Naamah: New Sanskrit NER Dataset via DBpedia and LLMs

Researchers have introduced Naamah, a large-scale synthetic Sanskrit Named Entity Recognition (NER) dataset comprising 102,942 sentences. The methodology combines entity extraction from DBpedia with a 24B parameter hybrid reasoning model to generate grammatically natural and diverse training data. The dataset benchmarks two transformer architectures: XLM RoBERTa and IndicBERTv2. This work addresses the scarcity of annotated resources for classical Sanskrit literature digitization.