On April 28, Hannah Devinney presented a talk on Gender Bias in Natural Language Processing at AI Sweden’s Swedish NLP Webinar series, which focuses on NLP development in Sweden and/or about the Swedish language.
The talk focuses on gendered aspects of “bias”: exploring what it is, how it manifests in NLP, the harms it causes, and what we as NLPers can do to combat these harms. Their presentation was followed by an engaging discussion on the nature and future of “unbiased” NLP. We hope that this talk will lead to increased awareness in the AI community of the importance of intersectional and inclusive models of gender for mitigating bias.
A recording of the talk can be found in the video embedded below.
On Monday the 29th of September, Hannah Devinney presented the EQUITBL project at the AI for Good Breakthrough Days main stage event. The project is one of three winners in the Breakthrough Days’ Gender Equity Track.
This interdisciplinary project explores ways of combining qualitative and quantitative methods in order to explore and understand how bias and stereotypes manifest themselves in large text collections, such as those commonly used to train machine learning models in language technology. We also develop tools for mitigating the detrimental effects bias, stereotyping, and underrepresentation can have when the ML models are integrated into AI systems used for decision making.
The project members are:
Hannah Devinney, Computing Science, Centre for Gender Studies, and LPCN, Umeå University
Henrik Björklund, Computing Science and LPCN, Umeå University
Jenny Björklund, Centre for Gender Research, Uppsala University