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:Efficient Constrained Inference and Structured Neu
 ral Networks for Semantic Role Labeling - Oscar Tä
 ckström\, Google
DTSTART;TZID=Europe/London:20151204T120000
DTEND;TZID=Europe/London:20151204T130000
UID:TALK62579AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/62579
DESCRIPTION:*Abstract:*\n\nI will describe some of our recent 
 advances in the prediction of predicate-argument s
 tructure in natural language text.\n\nFirst\, I wi
 ll describe a dynamic programming algorithm for ef
 ficient constrained inference in semantic role lab
 eling. The algorithm efficiently captures a majori
 ty of the structural constraints examined by prior
  work in this area\, which has resorted to either 
 approximate methods or slow integer linear program
 ming solvers. In addition\, it allows for structur
 ed learning\, with respect to constrained conditio
 nal likelihood\, which leads to improved predictio
 ns over a locally learned model.\n\nSecond\, I wil
 l describe how the potential functions in the grap
 hical model corresponding to the dynamic program c
 an be replaced with neural networks. In addition t
 o increased modeling power and automatically induc
 ed feature combinations\, this allows us to embed 
 phrasal arguments and semantic roles jointly in th
 e same vector space\, and provides a flexible fram
 ework for multi-task learning by the embedding of 
 semantic roles from multiple annotation schemes in
  a shared vector space.\n\nWith these advances\, b
 oth by themselves and combined\, we obtain state-o
 f-the-art results on both PropBank- and FrameNet-a
 nnotated datasets.\n\n*Short bio:*\n\nOscar Täckst
 röm is a research scientist at Google in New York\
 , where he works primarily on the semantic analysi
 s of text and question answering from structured k
 nowledge bases. Before joining Google in 2013\, he
  was a PhD student in the Computational Linguistic
 s group at Uppsala University and a research scien
 tist at the Swedish Institute of Computer Science.
  In his thesis\, he explored the use of incomplete
  and cross-lingual supervision for learning statis
 tical models in natural language processing. Toget
 her with Ryan McDonald and Jakob Uszkoreit\, he re
 ceived the IBM Best Student Paper Award at NAACL 2
 012.
LOCATION:FW26\, Computer Laboratory
CONTACT:Kris Cao
END:VEVENT
END:VCALENDAR