Collaborative learning has long been proved to be a crucial agent for enhancing students' social skills, problem-solving abilities and individual learning performance. Understanding how students move from one phase to another in their collaboration process can inform educators of how best to facilitate such learning. However, this is still an area needing more vigorous research. This study explores the process of students' co-construction of knowledge in a digital game-based learning environment using a sequential analysis approach, with the aim to identify the significant features of students' interactional transitions that would potentially lead to better learning. To achieve this aim, we designed a collaborative learning environment in which 24 sixth graders worked in 6 groups to compete in an Augmented Reality-based mathematic game that we designed. The results from the analysis of their interaction show that phases of interaction in a digital game-based learning environment transited differently in many ways from those in online text-based forums, and that task types, modalities, participant characteristics and learning goals all played a crucial role in determining how learners co-constructed knowledge.