Sujay
Hi all,
I am a masters student in the LTI and propose to work on a project connected with learning composition of tensorial representation of words. I am looking for team-mates so if you're interested in working on something along these lines please write to me at sja...@cs.cmu.edu. Here's a brief (first draft) description of what the project's goals are:
Over the last decade, statistical methods for natural language processing have used distributional vectors to represent word semantics. However this model is shallow, failing to capture anything beyond the shortest of dependancy ranges, and most discrete word senses. We have been recently working on a more comprehensive representation of word semantics that capture distributions over syntactic dependancies to form a tensor-like structure. We postulate that this representation lends itself better to the properties of compositionality (the linguistic theory of how complex meaning is formed from simpler sub-components), and the goal of this project will be to learn such a composition function from data.
Hope to hear from you guys.
Best,
Sujay.