Detection-Based Approaches: Methods, Theories and Applications

The five studies reported in this volume provide strong evidence of the usefulness of a detection-based approach for classifying texts written in English based on the first language (L1) of the writer. Such an approach not only provides support for theories concerning crosslinguistic influences, but also yields important methodological innovations that will prove beneficial for future research that investigates language transfer, second language (L2) acquisition, language pedagogy, language assessment and forensic linguistics. Historically, language detection methods have been employed to examine authorship attribution (e.g. whether Madison or Hamilton were the authors of the Federalist papers; Mosteller & Wallace, 1964). An important point of departure for the present volume is the notion that identifying an author’s L1 is a type of authorship attribution. It differs slightly from classic authorship studies because we are not interested in identifying a particular author, but rather a class of authors based on the sharing of a similar L1. In doing so, the chapters in this volume are designed to discover features of the L1 that are relatively invariant across the group, but vary within groups. The notion underlying such an approach is that the emergence of clear differences in the patterns of language use among different language groups writing in an L2 can provide evidence of interlingual transfer (Kubota, 1998). Additionally, the approaches found in this volume depend on machine-learning algorithms for group classification, setting up a signal or pattern detection framework. The studies in this volume demonstrate that such a framework provides advantages for identifying subtle linguistic differences that exist between L1 groups that may not be observable in qualitative analyses.