The beauty of nature or nature's beauty? The Genitive Alternation in EFL learners

In this study, I investigate how learners of English as a Foreign Language choose between the variants of the genitive alternation (the tail of the dog/the dog’s tail). The study of alternation phenomena is particularly interesting because they shed light on the different kinds of factors that lead speakers to choose between semantically equivalent structures (Marneffe et al. 2012: 28). While most studies consider native speakers, the present study contributes to the existing literature on alternation phenomena by focusing on learners of English as a Foreign Language. More specifically, I investigate whether learners of English differ in their use of the variants compared to native speakers, and if so, whether these differences are due to processing-related effects and/or learner variables such as proficiency level and mother tongue background. It is hypothesized that more advanced learners with similar structures in their mother tongue should better approximate target usage (see Wulff et al. 2018). The data stems from the Trinity Lancaster Corpus, a three-million word corpus that features intermediate to advanced learners of English from typologically diverse language backgrounds (e.g. Chinese, Spanish, Russian, Hindi) taking part in an official language exam (Gablasova et al. 2019). The multifactorial nature of the alternation I account for by annotating thousands of genitive observations for a wide range of constraining factors, such as the length, animacy, discourse status and overall frequency of the possessor and/or possessum (Heller et al. 2017). The same applies to genitive observations retrieved from the examiners in the corpus, all of whom are native speakers of British English. To these datasets I apply state-of-the-art statistical techniques, specifically mixed-effects logistic regression, where the different factors figure as predictors of the genitive variant. The regression models then reveal how learners deviate from target usage in terms of the predictors’ effect sizes compared across the native and learner samples (e.g. Röthlisberger et al. 2017, Wulff et al. 2014). In order to offer a theory-based account of learners’ use of the genitive variants, I analyze the results from the perspective of both the Probabilistic Grammar framework (Bresnan 2007), which is specifically tailored to deal with alternation phenomena, and compatible Usage-based approaches to Second Language Acquisition such as the CREED model (Ellis 2006). Analysis shows that the choice of genitive variant of more proficient learners indeed better approximates that of native speakers compared to lower-proficiency speakers, although both learners and native speakers as a whole are sensitive to largely the same set of constraints. Surprisingly however, the way in which the probabilistic grammar of low-proficiency learners differs from that of native speakers does not directly follow from the insights of usage-based approaches to Second Language Acquisition, which argue that more learnable probabilistic constraints should be acquired first.