IIIT Hyderabad at TAC 2008

This paper describes our participation at TAC 2008 in all the three tracks. For the Summarization Track we introduced two major features. First, a feature based on Information Loss if we don’t pick a particular sentence. Second, a language modeling extension that boosts novel terms and penalizes stale terms. During our post-TAC analysis we observed that a simple sentence position based summarizer leads to better short summaries than most official runs submitted this year. In the Opinion QA and Summarization Track for the rigid list questions, we have added some additional features to handle opinion expressed in the question. and for the squishy list questions in Opinion QA and Summarization Track, we leveraged on our existing Summarization engine and used a classification based approach to both finding opinionated sentences and also the polarity of the opinions. Finally, for the RTE track we explored a simple graph partition matching based approach.