Multilingualism research group, seminar 3

Speakers: Dr Dora Alexopoulou and Professor Margaret Deuchar

Factors favouring the production of code-switching by Welsh-English bilingual speakers

Professor Margaret Deuchar (Dept. of Theoretical and Applied Linguistics)

This paper presents new findings regarding the relationship between early bilingual acquisition and the use of code-switching in adulthood. Following the development of a method to extract 67,515 clauses from a Welsh-English corpus (www.bangortalk.org.uk) automatically, we used RBrul to determine which extralinguistic factors appeared to influence the production of bilingual versus monolingual clauses. We found that intraclausal code-switching was produced more frequently by those who had acquired Welsh and English in infancy than those who had acquired the two languages sequentially. There was also an independent effect of age on code-switching, suggesting a change in progress.  Explanations for these findings will be discussed. 

From applied language learning to linguistic theory

Dr Dora Alexopoulou (Dept. of Theoretical and Applied Linguistics)

There has been little, if any, interaction between applied language learning research and linguistic theory. The former focuses on the proecess of language learning in real life contexts while the latter is primarily concerned with the nature of linguistic knowledge. In this exploratory talk, I will sketch a line of inquiry aiming to show that empirical research from applied language learning can bring insights to a fundamental question of linguistic theory, namely the amount of prior linguistic/hierarchial knowledge that is necessary (or not) for successful language acquisition.

To do this, I will start from an unlikely place, the emergence of online EFL teaching platforms offering teaching and learning to students around the globe. The activity of these institutions results in unprecedented amounts of learner production data: data can come from rich task sets across the proficiency spectrum and learners from a variety of linguistic, educational and cultural backgrounds. Exploiting such datasets opens important opportunities for research. But at the same time, such datasets have all the pitfalls of big data: a range of variables standardly controlled for in carefully designed data collections (e.g. task sets) are not considered. Access to unprecedented numbers of learners is set against lack of rich learner metadata targeted in typical data collections. In addition, the very context of production poses arbitrary constraints (e.g. word limits on writings). Last, but not least, the size of such datasets brings new challenges for extracting information and addressing the noisy aspects of the data.

Using the EF-Cambridge Open Language Database (EFCAMDAT) as an example of a big data resource, I will show how Natural Language Processing (NLP) technology can help us address many of the methodological issues and, crucially support new generalisations about learner grammars that can provide insight to the question of whether learners arrive at generalisations on an item-by-item basis or draw from underspecified hierarchical structure.