Situated in cognitive interactionist theory and driven by task-based language teaching (TBLT), this study employed a multiple methods design to better address research questions regarding EFL learners' language use and perceptions about their language practices during task-based interaction in Second Life (SL). Findings showed that students perceived SL as a viable platform for language learning. Nine adult EFL learners worldwide were recruited to participate in this virtual course and used avatars to interact with peers via voice chat in simulated real-life tasks.

Quantitative results revealed that confirmation checks, clarification requests and comprehension checks were the three most frequently used strategies. Two strategies that had not been documented in previous SL research were found--metacognitive strategy and "spell out the word." Negotiation patterns were also identified: single-layered and multi-layered trigger-resolution sequences. Additionally, the interrelationship among task types, negotiation and strategies was established--jigsaw task prompted the most instances of negotiation and strategy use whereas opinion-exchange task triggered the least. Results also indicated that EFL students had a statistically significant improvement on syntactic complexity and variety as well as on linguistic accuracy across all measured levels.

Three core themes emerged from qualitative data: 1) perceptions about factors that impact virtual learning experience in SL, 2) attitudes toward learning English via avatars in SL, and 3) beliefs about the effects of task-based instruction on learning outcomes in SL. SL was endorsed as a promising learning environment owing to its conspicuous features, simulated immersion, augmented reality, tele/copresence and masked identities via avatars.

This study demonstrated that implementation of task-based instruction can be maximized by 3-D, simulated features in SL, as evidenced in that 1) convergent tasks with single-outcome conditions stimulate more cognitive and linguistic processes; 2) 3-D multimodal resources in SL provide additional visual and linguistic support; 3) pre-task planning can optimize the quality of learners' linguistic performance; 4) real-life tasks that capitalize on SL features, accommodate learners' cultural/world knowledge, and simulate real-life tasks can make a difference in their virtual learning experiences; and 5) avatar identities boost learners' sense of self-image and confidence.