On Measuring and Mitigating Biased Inferences of Word Embeddings

Word embeddings carry stereotypical connotations from the text they are trained on, which can lead to invalid inferences in downstream models that rely on them. We use this observation to design a mechanism for measuring stereotypes using the task of natural language inference. We demonstrate a reduction in invalid inferences via bias mitigation strategies on static word embeddings (GloVe). Further, we show that for gender bias, these techniques extend to contextualized embeddings when applied selectively only to the static components of contextualized embeddings (ELMo, BERT).

• Publication year

  2019

• Venue

  AAAI Conference on Artificial Intelligence

• Publication date

  2019-08-25

• Fields of study

  Computer Science

• Identifiers
  DOI 10.1609/AAAI.V34I05.6267arXiv 1908.09369
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

  2019cited by this paper
• Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them

  2019influential reference
• Gender Bias in Contextualized Word Embeddings

  2019cited by this paper
• Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings

  2019cited by this paper
• Attenuating Bias in Word Vectors

  2019influential reference
• A Transparent Framework for Evaluating Unintended Demographic Bias in Word Embeddings

  2019cited by this paper
• Learning Gender-Neutral Word Embeddings

  2018cited by this paper
• Gender Bias in Coreference Resolution

  2018cited by this paper
• Deep Contextualized Word Representations

  2018cited by this paper
• Annotation Artifacts in Natural Language Inference Data

  2018cited by this paper
• Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints

  2017cited by this paper
• Social Bias in Elicited Natural Language Inferences

  2017cited by this paper
• Faculty Opinions recommendation of Semantics derived automatically from language corpora contain human-like biases.

  2017cited by this paper
• A Decomposable Attention Model for Natural Language Inference

  2016influential reference
• Long Short-Term Memory-Networks for Machine Reading

  2016cited by this paper
• Semantics derived automatically from language corpora contain human-like biases

  2016cited by this paper
• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings

  2016influential reference
• Bidirectional Attention Flow for Machine Comprehension

  2016cited by this paper
• Recognizing Textual Entailment: Models and Applications Ido Dagan1, Dan Roth2, Mark Sammons2, and Fabio Massimo Zanzotto3 (1Bar-Ilan University, Israel, 2University of Illinois, Urbana, IL, and 3University of Rome “Tor Vergata,” Italy) Morgan & Claypool (Synthesis Lectures on Human Language Technolo

  2015cited by this paper
• A large annotated corpus for learning natural language inference

  2015cited by this paper
• GloVe: Global Vectors for Word Representation

  2014cited by this paper
• Distributed Representations of Words and Phrases and their Compositionality

  2013cited by this paper
• The PASCAL Recognising Textual Entailment Challenge

  2005cited by this paper

• IndicFairFace: Balanced Indian Face Dataset for Auditing and Mitigating Geographical Bias in Vision-Language Models

  2026cites this paper
• DeFrame: Debiasing Large Language Models Against Framing Effects

  2026cites this paper
• SAFARI: A Community-Engaged Approach and Dataset of Stereotype Resources in the Sub-Saharan African Context

  2026influential citation
• Bi-directional Bias Attribution: Debiasing Large Language Models without Modifying Prompts

  2026cites this paper
• A Unified Framework to Quantify Cultural Intelligence of AI

  2026cites this paper
• Detecting implicit biases of large language models with Bayesian hypothesis testing

  2025cites this paper
• Improving ethical sensitivity for ethical decision-making in conversational artificial intelligence

  2025cites this paper
• Analyzing the Safety of Japanese Large Language Models in Stereotype-Triggering Prompts

  2025cites this paper
• Language Models Predict Empathy Gaps Between Social In-groups and Out-groups

  2025cites this paper
• Structured Reasoning for Fairness: A Multi-Agent Approach to Bias Detection in Textual Data

  2025cites this paper
• Do LLMs exhibit demographic parity in responses to queries about Human Rights?

  2025cites this paper
• Textual Entailment is not a Better Bias Metric than Token Probability

  2025influential citation
• Exploring and Mitigating Gender Bias in Encoder-Based Transformer Models

  2025cites this paper
• Bias is a Math Problem, AI Bias is a Technical Problem: 10-year Literature Review of AI/LLM Bias Research Reveals Narrow [Gender-Centric] Conceptions of 'Bias', and Academia-Industry Gap

  2025cites this paper
• Investigating Intersectional Bias in Large Language Models using Confidence Disparities in Coreference Resolution

  2025cites this paper
• I Think, Therefore I Am Under-Qualified? A Benchmark for Evaluating Linguistic Shibboleth Detection in LLM Hiring Evaluations

