metadata
configs:
- config_name: bpcc-seed-latest
data_files:
- split: asm_Beng
path: bpcc-seed-latest/asm_Beng.tsv
- split: ben_Beng
path: bpcc-seed-latest/ben_Beng.tsv
- split: brx_Deva
path: bpcc-seed-latest/brx_Deva.tsv
- split: doi_Deva
path: bpcc-seed-latest/doi_Deva.tsv
- split: gom_Deva
path: bpcc-seed-latest/gom_Deva.tsv
- split: guj_Gujr
path: bpcc-seed-latest/guj_Gujr.tsv
- split: hin_Deva
path: bpcc-seed-latest/hin_Deva.tsv
- split: kan_Knda
path: bpcc-seed-latest/kan_Knda.tsv
- split: kas_Arab
path: bpcc-seed-latest/kas_Arab.tsv
- split: mai_Deva
path: bpcc-seed-latest/mai_Deva.tsv
- split: mal_Mlym
path: bpcc-seed-latest/mal_Mlym.tsv
- split: mar_Deva
path: bpcc-seed-latest/mar_Deva.tsv
- split: mni_Mtei
path: bpcc-seed-latest/mni_Mtei.tsv
- split: npi_Deva
path: bpcc-seed-latest/npi_Deva.tsv
- split: ory_Orya
path: bpcc-seed-latest/ory_Orya.tsv
- split: pan_Guru
path: bpcc-seed-latest/pan_Guru.tsv
- split: san_Deva
path: bpcc-seed-latest/san_Deva.tsv
- split: snd_Deva
path: bpcc-seed-latest/snd_Deva.tsv
- split: tam_Taml
path: bpcc-seed-latest/tam_Taml.tsv
- split: tel_Telu
path: bpcc-seed-latest/tel_Telu.tsv
- split: urd_Arab
path: bpcc-seed-latest/urd_Arab.tsv
- config_name: bpcc-seed-v1
data_files:
- split: asm_Beng
path: bpcc-seed-v1/asm_Beng.tsv
- split: ben_Beng
path: bpcc-seed-v1/ben_Beng.tsv
- split: brx_Deva
path: bpcc-seed-v1/brx_Deva.tsv
- split: doi_Deva
path: bpcc-seed-v1/doi_Deva.tsv
- split: gom_Deva
path: bpcc-seed-v1/gom_Deva.tsv
- split: guj_Gujr
path: bpcc-seed-v1/guj_Gujr.tsv
- split: hin_Deva
path: bpcc-seed-v1/hin_Deva.tsv
- split: kan_Knda
path: bpcc-seed-v1/kan_Knda.tsv
- split: kas_Arab
path: bpcc-seed-v1/kas_Arab.tsv
- split: mai_Deva
path: bpcc-seed-v1/mai_Deva.tsv
- split: mal_Mlym
path: bpcc-seed-v1/mal_Mlym.tsv
- split: mar_Deva
path: bpcc-seed-v1/mar_Deva.tsv
- split: mni_Mtei
path: bpcc-seed-v1/mni_Mtei.tsv
- split: npi_Deva
path: bpcc-seed-v1/npi_Deva.tsv
- split: ory_Orya
path: bpcc-seed-v1/ory_Orya.tsv
- split: pan_Guru
path: bpcc-seed-v1/pan_Guru.tsv
- split: san_Deva
path: bpcc-seed-v1/san_Deva.tsv
- split: snd_Deva
path: bpcc-seed-v1/snd_Deva.tsv
- split: tam_Taml
path: bpcc-seed-v1/tam_Taml.tsv
- split: tel_Telu
path: bpcc-seed-v1/tel_Telu.tsv
- split: urd_Arab
path: bpcc-seed-v1/urd_Arab.tsv
- config_name: daily
data_files:
- split: asm_Beng
path: daily/asm_Beng.tsv
- split: ben_Beng
path: daily/ben_Beng.tsv
- split: brx_Deva
path: daily/brx_Deva.tsv
- split: doi_Deva
path: daily/doi_Deva.tsv
- split: gom_Deva
path: daily/gom_Deva.tsv
- split: guj_Gujr
path: daily/guj_Gujr.tsv
- split: hin_Deva
path: daily/hin_Deva.tsv
- split: kan_Knda
path: daily/kan_Knda.tsv
- split: kas_Arab
path: daily/kas_Arab.tsv
- split: mai_Deva
path: daily/mai_Deva.tsv
- split: mal_Mlym
path: daily/mal_Mlym.tsv
- split: mar_Deva
path: daily/mar_Deva.tsv
- split: mni_Mtei
path: daily/mni_Mtei.tsv
- split: npi_Deva
path: daily/npi_Deva.tsv
- split: ory_Orya
path: daily/ory_Orya.tsv
- split: pan_Guru
path: daily/pan_Guru.tsv
- split: san_Deva
path: daily/san_Deva.tsv
- split: snd_Deva
path: daily/snd_Deva.tsv
- split: tam_Taml
path: daily/tam_Taml.tsv
- split: tel_Telu
path: daily/tel_Telu.tsv
- split: urd_Arab
