AI as Cultural Mediation: Agentic Sanskrit–English Translation with Linguistic Grounding

Jintao Ma, Junwen SHEN, Xinyue WANG, Leqi LIU, Dengkui Hou, lingxiang hu, nicolas turenne, Dun Li

Keywords: culturally grounded AI, cultural fidelity, symbolic grounding, linguistic heritage, interpretive translation, civilizational knowledge systems

TL;DR: A training-free, agentic translation framework that grounds LLMs in external symbolic resources to preserve cultural fidelity and prevent semantic drift in civilizational knowledge systems.

Abstract: Machine translation for civilizational knowledge systems, such as Sanskrit, presents challenges that extend far beyond mere data scarcity. In these contexts, purely statistical generative AI risks distorting culturally embedded meanings, ritual concepts, and historical genealogies through semantic drift. To address this, we propose a training-free, agentic decoding framework that grounds AI generation in external cultural and symbolic knowledge resources. By orchestrating tools such as traditional dictionary lookup, morpho-syntactic analysis, and strict glossary constraints at inference time, our system helps stabilize culturally situated meaning without requiring parameter updates. Experiments across 5 Sanskrit–English benchmarks—highlighting classical epics and religious prose—demonstrate that our approach not only yields consistent gains in standard metrics (BLEU and chrF) but significantly enhances cultural fidelity. By preserving sacred named entities and complex ritual terminology, these results demonstrate that linguistically grounded, agentic AI can successfully act as a cultural mediator, ensuring the preservation of meaning across civilizations rather than merely optimizing for resource constraints.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 78