  2025cites this paper
• EsCorpiusBias: The Contextual Annotation and Transformer-Based Detection of Racism and Sexism in Spanish Dialogue

  2025cites this paper
• Towards Fair Rankings: Leveraging LLMs for Gender Bias Detection and Measurement

  2025cites this paper
• Understanding Gender Bias in AI-Generated Product Descriptions

  2025cites this paper
• Can LLMs Solve My Grandma's Riddle? Evaluating Multilingual Large Language Models on Reasoning Traditional Bangla Tricky Riddles

  2025cites this paper
• Benchmarking and Pushing the Multi-Bias Elimination Boundary of LLMs via Causal Effect Estimation-guided Debiasing

  2025cites this paper
• Evaluate Bias without Manual Test Sets: A Concept Representation Perspective for LLMs

  2025cites this paper
• Assumed Identities: Quantifying Gender Bias in Machine Translation of Ambiguous Occupational Terms

  2025cites this paper
• PolBiX: Detecting LLMs' Political Bias in Fact-Checking through X-phemisms

  2025cites this paper
• Language Models That Walk the Talk: A Framework for Formal Fairness Certificates

  2025cites this paper
• GenderBench: Evaluation Suite for Gender Biases in LLMs

  2025cites this paper
• SHADES: Towards a Multilingual Assessment of Stereotypes in Large Language Models

  2025influential citation
• Exploring AI-Based Support in Speech-Language Pathology for Culturally and Linguistically Diverse Children

  2025cites this paper
• Deconstructing the ethics of large language models from long-standing issues to new-emerging dilemmas: a survey

  2024cites this paper
• Ask LLMs Directly, "What shapes your bias?": Measuring Social Bias in Large Language Models

  2024cites this paper
• ABLE: Agency-BeLiefs Embedding to Address Stereotypical Bias through Awareness Instead of Obliviousness

  2024cites this paper
• Recognizing Value Resonance with Resonance-Tuned RoBERTa Task Definition, Experimental Validation, and Robust Modeling

  2024cites this paper
• Large Language Model Bias Mitigation from the Perspective of Knowledge Editing

  2024cites this paper
• Hire Me or Not? Examining Language Model's Behavior with Occupation Attributes

  2024cites this paper
• “They only care to show us the wheelchair”: disability representation in text-to-image AI models

  2024cites this paper
• Believing Anthropomorphism: Examining the Role of Anthropomorphic Cues on Trust in Large Language Models

  2024cites this paper
• GeniL: A Multilingual Dataset on Generalizing Language

  2024cites this paper
• Fairness in Large Language Models: A Taxonomic Survey

  2024cites this paper
• Projective Methods for Mitigating Gender Bias in Pre-trained Language Models

  2024cites this paper
• Detecting Bias in Large Language Models: Fine-tuned KcBERT

  2024cites this paper
• Take Care of Your Prompt Bias! Investigating and Mitigating Prompt Bias in Factual Knowledge Extraction

  2024cites this paper
• Uncovering Stereotypes in Large Language Models: A Task Complexity-based Approach

  2024cites this paper
• SeeGULL Multilingual: a Dataset of Geo-Culturally Situated Stereotypes

  2024cites this paper
• “Flex Tape Can’t Fix That”: Bias and Misinformation in Edited Language Models

  2024cites this paper
• Improved BTM topic embedding method for Web text data extraction

  2024cites this paper
• A Note on Bias to Complete

  2024cites this paper
• From Prejudice to Parity: A New Approach to Debiasing Large Language Model Word Embeddings

  2024cites this paper
• Evaluating Unsupervised Dimensionality Reduction Methods for Pretrained Sentence Embeddings

  2024influential citation
• Deconstructing The Ethics of Large Language Models from Long-standing Issues to New-emerging Dilemmas

  2024cites this paper
• La recherche sur les biais dans les modèles de langue est biaisée : état de l’art en abyme [Bias Research for Language Models is Biased: a Survey for Deconstructing Bias in Large Language Models]

  2024cites this paper
• BenchmarkCards: Large Language Model and Risk Reporting

  2024cites this paper
• Measuring Machine Learning Harms from Stereotypes Requires Understanding Who Is Harmed by Which Errors in What Ways

  2024cites this paper
• MAFIA: Multi-Adapter Fused Inclusive Language Models

  2024cites this paper
• Stereohoax: a multilingual corpus of racial hoaxes and social media reactions annotated for stereotypes

  2024cites this paper
• Bias in Large Language Models: Origin, Evaluation, and Mitigation

