A lexical approach for Spanish question answering

Abstract

This paper discusses our system's results at the Spanish Question Answering task of CLEF 2007. Our system is centered in a full data-driven approach that combines information retrieval and machine learning techniques. It mainly relies on the use of lexical information and avoids any complex language processing procedure. Evaluation results indicate that this approach is very effective for answering definition questions from Wikipedia. In contrast, they also reveal that it is very difficult to respond factoid questions from this resource solely based on the use of lexical overlaps and redundancy. © 2008 Springer-Verlag Berlin Heidelberg.