Uma proposta corpus-driven para a classificação de disfluências segundo a estrutura informacional da fala espontânea

Abstract

The goal of this work is to develop a classification of disfluencies based on the informational structure of spontaneous speech. Disfluencies are understood as instances in which the speaker suspends, either temporarily or not, the flow of speech in order to restructure it structurally or functionally. They may be described etiologically - in terms of their causes in the relationship between language and thought (Chafe, 1980) - or ontologically, through various attempts to classify the phenomena themselves (Schegloff, 2013; Kosmala, 2024). In this work, we begin from the characterization of different types of disfluency and their motivations in relation to informational structure, as understood within the Language into Act Theory (LAcT) (Cresti, 2000; Moneglia & Raso, 2014). Based on corpus-driven analyses, eight classes of disfluencies and ten classes of motivations were defined, which can combine freely to form disfluency-motivation compounds. All disfluencies in a sample of ten texts from the C-ORAL-BRASIL I minicorpus (Raso & Mello, 2012) were manually annotated according to this classification. A total of 944 disfluency-motivation compounds were identified. The relative frequency of each class was described with respect to the architecture of the corpus and the informational units defined by L-AcT. Both disfluency types and motivations exhibit distinct distributions in relation to informational structure. The informational units that most frequently host disfluency phenomena are the Parenthetical (46.7%), the Bound Comment (40.0%), and the Topic (31.5%). Based on relative-frequency measures, we propose an organization of the phenomena at two levels: the locutive level (as in morphosyntactic repairs) and the informational-illocutionary level (as in repairs of informational units or illocution). Disfluencies occurring at the locutive level have no implications for the informational-illocutionary level; those occurring at the informational-illocutionary level, however, may condition additional disfluencies at the locutive level (such as filled pauses and repetitions). This proposal is supported by empirical evidence and demonstrates that even phenomena of fragmentation (disfluencies) act upon both the function and the form of language.