---
dataset_info:
  features:
  - name: word
    dtype: string
  - name: lemma
    dtype: string
  - name: id
    dtype: string
  splits:
  - name: train
    num_bytes: 1407836
    num_examples: 12554
  download_size: 521123
  dataset_size: 1407836
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: mit
task_categories:
- text-generation
language:
- ta
pretty_name: tamil_lemmatizer
size_categories:
- 10K<n<100K
---

A high-quality dataset curated for training and evaluating **Tamil lemmatization models**.
This dataset contains inflected Tamil word forms paired with their **base lemma**, enabling machine learning models to learn morphological normalization for Tamil.


## **Dataset Overview**

Tamil is a morphologically rich language where a single verb can appear in **hundreds of surface forms** due to tense, person, gender, number, case, and auxiliary markers.
A lemmatizer helps convert these complex inflected forms into the dictionary base form.

This dataset is designed for:

* Lemmatization model training
* NLP text normalization
* Morphological analysis
* Character-level seq2seq experiments
* Linguistic research in Tamil morphology

## **Features**
| Field      | Description                         |
| ---------- | ----------------------------------- |
| **word**   | Inflected Tamil word (input)        |
| **lemma**  | Base dictionary form (target)       |
| **id** | id number for Tamil word                |

 

### Load Example

```python
from datasets import load_dataset

ds = load_dataset("Hemanth-thunder/tamil-lemma-data")
print(ds)
```


## **Intended Use**

* Training LLM-based Tamil lemmatizers
* Building linguistic tools such as spell checkers, morphological analyzers
* Improving downstream NLP tasks: MT, ASR, sentiment, QA
* Academic and applied research

---

## **Motivation**

Traditional rule-based Tamil morphological analyzers are limited by ambiguity and coverage.
This dataset provides a **large, clean, aligned** mapping of word → lemma to train modern neural models such as:

* Seq2Seq (LSTM/GRU)
* Transformers
* Causal LLMs with adapters
* Character-level encoder–decoder models

---

## **License**

MIT License (recommended for open NLP datasets).
Update if you intend a different license.

---

## **Project Links**

* **PyPI:** [https://pypi.org/project/tamil-lemmatizer/](https://pypi.org/project/tamil-lemmatizer/)
* **GitHub:** [https://github.com/Hemanthkumar2112/tamil-lemmatizer](https://github.com/Hemanthkumar2112/tamil-lemmatizer)
* **Author Profile:** [https://huggingface.co/Hemanth-thunder](https://huggingface.co/Hemanth-thunder)

---

## **Citation**

Please cite this dataset as:

```
@dataset{hemanth_thunder_2025_tamil_lemmatizer,
  author       = {Hemanth-thunder},
  title        = {Tamil Lemmatizer Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/Hemanth-thunder/tamil-lemma-data}
}