Error Analysis of Machine-Generated Translations in TikTok Captions

Abstract

This study investigates the types and forms of translation errors produced by TikTok’s automatic translation feature, particularly in influencer captions translated between English and Indonesian. The research was conducted using a descriptive qualitative approach, focusing on sentence structure deviations such as inaccurate word order, missing or added words, literal idiom translation, clause shifts, and linkage failures. Data were collected from twenty influencer accounts with high engagement and analyzed using Vilar et al.’s (2006) error classification framework. The findings reveal that the most dominant errors involve incorrect word arrangement, omission of function words, and literal interpretation of idiomatic expressions, all of which distort sentence meaning and disrupt grammatical relationships between sentence elements. These results demonstrate that TikTok’s machine translation system still struggles to recognize flexible sentence patterns and informal language commonly found in digital communication. The study highlights the need to develop translation models capable of understanding non-standard expressions, contextual meaning, and mixed-language usage typical of social media platforms. This research contributes to the improvement of machine translation quality for Indonesian and provides insights for future development of more context-aware translation technologies.