BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Five Sources of Biases and Ethical Issues in NLP\, and What to Do 
 about Them - Dirk Hovy (Bocconi University)
DTSTART:20201023T110000Z
DTEND:20201023T120000Z
UID:TALK152380@talks.cam.ac.uk
CONTACT:Guy Aglionby
DESCRIPTION:Never before was it so easy to write a powerful NLP system\, n
 ever before did it have such a potential impact. However\, these systems a
 re now increasingly used in applications they were not intended for\, by p
 eople who treat them as interchangeable black boxes. The results can be si
 mple performance drops\, but also systematic biases against various user g
 roups. \n\nIn this talk\, I will discuss several types of biases that affe
 ct NLP models (based on Shah et al. 2020 and Hovy & Spruit\, 2016)\, what 
 their sources are\, and potential counter measures. \n- bias stemming from
  data\, i.e.\, selection bias (if our texts do not adequately reflect the 
 population we want to study)\, label bias (if the labels we use are skewed
 )\, and semantic bias (the latent stereotypes encoded in embeddings). \n- 
 biases deriving from the models themselves\, i.e.\, their tendency to ampl
 ify any imbalances that are present in the data.\n- design bias\, i.e.\, t
 he biases arising from our (the practitioners) decisions which topics to e
 xplore\, which data sets to use\, and what to do with them.\n\nAs a conseq
 uence\, we as NLP practitioners suddenly have a new role\, in addition to 
 researcher and developer: considering the ethical implications of our syst
 ems\, and educating the public about the possibilities and limitations of 
 our work.The time of academic innocence is over\, and we need to address t
 his newfound responsibility as a community.\n\nFor each bias\, I will prov
 ide real examples and discuss the possible ramifications for a wide range 
 of applications\, and the various ways to address and counteract these bia
 ses\, ranging from simple labeling considerations to new types of models.\
 nI conclude with some provocations for future directions.\n\n\nReference:\
 n- Deven Shah\, H. Andrew Schwartz\, & Dirk Hovy. 2020. Predictive Biases 
 in Natural Language Processing Models: A Conceptual Framework and Overview
 . In Proceedings of ACL. [https://www.aclweb.org/anthology/2020.acl-main.4
 68/]\n- Dirk Hovy & Shannon L. Spruit. 2016. The Social Impact of Natural 
 Language Processing. [https://www.aclweb.org/anthology/P16-2096.pdf]\n\n\n
 Bio:\nDirk Hovy is associate professor of computer science at Bocconi Univ
 ersity in Milan\, Italy. Before that\, he was faculty and a postdoc in Cop
 enhagen\, got a PhD from USC\, and a linguistics masters in Germany. He is
  interested in the interaction between language\, society\, and machine le
 arning\, or what language can tell us about society\, and what computers c
 an tell us about language. He has authored over 50 articles on these topic
 s\, including 3 best paper awards. He has organized one conference and sev
 eral workshops (on abusive language\, ethics in NLP\, and computational so
 cial science). Outside of work\, Dirk enjoys cooking\, running\, and leath
 er-crafting. For updated information\, see http://www.dirkhovy.com\n\nhttp
 s://cl-cam-ac-uk.zoom.us/j/92174303432?pwd=S2NLWE42VmhRdGE0dlRuMXFFb3FOZz0
 9\n\nMeeting ID: 921 7430 3432\nPasscode: 137181
LOCATION:Virtual (Zoom)
END:VEVENT
END:VCALENDAR