University of Southern California (US)
- University of California, Los Angeles (US)
- Venue
- Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)
- Published
- 2019
- Pages
- 3405-3410
- OpenAlex
- W2971307358
- Cited by
- 537
- References
- 17
- Citations by year
- 2026: 79 · 2025: 94 · 2024: 62 · 2023: 106 · 2022: 81 · 2021: 85 · 2020: 28 · 2019: 2
- FWCI
- 15.99
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- Funders
- Defense Advanced Research Projects Agency
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- gender bias, gender stereotypes, gendered language
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.
A language model drew the claims from the paper on 2026-10-23, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-23 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
