Adaptação linguística e acessibilidade em sistema com agente conversacional (chatbot)

Abstract

The innovation brought about by the possibility of providing access to health through technology has led to the emergence of the term "telehealth," which involves using information and communication technologies to promote advances in the spheres of health, care, and wellbeing (GROYER; CAMPBELL, 2019). Among these technologies are virtual assistants, or chatbots, which have played a growing role in the Health field. This thesis aimed to analyze the ANA chatbot system, a virtual assistant specialized in clinical and educational guidance on COVID-19, considering its accessibility at the: (1) textual level, related to language, as explored by the fields of Textual Adaptation and Accessible Audiovisual Translation; and (2) computational level, related to usability, explored by the field of Human-Computer Interaction. To this end, users with varied levels of schooling, with and without visual impairment, were contemplated. The research sought to investigate the accessibility of the chatbot language, considering that it was developed by professionals in the Health area, using domain taxonomies, which can generate problems for the effective understanding of an audience without domain in the area and with the expectation of a colloquial language. In addition, the study sought to ascertain problems related to usability, considering that mobile applications demand solutions for the needs of the public in general, and especially for the visually impaired public. The analysis consisted of evaluations based on comprehension test with participants, elaborated from language adaptation methodologies, provided by Doak et al. (1996), and from evaluations of human-machine interaction systems, provided by Lazar et al. (2017) and Petrie et al. (2006). The results obtained confirmed the need to adapt the language and content, since even the audience with an average level of education presented difficulty in understanding statements and terms. There were also problems related to the system's usability, especially for the visually impaired participants, who reported recurring obstacles in accessing it with a screen reader. From the results, guidelines for promoting the accessibility of conversational systems were listed. Adopting these guidelines and the results of an analysis of the texts based on Systemic-Functional Linguistics, a final version of the texts was elaborated to meet the requirements of textual accessibility and usability. It is expected that the results can contribute to the consolidation of guidelines for the production of future accessible conversational systems, considering several user profiles, having, therefore, a greater potential to reach the population, without being restricted to a public of higher educational level and without disabilities.