Advaith Siddharthan
Columbia University

as372@cs.columbia.edu


Title: Generating Referring Expressions in Open Domains

Time:Thursday, January 15, 11:30 - 12:30

Place:CS Conference Room in MUDD

Abstract:
We present an algorithm for generating referring expressions in open domains. Existing algorithms assume a classification of adjectives. This exists (or is feasable to construct) only for very restricted domains. Large scale adjective classification is not only unfeasable, it is also unnatural and unintuitive. Unlike nouns, which are classified hierarchically, the important relations for relating adjectives are antonymy and synonymy, not hyponymy. This largely follows from the purpose of adjectives - to describe or contrast entities.

Our algorithm relies on WordNet synonym and antonym sets and gives equivalent results on the (restricted domain) examples cited in the literature and good results for many other cases that prior approaches cannot handle. We believe that it is also the first algorithm that allows for the incremental incorporation of relations as well as attributes. Our algorithm also overcomes various other problems with existing alorithms.

We also present a corpus-based evaluation of our algorithm using the Penn Treebank. Evaluations are uncommon in this research area, and we feel that our ability to evaluate our algorithm on a standard corpus is by itself a vindication of our approach.

About the speaker: I am doing a PostDoc with the Natural Language Group at Columbia University, where I am applying my thesis work to the problem of summarization. My PhD is from the Natural Language and Information Processing Group at the Computer Lab, University of Cambridge. My supervisor was Ann Copestake. I've followed a rather erratic trajectory to get this far, having started out in physics a while back. My PhD thesis was on syntactic simplification and text cohesion; I was looking at the problem of simplifying newspaper text to make it accessible to people with reading difficulties. In particular, I was interested in analysing the discourse level aspects of syntactically rewriting text. My publications.