A statistical method for short answer extraction
Gideon S. Mann · 2001
This paper presents a simple, general method for using the Mutual Information (MI) statistic trained on unannotated trivia questions to estimate question class/semantic tag correlation. This MI method and a variety of question classifiers and semantic taggers are used to build short-answer extractors that show improvement over a hand-built match module using a similar question classifier and semantic tagger.