Automatic Critiquing of Novices’ Scientific Writing Using Argumentative Zoning

Exploring Attitude and Affect in Text: Theories and Applications

Volume

Papers from the 2004 AAAI Spring Symposium

Contents

Scientific writing can be hard for novice writers, even in their own language. We present a system that applies Argumentative Zoning (AZ), a method of determining argumentative structure in texts, to the task of advising novice writers on their writing. We address this task by automatically determining the rhetorical/argumentative status and the implicit author stance of a given sentence in the text. Then users can be advised, for instance that a different order of sentences might be more advantageous, or that certain argumentative 'moves’ are missing. In implementing such a system, we had to port the feature detection stage of Argumentative Zoning from English to Portuguese, as our system is designed for Brazilian PhD theses in Computer Science. In this paper we report on the overall system, the porting exercise, a human annotation experiment to verify the reproducibility of our annotation scheme and an intrinsic evaluation of the AZ-part of our system.