ARTIFICIAL INTELLIGENCE AND HUMAN EXPERTISE IN SUBTITLING: A MULTIDIMENSIONAL QUALITY ASSESSMENT OF AI-GENERATED ROMANIAN SUBTITLES
ARTIFICIAL INTELLIGENCE AND HUMAN EXPERTISE IN SUBTITLING: A MULTIDIMENSIONAL QUALITY ASSESSMENT OF AI-GENERATED ROMANIAN SUBTITLES
Author(s): Marcela Alina FărcașiuSubject(s): Language and Literature Studies, Economy, Translation Studies, ICT Information and Communications Technologies
Published by: Editura Universităţii Vasile Goldiş
Keywords: audiovisual translation (AVT); subtitling; artificial intelligence; machine translation post-editing (MTPE); Multidimensional Quality Metrics (MQM);
Summary/Abstract: This study examines the impact of artificial intelligence on subtitling by comparing AI-generated Romanian subtitles with professionally produced human subtitles for a two-minute clip from The Devil Wears Prada. Using the Multidimensional Quality Metrics (MQM) 2.0 framework, the analysis identifies 48 errors across five major dimensions, including accuracy, fluency, style, terminology, and design. The results show systemic weaknesses in subtitle segmentation, idiomatic translation, domain-specific terminology, and register control, producing a Raw Quality Score far below professional standards. The findings highlight the continued necessity of human post-editing and support a collaborative human–AI model for future subtitling workflows in practice today.
Journal: Studii de Ştiinţă şi Cultură
- Issue Year: XXII/2026
- Issue No: 2
- Page Range: 116-132
- Page Count: 17
- Language: English
