Thursday, 2nd November - 10am Jasmijn Bastings : Seminar

Title: Towards Gender-Fair Language Technology

Abstract: 

The past year, not a week has gone by without AI being in the news. This is true especially for large language models powering chat bots. Users are generating essays, recipes, and complete fantasy worlds that they interact with. At the same time, automatic translation remains one of the most popular use cases in language technology. While some consider translation "solved", we have seen -- and are still seeing -- many examples of translation services failing; quite often so because of gender bias. In this talk I will discuss work from an ongoing collaboration where we take stock of the current situation in machine translation with respect to gender bias. We will cover attempts to evaluate and mitigate it, discuss why it is such a hard problem that is bigger than technological solutions (and rather should be studied under a sociotechnical lens), and why the status quo doesn't seem to have improved over the past few years, especially if we consider that most prior work considers gender to be a binary. Finally, we will conclude by looking ahead towards mitigations that recent advances in NLP might enable.  

Bio:

Dr Jasmijn Bastings is a senior research scientist at Google DeepMind. They hold a PhD and MSc in artificial intelligence (AI) from the University of Amsterdam, and a BSc in AI with a minor in linguistics from Utrecht University. Jasmijn has worked on various topics within natural language processing, including automatic translation (e.g., Linguistically-informed NMT, Joey NMT) and interpretability (e.g., rationalizations, the Language Interpretability Tool), and has more recently been focusing on how to make language technology, in particular machine translation, more inclusive and gender fair.

 vCal  iCal