Home

Awesome

Bias in NLP

This is a collection of natural language processing papers that deal with bias (mostly gender bias). The list is by no means complete and is just a way to keep up with the large amount of papers in that area. If you miss a paper, please add it.

Papers

Towards Detection of Subjective Bias using Contextualized Word Embeddings
WebConf2020 - Paper, Code
Note: Wikineutrality Corpus.

Joint Multiclass Debiasing of Word Embeddings
ISMIS2020 - Paper, Code
Note: Hard and Soft WEAT

Towards Debiasing Sentence Representations
ACL2020 - Paper, Code
Note: Sentence-level debiasing. Difference between pretraining and finetuning.

Neutralizing Gender Bias in Word Embedding with Latent Disentanglement and Counterfactual Generation
arxiv2020 - Paper
Note: Counterfactual generation.

Unsupervised Discovery of Implicit Gender Bias
arxiv2020 - Paper, Code
Note: Unsupervised bias detection from comments.

StereoSet: Measuring stereotypical bias in pretrained language models
arxiv2020 - Paper, Code
Note: Benchmark and Dataset for measuring bias in 4 domains (gender, profession, race, religion).

Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation
ACL2020 - Paper, Code
Note: Double Hard Debias: mitigigate dataset and then do debiasing

Gender Bias in Multilingual Embeddings and Cross-Lingual Transfer
ACL2020 - Paper
Note: Bias in multilingual embeddings depends on the alignment direction.

Scalable Cross Lingual Pivots to Model Pronoun Gender for Translation
arxiv2020 - Paper
Note: Gender labels for pronouns in MT English-Spanish.

Detecting Emergent Intersectional Biases: Contextualized Word Embeddings Contain a Distribution of Human-like Biases
arxiv2020 - Paper
Note: CEAT

OSCaR: Orthogonal Subspace Correction and Rectification of Biases in Word Embeddings
arxiv2020 - Paper
Note: Preserve semantic meaning of embeddings.

Investigating Gender Bias in BERT
arxiv2020 - Paper
Note: Identify one gender direction per BERT layer.

Type B Reflexivization as an Unambiguous Testbed for Multilingual Multi-Task Gender Bias
arxiv2020 - Paper, Code
Note: Multilingual multitask dataset across 4 languages.

Towards Debiasing NLU Models from Unknown Biases
arxiv2020 - Paper, Code
Note: Unsupervised bias detection.

Robustness and Reliability of Gender Bias Assessment in Word Embeddings: The Role of Base Pairs
arxiv2020 - Paper, Code
Note: Choice of base pairs is relevant.

LOGAN: Local Group Bias Detection by Clustering
arxiv2020 - Paper
Note: Identify biases through clustering.

Exploring the Linear Subspace Hypothesis in Gender Bias Mitigation
arxiv2020 - Paper
Note: Verify whether non-linear debiasing helps. It seems not.

Unmasking Contextual Stereotypes: Measuring and Mitigating BERT’s Gender Bias
GeBNLP2020 - Paper, Code
Note: Verify gender debiasing techniques in German.

Language (Technology) is Power: A Critical Survey of “Bias” in NLP
arxiv2020 - Paper
Note: Metastudy: survey of 146 gender bias papers

Pick a Fight or Bite your Tongue: Investigation of Gender Differences in Idiomatic Language Usage
arxiv2020 - Paper
Note: Idiomatic expressions depending on the speaker.

Evaluating Bias In Dutch Word Embeddings
GeBNLP2020 - Paper, Code
Note: Examining bias in Dutch (using WEAT)

Analyzing Gender Bias within Narrative Tropes
arxiv2020 - Paper, Code
Note: Analyze bias using tropes

Neural Machine Translation Doesn’t Translate Gender Coreference Right Unless You Make It
GeBNLP2020 - Paper, Code
Note: Incorporate explicit word-level gender tags.

The Gap on GAP: Tackling the Problem of Differing Data Distributions in Bias-Measuring Datasets
NeurIPS 2020 - Paper, Code
Note: Distances in GAP play a role.

