exentai/Srilankan-Tamil-NER

Sri Lankan Tamil NER

Dataset Description

A high-quality Named Entity Recognition (NER) dataset for the Tamil language, focused on Sri Lankan (Jaffna) regional context. This dataset contains 10,195 carefully annotated sentences covering entities that are commonly missed by general-purpose Indic NER models — including Sri Lankan place names, Tamil organizations, and regional Tamil personalities.

Built and curated by ExentAI as part of a Tamil NLP research pipeline.

• Language: Tamil (ta)
• Task: Token Classification → Named Entity Recognition
• Total Sentences: 10,195
• Label Format: IOB2
• Region Focus: Sri Lanka
• Max Sentence Length: 50 tokens (strictly enforced)
• BIO Label Errors: 0 (fully validated)

---

Dataset Details

Sources & Composition

Source	Sentences
Books V1–V15 (Tamil literary & historical texts)	6,393
Final_Tamil Corpus (regional news & documents)	3,802
Grand Total	10,195

Entity Distribution

Overall entity distribution across the full dataset:

Entity Type	Exact Count
B-PER (Person)	4,533
B-LOC (Location)	8,110
B-ORG (Organization)	3,369
Total Entities	16,012

(Note: 50% of the sentences contain at least one named entity).

Label Schema

Follows the Naamapadam schema exactly — fully compatible with ai4bharat/IndicNER:

ID	Label	Description
0	O	Outside — not an entity
1	B-PER	Beginning of a Person name
2	I-PER	Inside a Person name
3	B-LOC	Beginning of a Location
4	I-LOC	Inside a Location
5	B-ORG	Beginning of an Organization
6	I-ORG	Inside an Organization

---

Dataset Structure

Data Fields

{
  "id":    "0",                                      # Unique sentence ID (string)
  "words": ["யாழ்ப்பாணம்", "உள்ள", "..."],          # List of word tokens
  "ner":   [3, 0, ...]                               # List of integer label IDs per word
}

Dataset Splits

Split	Sentences
Train	~8,156
Validation	~1,020
Test	~1,019
Total	10,195

---

Usage

Load with 🤗 Datasets

from datasets import load_dataset

dataset = load_dataset("exentai/Srilankan-Tamil-NER")
print(dataset)
# DatasetDict({
#     train: Dataset({features: ['id', 'words', 'ner'], num_rows: ~8156}),
#     validation: Dataset({features: ['id', 'words', 'ner'], num_rows: ~1020}),
#     test: Dataset({features: ['id', 'words', 'ner'], num_rows: ~1019})
# })

Label Mapping

LABEL_NAMES = ["O", "B-PER", "I-PER", "B-LOC", "I-LOC", "B-ORG", "I-ORG"]
ID2LABEL = {i: label for i, label in enumerate(LABEL_NAMES)}
LABEL2ID = {label: i for i, label in enumerate(LABEL_NAMES)}

Fine-Tune with Transformers

from transformers import AutoTokenizer, AutoModelForTokenClassification

tokenizer = AutoTokenizer.from_pretrained("ai4bharat/IndicNER")
model = AutoModelForTokenClassification.from_pretrained(
    "ai4bharat/IndicNER",
    num_labels=len(LABEL_NAMES),
    id2label=ID2LABEL,
    label2id=LABEL2ID,
    ignore_mismatched_sizes=True
)

---

Example

Input Sentence:

யாழ்ப்பாண போதனா வைத்தியசாலையில் டாக்டர் கந்தசாமி அவர்கள் சுகாதார அமைச்சு அதிகாரிகளுடன் ஆலோசனை நடத்தினார்.

Annotations:

Word	Tag
யாழ்ப்பாண	B-LOC
போதனா	B-ORG
வைத்தியசாலையில்	I-ORG
டாக்டர்	B-PER
கந்தசாமி	I-PER
அவர்கள்	O
சுகாதார	B-ORG
அமைச்சு	I-ORG
அதிகாரிகளுடன்	O
ஆலோசனை	O
நடத்தினார்.	O

---

Data Quality

• ✅ Zero BIO errors — all IOB2 tag sequences fully validated
• ✅ Orphan I- tags fixed — continuation tags without a matching B- are corrected
• ✅ B-B chain fixes applied — consecutive B- tags for same entity type merged correctly
• ✅ Max 50 tokens per sentence — strictly enforced during preprocessing

---

Intended Use

• Fine-tuning Indic NER models on Sri Lankan Tamil text
• Improving entity detection for Jaffna-domain Tamil documents
• Research in low-resource Tamil NLP
• Building Tamil document processing and knowledge graph pipelines

---

Curation & Annotation

• Sources: 15 Tamil literary/historical books + regional Tamil corpus
• Annotation Method: GPT-assisted CoNLL-format annotation with automated BIO validation
• Preprocessing: Sentence splitting, token length enforcement, BIO error correction
• Organization: ExentAI

---

Citation

If you use this dataset in your research, please cite:

@dataset{exentai_srilankan_tamil_ner,
  author    = {ExentAI},
  title     = {Sri Lankan Tamil NER},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/exentai/Srilankan-Tamil-NER}
}

---

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.