path: daily/urd_Arab.tsv
- config_name: daily
data_files:
- split: asm_Beng
path: wiki/asm_Beng.tsv
- split: ben_Beng
path: wiki/ben_Beng.tsv
- split: brx_Deva
path: wiki/brx_Deva.tsv
- split: doi_Deva
path: wiki/doi_Deva.tsv
- split: gom_Deva
path: wiki/gom_Deva.tsv
- split: guj_Gujr
path: wiki/guj_Gujr.tsv
- split: hin_Deva
path: wiki/hin_Deva.tsv
- split: kan_Knda
path: wiki/kan_Knda.tsv
- split: kas_Arab
path: wiki/kas_Arab.tsv
- split: mai_Deva
path: wiki/mai_Deva.tsv
- split: mal_Mlym
path: wiki/mal_Mlym.tsv
- split: mar_Deva
path: wiki/mar_Deva.tsv
- split: mni_Mtei
path: wiki/mni_Mtei.tsv
- split: npi_Deva
path: wiki/npi_Deva.tsv
- split: ory_Orya
path: wiki/ory_Orya.tsv
- split: pan_Guru
path: wiki/pan_Guru.tsv
- split: san_Deva
path: wiki/san_Deva.tsv
- split: snd_Deva
path: wiki/snd_Deva.tsv
- split: tam_Taml
path: wiki/tam_Taml.tsv
- split: tel_Telu
path: wiki/tel_Telu.tsv
- split: urd_Arab
path: wiki/urd_Arab.tsv
- config_name: comparable
data_files:
- split: asm_Beng
path: comparable/asm_Beng.tsv
- split: ben_Beng
path: comparable/ben_Beng.tsv
- split: brx_Deva
path: comparable/brx_Deva.tsv
- split: guj_Gujr
path: comparable/guj_Gujr.tsv
- split: hin_Deva
path: comparable/hin_Deva.tsv
- split: kan_Knda
path: comparable/kan_Knda.tsv
- split: mai_Deva
path: comparable/mai_Deva.tsv
- split: mal_Mlym
path: comparable/mal_Mlym.tsv
- split: mar_Deva
path: comparable/mar_Deva.tsv
- split: mni_Beng
path: comparable/mni_Beng.tsv
- split: npi_Deva
path: comparable/npi_Deva.tsv
- split: ory_Orya
path: comparable/ory_Orya.tsv
- split: pan_Guru
path: comparable/pan_Guru.tsv
- split: san_Deva
path: comparable/san_Deva.tsv
- split: tam_Taml
path: comparable/tam_Taml.tsv
- split: tel_Telu
path: comparable/tel_Telu.tsv
- split: urd_Arab
path: comparable/urd_Arab.tsv
- config_name: ilci
data_files:
- split: asm_Beng
path: ilci/asm_Beng.tsv
- split: ben_Beng
path: ilci/ben_Beng.tsv
- split: brx_Deva
path: ilci/brx_Deva.tsv
- split: gom_Deva
path: ilci/gom_Deva.tsv
- split: guj_Gujr
path: ilci/guj_Gujr.tsv
- split: hin_Deva
path: ilci/hin_Deva.tsv
- split: kan_Knda
path: ilci/kan_Knda.tsv
- split: mal_Mlym
path: ilci/mal_Mlym.tsv
- split: mar_Deva
path: ilci/mar_Deva.tsv
- split: mni_Beng
path: ilci/mni_Beng.tsv
- split: mni_Mtei
path: ilci/mni_Mtei.tsv
- split: npi_Deva
path: ilci/npi_Deva.tsv
- split: pan_Guru
path: ilci/pan_Guru.tsv
- split: tam_Taml
path: ilci/tam_Taml.tsv
- split: tel_Telu
path: ilci/tel_Telu.tsv
- split: urd_Arab
path: ilci/urd_Arab.tsv
- config_name: nllb-seed
data_files:
- split: kas_Arab
path: nllb_seed/kas_Arab.tsv
- split: kas_Deva
path: nllb_seed/kas_Deva.tsv
- split: mni_Beng
path: nllb_seed/mni_Beng.tsv
- config_name: massive
data_files:
- split: ben_beng
path: massive/ben_Beng.tsv
- split: hin_Deva
path: massive/hin_Deva.tsv
- split: kan_Knda
path: massive/kan_Knda.tsv
- split: mal_Mlym
path: massive/mal_Mlym.tsv
- split: tam_Taml
path: massive/tam_Taml.tsv
- split: tel_Telu
path: massive/tel_Telu.tsv
- split: urd_Arab
path: massive/urd_Arab.tsv
- config_name: nllb-filtered
data_files:
- split: asm_Beng
path: nllb_filtered/asm_Beng.tsv
- split: ben_Beng
path: nllb_filtered/ben_Beng.tsv
- split: guj_Gujr
path: nllb_filtered/guj_Gujr.tsv
