It is well known that interaction is a crucial source of input and serves as a pre- requisite for learning (Ellis 2008), and that L2 users’ engagement levels in interactive tasks enhance learning (Tudor 2001; van Lier 2008). Recent interaction research has shown that drawing on L2 users’ entire linguistic repertoires can enhance students’ levels of engagement in class discussions (cf. Author 2013), and that flexible, multilingual practices enhances learning (Hall & Cook 2012; Hornberger 2002). We also know from a number of recent studies that multilinguals naturally activate other languages known to them as they engage with L2/L3/L4 learning tasks (Cenoz & Gorter 2011; Falk & Bardel 2010; García 2013).

Building on this research and framed by the Interaction Hypothesis (Gass 2003) and task-based language teaching (e.g. van den Branden 2007), the present study presents a detailed, qualitative analysis of teacher-led interaction aimed to enhance the learning of L2 morphosyntax in three bilingual classrooms at a Swedish university. It forms part of longitudinal study, combining ethnographic data collection with an experimental design. Three groups of students (CEFR level B2) were taught L2 grammar over one semester by the same instructor, who also audio- recorded the lessons. Two of the three groups were formed by matched-pair random assignment, while the third group was an intact group. In two of the groups, bilingual grammar tasks were used, typically involving L1-to-L2 translation tasks, whereas in the third group exactly the same grammar structures were covered using tasks without involving L1-L2 comparison or translation, except for the end-of-course exam preparation session. The analysis focuses on examining student and teacher agency as an effect of the whether the tasks were bilingual (L1-to-L2 translation) or monolingual (in L2 only), and as an effect of properties of the three different groups. The data were examined using Nexus Analysis (Scollon & Scollon 2004), which is (critical) discourse analysis of ethnographic data, focusing on social actions in groups (in this case student-initiated communicative turns). The results reveal that student interaction patterns in the three different groups differed as an effect of the different tasks. Consistently across all three groups, student-initiated turns were more common when students drew on their L1, and there was a tendency for students who otherwise remained quiet to initiate discussion in the L2 when having the opportunity to draw on their L1. The analysis also shows a certain amount of variation between the groups when engaging with the same task. The nexus analysis of the interaction patterns presented offers a socio-culturally context-sensitive way of understanding why and how teacher-student and student-teacher interaction developed and differed between the three groups.