Engagement has gained increasing interest in second language (L2) research (Hiver et al., 2024), alongside the integration of tasks with technology for language learning (González-Lloret & Ortega, 2014). Yet research systematically investigating L2 engagement in technology-mediated task-based settings remains limited, particularly in metaverse platforms that enable immersive, AI-enhanced remote interaction as potential alternatives to conventional synchronous computer-mediated communication (SCMC). Technology-mediated tasks can function as out-of-class activities supplementing classroom instruction, and procedural repetition can impact engagement and learning outcomes. Examining how these task implementation features influence engagement and learning is essential, as is examining the affective pathways through which these effects occur, given that the mediating roles of task-specific emotions and willingness to communicate (WTC) remain underexplored in these contexts. The dissertation has three goals: (1) examine how modality (video-based SCMC vs. virtual 3D) and procedural repetition influence L2 learners’ cognitive, behavioral, agentive, and social engagement and learning outcomes across vocabulary, grammar, and pragmatics; (2) investigate the mediating effects of task-specific emotions (enjoyment, boredom, anxiety) and WTC on engagement and learning outcomes; and (3) analyze the relationship between engagement and learning outcomes. Ninety Korean college EFL learners were assigned to video-based SCMC or virtual 3D conditions, completing three oral role-play tasks in pairs across four sessions alongside post-task surveys, pre- and posttests, a speaking proficiency test, and stimulated recall and perception interviews. Engagement was measured through discourse-analytic indicators, self-reported surveys, and platform-specific technology use, and learning outcomes through vocabulary knowledge and word association tests, wh-question accuracy, and oral discourse completion tests targeting suggestions. The findings revealed that modality and procedural repetition differentially influenced engagement. The virtual 3D group demonstrated greater agentive and behavioral engagement over time, while the video SCMC group showed higher cognitive engagement in certain measures. Procedural repetition consistently enhanced all four self-reported task engagement dimensions through increased task enjoyment and WTC, and partially through reduced task anxiety. Both groups showed comparable vocabulary and grammar gains, while the virtual 3D group outperformed in pragmatic development. Affective states did not mediate learning outcomes, and only social engagement predicted vocabulary learning for the virtual 3D group.