BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Probabilistic models of similarity and plausibility in context - D
 iarmuid Ó Séaghdha\, University of Cambridge
DTSTART:20111202T120000Z
DTEND:20111202T130000Z
UID:TALK34840@talks.cam.ac.uk
CONTACT:Ekaterina Kochmar
DESCRIPTION:The distributional approach in its many guises is the most pop
 ular\nparadigm for current research on lexical semantics. In this talk I'l
 l\ndescribe a framework for distributional semantics based on latent\nvari
 able probabilistic models of co-occurrence (aka "topic models").\nThese mo
 dels can answer a variety of semantic questions about how a word\ninteract
 s with its context\; I will focus on questions about\nco-occurrence plausi
 bility and about similarity between words in the\ndisambiguating context o
 f a sentence or syntactic structure. Modelling\nplausibility corresponds t
 o the well-known task of selectional\npreference learning\; in-context sim
 ilarity is fundamental to\ndisambiguation tasks such as lexical substituti
 on. I will show that\nrelatively simple topic models give very good perfor
 mance across a range\nof lexical semantic evaluation settings.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR