Silences and disfluencies in a corpus of spoken Italian produced by patients with Alzheimer's Disease

The corpus of semi-spontaneous speech “Corpus di italiano parlato patologico - Malattia di Alzheimer” (CIPP-ma) produced by Italian subjects with Cognitive Impairment evolved in AD (divided on the basis of the MMSE score) was analyzed to test the recurrence of disfluency phenomena in the patients' dialogues. Speech samples were based on two different tasks: picture description and semi-structured interview. The disfluencies annotated are word-related phenomena, verbal and non-verbal vocal phenomena, and silent pauses. The unit of analysis was identified by the XML tag “pause”, a threshold was set to include pauses less than or greater than one second. Furthermore, an analysis focused on disfluency ‘chains’ (a cluster of several disfluent phenomena occurring in a sequence that constitutes a coherent unit within a turn) was carried out, identifying all possible disfluency permutations. Our data confirmed the importance of silent pauses and disfluency chains as the discriminating cue between neurotypical and pathological, with particular differences related to tasks.