Conversation in the Wild: Automating Discourse Analysis for Assessment of Cognitive-Communication Impairment

Acquired brain injury often causes cognitive-communication impairments (CCI), but speech-language pathologists lack a standardized normed assessment for conversational communication. Large-scale quantification and characterization of natural conversation patterns will inform diagnosis of CCI and facilitate identification of evidence-based treatment targets. The objective of this project is to automate measurement of clinically important interactional and structural features of conversational stories. To do this, we applied Natural Language Processing tools to the StoryCorps Archive, a very large corpus of minimally constrained conversational stories. Our analysis revealed altered interactional dynamics and structural features of conversational stories told by people with memory loss, in comparison to control groups. This research demonstrates the feasibility of automating clinical feature measurement for ecologically valid analysis of conversational story-telling.