A legendagem automática em vídeos de popularização científica: um estudo de caso

Abstract

The use of videos in the educational environment has grown in parallel with the increased production of materials aimed at disseminating science or even teaching scientific concepts. A large proportion of these materials are in English, which requires the use of subtitles for audiences who cannot understand this language. This study sought to compare auto-captions (English-language) and machine-translated (Portuguese-language) subtitles with human edited or post-edited subtitles (for non-locale-specific Portuguese or localized Brazilian or European Portuguese). Two videos were selected, with an excerpt of each being extracted for analysis: 377 seconds of "How big is a mole?", containing 35 segments; and 87 seconds of the "How small is an atom?", containing 37 segments. The subtitle transcripts were aligned in tables and organized according to the auto-captions in English. The classification proposed by Koponen (2010) was used to analyze both the concepts (omitted, added, not translated, translated correctly, translation error, and replaced) and the general (in)adequacies of the subtitles. The analysis resulted in 51 classification entries; the adequacies and inadequacies of both the machine and humans indicate that the machine-generated subtitles can convey the gist of the information and meet the immediate interests of understanding an online video.