Skip to content
conftrace_
2018 NAACL NAACL 2018

Gender Bias in Coreference Resolution

Abstract

AbstractWe present an empirical study of gender bias in coreference resolution systems. We first introduce a novel, Winograd schema-style set of minimal pair sentences that differ only by pronoun gender. With these “Winogender schemas,” we evaluate and confirm systematic gender bias in three publicly-available coreference resolution systems, and correlate this bias with real-world and textual gender statistics.

🌉 Interdisciplinary Bridge - Artificial Intelligence and Natural Language Processing
🧭 Keyword Pioneer - bias evaluation
🐣 Hot Topic Early Bird - gender bia

Authors

Rachel Rudinger, Jason Naradowsky, Brian Leonard, Benjamin Van Durme

Topics

Machine Learning > Application Areas > Fairness Natural Language Processing > Understanding > Coreference Resolution Natural Language Processing > Applications > Natural Language Inference Artificial Intelligence > Core AI > Fairness

Keywords

natural language processing coreference resolution empirical evaluation gender bia bias evaluation minimal pair winograd schema
Download PDF

More from NAACL 2018

All NAACL 2018 papers →
Neural Text Generation in Stories Using Entity Representations as Context
Recurrent Neural Networks as Weighted Language Recognizers
Enhanced Word Representations for Bridging Anaphora Resolution
Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods
Integrating Stance Detection and Fact Checking in a Unified Corpus
Is Something Better than Nothing? Automatically Predicting Stance-based Arguments Using Deep Learning and Small Labelled Dataset