Error analysis in automatic speech recognition and machine translation

• Autor(a) principal: Loomans, Nicolaas Dirk Petrus
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10451/57155
• topic: Domínio/Área Científica::Humanidades::Línguas e Literaturas

Automatic speech recognition and machine translation are well-known terms in the translation world nowadays. Systems that carry out these processes are taking over the work of humans more and more. Reasons for this are the speed at which the tasks are performed and their costs. However, the quality of these systems is debatable. They are not yet capable of delivering the same performance as human transcribers or translators. The lack of creativity, the ability to interpret texts and the sense of language is often cited as the reason why the performance of machines is not yet at the level of human translation or transcribing work. Despite this, there are companies that use these machines in their production pipelines. Unbabel, an online translation platform powered by artificial intelligence, is one of these companies. Through a combination of human translators and machines, Unbabel tries to provide its customers with a translation of good quality. This internship report was written with the aim of gaining an overview of the performance of these systems and the errors they produce. Based on this work, we try to get a picture of possible error patterns produced by both systems. The present work consists of an extensive analysis of errors produced by automatic speech recognition and machine translation systems after automatically transcribing and translating 10 English videos into Dutch. Different videos were deliberately chosen to see if there were significant differences in the error patterns between videos. The generated data and results from this work, aims at providing possible ways to improve the quality of the services already mentioned.