Academic word difficulty and multidimensional lexical sophistication: An English‐for‐academic‐purposes‐focused conceptual replication of Hashimoto and Egbert (2019)

Abstract

This article presents a conceptual replication of Hashimoto and Egbert (https://doi.org/10.1111/lang.12353), a study that featured multivariate models where lexical sophistication variables accounted for word difficulty (yes-no recognition) better than frequency alone among learners of English as a second or foreign language from North America. This current study (n words = 88; n people = 128) conceptually replicated Hashimoto and Egbert with data from three Asian university English-for-academicpurposes sites. Methodological differences included a more conservative lexical sophistication operationalization process and avoidance of stepwise regression. Like the original study, the replication's findings favored multivariate models over frequency, which predicted 36% of word difficulty's variance alone. In a multiple regression model accounting for word difficulty, R 2 = .52, frequency accounted for 17% of the predicted variance with age of acquisition (AoA: 18%) and word naming reaction time (WN_RT: 16%) also being significant predictors. This replication also extended the testing approach by using a mixed-effect model, involving person and site intercepts as random effects. The model's ability to predict word difficulty fell, marginal R 2 = .22, conditional R 2 = .40, but frequency, AoA, and WN_RT remained the strongest predictors. Taken together, this replication successfully supports the original study's more-than-frequency conclusion while highlighting the need for further research into the area.

Semantic variables and their application in L2 research

“…To explore which lexical characteristics predict the probability an L2 learner will recognize a word, Vitta et al (2023) administered a yes-no test to Arabic and Japanese L2 learners of English. In this testing format, participants indicated whether they do or do not know a word by clicking yes or no.…”

“…Anible (2020) exemplifies the most typical use of semantic variables which is controlling for semantic effects by creating matched lists of words between conditions. Unlike Vitta et al (2023) in which concreteness freely varied, Anible controlled for word frequency, word length, concreteness, and imageability across the four conditions. The imageability ratings across each condition sit around an average rating of 500 with a typical standard deviation of around 75 points.…”