The role of statistical learning in spelling acquisition: A comparison of children's spelling with a simulated reference speller

In the domain of spelling acquisition, statistical learning refers to the capability of finding and integrating spelling patterns into one’s own knowledge, to then access these patterns as a supplement in spelling production (Treiman & Kessler, 2014). In this study I adapted a two-step regression approach (e.g., Gries & Deshors, 2014; Gries & Deshors, 2020), first modeling a reference speller who emulates statistical learning of patterns related to German vowel-lengthening as well as syllable-separating <h>. After concluding this step, I compared spelling predictions generated by this reference speller to empirical data of German primary school students, provided in the Litkey Corpus (Laarmann-Quante et al., 2019). Of special interest was the question whether children showing varying degrees of orthographic proficiency also vary in their usage of spelling patterns. I hypothesized that better spellers would be more receptive to these patterns and therefore use them in their writing more systematically, even when producing errors. Due to technical as well as conceptual shortcomings, the statistical analysis did not provide evidence for this hypothesis, though in a subsequent qualitative inspection of specific errors, there were traces suggesting that a follow-up investigation could be more successful, if implemented with care. Throughout this report, I provide detailed descriptions of the methodology, pointing out areas where adjustments are in order. unknown unknown