Relative complexity in a model of word difficulty: The role of loanwords in vocabulary size tests

Abstract

Recent studies have shown that the frequency effect, although long used as a guide to word difficulty, fails to explain all variance in learner word knowledge. As such, a “more than frequency” conclusion has been offered to explain how lexical sophistication accounts for word difficulty. This study presents a multiple regression model of word-learning difficulty from a data set of monolingual Japanese first language (L1) learners. Vocabulary Size Test (VST) scores of 2,999 L1 Japanese university students were converted to logit scores to determine the word-learning difficulty of 80 target words. Five lexical sophistication variables were found to correlate with word-learning difficulty (frequency, cognate status, age of acquisition, prevalence, and polysemy) above a practical significance threshold. These were subsequently entered into a regression model with the logit scores as the dependent variable. The model (R2 = .55) indicates that three lexical sophistication variables significantly predicted VST scores: frequency (ß = -.28, p = .029), cognateness (ß = -.24, p = .005), and prevalence (ß = 0.22, p = .040). Despite suggestions that complexity studies be interpreted considering what is understood about the construct of linguistic complexity, researchers have rarely made explicit the differences between absolute and relative complexity variables. As some variables can be shown to vary in complexity according to the L1 population, these must be considered in discussions of test generalizability. Although frequency will continue to be the primary criterion for the selection of lexical items for teaching and testing, the cognate status of words can be used to predict the potential learning burden of the word more precisely for learners of different L1 backgrounds.