Linking text readability and learner proficiency using linguistic complexity feature vector distance

摘要

How can we identify authentic reading material that matches the learner's proficiency and fosters their language development? Traditionally, this involves assigning a one-dimensional label to the text that identifies the grade or proficiency level of the learners that the text is intended for. Such an approach is inadequate given that both the text complexity and proficiency constructs are multi-dimensional in nature. We propose to instead link readers and texts through multidimensional vectors characterizing the linguistic complexity of the reading material and that of texts written by the learners as proxy of their proficiency level. We first validate the approach using a leveled reading corpus by showing that vector distances computed on the complexity representations can serve the function of the traditional labels. We then highlight the advantage of the multi-dimensional approach using data from a continuation writing task, showing that it makes it possible to study individual complexity dimensions and to explore different degrees of challenge for different dimensions. Our approach essentially makes it possible to empirically investigate the +1 of Krashen's i+1, the challenge that best fosters development given the learner's interlanguage. On the practical side, we discuss an ICALL system demonstrating the viabilityof the approach in real-life..