Carolyn Rosé

Professor, Language Technologies Institute
Human-Computer Interaction Institute

Research Area

Computer-Supported Collaborative Learning/MOOCs, Information Retrieval, Text Mining and Analytics, Language Technologies for Education, Natural Language Processing and Computational Linguistics

Education

Research

My research focuses on better understanding the social and pragmatic nature of conversation, and using this understanding to build computational systems that improve the efficacy of conversation both between people, and between people and computers. In order to pursue these goals, I invoke approaches from computational discourse analysis and text mining, conversational agents, and computer-supported collaborative learning. I ground my research in the fields of language technologies and human-computer interaction, and am fortunate to work closely with students and post-docs from the LTI and the Human-Computer Interaction Institute, as well as to direct my own lab, TELEDIA. My group’s highly interdisciplinary work, published in 160 peer-reviewed publications, is represented in the top venues in five fields: language technologies, learning sciences, cognitive science, educational technology and human-computer interaction.

An exciting direction of my group's work is spearheading a satellite working group to support social interaction for learning in MOOCs with EdX called DANCE. My research toward this end has birthed and substantially contributed to the growth of two thriving interrelated research areas: automated analysis of collaborative learning processes and dynamic support for collaborative learning. Both areas use intelligent conversational agents to support collaborative learning in a context-sensitive way.

All of my work draws insight from rich theoretical models from sociolinguistics and discourse analysis, and pares them down to precise operationalizations that capture the most important essence for achieving impact. I always start by investigating how conversation works and formalizing this understanding in models that are precise enough to be reproducible and that demonstrate explanatory power in connection with outcomes with real-world value. The next step is to adapt, extend and apply machine learning and text mining technologies that leverage this deep understanding to build computational models capable of automatically applying these constructs to naturally occurring language interactions. Finally, with the technology to automatically monitor naturalistic language communication in place, we can build interventions with real-world benefits.

This approach leads to three aspects included in each project:

• Basic research on discourse analysis to identify conversational constructs that predict important group outcomes such as learning, knowledge transfer or motivation.
• Basic research on text classification technology for automated analysis of conversational constructs identified under research on discourse analysis, as well as tools to enable other researchers to do the same.
• Basic research on conversational agent technology and summarization that eases development of interventions triggered by automatic analyses from basic research on text classification that either enables human facilitators to offer support, directly provide feedback to groups or influence group participation in positive ways.