Abstract
Dyadic interactions among humans are marked by speakers continuouslyinfluencing and reacting to each other in terms of responses and behaviors,among others. Understanding how interpersonal dynamics affect behavior isimportant for successful treatment in psychotherapy domains. Traditionalschemes that automatically identify behavior for this purpose have often lookedat only the target speaker. In this work, we propose a Markov model of how atarget speaker's behavior is influenced by their own past behavior as well astheir perception of their partner's behavior, based on lexical features. Apartfrom incorporating additional potentially useful information, our model canalso control the degree to which the partner affects the target speaker. Weevaluate our proposed model on the task of classifying Negative behavior inCouples Therapy and show that it is more accurate than the single-speakermodel. Furthermore, we investigate the degree to which the optimal influencerelates to how well a couple does on the long-term, via relating torelationship outcomes