The development of probabilistic grammars in spoken English as a Foreign Language

Most previous research on alternation phenomena in learners of English as a Foreign Language (EFL) investigated the influence of learners' mother tongue on the choice of variant. In this project, I instead focused on how proficiency level modulates EFL learners' choice of variant for the future marker alternation (he will read the newspaper vs. he is going to read the newspaper) and the genitive alternation (the dog's tail vs. the tail of the dog), which revealed whether and how learners at similar stages of language acquisition differ from native speakers regardless of their mother tongue background. Methodologically, I collected 3,616 future markers and 2,302 genitive observations from the Trinity Lancaster Corpus, a three-million-word corpus consisting of recordings from an official language exam between a native speaker of British English and low-intermediate (B1) to advanced (C2) learners of English from several mother tongue backgrounds, including Chinese, Hindi, Russian, Spanish and Italian. The observations from both alternations were annotated for the relevant constraints guiding the choice between their variants, such as clause type, clause length, subject grammatical person and animacy for the future markers, and the animacy, length, sibilancy and thematicity of the possessor for the genitive variants. Both annotated datasets were then analyzed using mixed-effects logistic regression, where the constraints were allowed to interact with proficiency level. Since corpus-based regression models primarily capture language use rather than implicit grammatical knowledge, the corpus-based study on the genitive alternation was complemented with a rating task experiment, where 25 native speakers and 101 EFL learners rated the naturalness of the genitive variants in authentic corpus excerpts. Results show that native speakers allow for considerable variation between will and be going to. By contrast, low-proficiency learners rely almost exclusively on will, the first future marker they acquire, whereas the use of be going to is restricted by additional constraints such as the temporal proximity of the future event. At the same time, these low-proficiency learners are less sensitive to certain constraints that influence native speakers' choice of genitive variant, such as possessor definiteness and possessor animacy, because learners have difficulty learning how these constraints are associated with the distribution of the genitive variants in their linguis