Generating diagnostic multiple choice comprehension cloze questions - Robotics Institute Carnegie Mellon University

Generating diagnostic multiple choice comprehension cloze questions

Jack Mostow and Hyeju Jang
Workshop Paper, 7th Workshop on Building Educational Applications Using NLP (NAACL HLT '12), pp. 136 - 146, June, 2012

Abstract

This paper describes and evaluates DQGen, which automatically generates multiple choice cloze questions to test a child's comprehension while reading a given text. Unlike previous methods, it generates different types of distracters designed to diagnose different types of comprehension failure, and tests comprehension not only of an individual sentence but of the context that precedes it. We evaluate the quality of the overall questions and the individual distracters, according to 8 human judges blind to the correct answers and intended distracter types. The results, errors, and judges' comments reveal limitations and suggest how to address some of them.

BibTeX

@workshop{Mostow-2012-122062,
author = {Jack Mostow and Hyeju Jang},
title = {Generating diagnostic multiple choice comprehension cloze questions},
booktitle = {Proceedings of 7th Workshop on Building Educational Applications Using NLP (NAACL HLT '12)},
year = {2012},
month = {June},
pages = {136 - 146},
}