BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:NLP Reading Group: Cross-Cutting Models of Lexical Semantics - Dia
 rmuid Ó Séaghdha (University of Cambridge)
DTSTART:20120202T120000Z
DTEND:20120202T130000Z
UID:TALK36185@talks.cam.ac.uk
CONTACT:Jimme Jardine
DESCRIPTION:@article{reisinger2003cross\,\n  title={Cross-Cutting Models o
 f Lexical Semantics}\,\n  author={Reisinger\, J. and Mooney\, R.}\,\n  jou
 rnal={Journal of Artificial Intelligence Research}\,\n  volume={18}\,\n  p
 ages={1--44}\,\n  year={2003}\n}\n\nContext-dependent word similarity can 
 be measured over multiple cross-cutting dimensions. For example\, lung and
  breath are similar thematically\, while authoritative and superficial occ
 ur in similar syntactic contexts\, but share little semantic similarity. B
 oth of these notions of similarity play a role in determining word meaning
 \, and hence lexical semantic models must take them both into account. Tow
 ards this end\, we develop a novel model\, Multi-View Mixture (MVM)\, that
  represents words as multiple overlapping clusterings. MVM finds multiple 
 data partitions based on different subsets of features\, subject to the ma
 rginal constraint that feature subsets are distributed according to Latent
  Dirichlet Allocation. Intuitively\, this constraint favors feature partit
 ions that have coherent topical semantics. Furthermore\, MVM uses soft fea
 ture assignment\, hence the contribution of each data point to each cluste
 ring view is variable\, isolating the impact of data only to views where t
 hey assign the most features. Through a series of experiments\, we demonst
 rate the utility of MVM as an inductive bias for capturing relations betwe
 en words that are intuitive to humans\, outperforming related models such 
 as Latent Dirichlet Allocation. 
LOCATION:GS15\, Computer Laboratory
END:VEVENT
END:VCALENDAR