Padrão informacional de stanzas de pacientes com esquizofrenia

• Ano de defesa: 2022
• Autor(a) principal: José Carlos da Costa Júnior
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; FALE - FACULDADE DE LETRAS; Programa de Pós-Graduação em Estudos Linguísticos; UFMG
• Palavras-chave em Português: Esquizofrenia; Padrão informacional; Teoria da Língua em Ato; Comparação entre corpora orais; Processamento de linguagem natural - PLN; Atos de fala (Linguística); Linguística
• Link de acesso: http://hdl.handle.net/1843/44004

Goals: We make a comparison between the informational pattern and speech disfluencies of patients with schizophrenia and participants without this mental illness. The informational pattern represents the different arrays of information units that are found in utterances (CRESTI, 2000; RASO, 2012; MONEGLIA, RASO, 2014). We also offer a preliminary description for speech disfluencies, lexicon and part of speech of a representative sample of patients. Hypotheses: I) patients with schizophrenia could have a less varied and less complex informational pattern - given for the textual units frequency- as it was already suggested in Cresti et al (2015); II) these patients could make more speech disfluencies, as larger filled and silent pauses or more speech reformulations or interrupted utterances than people without schizophrenia. Methodology: We use 12 informational tagged transcriptions from two oral corpora - 6 of each -, namely C-ORAL-ESQ (FERRARI, ROCHA, RASO, forthcoming), which is representative of patients with schizophrenia, and C-ORAL-BRASIL minicorpus (RASO, MELLO, 2012), which is representative of spoken Brazilian Portuguese. Comparison between the two groups was based on the number of Bound Comments - COBs - a textual unit whose presence defines a terminated unit of speech as a stanza. We compare stanzas with the same number of COBs and we count other textual units around, as these last ones could suggest more complexity in speech elaboration. For this purpose, it was developed Python (VAN ROSSUM, DRAKE, 1995) applications, which are free available for users, that extract text transcriptions and normalize their spelling conventions, tag part of speech, detail word frequency and morphology, make statistical analysis, build a balanced research sample and generate charts of all variables compared automatically, among other implementations. This was built mainly with Pandas (MCKINNEY, 2010), for statistical analysis and automation; Natural Language Toolkit (BIRD et al, 2011) and buil-in libraries like re (VAN ROSSUM, 2020) for natural language processing and text mining. Results: Patients with schizophrenia made less: unique patterns, total textual units (p =0,001); Topics (p=0,01); Parenthetics (p=0,03); Locutive Introducers (p=0,01); Multiple Comments (p=0,04) among other significant results in Mann Whitney U test. Interrupted utterances were more frequent in C-ORAL-BRASIL, although patients with schizophrenia made less textual units before their interruption. Retractings, for their turn, presented different frequency patterns in both groups, though they were less frequent in patient’s speech in the most relevant samples, with similar number of syllables and distribution of the first phonetic segment. Regarding pauses, it was observed that patients made them longer than no patients, both for silent (p<0,01) as for filled pauses, though these last ones had no statistical significance. Therefore, we claim that patients with schizophrenia have less elaborated speech from the point of view of the information pattern – hence less varied melodically - as well as they have more speech disfluencies in the considered variables overall.