AraWEAT: Multidimensional Analysis of Biases in Arabic Word Embeddings
arxiv2020 - Paper, Code
Note: Arabic WEAT.

Characterising Bias in Compressed Models
arxiv2020 - Paper
Note: Bias in compressed model is large. Provide method to identify biased examples.

Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them
NAACL2019 - Paper, Code
Note: Debiasing by setting dimensions to zero ist not effective

Equalizing Gender Bias in Neural Machine Translation with Word Embeddings Techniques
GeBNLP 2019 - Paper
Note: Spanisch-Englisch translation with occupations.

Evaluating the Underlying Gender Bias in Contextualized Word Embeddings
GeBNLP 2019 - Paper
Note: Cointextualized embeddings are less biased than static ones.

Mitigating Gender Bias in Natural Language Processing: Literature Review
ACL2019 - Paper
Note: Survey

What's in a Name? Reducing Bias in Bios without Access to Protected Attributes
NAACL2019 - Paper
Note: Work on biographies.

Assessing Social and Intersectional Biases in Contextualized Word Representations
NeurIPS2019 - Paper
Note: Strong bias in contextualized embeddings. Bias not always visible on sentence level.

It’s All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution
EMNLP2019 - Paper
Note: Counterfactual Data Substitution (CDS)

Good Secretaries, Bad Truck Drivers? Occupational Gender Stereotypes in Sentiment Analysis
GeBNLP 2019 - Paper, Code
Note: Dataset of 800 sentences analysed with sentiment analysis.

Automatic Gender Identification and Reinflection in Arabic
GeBNLP 2019 - Paper
Note: Arabic English Translation with focus on getting the pronouns right.

Gendered Ambiguous Pronouns Shared Task: Boosting Model Confidence by Evidence Pooling
GeBNLP 2019 - Paper, Code
Note: Shared task winner GAP

Gendered Ambiguous Pronouns (GAP) Shared Task at the Gender Bias in NLP Workshop 2019
GeBNLP 2019 - Paper, Code
Note: GAP shared task description

Conceptor Debiasing of Word Representations Evaluated on WEAT
GeBNLP 2019 - Paper
Note: Proposes Conceptor Debiasing.

On Measuring Gender Bias in Translation of Gender-neutral Pronouns
GeBNLP 2019 - Paper, Code
Note: Gender bias in pronoun translation Korean-English

Measuring Gender Bias in Word Embeddings across Domains and Discovering New Gender Bias Word Categories
GeBNLP 2019 - Paper, Code
Note: Clustering method for discovering new biases.

The Role of Protected Class Word Lists in Bias Identification of Contextualized Word Representations
GeBNLP 2019 - Paper
Note: Uses conceptor debiasing

The Woman Worked as a Babysitter: On Biases in Language Generation
EMNLP2019 - Paper, Code
Note: Regard and Sentiment. Annotations released.

Exploring Human Gender Stereotypes with Word Association Test
EMNLP2019 - Paper, Code
Note: Word association graphs

Gender-preserving Debiasing for Pre-trained Word Embeddings
ACL2019 - Paper, Code
Note: Differentiate between bias and gender information.

Quantifying Social Biases in Contextual Word Representations
GeBNLP 2019 - Paper
Note: Template based method to quantify bias.

Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting
ACM Fat 2019 - Paper
Note: Analyze effects of bias.

Gender Bias in Neural Natural Language Processing
Logic, Language, and Security. Springer. 2018 - Paper
Note: Counterfactual Data Augmentation (CDA). Clear definition of Bias. Evaluates on coreference resolution and language modelling.

Gender Bias in Coreference Resolution
NAACL2018 - Paper, Code
Note: Windogender schemes.

Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
NeurIPS2016 - Paper, Code
Note: Among the first to address gender bias

Rejecting the Gender Binary: A Vector-Space Operation.
2015 - Paper
Note: Blog post: first to propose to remove gender dimension

TODOS