AI Paper Atlas

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

AAAI 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
train on EaMaTa-translated FewNERD-style data instead of the distantly supervised TAFSIL dataDECENT with XLM-R base · TAFSIL Sanskrit testmicro F1: 40.2 -> 43.9Improved
train on EaMaTa data instead of the larger TAFSIL dataDECENT with XLM-R base · TAFSIL Hindi testmicro F1: 70.0 -> 70.9Improved
train on EaMaTa data instead of MultiCoNER2 train dataDECENT with XLM-R base · MultiCoNER2 Hindi testmicro F1: 71.5 -> 72.8Improved
train on EaMaTa data instead of MultiCoNER2 train dataDECENT with XLM-R base · MultiCoNER2 Bengali testmicro F1: 70.6 -> 71.2Improved
add Stage 2 entity-anchor mismatch filtering (ablation)EaMaTa cleaning · SampurNER (several languages)F1: +2.8% averageImproved
add Stage 3 entity-count mismatch filtering (ablation)EaMaTa cleaning · SampurNER (several languages)F1: +2.4% averageImproved
use IndicBERTv2 (pre-trained on all 22 languages) instead of mBERTSampurNER fine-tuning · Santali silver testmicro F1: 22.8 -> 51.2Improved
use IndicBERTv2 instead of mBERTSampurNER fine-tuning · Odia silver testmicro F1: 16.8 -> 64.7Improved
use IndicBERTv2 instead of mBERTSampurNER fine-tuning · Hindi silver testmicro F1: 61.4 -> 62.4Improved
fine-tune mBERT on Bengali (seen) rather than Assamese (unseen), zero-shot to the other languagemBERT · Assamese/Bengali zero-shotmicro F1: Assamese->Bengali 49; Bengali->Assamese 28Mixed
fine-tune on languages sharing script or family rather than all 22 (ablation)IndicBERTv2 · cross-lingual zero-shotmicro F1Improved
train on all 22 languages versus only Perso-Arabic-script languages (ablation)IndicBERTv2 · Santali zero-shotmicro F1Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy