News Headline Generation based on Linguistic Methods

R. Wang, N. Stokes, W. Doran, E. Newman, J. Carthy, and J. Dunnion (Ireland)


Headline, Information Retrieval, summarisation, lexical chaining


In this paper, we present a new system T-Lex, which system generates a headline for each news story. The system employs lexical chaining and a parse-trim approach. In recent years a number of researchers have used lexical chains in Information Retrieval and related areas. We have used this technique for Topic Detection and Tracking (TDT) and text summarisation. Currently, we are focusing on news story headline generation. The parse-trim approach was used in the Topiary system developed at the University of Maryland. This approach uses linguistically-motivated heuristics to guide the choice of potential headlines. We present our experimental results that demonstrate the effectiveness of our new system over our previous system, Lex, and the Topiary system. The performance of these systems is evaluated using the Recall-Oriented Understudy for Gisting Evaluation (ROUGE) evaluation suite from the Document Understanding Conference (DUC) 2004 news stories collection.

Important Links:

Go Back