Probabilistic Latent Semantic Analysis of Verbal and Behavioral Data in Dynamic Mission Operations

Recent efforts to automatically assess cognitive state from communications data using Latent Semantic Analysis (LSA) have produced impressive results (Gorman, Foltz, et al., 2003; Freeman, Thompson, and Cohen, 2000). However, these efforts (1) leveraged only the linguistic data available in simulations, not the rich behavioral data (such as maneuvers, sensor use, and weapons use), and (2) they failed to use new advances on the LSA algorithm. We propose to combine linguistic and behavioral data as input to a recently developed, highly robust statistical technique - Probabilistic Latent Semantic Analysis (pLSA) - to assess performance in Distributed Mission Operations. In particular, we propose to develop a model of knowledge and skills associated with F16 and AWACS training. The proposed model will take linguistic and behavioral input in the form of verbal exercise communications and discrete behavioral measures. This integrated linguistic/behavioral model will output assessments of trainee knowledge and skills. It will be the foundation for the development, in Phase II, of a performance assessment module for the DMO entitled DECA2DE: Dynamic Embedded Communication & Action Assessment in Distributed Environments.