Answers are grounded in this paper and cite their sources.
This study investigates how AI translation systems handle the Indonesian gender terms wanita and perempuan, which carry distinct ideological and historical meanings. Using corpus analysis and translation experiments with Google Translate, Gemini, Copilot, and ChatGPT, the research finds that these systems tend to homogenize both terms into “women,” thereby reducing socio-political nuance. Corpus evidence shows that wanita is often linked to formal and domestic contexts, while perempuan is more frequently associated with political and activist discourse. Yet translation outputs consistently collapse these distinctions, reflecting patriarchal distributions embedded in training data. To interpret this phenomenon, the study applies the Algorithmic Gender Representation Paradigm (AGRP), which situates bias within three dimensions: ideological residues from past regimes, the homogenizing tendencies of algorithmic logic, and the absence of ethical auditing sensitive to local contexts. The findings highlight that homogenization is conditional rather than uniform, with occasional partial distinctions reintroduced in activist discourse. These results underscore the implications for bilingual feminist scholarship: collapsing wanita and perempuan into a single English equivalent risks silencing feminist narratives and obscuring the ideological struggle encoded in language.
No takes yet. Share an insight, caveat, or question.
Yumasdaleni et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: