Using regressions to explore deviations between corpus data and a standard/target: two suggestions
Stefan Τh. Gries, Sandra C. Deshors · Corpora · 2014
The main goal of this study is to develop more appropriate ways to study variation between corpus data that instantiate a linguistic standard or target on the one hand, and corpus data that are compared to that standard, or that represent speakers that may aspire to approximate the target (such as second- or foreign-language learners). Using the example of SLA/FLA research, we first, briefly, discuss a highly influential model, Granger's (1996) Contrastive Interlanguage Analysis (CIA), and the extent to which much current research fails to exploit this model to its full potential. Then, we outline a few methodological suggestions that, if followed, can elevate corpus-based analysis in SLA/FLA to a new level of precision and predictive accuracy. Specifically, we propose that, and exemplify how, the inclusion of statistical interactions in regressions on corpus data can highlight important differences between native speakers (NS) and learners/non-native speakers (NNS) with different native linguistic (L1) backgrounds. Secondly, we develop a two-step regression procedure that answers one of the most important questions in SLA/FLA research – ‘What would a native speaker do?’ – and, thus, allows us to study systematic deviations between NS and NNS at an unprecedented degree of granularity. Both methods are explained and exemplified in detail on the basis of over 5,000 uses of may and can produced by NSs of English and French and Chinese learners of English.