BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Accurate CCG Parsing with Approximate Language Intersection and Ta
 sk-specific Optimization - Michael Auli
DTSTART:20110506T110000Z
DTEND:20110506T120000Z
UID:TALK31197@talks.cam.ac.uk
CONTACT:Thomas Lippincott
DESCRIPTION:Combinatory Categorial Grammar (CCG) parsing is a longstanding
  problem in computational linguistics\, due to the complexities associated
  with its mild context-sensitivity. Via an oracle experiment\, we show tha
 t the upper bound on accuracy of a CCG parser is significantly lowered whe
 n its search space is pruned using a supertagger\, though the supertagger 
 also prunes many bad parses.\n\nInspired by this analysis\, we design a si
 ngle model with \nboth supertagging and parsing features\, rather than sep
 arating them into distinct models chained together in a pipeline. To overc
 ome the resulting complexity\, we experiment with two approximation algori
 thms for language intersection: loopy belief propagation and dual decompos
 ition.\n\nThe second part of this talk deals with task-specific optimisati
 on of parsing models. We adopt the softmax-margin training objective which
  minimises a bound on expected risk for a given loss function but requires
  the loss to decompose over the predicted structure\, which is not true of
  F-measure. We present a novel dynamic programming algorithm which allows 
 us to use it with F-measure leading to substantial gains in accuracy on CC
 GBank.\n\nEach of the presented methods improves over the state-of-the-art
 . Moreover\, the improvements are additive\, obtaining the best reported r
 esults on this task. Our algorithms are general and we expect them to appl
 y to other parsing problems\, including lexcalized tree adjoining grammar 
 and context-free grammar. 
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR