Using machine learning to predict children’s reading comprehension from linguistic features extracted from speech and writing.

Abstract

Advances in machine learning (ML) are poised to contribute to our understanding of the linguistic processes associated with successful reading comprehension, which is a critical aspect of children’s educational success. We used ML techniques to investigate and compare associations between children’s reading comprehension and 260 linguistic features extracted from their speech and writing. Language samples were gathered from 99 linguistically diverse children in grades 4–6 using Talk2Me, Jr., an online language and literacy assessment platform. Lexical and syntactic features were extracted via a consolidated natural language processing pipeline. We compared five machine learning models predicting reading comprehension from the linguistic features and then, using the best models, analyzed the 20 top predictive features for both the oral-elicited and text-elicited data. The findings suggest that variance in children’s reading comprehension can be predicted by grammatical and lexical features extracted from productive written and spoken language. The study highlights how ML methodologies can enable nuanced examination of the language processes associated with reading comprehension. (PsycInfo Database Record (c) 2021 APA, all rights reserved)