In order to respond correctly to a free form factual question given a large collection of texts, one needs to be able to understand the question to a level that allows determining some of the constraints the question imposes on a possible answer. These constraints may include a semantic classification of the answer and may even suggest taking different strategies in looking for and in veriftying an answer.

We define Question Classification(QC) here to be the task that, given a question, maps it to one of k classes, which provide a semantic constraint on the sought-after answer. This classification, potentially with other constraints on the answer, will be used by a downstream process which selects a correct answer from among several candiates.

We define a two-layered taxonomy of semantic classes of answers. The hierarchy contains 6 coarse classes(ABBREVIATION, ENTITY, DESCRIPTION, HUMAN, LOCATION and NUMERIC VALUE) and 50 fine classes.

The actual classifier makes use of a sequence of two simple classifiers, each utilizing the Winnow algorithm within SNoW. One for classifying questions into coarse classes and one for classifying fine classes. A feature extractor automatically extracts the same features for each classifier.