TY - JOUR
AU - 
AB - REDDITBIAS: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Models 1 1 2 1 Soumya Barikeri, Anne Lauscher, Ivan Vulic, ´ and Goran Glavas ˇ Data and Web Science Research Group University of Mannheim soumyabarikeri@gmail.com,fanne, gorang@informatik.uni-mannheim.de Language Technology Lab University of Cambridge iv250@cam.ac.uk Abstract decessors (Bolukbasi et al., 2016; Caliskan et al., 2017; Dev and Phillips, 2019; Gonen and Gold- Text representation models are prone to exhibit berg, 2019; Lauscher et al., 2020a, inter alia). Hav- a range of societal biases, reflecting the non- ing models that capture or even amplify human controlled and biased nature of the underlying biases brings about further ethical challenges to the pretraining data, which consequently leads to society (Henderson et al., 2018), since stereotyp- severe ethical issues and even bias amplifica- tion. Recent work has predominantly focused ing minoritized groups is a representational harm on measuring and mitigating bias in pretrained that perpetuates societal inequalities and unfairness language models. Surprisingly, the landscape (Blodgett et al., 2020). Human biases are in all of bias measurements and mitigation resources likelihood especially harmful if encoded in con- and methods for conversational language mod- versational AI systems, like the recent DialoGPT els is still 
TI - RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Models
JF - Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
DO - 10.18653/v1/2021.acl-long.151
DA - 2021-01-01
UR - https://www.deepdyve.com/lp/unpaywall/redditbias-a-real-world-resource-for-bias-evaluation-and-debiasing-of-RaPrJsdk1C
DP - DeepDyve
ER -