Predicting Difficulty of English Derivations for L2 learners using Usage-Based Metrics: A Comparison to Word Frequency Measures

Usage-based models (Bybee, 2006; Tomasello, 2003) suggest that multiple derivations of a given headword cannot be considered equally difficult, because the frequency and distributions of their various affixes influence the likelihood that an L2 learner can determine their meaning. Author (2025) proposed usage-based metrics to describe derivations in relation to affix frequency, semantics, and form. Such metrics may improve frequency-based models of derivational difficulty. In this study, multiple regressions were run to predict difficulty of derived forms for 446 Japanese EFL learners using five such metrics: affix type frequency, affix type–token ratio, relative entropy, contingency, and proportion of known basewords, along with the more common predictors of derivational (lemma) and word family frequency. Lemma and word family frequency were found to make contributions unique to one another when explaining variance, and two usage-metrics (affix frequency and relative entropy) appeared to explain residuals to the word frequency-based predictors, and consequently were highly significant ( p < .001) in the final model, with a reciprocal suppression effect heightening their significance after frequency measures were added. An additional major finding was that, rather than being competing measures of word frequency that researchers must choose between, lemma and word-family frequency explain unique variance, and their concurrent use improves predictions of derivation difficulty.