Pelos caminhos discursivos e da inteligência artificial em um laboratório virtual para ensino de língua inglesa

Abstract

This dissertation aims to understand the discursive functioning of/in the artificial intelligence (AI) system of ELLA - English Language Learning Laboratory, a virtual laboratory for learning English as a foreign language (FL) that is being developed by the Federal University of Uberlandia and has the participation of researchers from the Federal Institute of Triangulo Mineiro. Teachers and researchers from several areas participate in the team, which made it possible for the themes and concerns related to Applied Linguistics, discourse, language itself, and the AI to be problematized and brought to the surface of the discussions inside the laboratory. It is from within these discussions that the theme of this dissertation emerges. ELLA is a freeware, whose AI system is designed to interact with the student giving feedback when she/he orally enunciates in English. In the process of creating the laboratory, theories in Applied Linguistics, Discourse Analysis, Programming, Artificial Intelligence, Decolonial Thinking, among others, are mobilized, and from these theories we start to question how the discursive dimension of and in programming works in the conditions of the laboratory. The research questions we aim to answer in the study are: how will the natural language processor (NLP) of the AI system behave when transcribing spoken English as FL, enunciated by subjects in initial learning processes? That is, will features (phonological, lexical, syntactic) of the Portuguese as mother tongue of these subjects be a hindrance in the NLP transcriptions? Does this processor rely on semantic and discourse relations for transcription? How will the discourse dimension work in the human-AI conversation system in ELLA? To answer these questions, we have set as a general goal to investigate how the discourse functioning in ELLA may occur. The specific goals are to describe and analyze the initial parsing model of this NLP (syntactic analysis of English at the AI level); and to perform tests with this processor and verify its level of transcription accuracy; identify which physical aspects (sound quality, voice types, noises, pronunciations, etcetera) may affect the processing; predict, from the results obtained for the other objectives, how the discursive dimension in the processes of sense production may occur between student and conversation tool with the AI in ELLA, in a first version of the laboratory. The dissertation is divided in four parts, the first one, "On the paths of Artificial Intelligence and ELLA", brings an evolutionary overview of AI and presents in more detail the virtual laboratory, both its structural part and its functioning one. The second part, "On the Analytical Paths", brings the main concepts and conceptions of language, discourse and ideology in a beat between natural and computational language, besides problematizing NLP, through Deepspeech, which performs automatic speech recognition, transforming it into text. The third part, "On the paths of the Algorithm", brings a problematization about decolonial programming, with rules that guide ELLA's AI and the ways thought by the team to provide feedback to the student. The fourth and last part, "On the Conclusive Paths", brings the final considerations that answer the research questions, as well as the interpellations and the questions made during the writing of the dissertation.