BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:NLIP Seminar Series
SUMMARY:[RESCHEDULED] Typological Feature Prediction and B
 linding for Cross-Lingual NLP - Johannes Bjerva (A
 alborg University)
DTSTART;TZID=Europe/London:20210305T120000
DTEND;TZID=Europe/London:20210305T130000
UID:TALK157561AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/157561
DESCRIPTION:I will discuss the usefulness of typological featu
 res in NLP\, particularly in cross-lingual setting
 s. On the one hand\, typological features from dat
 abases such as the World Atlas of Language Structu
 res (WALS) seem promising for cross-lingual NLP\, 
 as annotations for useful aspects of language exis
 t even for very low-resource languages. Furthermor
 e\, missing features in WALS can be predicted with
  relatively high success\, and has been the focus 
 of much recent work (e.g. in the SIGTYP 2020 share
 d task). When it comes to application of these fea
 tures\, however\, previous work has only found min
 or benefits from using typological information in 
 actual NLP modelling. In recent work (to appear at
  EACL 2021)\, we hypothesised that these minor gai
 ns might stem from that a model trained in a cross
 -lingual setting picks up on typological cues from
  the input data\, thus overshadowing the utility o
 f explicitly using such features. We verify this h
 ypothesis by blinding a model to typological infor
 mation\, and investigate how cross-lingual sharing
  and performance is impacted. While this sheds som
 e light on the matter\, I want to further explore 
 the question of the usefulness of typological feat
 ure prediction in general\, and the use of such fe
 atures in NLP.
LOCATION:Virtual (Zoom)
CONTACT:James Thorne
END:VEVENT
END:VCALENDAR