  2024cites this paper
• BiasWipe: Mitigating Unintended Bias in Text Classifiers through Model Interpretability

  2024cites this paper
• Large Language Models Still Exhibit Bias in Long Text

  2024cites this paper
• LLMs are Biased Teachers: Evaluating LLM Bias in Personalized Education

  2024cites this paper
• BenchmarkCards: Standardized Documentation for Large Language Model Benchmarks

  2024cites this paper
• Measuring machine learning harms from stereotypes: requires understanding who is being harmed by which errors in what ways

  2024cites this paper
• Collapsed Language Models Promote Fairness

  2024cites this paper
• Extracting Features from Text Flows based on Semantic Similarity for Text Classification: an Approach Inspired by Audio Analysis

  2024cites this paper
• Leveraging Temporal Trends for Training Contextual Word Embeddings to Address Bias in Biomedical Applications: Development Study

  2024cites this paper
• Pro-Woman, Anti-Man? Identifying Gender Bias in Stance Detection

  2024cites this paper
• Debiasing Large Language Models with Structured Knowledge

  2024cites this paper
• Identifying and Mitigating Social Bias Knowledge in Language Models

  2024cites this paper
• Towards trustworthy LLMs: a review on debiasing and dehallucinating in large language models

  2024cites this paper
• Interpretations, Representations, and Stereotypes of Caste within Text-to-Image Generators

  2024cites this paper
• Fairness Definitions in Language Models Explained

  2024influential citation
• Do Generative AI Models Output Harm while Representing Non-Western Cultures: Evidence from A Community-Centered Approach

  2024influential citation
• Exploring Changes in Nation Perception with Nationality-Assigned Personas in LLMs

  2024cites this paper
• Evaluating Short-Term Temporal Fluctuations of Social Biases in Social Media Data and Masked Language Models

  2024cites this paper
• Exploring Safety-Utility Trade-Offs in Personalized Language Models

  2024cites this paper
• GECOBench: a gender-controlled text dataset and benchmark for quantifying biases in explanations

  2024cites this paper
• Debiasing with Sufficient Projection: A General Theoretical Framework for Vector Representations

  2024cites this paper
• A Survey on Fairness in Large Language Models

  2023cites this paper
• Nationality Bias in Text Generation

  2023cites this paper
• Counter-GAP: Counterfactual Bias Evaluation through Gendered Ambiguous Pronouns

  2023cites this paper
• BiasTestGPT: Using ChatGPT for Social Bias Testing of Language Models

  2023cites this paper
• A machine learning approach to identifying suicide risk among text-based crisis counseling encounters

  2023cites this paper
• A hunt for the Snark: Annotator Diversity in Data Practices

  2023cites this paper
• A Multilingual Dataset of Racial Stereotypes in Social Media Conversational Threads

  2023cites this paper
• Building Stereotype Repositories with Complementary Approaches for Scale and Depth

  2023cites this paper
• VERB: Visualizing and Interpreting Bias Mitigation Techniques Geometrically for Word Representations

  2023influential citation
• Large Language Model Programs

  2023cites this paper
• Biases in Large Language Models: Origins, Inventory, and Discussion

  2023cites this paper
• SeeGULL: A Stereotype Benchmark with Broad Geo-Cultural Coverage Leveraging Generative Models

  2023cites this paper
• Keeping Up with the Language Models: Robustness-Bias Interplay in NLI Data and Models

  2023cites this paper
• "I wouldn't say offensive but...": Disability-Centered Perspectives on Large Language Models

  2023cites this paper
• Sociodemographic Bias in Language Models: A Survey and Forward Path

  2023cites this paper
• Visual Exploration of Indirect Bias in Language Models

  2023cites this paper
• Prompt Tuning Pushes Farther, Contrastive Learning Pulls Closer: A Two-Stage Approach to Mitigate Social Biases

  2023influential citation
• Adjective Scale Probe: Can Language Models Encode Formal Semantics Information?

  2023cites this paper
• Unlearning Bias in Language Models by Partitioning Gradients

  2023cites this paper
• Gender-tuning: Empowering Fine-tuning for Debiasing Pre-trained Language Models

  2023cites this paper
• Mitigating Unintended Bias in Masked Language Models

  2023cites this paper
• Building Socio-culturally Inclusive Stereotype Resources with Community Engagement

  2023cites this paper
• Curriculum learning and evolutionary optimization into deep learning for text classification

  2023cites this paper
• CALM : A Multi-task Benchmark for Comprehensive Assessment of Language Model Bias

  2023influential citation
• Bias and Fairness in Large Language Models: A Survey

  2023influential citation