A Computational Theory of Complex Problem Solving Using Latent Semantic Analysis

The literature of complex problem solving and system control has focused on how to improve the adaptation of operators to new, unpredictable circumstances. The present work reviews the main methodologies and assumptions that are currently being used in complex, dynamic task to answer questions regarding the adaptability problem, i.e. the work on DuressII (Vicente and Collaborators) and on Firechief (Cañas and collaborators). Some methodological problems for Cañas et al. analysis assumptions that could have important consequences in the results obtained are discussed. This study proposes Latent Semantic Analysis (LSA) as an alternative that remedies some of the flaws and adds some interesting new possibilities of analysis, such as coherence measures to assess performance changes in a similar vein as the Within-trial Trajectory Deviation (WTD) used in continuous systems such as DuressII. The study uses an LSA corpus created from the experimental data generated by past experiments in Firechief on adaptation to unpredictable task changes to replicate and extend the results previously obtained. The new LSA approach and results obtained are discussed. The fact that results from both microworlds could be explained by LSA with no modifications in its basic assumptions promises a future common theory and method of complex problem solving.