Maarten Sap

Risk of Racial Bias in Hate Speech Detection

There are three files:

Founta dataset with dialect extracted

founta_all_dial.csv

Davidson dataset with dialect extracted

davidson_dial.csv

Mturk re-annotations with no/dialect/race priming:

Collected with this MTurk template, sap2019risk_mTurkExperiment.csv contains the annotations from our pilot study, with the following columns:

• annotator demographic information: annotatorAge, annotatorGender, annotatorMinority, annotatorPolitics, annotatorRace
  • since they had to enter it every time, they might not have entered the exact same data every time – I recommend taking the most frequent value
  • note the “decline to answer” value for age was 100, so remove those when computing demographic stats!
• main questions about the offensiveness of the post: intentYN, offensive2anyoneYN, offensive2youYN
  • intent had 4 possible values, the offensiveYN questions had 3 (not counting “don’t understand”)
• additional race/dialect flagging: dialectIsWrong, raceIsWrong
  • simple checkbox answer to see whether workers disagreed with the dialect/race inferred by SuLin’s model
• WorkerIdHashed
• text of the post: tweet
• experimental condition: condition
• condition specific text shows to users: dialect, username
• original data labels: davidson_label,founta_label

Citations

• Racial bias paper: Maarten Sap, Dallas Card, Saadia Gabriel, Yejin Choi & Noah A Smith (2019). The Risk of Racial Bias in Hate Speech Detection. ACL
• Dialect extraction: Su Lin Blodgett, Lisa Green, and Brendan O’Connor. 2016. Demographic dialectal variation in social media: A case study of African-American English. In EMNLP.
• Davidson dataset: Thomas Davidson, Dana Warmsley, Michael W. Macy, and Ingmar Weber. 2017. Automated hate speech detection and the problem of offensive language. In ICWSM.
• Founta dataset: Antigoni-Maria Founta, Constantinos Djouvas, Despoina Chatzakou, Ilias Leontiadis, Jeremy Blackburn, Gianluca Stringhini, Athena Vakali, Michael Sirivianos, and Nicolas Kourtellis. 2018. Large scale crowdsourcing and characterization of twitter abusive behavior. In ICWSM.

Annotators with Attitudes

Download annotated data: annWithAttitudes.tgz

Qual file: annWithAttitudes-Qual.html

Large-scale question: annWithAttitudes-LargeScale.html

Citation

Maarten Sap, Swabha Swayamdipta, Laura Vianna, Xuhui Zhou, Yejin Choi & Noah A. Smith (2022) Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detection. NAACL.