Department of Comptuer Science - Columbia University

Natural Language Applications Across Genres: From News to Novels

Much research in the natural language field has been carried out on news and there is a need for applications in this genre. In earlier work, we developed a robust news summarization system, called Newsblaster, that provides a browsing interface to news on the web. We extended this research, developing techniques for generating responses to open-ended question answering, enabling the generation of a biography of a queried person or a description of an event. While these news applications were difficult and raised many research chalenges, we began working with weblogs and multilingual input as well. In this talk, I will discuss the issues that arise when question answering systems must handle noisy input. I will also show how the need for new applications arise for new genres and touch on research that we are currently doing on identifying persuasion on weblogs. Finally, I will turn to our most recent research, where we have moved to a yet more difficult genre, the novel, and discuss how we can use natural language technology to investigate theories that have been proposed in comparative literature.