A Categorical Compositional Distributional Model of Meaning and some Supporting Experiments

Add to your list(s) Download to your calendar using vCal

• Mehrnoosh Sadrzadeh, University of Oxford
• Friday 20 May 2011, 16:00-17:00
• UG40 Computer Science.

If you have a question about this talk, please contact Paul Levy.

Words are building blocks of sentences yet meaning of a sentence goes well beyond meanings of the words therein. Good models of word meaning (vector space distributions) are not compositional and good models of sentence meaning (grammatical structure) say nothing about meanings of words. We used a compact categorical setting to bring the two together: the grammar is represented in a Lambek Pregroup, which is `quantized’ via a monoidal functor to the category of word meanings: FVect. This is joint work with Clark-Coecke-Preller and has been covered by the New Scientist, since it uses methods similar to categorical quantum mechanics of Abramsky-Coecke.

One on-the-nose consequence is a solution to a long-standing problem in computational linguistics: measuring sentence synonymy. The solution is a formula: application of the functor to the tensor products of the meanings of words. But does this abstract meaning mean anything? In recent work, with E.Grefenstette, we implemented part of the setting (transitive sentences), performed experiments on a disambiguation task and provided some nicely positive answers!

This talk is part of the Theoretical computer science seminar series.

This talk is included in these lists:

• Computer Science Departmental Series
• Computer Science Distinguished Seminars
• Theoretical computer science seminar
• UG40 Computer Science
• computer sience

Note that ex-directory lists are not shown.