Title: Automatic Factual Question Generation

Time: Monday, September 20, 2010, 11 am

Location: Room S2|02|C110 (Hochschulstr. 10)

Speaker: Michael Heilman, Carnegie Mellon University, USA

Abstract

In this talk, we focus on question generation for the creation of educational materials for reading practice and assessment. Our goal is to create an automated system that can take as input a text (e.g., an article that a student might read for homework or during class), and produce as output a ranked list of questions for assessing knowledge of the information in the text.

Automated question generation raises a number of challenges related to the analysis of lexical items and syntactic constructions. Examples include the variety of syntactic constructions in which factual information can be presented (e.g., relative clauses, participial phrases, conjunctions, etc.), complex linguistic constraints on WH movement (i.e., island constraints), and the mapping of potential answers to appropriate WH words (e.g., that “linguist” should lead to a “who” question).

We describe an implemented question generation system that leverages existing natural language processing tools and formalisms to solve these challenges, and we present results from experimental evaluations of the system's output. We conclude by discussing our ongoing and potential future work in question generation, including the development of a user interface for teachers as well as possible steps toward deeper linguistic analyses and more cognitively challenging questions.