Within Dialogue Modeling research in AI and NLP, considerable attention has been spent on “dialogue state tracking (DST)”, which is the ability to update the representations of the speaker’s (user’s) needs at each turn in the dialogue, by taking into account the past dialogue moves and history. This is used to help arrive at the most useful dialogue policy (in an ML model)  to predict next state/actions that should be performed in the dialogue. Less studied but just as important to dialogue modeling, however, is “common ground tracking (CGT)”, which identifies the shared belief space held by all of the participants in a task-oriented dialogue. Carlos will be annotating multimodal classroom interactions, and examining important features to train into CGT models to both identify the current set of beliefs, as well as predict upcoming common ground, e.g., where the dialogue will go (the “questions under discussion,” or QUDs). The goal is to provide a more informative snapshot of the dialogue situation, after each action in the task, to develop a policy incorporating shared beliefs in addition to past dialogue history. Annotations performed over video will be operationalized as features to use in deep neural network models for recognizing and predicting common ground.