BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning Latent Syntactic Representations with Joint Models - Jaso
 n Naradowsky\, UCL
DTSTART:20150313T120000Z
DTEND:20150313T130000Z
UID:TALK57213@talks.cam.ac.uk
CONTACT:Tamara Polajnar
DESCRIPTION:A human listener\, charged with the difficult task of mapping 
 language to meaning\, must infer a rich hierarchy of linguistic structures
  before reaching an understanding of what was spoken.  Much in the same ma
 nner\, developing complete natural language processing systems requires th
 e processing of many different layers of linguistic information in order t
 o solve complex tasks\, like answering a question or translating a documen
 t.  \n\nOne such "pre-requisite" layer\, syntactic structure\, has proven 
 to be useful information for a wide variety of downstream NLP tasks\, and 
 research in the supervised training of syntactic parsers has received sign
 ificant attention by the NLP community for decades.  However\, training su
 ch parsers requires access to large corpora of syntactically-annotated dat
 a\, which are both costly to produce and unavailable in many widely spoken
  languages.  Even when available\, the domain of interest often differs fr
 om the domain of the syntactic data\, often newswire\, leading to domain d
 rift and lower performance on the downstream task.\n\nTo address these iss
 ues we present a general framework for constructing and reasoning with joi
 nt graphical models.  In a joint model individual component models are cou
 pled and inference is performed globally\, allowing the beliefs of one mod
 el to influence the other and vice versa.  While joint inference is tradit
 ionally pursued to limit the propagation of errors between components\, he
 re we utilize it for a different purpose: to train syntactic models withou
 t syntactically-annotated data.   We propose a novel marginalization-based
  training method in which end task annotations are used to guide the induc
 tion of a constrained latent syntactic representation\, with the resulting
  syntactic distribution being specially-tailored for the desired end task.
    We find that across a number NLP tasks (semantic role labeling\, named 
 entity recognition\, relation extraction) this approach not only offers pe
 rformance comparable to the fully supervised training of the joint model (
 using syntactically-annotated data)\, but in some instances even improves 
 upon it by learning latent structures which are more appropriate for the t
 ask.\n\nThis is joint work with Mark Johnson\, Sebastian Riedel\, and Davi
 d A. Smith
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR