BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Beyond MaltParser -- Recent Advances in Transition-Based Dependenc
 y Parsing - Joakim Nivre\, Uppsala University
DTSTART:20120229T120000Z
DTEND:20120229T130000Z
UID:TALK35061@talks.cam.ac.uk
CONTACT:Ekaterina Kochmar
DESCRIPTION:The transition-based approach to dependency parsing has become
  popular thanks to its simplicity and efficiency. Systems like MaltParser 
 achieve linear-time parsing\nwith projective dependency trees using locall
 y trained classifiers to predict the next parsing action and greedy best-f
 irst search to retrieve the optimal parse tree\, assuming that the input s
 entence has been\nmorphologically disambiguated using a part-of-speech tag
 ger. In this talk\, I survey recent developments in\ntransition-based depe
 ndency parsing that address some of the limitations of the basic transitio
 n-based approach. First\, I discuss different methods for extending the co
 verage to non-projective trees\, which are required\nfor linguistic adequa
 cy in many languages.\nSecondly\, I show how globally trained classifiers 
 and beam search can be used to mitigate error propagation\nand enable rich
 er feature representations. Finally\, I present a model for joint tagging 
 and parsing that leads\nto improvements in both tagging and parsing accura
 cy as compared to the standard pipeline approach.\n\n
LOCATION:FW11\, Computer Laboratory
END:VEVENT
END:VCALENDAR