Using syntactic distributional patterns for data-driven answer extraction from the Web

Abstract

In this work, a data-driven approach for extracting answers from web-snippets is presented. Answers are identified by matching contextual distributional patterns of the expected answer type(EAT) and answer candidates. These distributional patterns are directly learnt from previously annotated tuples {question, sentence, answer}, and the learning mechanism is based on the principles language acquisition. Results shows that this linguistic motivated data-driven approach is encouraging. © Springer-Verlag Berlin Heidelberg 2006.