morphTok

This repository contains the morphological segmentation data released as part of the paper:

Morphtok: Morphologically Grounded Tokenization for Indian Languages

Description

The dataset provides word-level morphological segmentations for Indian languages. Each entry consists of a surface word form and its morpheme-level decomposition, represented using the + delimiter.

Currently supported languages:

• Hindi (hindi.csv)
• Marathi (marathi.csv)
• Tamil (tamil.csv)

Additional languages may be added in future releases.

print(ds)

Data Format

Each CSV file contains the following columns:

• word: surface form
• segmentation: morpheme-level segmentation
• language: language code

Usage

from datasets import load_dataset

ds = load_dataset(
"karthika95/morphTok",
data_files={
"hindi": ["hindi.csv"],
"marathi": ["marathi.csv"],
"tamil": ["tamil.csv"],
}
)
print(ds)

Acknowledgements

We thank BharatGen for providing resources and access to domain experts who supported the curation of this dataset.

BibTeX

@article{brahma2025morphtok, title={Morphtok: Morphologically grounded tokenization for indian languages}, author={Brahma, Maharaj and Karthika, NJ and Singh, Atul and Adiga, Devaraj and Bhate, Smruti and Ramakrishnan, Ganesh and Saluja, Rohit and Desarkar, Maunendra Sankar}, journal={arXiv preprint arXiv:2504.10335}, year={2025} }