University of Cambridge > Talks.cam > NLIP Seminar Series > Are you losing Structures in Distributional Vectors? Smoothed Distributed Tree Kernels and the Convolution Conjecture

Are you losing Structures in Distributional Vectors? Smoothed Distributed Tree Kernels and the Convolution Conjecture

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

If you have a question about this talk, please contact Tamara Polajnar.

Syntax and words contribute to construct meaning of sentences. But, compositional distributional semantics models (CDSMs) seem to compose word meaning forgetting syntactic structures. Distributional vectors for sentences appear to carry only semantic information.

In this talk, I propose the Convolution Conjecture and a novel structure-informed CDSM : the Smoothed Distributed Tree Kernel. The Convolution Conjecture postulates an unexpected equivalence between semantic-driven CDS Ms and structure-driven convolution kernels. The conjecture thus suggests that structures are still encoded in distributional vectors. Building on this conjecture, we proposed the Smoothed Distributed Tree Kernel that combines structural syntactic information and distributional semantics by clearly separating the two. Our structure-informed CDSM is based on Distributed Tree Kernels that embed syntactic structures in small vectors. We believe that the Convolution Conjecture and our Smoothed Distributed Tree Kernel could help in defining a novel class of structure-informed CDS Ms.

This talk is part of the NLIP Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2017 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity