Editorial

Abstract

May is here and with it comes a new issue of ReCALL for 2022. This time, we have seven papers covering three broad topics: the first involves automatic L2 proficiency assessment, the second, mobile-assisted language learning (MALL), and the third, L2 vocabulary acquisition, particularly the role of audiovisual materials. Several papers draw on Paivio’s (1971) dual processing theory, according to which “knowledge representation in verbal and visual modes may facilitate processing and therefore aid understanding and retention of knowledge more effectively than representations depending on a single mode” (Sato, Lai & Burden). There is also some overlap between the MALL and vocabulary studies, since the MALL meta-analysis by Burston and Giannakou reveals that by far the most common MALL learning objective is, in fact, lexical learning. Similarly, Lin’s paper on a new web-based app focusing on formulaic expressions in YouTube videos sits at the intersection of MALL and learning of vocabulary and phraseology. As you will read, the studies include a range of methodologies and research designs, from meta-analysis (Burston & Giannakou; Yu & Trainin), to survey (Puebla, Fievet, Tsopanidi & Clahsen), experimental study (Dziemianko; Sato et al.) and computer modelling (Gaillat et al.), through to research and development (Lin). Our first paper shows that collaborative research between universities in Paris and Galway is making headway in the complex area of automatic L2 proficiency assessment by developing AI systems to analyse learners’ writing samples and assign them to appropriate CEFR proficiency levels. The research by Thomas Gaillat, Andrew Simpkin, Nicolas Ballier, Bernardo Stearns, Annanda Sousa, Manon Bouyé and Manel Zarrouk focuses on machine learning from a large corpus of Cambridge and Education First essays, using linguistic microsystems constructed around L2 functions such as modals of obligation, expressions of time, and proforms, for instance, in addition to more traditional measures of complexity involving lexis, syntax, semantics, and discourse features. After training on some 12,500 English L2 texts written by around 1,500 L1 French and Spanish examinees, which had been assigned to one of the six CEFR levels by human raters, the AI system reached 82% accuracy in identifying writers’ proficiency levels. It also identified specific microsystems associated with learners at level A (nominals, modals of obligation, duration, quantification), level B (quantifiers and determiners), and level C (proforms and should/will). External validation for the model was less successful, however: only 51% of texts from the ASAG corpus (a different set of graded short answers) were correctly identified using logistic regression, rising to 59% with a more sophisticated elastic net method. Moving on to the MALL papers, Jack Burston and Konstantinos Giannakou report on an extensive meta-analysis of a large number of studies published over the past quarter century in established CALL and education technology journals as well as in graduate theses. Their work focuses on research on learning outcomes and shows that around half the studies reporting learning effects were conducted at university level, most in interventions lasting 8–14 weeks, and most frequently in Asia, the Gulf States, and the US. Of the studies reviewed here, 95% had English as the target language, and, as noted, by far the most common learning objective