We introduce SampurNER, a fine-grained named entity recognition (FgNER) dataset encompassing all 22 scheduled Indian languages spoken by more than two billion people across various countries. While manual annotation for FgNER resources is often labor-intensive and expensive, distant supervision methods have been employed as a viable solution. However, such datasets are often noisy, with entity mentions tagged with multiple types, requiring computationally intensive noise-aware models for effective FgNER. Moreover, resources for both coarse-grained and fine-grained named entity recognition tasks in Indian languages remain scarce. To address this, we propose an entity-anchored machine translation (EaMaTa) framework that leverages the largest manually annotated English FgNER dataset, FewNERD, to create a large-scale FgNER dataset in 22 languages. On average, the dataset comprises over 153k sentences, 354k entities, and 3.3M tokens in each language. The languages covered are: Assamese (as), Bengali (bn), Bodo (brx), Dogri (doi), Gujarati (gu), Hindi (hi), Kannada (kn), Kashmiri (ks), Konkani (gom), Maithili (mai), Malayalam (ml), Manipuri (mni), Marathi (mr), Nepali (ne), Odia (or), Punjabi (pa), Sanskrit (sa), Santali (sat), Sindhi (sd), Tamil (ta), Telugu (te), and Urdu (ur). Various rigorous analyses and human evaluations confirm the high quality of the dataset and demonstrate the effectiveness of the entity-anchored machine translation framework with up to 9% increase in F1-score against the current state-of-the-art. Additionally, we extend our analysis to zero-shot, multilingual, and cross-lingual settings, investigating the influence of language family and script similarity on cross-lingual FgNER performance.

Several studies have used distant supervision to create resources for fine-grained entity recognition (FgER) to mitigate the challenges of manual annotation.However, most of these methods are primarily developed for English and cannot be efficiently adapted to many other languages, including Indian languages.Moreover, the emergence of new and unseen entity types deteriorates the performance of the supervised models trained on FgER datasets with different predefined sets of entity types.This work introduces TAFSIL, a taxonomy-adaptable FgER framework to create FgER datasets in six Indian languages.The chosen languages are spoken by more than a billion speakers across various countries.TAFSIL utilizes the high interlink between the knowledge base WikiData and linked corpora Wikipedia through multi-stage heuristics and improves annotation through fuzzy match and quality sentence selection.TAFSIL enables us to create datasets of a total size of around three million samples for six languages Hindi (Hi), Marathi (Mr), Sanskrit (Sa), Tamil (Ta), Telugu (Te), and Urdu (Ur) belonging to two language families Indo-European and Dravidian.We evaluate the robustness of TAFSIL by creating various datasets in four taxonomies FIGER, OntoNotes, HAnDS, and MultiCoNER2.Our extensive experiments suggest the sound quality of the datasets as there is a relative improvement of 83% in average F1 score over zero-shot performance across various FgER state-of-the-art models.The resource is publicly available at https://huggingface.co/datasets/prachuryyaI- ITG/TAFSIL.