The Findings Group, LLC

Authors: Amanda E. Griffith, Gloria Ashiya Katuka, Joseph B. Wiggins, Kristy Elizabeth Boyer, Jason Freeman, Brian Magerko, Tom Mcklin

Abstract

Many important forms of collaborative learning are co-creative in nature. AI systems to support co-creativity in learning are highly underinvestigated, and very little is known about the dialogue mechanisms that support learning during collaborative co-creativity. To address this need, we analyzed the structure of collaborative dialogue between pairs of high school students who co-created music by writing code. We used hidden Markov models to analyze 68 co-creative dialogues consisting of 3,305 total utterances. The results distinguish seven hidden states: three of the hidden states are characterized by conversation, such as social, aesthetic, or technical dialogue. The remaining four hidden states are characterized by task actions including code editing, accessing the curriculum, running the code successfully, and receiving an error when running the code. The model reveals that immediately after the pairs ran their code successfully, they often transitioned into the aesthetic or technical dialogue state. However, when facing code errors, learners were unlikely to transition into a conversation state. In the few cases where they did transition to a conversation state, this transition was almost always to the technical dialogue state. These findings reveal processes of human co-creativity and can inform the design of intelligent co-creative agents that support human collaboration and learning.

Citation:
Griffith, Amanda & Katuka, Gloria & Wiggins, Joseph & Boyer, Kristy & Freeman, Jason & Magerko, Brian & Mcklin, Tom. (2021). Discovering Co-creative Dialogue States During Collaborative Learning. 10.1007/978-3-030-78292-4_14.