Translation Quality Of Capcut Pro’s Auto-Translated Captions In Indonesian Audiovisual Content(A Study of Instagram Reels from @fadiljaidi)

Abstract

The development of artificial intelligence and machine translation technology has significantly transformed audiovisual communication within digital media platforms. One of the most widely used applications is CapCut Pro, which provides an auto-translated captions feature that automatically generates multilingual subtitles for audiovisual content. Despite its practical advantages, the translation quality of machine-generated subtitles remains problematic, particularly in informal conversational discourse commonly found on social media platforms. This study aims to analyze the translation quality of CapCut Pro's auto-translated captions in Indonesian audiovisual content, specifically Instagram Reels from @fadiljaidi. The study analyzes 53 subtitle data derived from one Instagram Reel video. The study employs a qualitative descriptive method using Translation Quality Assessment (TQA) principles, audiovisual translation theory, and machine translation analysis. The data consist of Indonesian spoken utterances and English subtitles automatically generated by CapCut Pro. The findings reveal that the generated subtitles demonstrate varying levels of translation quality categorized into accurate, moderately accurate, and inaccurate translations. Specifically, 34% of subtitles are categorized as accurate, 45% as moderately accurate, and 21% as inaccurate. Most subtitle outputs fall into the moderately accurate category, indicating that the system generally succeeds in transferring basic semantic meaning but still experiences contextual and pragmatic limitations. The dominant translation issues identified in the study include pragmatic shifts, lexical shifts, literal translation, grammatical awkwardness, and untranslated expressions. The findings further indicate that CapCut Pro's machine translation system struggles to interpret conversational Indonesian discourse particles, colloquial expressions, and culturally contextual utterances naturally in English. Although the feature improves multilingual accessibility in digital audiovisual communication, machine-generated subtitles still require human linguistic evaluation to ensure contextual accuracy and communicative naturalness.