Morphemes in the wild: Modelling affix learning from the noisy landscape of natural text

Abstract

Morphological knowledge serves as a powerful heuristic for vocabulary growth and contributes significantly to the speed and efficiency of reading. While research has long sought to explain how the knowledge of derivational morphology is acquired, previous approaches have struggled to capture the nuanced and complex ways in which derivational morphemes are used in written language, particularly that these morphemes contribute to meaning in a graded manner and that noise introduced by misleading forms (e.g., deliver) can impede learning. Our approach builds on earlier insights but moves beyond them by combining a large-scale analysis of vocabulary used in 1,200 popular books with computational modelling to explore how learning of derivational affixes may occur from text containing naturally occurring noise. We use a compositional distributional semantic model to investigate what can be learned about the meanings of individual English prefixes and suffixes through reading and evaluate the model’s performance against data from 120 adults in a lexical processing task. Our findings demonstrate that, despite the presence of noise, natural text contains sufficient structure to support the extraction of core affix semantics, and that readers are attuned to the complex patterns that shape affix use in the wild. This work contributes a new dimension to a more principled and psychologically grounded account of morpheme learning, and we discuss both this contribution and the broader insights it offers for language research.