Are we biased on bias? Characterizing social bias research in the ACL community
Submission Type: Regular Long Paper
Submission Track: Ethics in NLP
Submission Track 2: Theme Track: Large Language Models and the Future of NLP
Keywords: Social Bias, Survey, Ethics
TL;DR: This survey gives a quantitive and qualitative of the social bias research in the ACL community
Abstract: Recent events in business, politics and society have shed light on the importance and potential dangers of Natural Language processing (NLP) in the real world. NLP applications have gained unprecedented popularity not just among scientist and practitioners, but also the general public. As we develop new methodologies and curate new benchmarks and datasets it is more important than ever to consider the implications and societal impact of our work. In this paper, we characterize the landscape of societal bias research within the ACL community and provide a quantitative and qualitative survey by analyzing an categorized corpus of \textit{348} papers. More specifically, we present a definition of social bias based on ethical principals and investigate (i) types of bias, (ii) languages, and (iii) type of paper. We find that there is significantly more work on gender biases and English than other languages. Finally, we discuss the possible causes behind our findings and provide pointers to future opportunities.
Submission Number: 3993
Loading