Segmentation of the Speech Flow for the Evaluation of Spontaneous Productions in Pathologies Affecting the Language Capacity. 4 Case Studies of Schizophrenia

This paper aims to present a multi-level analysis of spoken language, which is carried out through Praat software for the analysis of speech in its prosodic aspects. The main object of analysis is the pathological speech of schizophrenic patients with a focus on pausing and its information structure. Spoken data (audio recordings in clinical settings; 4 case studies from CIPPS corpus) has been processed to create an implementable layer grid. The grid is an incremental annotation with layers dedicated to silent/sounding detection; orthographic transcription with the annotation of different vocal phenomena; Utterance segmentation; Information Units segmentation. The theoretical framework we are dealing with is the Language into Act Theory and its pragmatic and empirical studies on spontaneous spoken language. The core of the analysis is the study of pauses (signaled in the silent/sounding tier) starting from their automatic detection, then manually validated, and their classification based on duration and position inter/intra Turn and Utterance. In this respect, an interesting point arises: beyond the expected result of longer pauses in pathological schizophrenic than non-pathological, aside from the type of pause, analysis shows that pauses after Utterances are specific to pathological speech when >500 ms.