In this study of affordances for second language (L2) learning in World of Warcraft (WoW) group play, we compared three gameplay episodes spanning a semester-long course. Applying multimodal analysis framed by ecological, dialogical and distributed (EDD) views (Zheng and Newgarden, forthcoming), we explored four English as a second language learners' verbalizations and avatar actions. Players learned to take skilled linguistic action as they coordinated recurrent WoW gameplay activities (questing, planning next moves, traveling, learning a skill, etc.). Frequent activities matched Common European Framework of Reference (CEFR) speaking proficiency descriptors, used widely in L2 teaching and learning (L2TL), providing evidence that players engaged in the types of communicative activities interaction-oriented classroom approaches develop. However, in the WoW context, interactions were not planned, but emerged as players dynamically directed the course of play. Furthermore, modalities of avatar-embodiment and conversing over Skype allowed players to flexibly integrate language and actions to co-act toward game goals, discuss non-game topics during play, or demonstrate comprehension with avatar actions alone, an affordance for less verbal players. This research builds on previous work (Zheng, Newgarden & Young, 2012) relating WoW's multiplayer activities and L2 learners' skilled linguistic actions. We refer to Chemero's (2009) model of the animal-environment system to explain how L2 learners develop abilities to take skilled linguistic action by acting on affordances in WoW. The EDD framework presented may enable other researchers to account for more of the complexities involved in L2 learning in multimodal, multiplayer virtual environments.