This research seeks to improve collaborative analysis in fields such as criminal justice, intelligence, and epidemiology by developing tools that capture and represent essential elements of analytic conversations. The work has three major aims: (a) understanding how team hypotheses, insights, and problem orientation are reflected in their conversations; (b) developing and testing natural language processing (NLP) techniques to detect teams? hypotheses, insights, and problem orientation; and (c) developing and testing methods to communicate the results of NLP analyses to provide teams with feedback on their own and other teams? reasoning processes. These goals are addressed through behavioral studies, NLP research, and tool development and evaluation. Intellectual merit: The project provides unique contributions in four areas: (a) understanding how team analytical processes are evidenced in team communication; (b) advancing the state-of-the art in NLP by developing discourse analysis techniques for use in collaborative analysis applications; (c) applying NLP techniques to the support of team analytical processes; and (d) designing interfaces to provide feedback within and across teams. The research also furthers the training and education of undergraduate and graduate students. Broader impact: The project has the potential to improve collaborative investigative analysis in many fields of critical importance to society, including criminal justice, intelligence, science, and epidemiology. The results will provide new tools for analysts, recommendations for organizational practices to improve the quality of collaborative analysis, new methods for training professional analysts, and new learning tools for graduate programs in fields such as epidemiological analysis and criminal justice.
Collaborative analysis of information plays an import role in societal domains such as crime solving, epidemiology, and national intelligence. The key research goal of this project was to improve collaborative analysis by automating or supporting the capture and representation of essential elements of analytic conversations and making these elements available to collaborating team members (e.g. for retrospective analysis), and to other teams, to facilitate their problem solving efforts. The project had three major aims: (a) understanding how team hypotheses, insights, and problem orientation are reflected in conversations among team members; (b) developing and testing natural language processing (NLP) techniques to detect teams’ hypotheses, insights, and problem orientation; and (c) developing and testing methods to communicate the results of NLP analyses to provide teams with feedback on their own and other teams’ reasoning processes. We addressed these aims through an interdisciplinary collaboration between personnel at Cornell University and Carnegie Mellon University. First, we conducted laboratory studies to identify problems in communication between analysts as well as errors in reasoning such as cognitive tunneling (focusing attention prematurely on one suspect or explanation and thereby overlooking other key suspects or explanations). Second, we developed new natural language processing (NLP) methods that could automatically analyze conversational data to determine if disagreements among participants were arising and to highlight information that participants needed to remember or were required to identify if a good solution to the problem was going to be constructed. Third, we used human-computer interaction (HCI) techniques to design and test new interfaces that facilitating sharing of information between analysts and reduced cognitive tunneling. The project had important broader impacts in a number of areas. It created new knowledge about how analysts collaborate, new techniques for automatic summarization of text, and new interface designs that can be employed for a variety of analytic tasks. All results were published at highly-refereed, top-tier conferences and workshops in NLP and HCI, which is a sign of their significance and impact on the field. The project also provided training in STEM disciplines for a large number of students, including four doctoral students, several Master's of Engineering students, and over a dozen undergraduates. The PIs organized several outreach and training workshops to provide training to the wider academic community. We hope that through our dissemination efforts, the project will help improve collaborative analysis in areas of societal concern, such as crime and epidemiology.
