Prof Emily Bender Seminar
Date: Thursday, 5th July 2018, 2:00pm - 3:00pm
Venue: 9 Wally's Walk (Building E6A), Level 3, Room 357, Macquarie University
Speaker: Professor Emily Bender, Department of Linguistics, University of Washington
Host: A/Professor Mark Dras
Topic: Articulating how our NLP data and systems do and don't represent the world: Toward mitigating bias and enabling better science
Joint work with Batya Friedman, Information School, University of Washington
Abstract
As the technology we work on becomes broadly used, impacting the lives of both direct and indirect stakeholders, we as NLP technologists have a responsibility to think critically about the real world effects of the design decisions we make as we build systems. In this talk, I look in particular at the problems that arise when there is a mismatch between the datasets used for training and testing NLP systems and the contexts of system deployment. I present a proposed professional practice, drawing on value sensitive design (Friedman et al 2006), which should help us as a field engage with the ethical issues of exclusion, overgeneralization and underexposure (Hovy & Spruitt 2016). This professional practice, called "data statements", will bring our datasets and the populations they represent into better focus and as a result position us to better understand and describe our results and do better science and engineering.
Bio
Emily M. Bender is a Professor of Linguistics and Adjunct Professor of Computer Science and Engineering at the University of Washington, where she has been Faculty Director of the Professional Masters in Computational Linguistics (CLMS) program since 2005. Her research interests include multilingual grammar engineering, the relationship between linguistics and computational linguistics, computational semantics, variation within and across languages, and methodology for engaging with ethical practice in NLP.