BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Cross-lingual Learning 2.0 - Anders Søgaard (University of Copenh
 agen)
DTSTART:20170427T100000Z
DTEND:20170427T110000Z
UID:TALK71801@talks.cam.ac.uk
CONTACT:Mohammad Taher Pilehvar
DESCRIPTION:I survey the last decade of work in cross-lingual transfer of 
 NLP models. I briefly present the methods\, but also the assumptions that 
 have been made about the resources available to us when developing models 
 for low-resource languages. I will discuss three trends: a) Cross-lingual 
 learning is applied to more complex tasks\, including frame semantics\, di
 scourse parsing\, machine translation\, and question answering. b) We make
  fewer and fewer assumptions. c) We have started exploring synergies betwe
 en multiple transfer problems\, potentially leading to “universal” mod
 els. Finally\, I will discuss work in progress in our group to guide or re
 gularize multi-task cross-lingual learning using auxiliary loss from readi
 ly available data.
LOCATION:GR06-7\, English Faculty
END:VEVENT
END:VCALENDAR