BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:[RESCHEDULED] Typological Feature Prediction and Blinding for Cros
 s-Lingual NLP - Johannes Bjerva (Aalborg University)
DTSTART:20210305T120000Z
DTEND:20210305T130000Z
UID:TALK157561@talks.cam.ac.uk
CONTACT:James Thorne
DESCRIPTION:I will discuss the usefulness of typological features in NLP\,
  particularly in cross-lingual settings. On the one hand\, typological fea
 tures from databases such as the World Atlas of Language Structures (WALS)
  seem promising for cross-lingual NLP\, as annotations for useful aspects 
 of language exist even for very low-resource languages. Furthermore\, miss
 ing features in WALS can be predicted with relatively high success\, and h
 as been the focus of much recent work (e.g. in the SIGTYP 2020 shared task
 ). When it comes to application of these features\, however\, previous wor
 k has only found minor benefits from using typological information in actu
 al NLP modelling. In recent work (to appear at EACL 2021)\, we hypothesise
 d that these minor gains might stem from that a model trained in a cross-l
 ingual setting picks up on typological cues from the input data\, thus ove
 rshadowing the utility of explicitly using such features. We verify this h
 ypothesis by blinding a model to typological information\, and investigate
  how cross-lingual sharing and performance is impacted. While this sheds s
 ome light on the matter\, I want to further explore the question of the us
 efulness of typological feature prediction in general\, and the use of suc
 h features in NLP.
LOCATION:Virtual (Zoom)
END:VEVENT
END:VCALENDAR