On over- and underuse in learner corpus research and multifactoriality in corpus linguistics more generally

This paper critically discusses how corpus linguistics in general, but learner corpus research in particular, has been dealing with all sorts of frequency data in general, but over- and underuse frequencies in particular. I demonstrate on the basis of learner corpus data the pitfalls of using aggregate data and lacking statistical control that much work is unfortunately characterized by. In fact, I will demonstrate that monofactorial methods have very little to offer at all to research on observational data. While this paper is admittedly very didactic and methodological, I think the discussion of the empirical data offered here – a reanalysis of previously published work – shows how misleading many studies potentially and provides far-reaching implications for much of corpus linguistics and learner corpus research. Ideally/maximally, this paper together with Paquot & Plonsky (2017, Intntl. J. of Learner Corpus Research) would lead to a complete revision of how learner corpus linguists use quantitative methods and study over-/underuse; minimally, this paper would stimulate a much-needed discussion of currently lacking methodological sophistication.