Abstract

Automatic essay grading (AEG) is a process in which machines assign a gradeto an essay written in response to a topic, called the prompt. Zero-shot AEG iswhen we train a system to grade essays written to a new prompt which was notpresent in our training data. In this paper, we describe a solution to theproblem of zero-shot automatic essay grading, using cognitive information, inthe form of gaze behaviour. Our experiments show that using gaze behaviourhelps in improving the performance of AEG systems, especially when we provide anew essay written in response to a new prompt for scoring, by an average ofalmost 5 percentage points of QWK.

 

Quick Read (beta)