BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Hidden Biases. Ethical Issues in NLP\, and What to Do about Them -
  Dirk Hovy\, Bocconi University in Milan\, Italy
DTSTART:20200123T110000Z
DTEND:20200123T120000Z
UID:TALK138082@talks.cam.ac.uk
CONTACT:Qianchu Liu
DESCRIPTION:Texts reflect the authors' demographic properties and biases\,
  which in turn get magnified by statistical NLP models. This has unintende
 d consequences for our analysis: if we do not pay attention to the biases 
 contained\, we can easily draw the wrong conclusions\, and create disadvan
 tages for our users. \n\nIn this talk\, I will discuss several types of bi
 ases that affect NLP models\, what their sources are\, and potential count
 er measures. \n- bias stemming from data\, i.e.\, selection bias (if our t
 exts do not adequately reflect the population we want to study)\, label bi
 as (if the labels we use are skewed)\, and semantic bias (the latent stere
 otypes encoded in embeddings). \n- biases deriving from the models themsel
 ves\, i.e.\, their tendency to amplify any imbalances that are present in 
 the data.\n- design bias\, i.e.\, the biases arising from our (the researc
 hers) decisions which topics to analyze\, which data sets to use\, and wha
 t to do with them.\n\nFor each bias\, I will provide examples and discuss 
 the possible ramifications for a wide range of applications\, and who vari
 ous ways to address and counteract these biases\, ranging from simple labe
 ling considerations to new types of models.
LOCATION:GR04\, Faculty of English\, 9 West Rd (Sidgwick Site)
END:VEVENT
END:VCALENDAR