Automatic recognition of prosodic patterns in semantic verbal fluency tests - an animal naming task for edutainment applications

This paper automatically detects prosodic patterns in the domain of semantic fluency tests. Verbal fluency tests aim at evaluating the spontaneous production of words under constrained conditions. Mostly used for assessing cognitive impairment, they can be used in a plethora of domains, as edutainment applications or games with educational purposes. This work discriminates between list effects, disfluencies, and other linguistic events in an animal naming task. Recordings from 42 Portuguese speakers were automatically recognized and AuToBI was applied in order to detect prosodic patterns, using both European Portuguese and English models. Both models allowed to differentiate list effects from the other events, mostly represented by the tunes: L* H/L(-%) (English models) or L*+H H/L(-%) (Portuguese models). However, English models proved to be more suitable because they rely in substantial more training material.