Machine learning for spelling acquisition - How accurate is the prediction of specific spelling errors in German primary school students?

Abstract

Background: In Germany, 30 % of students demonstrate insufficient spelling skills at the end of primary school – partly owing to the challenge for teachers to manage a variety of student learning needs. Digital tools using machine learning (ML) have already proven to be useful in enabling teachers to individualise students’ learning. However, there are still no suitable approaches for demographics not yet proficient in spelling. Objectives: With an aim to adapt ML for students of all proficiencies, we investigate how accurately spelling errors can be predicted across different literacy skill levels, which specific errors are predicted best (and worst), and the content-related reasons for these errors.Methods: We developed a web application for spelling that was used by N = 685 first and second graders in Bavaria, Germany. With the 18,133 different misspellings made, we trained six ML models and compared their performances.Results and conclusions: Comparing all ML models employed in this work, the random forest performed best on average as a predictor of spelling errors. Errors at the syllable and morpheme level were predicted best, and errors at the basic phoneme-grapheme level were predicted slightly less accurately. Confusions often concerned cases that are considered linguistically ambiguous or occurred in complex error entanglements.Implications: The results show that ML can be a useful tool to investigate specific spelling deficiencies in demographics not yet proficiently literate. ML could assist in individualising spelling acquisition, such as by automatically providing students individualised feedback or by providing teachers with diagnostic information about their students’ learning progress.