BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:General-Purpose Representation Learning from Words to Sentences - 
 Felix Hill (University of Cambridge)
DTSTART:20160304T110000Z
DTEND:20160304T120000Z
UID:TALK63430@talks.cam.ac.uk
CONTACT:Kris Cao
DESCRIPTION:Real-valued vector representations of words (aka embeddings) t
 hat are trained on naturally occurring data by optimising general-purpose 
 objectives are useful for a range of downstream language tasks.  However\,
  the picture is less clear for larger linguistic units such as phrases or 
 sentences.  Phrases and sentences typically encode the facts and propositi
 ons that constitute the 'general knowledge' missing from many NLP systems 
 at present\, so the potential benefit of making representation-learning wo
 rk for these units is huge.  I will present a systematic comparison of dif
 ferent ways of inducing such representations with neural language models f
 rom unlabelled data. The study demonstrates clear and interesting differen
 ces between the representations learned by different methods\; in particul
 ar\, more elaborate or computationally expensive methods are not necessari
 ly best.  I'll also discuss a key challenge facing all research in unsuper
 vised or representation learning for NLP - the lack of robust evaluations.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR