Transcrição automática de entrevistas e anotação Universal Dependencies no Corpus Roda Viva

Abstract

This article presents the research about the automatic transcription of four interviews extracted from the Roda Viva Corpus, formed by 713 interviews from the Roda Viva Program, on TV Cultura. The original interviews were transcribed by journalists, thus acquiring the status of written text, and also presents interventions, such as encyclopedic information about facts and people mentioned. In order to work with oral text, this research carried out a pilot work of automatic transcription of four of these interviews, using the Whisper tool. Subsequently, the interviews were automatically annotated with the formalization of Universal Dependencies and manually reviewed by the Arborator Grew ElizIA tool. Through this work, it was possible to note the syntactic differences present in the original corpus and in the transcribed interviews.