Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces grounded glossary generation, a novel NLP task that requires models to extract meaningful Sanskrit phrases and their translations from sloka-translation pairs, effectively formalizing traditional commentary practices. A benchmark of 31,316 sloka-translation-glossary triples from the Valmiki Ramayana and Srimad Bhagavatam was created, utilizing Jaccard and Meaning Faithfulness metrics to evaluate performance. The findings reveal that instruction fine-tuning significantly improves outcomes over prompting methods, while challenges in morphological modeling, particularly with sandhi and samasa compounds, hinder accurate lexical decomposition.
Instruction fine-tuning outperforms traditional prompting in generating accurate Sanskrit glossaries, revealing critical insights into morphological modeling challenges.
We introduce grounded glossary generation, a structured task requiring models to recover semantically meaningful Sanskrit phrases and produce translation-grounded meanings from a sloka-translation pair, formalizing the traditional patha commentary practice as an evaluable NLP objective. We construct a benchmark of 31,316 sloka-translation-glossary triples from the Valmiki Ramayana and Srimad Bhagavatam, paired with two metrics: Jaccard for phrase recovery and Meaning Faithfulness for semantic consistency. Across zero-shot, few-shot, and instruction fine-tuned variants of Gemma-3n-E4B, Gemma-3-12B, Phi-4, and Qwen3.5-9B, instruction fine-tuning substantially outperforms prompting, while explicit segmentation yields gains. Error analysis identifies over-segmentation of sandhi and samasa compounds as the dominant failure mode, pointing to morphological modeling as the key bottleneck for faithful Sanskrit lexical decomposition.