Date(s) - 27/06/2013
4:00 pm - 5:00 pm
I will present work aimed at automating the assessment of differences between schizophrenic patients and healthy controls in personal emotional narratives and evoked emotion production.
First I present a comprehensive analysis of lexical use in persons with schizophrenia and healthy controls, performed on autobiographical narratives of emotional experiences of five basic emotions. We identified a number of significant differences and confirmed that these differences are a stable indicator of medical status. A classifier using only a small number of highly predictive features is capable of predicting subject status with accuracies much higher than chance. Emotions of anger and happiness revealed greater differences between the two groups, and automated clinical status prediction based on these narratives had higher accuracy.
Next I will discuss a method for speaker-sensitive emotion recognition from voice developed for the purpose of assessing affect blunting and inappropriate affect. Emotion recognition is modeled as a ranking problem where rankers are created for each emotion and each user is treated as a query. I will present experiments on non-clinical datasets which indicates that the proposed method considerably improves the recognition of pure emotion in both acted and spontaneous speech