Accuracy And Meaning Shift in Machine Translation of Indonesian Political Discourse Into English

Abstract

This study examines the performance of neural machine translation (NMT) in rendering political and social discourse on X during Indonesia’s 2024 general election, with a focus on translation shifts, functional equivalence, and the maintenance of pragmatic force. Drawing on the frameworks of Catford (1965), Nida and Taber (1969), and House (2015), the study conducts a qualitative analysis of political and personal microtexts containing sociocultural lexis, institutional metaphors, and colloquial and sarcastic expressions. Data were collected through purposive sampling of high-engagement posts and comment threads, and source texts in Indonesian were systematically compared with their English outputs produced by Google Translate.The findings indicate that unit shifts and structure shifts are the dominant patterns in NMT, particularly in the normalization of syntax and the realization of culturally embedded concepts that lack direct lexical equivalents. From a functional equivalence perspective, equivalent effect is largely preserved in formal and institutional discourse characterized by universal terminology and normative registers, whereas colloquial and affective texts exhibit substantial degradation due to lexical normalization and the omission of pragmatic markers. Analysis of pragmatic force further reveals a recurrent shift from interpersonal to referential function in informal social media texts, resulting in the neutralization of critique and ideological stance.The study contributes to Translation Studies and political discourse analysis by extending shift-based and functionalist models to algorithmic mediation in social media contexts and highlighting the need for pragmatically informed validation when NMT is used for cross-linguistic monitoring of political discourse.