An Attempt at Describing Learning Curves for a Complex Sub-Task in Reading Acquisition

Abstract

Learning curves are a well-known phenomenon in learning and describe the oscillation between correct and incorrect performance that precedes mastery. It demonstrates that making mistakes is part of the learning process. It is equally clear that these learning curves are highly individual and therefore pose a challenge in their description and direct comparison. With the ability to collect large amounts of data through learning games with adaptively generated content, it is now possible to take a novel look at this process. Literacy games were deployed in a school setting for the iRead EU Horizon 2020 Project. The Navigo app delivered a complex task of practising to distinguish vowel length in bi-syllabic words in German to pupils in the disguise of a game. Pupils' engagement with the game resulted in the largest longitudinal corpus that has ever been collected for this sort of task from 251 pupils in German elementary schools. The resulting data exhibits learning curves as trajectories, depicting response time and correctness across several weeks for each pupil’s playing sessions. The work presented here attempts to (a) model and parameterise these curves, (b) automate their classification into common forms across a larger population, and (c) detect mastery. In doing so, we propose a method of learning curve representation and interpretation and apply it to the data. A describable pattern of cognitive processing seems to be observable and common to all curves that may allow a prediction of mastery and ability for skill transfer to other environments for a subset of the players. As a result, we were able to describe five general types of common progressions. These findings are in part supported by additional data from pre- and post-tests in the form of paper-and-pencil activities. The work presented here should serve to demonstrate the importance of using large scale input data for training literacy skills rather than a few examples as is the norm in static schoolbooks.