itm-data
A translation dataset (English β 11 Indic languages: Hindi, Tamil, Bengali, Telugu, Malayalam, Gujarati, Punjabi, Kannada, Marathi, Assamese, Odia). Training is work in progress.
The files are derived from the sources listed below. Each source keeps its own licence and terms; the table is the licence statement for this dataset.
Sources and licences
| Source | Content used | Licence (as stated by the source) | Link |
|---|---|---|---|
| BPCC (AI4Bharat) β mined: NLLB-filtered, Samanantar-filtered, Samanantar++, Comparable | parallel sentences | CC0 packaging; AI4Bharat does not own the underlying text, original-source terms apply | https://huggingface.co/datasets/ai4bharat/BPCC |
| BPCC (AI4Bharat) β existing seed: ILCI, MASSIVE | parallel sentences | CC0 packaging; original-source terms apply (MASSIVE: CC-BY-4.0) | https://huggingface.co/datasets/ai4bharat/BPCC |
| BPCC (AI4Bharat) β BPCC-H-Wiki, BPCC-H-Daily, BPCC-Seed | parallel sentences | CC-BY-4.0 | https://huggingface.co/datasets/ai4bharat/BPCC |
| Pralekha (AI4Bharat) β train split | parallel documents | CC-BY-4.0 | https://huggingface.co/datasets/ai4bharat/Pralekha |
| Shiksha (SPRING Lab, IIT Madras) β train split, English-source pairs | parallel sentences | CC-BY-4.0 | https://huggingface.co/datasets/SPRINGLab/shiksha |
| OPUS wikimedia (Wikimedia content translations) | parallel sentences | CC-BY-SA-4.0 | https://opus.nlpl.eu/wikimedia |
| OPUS translatewiki (translatewiki.net) | parallel UI strings | CC-BY-3.0 | https://opus.nlpl.eu/translatewiki |
| OPUS GNOME (GNOME localisation) | parallel UI strings | same licence as the original GNOME sources (per OPUS) | https://opus.nlpl.eu/GNOME |
| OPUS Tatoeba (tatoeba.org) | parallel sentences | CC-BY-2.0-FR | https://opus.nlpl.eu/Tatoeba |
OPUS data is used via the OPUS collection (J. Tiedemann, 2012, Parallel Data, Tools and Interfaces in OPUS, LREC 2012).
Upstream corpora inside BPCC-Mined include, among others, Samanantar (whose own components include IITB EnβHi, CC-BY-NC-4.0; PMIndia, CC-BY-4.0; WikiMatrix, CC-BY-SA), NLLB-derived bitext (ODC-BY plus the terms of the original sources) and web/Wikipedia text. Users are responsible for complying with the terms of each original source.
Citation
@article{gala2023indictrans,
title={IndicTrans2: Towards High-Quality and Accessible Machine Translation Models for all 22 Scheduled Indian Languages},
author={Jay Gala and Pranjal A Chitale and A K Raghavan and Varun Gumma and Sumanth Doddapaneni and Aswanth Kumar M and Janki Atul Nawale and Anupama Sujatha and Ratish Puduppully and Vivek Raghavan and Pratyush Kumar and Mitesh M Khapra and Raj Dabre and Anoop Kunchukuttan},
journal={Transactions on Machine Learning Research},
year={2023},
url={https://openreview.net/forum?id=vfT4YuzAYA}
}
- Downloads last month
- -