Accurate modelling of language learning tasks and students using representations of grammatical proficiency
Accepted version
Peer-reviewed
Repository URI
Repository DOI
Change log
Authors
Abstract
Adaptive learning systems aim to learn the relationship between curriculum content and students in order to optimise a student’s learning process. One form of such a system is content recommendation in which the system attempts to predict the most suitable content to next present to the student. In order to develop such a system, we must learn reliable representations of the curriculum content and the student. We consider this in the context of foreign language learning and present a novel neural network architecture to learn such representations. We also show that by incorporating grammatical error distributions as a feature in our neural architecture, we can substantially improve the quality of our representations. Different types of grammatical error are automatically detected in essays submitted by students to an online learning platform. We evaluate our model and representations by predicting student scores and grammatical error distributions on unseen language tasks. We also discuss further uses for our model beyond content recommendation such as inferring student knowledge components for a given domain and optimising spacing and repetition of content for efficient long term retention.