Home › Categories › Business & Finance› Enforcement

Enforcement.pdf

sap2019risk.pdf

🔗 Source: homes.cs.washington.edu

📊 Size: 4.59 MB

👤 Author: Maarten Sap ; Dallas Card ; Saadia Gabriel ; Yejin Choi ; Noah A. Smith

⬇️ Downloads: 100

Summary

Researchers investigated racial bias in hate speech detection models, finding that annotators' insensitivity to African American English (AAE) dialect differences leads to biased models that amplify harm against minority populations. They uncovered correlations between AAE surface markers and toxicity ratings in widely-used datasets, and showed that models trained on these datasets acquire and propagate these biases, labeling AAE tweets and tweets from self-identified African Americans as offensive up to two times more often. To mitigate this bias, they proposed dialect and race priming, finding that annotators are less likely to label AAE tweets as offensive when made aware of the dialect. The study highlights the need to consider dialect and social identity when developing toxic language detection tools to avoid unintended negative impacts.

Download File

Description

Researchers investigated racial bias in hate speech detection models, finding that annotators' insensitivity to African American English (AAE) dialect...

Technical Information

File Format: PDF
File Size: 4.59 MB
Pages: 11
Language: EN
Author: Maarten Sap ; Dallas Card ; Saadia Gabriel ; Yejin Choi ; Noah A. Smith
Total Downloads: 100
Last Updated: 1 week ago

Document Overview

This PDF document about enforcement provides comprehensive information and guidance. Whether you're a beginner or advanced user, this resource offers valuable insights into enforcement.