CoPiT improves low-resource Mongolian translation by routing through Cyrillic

Researchers propose CoPiT, a pivot-based translation pipeline that addresses the data scarcity of Traditional-script Mongolian by routing translations through the more resource-rich Cyrillic script. This approach resolves script-induced ambiguity to enable stable and accurate meaning transfer.

• CoPiT consistently outperforms direct translation across multiple backbone models and target languages.
• The method achieves substantial absolute BLEU improvements with consistent 1.5-1.6x COMET gains.
• These gains allow strong open-source models to match or outperform GPT-4.1 under comparable evaluation settings.
• CoPiT enables the construction of synthetic parallel data directly from Traditional-script text.
• The authors release a new multi-script parallel dataset covering Mongolian in both scripts alongside English, Korean, and Russian.

CoPiT mitigates data scarcity in realistic low-resource scenarios by leveraging internal resource hierarchies, allowing open-source models to compete with proprietary systems.