BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Functional Distributional Semantics: Learning Linguistically Infor
 med Representations from a Precisely Annotated Corpus - Guy Emerson\, NLIP
 \, University of Cambridge
DTSTART:20180608T110000Z
DTEND:20180608T120000Z
UID:TALK104647@talks.cam.ac.uk
CONTACT:Andrew Caines
DESCRIPTION:The aim of distributional semantics is to design computational
  techniques that can automatically learn the meanings of words from a body
  of text. The twin challenges are: how do we represent meaning\, and how d
 o we learn these representations? The current state of the art is to repre
 sent meanings as vectors -- but vectors do not correspond to any tradition
 al notion of meaning. In particular\, there is no way to talk about *truth
 *\, a crucial concept in logic and formal semantics.\n\nIn this dissertati
 on\, I develop a framework for distributional semantics which answers this
  challenge. The meanings of words are not represented as vectors\, but as 
 *functions*\, which map from entities to probabilities of truth. Such a fu
 nction can be interpreted both in the machine learning sense of a classifi
 er\, and in the formal semantic sense of a truth-conditional function. Thi
 s simultaneously allows both the use of machine learning techniques to exp
 loit large datasets\, and also the use of formal semantic techniques to ma
 nipulate the learnt representations. I define a probabilistic graphical mo
 del\, which incorporates a probabilistic generalisation of model theory (a
 llowing a strong connection with formal semantics)\, and which generates s
 emantic dependency graphs (allowing it to be trained on a corpus). This gr
 aphical model provides a natural way to model logical inference\, semantic
  composition\, and context-dependent meanings\, where Bayesian inference p
 lays a crucial role. I demonstrate the feasibility of this approach by tra
 ining a model on WikiWoods\, a parsed version of the English Wikipedia\, a
 nd evaluating it on three tasks. The results indicate that the model can l
 earn information not captured by vector space models.
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR