Research map / Label-efficient and robust learning / Recognition and long-tail
SampurNER: Fine-Grained Named Entity Recognition Dataset for 22 Indian Languages
Prachuryya Kaushik, Ashish Anand
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on EaMaTa-translated FewNERD-style data instead of the distantly supervised TAFSIL data | DECENT with XLM-R base · TAFSIL Sanskrit test | micro F1: 40.2 -> 43.9 | Improved |
| train on EaMaTa data instead of the larger TAFSIL data | DECENT with XLM-R base · TAFSIL Hindi test | micro F1: 70.0 -> 70.9 | Improved |
| train on EaMaTa data instead of MultiCoNER2 train data | DECENT with XLM-R base · MultiCoNER2 Hindi test | micro F1: 71.5 -> 72.8 | Improved |
| train on EaMaTa data instead of MultiCoNER2 train data | DECENT with XLM-R base · MultiCoNER2 Bengali test | micro F1: 70.6 -> 71.2 | Improved |
| add Stage 2 entity-anchor mismatch filtering (ablation) | EaMaTa cleaning · SampurNER (several languages) | F1: +2.8% average | Improved |
| add Stage 3 entity-count mismatch filtering (ablation) | EaMaTa cleaning · SampurNER (several languages) | F1: +2.4% average | Improved |
| use IndicBERTv2 (pre-trained on all 22 languages) instead of mBERT | SampurNER fine-tuning · Santali silver test | micro F1: 22.8 -> 51.2 | Improved |
| use IndicBERTv2 instead of mBERT | SampurNER fine-tuning · Odia silver test | micro F1: 16.8 -> 64.7 | Improved |
| use IndicBERTv2 instead of mBERT | SampurNER fine-tuning · Hindi silver test | micro F1: 61.4 -> 62.4 | Improved |
| fine-tune mBERT on Bengali (seen) rather than Assamese (unseen), zero-shot to the other language | mBERT · Assamese/Bengali zero-shot | micro F1: Assamese->Bengali 49; Bengali->Assamese 28 | Mixed |
| fine-tune on languages sharing script or family rather than all 22 (ablation) | IndicBERTv2 · cross-lingual zero-shot | micro F1 | Improved |
| train on all 22 languages versus only Perso-Arabic-script languages (ablation) | IndicBERTv2 · Santali zero-shot | micro F1 | Hurt |
Closest papers by meaning
- MCL-NER: Cross-Lingual Named Entity Recognition via Multi-View Contrastive Learning (AAAI 2024 · similarity 0.88)
- UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition (ICLR 2024 · similarity 0.88)
- GuideNER: Annotation Guidelines Are Better than Examples for In-Context Named Entity Recognition (AAAI 2025 · similarity 0.86)
- PaDeLLM-NER: Parallel Decoding in Large Language Models for Named Entity Recognition (NeurIPS 2024 · similarity 0.86)
- Small Language Model Makes an Effective Long Text Extractor (AAAI 2025 · similarity 0.86)
- Multi-Grained Query-Guided Set Prediction Network for Grounded Multimodal Named Entity Recognition (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
