A relação entre produção e percepção de pistas prosódicas na segmentação de narrativas espontâneas

• Ano de defesa: 2017
• Autor(a) principal: Silva, Ebson Wilkerson Rocha da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Alagoas; Brasil; Programa de Pós-Graduação em Letras e Linguística; UFAL
• Palavras-chave em Português: Discurso narrativo; Prosódia; Segmentação - Percepção; Prosody; Segmentation - Perception; Narrative speech; CNPQ::LINGUISTICA, LETRAS E ARTES::LINGUISTICA
• Link de acesso: http://www.repositorio.ufal.br/handle/riufal/2963

This study is focused on the relation between prosody and the segmentation of the narrative discourse. It is known that prosodic elements, such as pause, duration and boundary tones contribute to the semantic identification of discursive units, in order to resolve the intentions of the speaker to the audience and to facilitate the communication process. The corpus of the dissertation has a set of 4 (four) narratives - collected from Oliveira (2000) - orthographically transcribed without punctuation and paragraphs. Each narrative has got four different presentation conditions, namely: (i) narrative orthographic transcription only; (ii) transcript and audio; (iii) narrative audio only and (iv) narrative filtered audio, resulting in a dyslexicalized (unintelligible) version of the narrative. The narratives were presented to 48 participants in total. Their main task was to identify the points of the narrative in which the speaker intended to finish a communicative unit; after the perception experiment, the narrative locations in which at least 41% of the participants agreed to have an end of a communicative unit were marked. Another step of the project consisted in using scripts to perform an automatic segmentation of the narratives, SG Detector and Beat Extractor were respectively used to segment into stress groups and to segment the narratives into vowels onset units. The goal of this dissertation is (i) to investigate the agreement between intention-based segmentation and automatic segmentation and (ii) to verify prosodic aspects perceived by the participants but not recognized by automatic segmentation. The results suggest that prosodic information facilitates the listener’s perception to how the speaker constructs the narrative text. The pauses related to an agreement between the participants and the script are more frequent and longer than in places where there is no agreement. Regarding the duration of the vowels onset unit, it was verified that the mean duration prior to the marking is also higher in parts where there is agreement.