Predicting misreadings from gaze in children with reading difficulties

Abstract

We present the first work on predicting reading mistakes in children with reading difficulties based on eye-tracking data from real-world reading teaching. Our approach employs several linguistic and gaze-based features to inform an ensemble of different classifiers, including multi-task learning models that let us transfer knowledge about individual readers to attain better predictions. Notably, the data we use in this work stems from noisy readings in the wild, outside of controlled lab conditions. Our experiments show that despite the noise and despite the small fraction of misreadings, gaze data improves the performance more than any other feature group and our models achieve good performance. We further show that gaze patterns for misread words do not fully generalize across readers, but that we can transfer some knowledge between readers using multitask learning at least in some cases. Applications of our models include partial automation of reading assessment as well as personalized text simplification.

Predictive Model for Dyslexia from Eye Fixation Events

“…Jothi Prabha A, Bhargavi R, Harish B more fixations and more regression errors when compared to the control group. Pause times and fixation duration were observed to be primary predictors for reading [13].…”

Text mining in education

“…(Dun, Wang, Wang, & Hao, 2017) This paper reports on a project to examine how and why automated content analysis could be used to assess precis writing by university students. (Bingel, Barrett, & Klerke, 2018) Document…”

Proceedings of the Beyond Vision and LANguage: inTEgrating Real-world kNowledge (LANTERN)

“…The latter study represents a group of studies where individual readers' records are available at training time (i.e., multiple copies of the data with annotations obtained from different reading behaviour) rather than learning from the aggregate of multiple readers. This approach which involves a minimal level of aggregation is frequently applied where individual readers' cognition is of primary interest, such as categorizing individual language skill level or behaviour (Martínez-Gómez et al, 2012;Matthies and Søgaard, 2013;Augereau et al, 2016;Bingel et al, 2018). Noticeably, the opposite approach of using maximally aggregated type-level representations which average all readings across all occurrences and all participants, has also been shown to contribute to improvements (Barrett et al, 2016;Bingel et al, 2018;.…”

“…This approach which involves a minimal level of aggregation is frequently applied where individual readers' cognition is of primary interest, such as categorizing individual language skill level or behaviour (Martínez-Gómez et al, 2012;Matthies and Søgaard, 2013;Augereau et al, 2016;Bingel et al, 2018). Noticeably, the opposite approach of using maximally aggregated type-level representations which average all readings across all occurrences and all participants, has also been shown to contribute to improvements (Barrett et al, 2016;Bingel et al, 2018;. The effect of these two different views (global vs local) on the same task hence remained unexplored and is a gap we seek to fill in this paper.…”