metadata

task_categories:
  - translation
tags:
  - tamil
  - intralingual
  - poetry
  - benchmark
size_categories:
  - n<1K
license: cc-by-nc-2.0

Tamil Intralingual Translation Benchmark Dataset (Draft)

Dataset Description

This dataset serves as a benchmark for intralingual translation tasks in Tamil, specifically targeting the conversion of Classical Tamil poetry into Modern Tamil prose. It provides carefully curated samples from literary works and designed to test the ability of machine learning models to retain the semantic essence and stylistic nuances of classical texts while making them accessible to contemporary readers.

Key Features

Task: Translation of Classical Tamil poetry into Modern Tamil.
Input: Classical Tamil poetry (original and preprocessed forms).
Target: Modern Tamil prose translations (original and preprocessed forms).
Metadata: Contextual task descriptions, word counts for input and target, and the source literary work (Nool).

Dataset Details

Column	Description
`Nool`	Source literary work (e.g., "Ainthinai Aimpathu").
`InputOrig`	Classical Tamil poetic excerpt in its original form.
`GroundTruthOrig`	Modern Tamil prose translation in its original form.
`Input`	Preprocessed Classical Tamil poetic text for model input.
`GroundTruth`	Preprocessed Modern Tamil prose text for evaluation.
`Instructions`	Contextual instructions for the task (e.g., "Tamil Poetry intralingual").
`No Of Words-Input`	Number of words in the `Input` text.
`No Of Words-GroundTruth`	Number of words in the `GroundTruth` text.

Sample Content

Example entry:

InputOrig: உள்ளார்கொல் காதலர் ஒண்டொடி நம்திறம்...
GroundTruthOrig: ஒளிமிக்க அழகிய வளையல் அணிந்த தலைவியே! இந்த மேகக் கூட்டமானது...
Instructions: Tamil Poetry intralingual
No Of Words-Input: 15
No Of Words-GroundTruth: 50

Dataset Use Cases

Benchmarking the performance of intralingual translation systems.
Researching stylistic and semantic transformations in Tamil NLP.

Supported Tasks

Translation: Classical Tamil → Modern Tamil.

Languages

Source Language: Tamil (Classical Tamil).
Target Language: Tamil (Modern Tamil).

Dataset Size

Total samples: < 1K.

Evaluation

The dataset is designed for evaluation using:

BLEU: Measures translation accuracy.
ROUGE: Evaluates textual overlap between input and target.
Human Evaluation: Assesses semantic retention and readability.

Limitations

The dataset is in a draft state and may not cover all Classical Tamil literary works.
Limited to less than 1,000 samples, making it more suitable for benchmarking than large-scale training.

Citation

If you use this dataset, please cite it as:

@dataset{akdiwahar_2025,
      title={Tamil Intralingual Translation Benchmark Dataset (Draft)}, 
      author={Diwahar A K},
      year={2023},
      publisher={Hugging Face},
      note={Available at https://huggingface.co/datasets/username/tamil_intralingual_translation} 
}

@dataset{your_name_2025, title={Tamil Intralingual Translation Benchmark Dataset (Draft)}, author={Diwahar A K}, year={2025}, publisher={Hugging Face}, note={Available at https://huggingface.co/datasets/username/tamil_intralingual_translation} }

License

This dataset is released under [cc-by-nc-2.0]. Ensure compliance with the terms of use.

Acknowledgments

Special thanks to the contributors and scholars who assisted in the curation and validation of this dataset.