Sintaxe da fala, probabilidade e cognição: uma proposta integrada de investigação em estudo sobre NPs baseado em corpus

Abstract

This dissertation consists of three broad complementary objectives: (i) to propose a theoretical reflection and to provide necessary subsidies for a more detailed investigation of spoken syntax and (ii) to propose a case study based on what was reflected in theoretical terms, using the following methods derived from statistics and (iii) propose a domain for spoken syntax. The starting point is Language into Act Theory [L-AcT] (CRESTI, 2000) - a pragmatic theory that individualizes utterance as the reference unit of spontaneous speech. L-AcT assumes that the information structure of the utterance is signaled by prosodic breaks, which are assigned different pragmatic units with distinct functions. Topic (TOP) is one of the information units and it is defined as the scope in which the illocutionary force is to be applied. Cresti (2014) says that any kind of syntactic relation can only take place if the locutive content of the utterance is arranged within the same information unit, whereas if such content is distributed between two units, the syntactic relations cease and what emerge are the pragmatic relations arising from the respective functions of each informational unit. However, this criterion may not always occur in speech. And this is what leads us to a probabilistic overview for the study of syntax, i.e., what are the chances that elements can establish syntactic relations between information units? In addition, there needs to be some theoretical foundation that takes into account the constitutive aspects of syntactic relations, which are inherently cognitive-computational, and that is compatible with speech data, including the probabilistic dimension of language. Cognitive Grammar [CG] (LANGACKER, 2008) is a theory compatible with empirical and experimental research and has been tested along these lines. From the proposed discussion and aiming to answer the above question, we seek to apply what was reflected in a case study. Our aim was to investigate the odds that a NP in TOP would establish a subject-verb relation given some selected variables. We used three minicorpora labeled according to L-AcT parameters, extracting data from English, Brazilian Portuguese and Spanish. The statistical method used was a mixed logistic regression model with crossover random effects. Quantitative analyzes were conducted with the aid of R software (R CORE TEAM, 2018). However, we only achieved statistically significant results with Spanish data. Regression analysis with Spanish data, however, showed that there is a greater likelihood that an NP in TOP will establish a subject relationship with COM content if such NP is defined, animated, and given. Subsequently, the results were interpreted according to the CG. A provisional analysis pointed out that for the cases in which the NP is subject, there would be a congruence in the trajector alignment in TOP, both at the sentence level and at the discursive level, whereas in the non-syntactic relationship alignment only appears at the discursive level between the TOP and COM units. A conditional inferece tree model was also applied to the Spanish data. The application of the model showed that animated NPs are the subject of the COM verbs, whereas inanimate NPs have a greater interaction with the other variables. If the NPs are inanimate and the verb of the next unit is of the modulation and relational verbal classes, the NPs tend to be subject. If the verbs are from the other classes (existential, material, mental and verbal), there is still interaction with the type of TOP, i.e., if the verbs belong to the above classes and if the TOP is type 2, the NPs tend to subject, whereas if TOP is type 1, 3 or flat, NPs tend to exhibit no syntactic relation. Finally, through empirical and experimental arguments, it was proposed that the constituent is the domain of spoken syntax.