Automating intercultural communication: pragmatic failures and cultural bias in AI-mediated Indonesian–English translation

Abstract

AI-mediated translation increasingly facilitates Indonesian–English communication, yet fluent outputs may obscure pragmatic failure and culturally reductive representation in institutionally produced public discourse across tourism, education, public information, and everyday digital interaction in Indonesia. This study examines how automated translation preserves interpersonal meaning, represents culturally embedded Indonesian expressions, and maintains consistency across systems and repeated generations. A comparative qualitative corpus design analysed twelve Indonesian source segments from verified public institutional documents using a Pragmatic Equivalence Matrix, Cultural Representation Audit, and Cross-System Stability Test across forty-eight matched outputs. Findings showed that interactional stance permeated all source segments, while audience alignment, mitigated directives, and figurative meanings created recurrent pressure points for automated translation. Cultural vulnerability centred on mudik, gotong royong, culinary naming, gift practices, and ritual symbolism, with generalisation and Anglocentric reframing compressing religious, communal, regional, and symbolic associations. Cross-system comparison indicated 56.3% strategy agreement, 60.4% lexical or framing variation, 37.5% recurrent pragmatic shifts, 31.3% repeated cultural reframing, and 70.8% severity consistency, demonstrating that stable surface quality did not ensure intercultural adequacy. This study contributes an integrated evaluative framework that reconceptualises AI translation quality as the combined preservation of pragmatic force, cultural accountability, and cross-run stability rather than lexical accuracy and fluency alone.