Question Classification and Answering from Procedural Text in English
Somnath Banerjee, Sivaji Bandyopadhyay · 2012
Linguistic patterns reflect the regularities of Natural Language and the applicability of such linguistic patterns is acknowledged in several Natural Language Processing tasks. Many question classification systems depend on patterns that are extracted from already framed questions. In this paper, we have investigated possible question categories and question patterns for procedural text documents in English and proposed seven question classes. More than six thousands questions of different domains, e.g., cooking recipes, electronics, home and maintenance, medical etc have been collected from Yahoo answers as experimentation corpus. Annotators reached almost perfect agreement of 94.6% at kappa scale. A procedural question answering system has been developed to verify the proposed question classes. The evaluation reveals that the proposed classes are a good approach to deal with Question Answering for procedural text questions. The procedural question answering system has achieved overall 95.08%, 86.95% and 90.84 precision, recall and F-measure value respectively.