Abstract

The automatic analysis of learner language can play a role in the annotation of learner corpora and in intelligent language tutoring systems. In this talk, I first want to raise some questions about the nature of the linguistic categories which are appropriate for learner language under different perspectives, and which role the context, explicit tasks, and learner modeling play for the interpretation of learner language. Then the talk moves from analyzing form to evaluating aspects of meaning. I discuss our work in the CoMiC project on automatically evaluating the meaning of learner answers to reading comprehension questions, for which we explore which linguistic representations and comparison strategies are effective and robust enough to evaluate meaning in the face of significant well-formed and ill-formed variation.