The term data visualization is often used to describe computer software or its products. I argue that learning sciences research can benefit from conceptualizing data visualization instead as a mode of shared sense-making, in which the co-construction of meaning is mediated by visual data artifacts. This process can be studied at sociocultural and microgenetic levels of analysis, for purposes of better understanding the multiple literacies and trajectories of learning in which visual data are employed. In this talk I present a microgenetic analysis of these social processes of data visualization in groups of 6th-grade students, evidenced in their emergent discourse practices in small group work over the course of an extended science investigation using visual data. A comparison of two students' patterns of participation in their respective small groups -- each a student with low levels of talk in the group overall -- highlights differences in the construction of roles for non-dominant students with respect to domain discourse around data. Implications for science learning are interpreted in the context of particular concepts and skills evidenced by each student in post interviews.