- split: hin_Deva
path: nllb_filtered/hin_Deva.tsv
- split: kan_Knda
path: nllb_filtered/kan_Knda.tsv
- split: kas_Arab
path: nllb_filtered/kas_Arab.tsv
- split: mai_Deva
path: nllb_filtered/mai_Deva.tsv
- split: mal_Mlym
path: nllb_filtered/mal_Mlym.tsv
- split: mar_Deva
path: nllb_filtered/mar_Deva.tsv
- split: npi_Deva
path: nllb_filtered/npi_Deva.tsv
- split: ory_Orya
path: nllb_filtered/ory_Orya.tsv
- split: pan_Guru
path: nllb_filtered/pan_Guru.tsv
- split: san_Deva
path: nllb_filtered/san_Deva.tsv
- split: tam_Taml
path: nllb_filtered/tam_Taml.tsv
- split: tel_Telu
path: nllb_filtered/tel_Telu.tsv
- split: urd_Arab
path: nllb_filtered/urd_Arab.tsv
- config_name: samanantar-filtered
data_files:
- split: asm_Beng
path: samanantar_v0.3_filtered/asm_Beng.tsv
- split: ben_Beng
path: samanantar_v0.3_filtered/ben_Beng.tsv
- split: guj_Gujr
path: samanantar_v0.3_filtered/guj_Gujr.tsv
- split: hin_Deva
path: samanantar_v0.3_filtered/hin_Deva.tsv
- split: kan_Knda
path: samanantar_v0.3_filtered/kan_Knda.tsv
- split: mal_Mlym
path: samanantar_v0.3_filtered/mal_Mlym.tsv
- split: mar_Deva
path: samanantar_v0.3_filtered/mar_Deva.tsv
- split: ory_Orya
path: samanantar_v0.3_filtered/ory_Orya.tsv
- split: pan_Guru
path: samanantar_v0.3_filtered/pan_Guru.tsv
- split: tam_Taml
path: samanantar_v0.3_filtered/tam_Taml.tsv
- split: tel_Telu
path: samanantar_v0.3_filtered/tel_Telu.tsv
- config_name: samanantar++
data_files:
- split: asm_Beng
path: samanantar_v2/asm_Beng.tsv
- split: ben_Beng
path: samanantar_v2/ben_Beng.tsv
- split: guj_Gujr
path: samanantar_v2/guj_Gujr.tsv
- split: hin_Deva
path: samanantar_v2/hin_Deva.tsv
- split: kan_Knda
path: samanantar_v2/kan_Knda.tsv
- split: mal_Mlym
path: samanantar_v2/mal_Mlym.tsv
- split: mar_Deva
path: samanantar_v2/mar_Deva.tsv
- split: npi_Deva
path: samanantar_v2/npi_Deva.tsv
- split: ory_Orya
path: samanantar_v2/ory_Orya.tsv
- split: pan_Guru
path: samanantar_v2/pan_Guru.tsv
- split: tam_Taml
path: samanantar_v2/tam_Taml.tsv
- split: tel_Telu
path: samanantar_v2/tel_Telu.tsv
- split: urd_Arab
path: samanantar_v2/urd_Arab.tsv
BPCC Dataset
Training
Bharat Parallel Corpus Collection (BPCC) is a comprehensive and publicly available parallel corpus that includes both existing and new data for all 22 scheduled Indic languages. It is comprised of two parts: BPCC-Mined and BPCC-Human, totaling approximately 230 million bitext pairs. BPCC-Mined contains about 228 million pairs, with nearly 126 million pairs newly added as a part of this work. On the other hand, BPCC-Human consists of 2.2 million gold standard English-Indic pairs, with an additional 644K bitext pairs from English Wikipedia sentences (forming the BPCC-H-Wiki subset) and 139K sentences covering everyday use cases (forming the BPCC-H-Daily subset). It is worth highlighting that BPCC provides the first available datasets for 7 languages and significantly increases the available data for all languages covered.
You can find the contribution from different sources in the following table:
| BPCC-Mined | Existing | Samanantar | 19.4M |
| NLLB | 85M | ||
| Newly Added | Samanantar++ | 121.6M | |
| Comparable | 4.3M | ||
| BPCC-Human | Existing | NLLB | 18.5K |
| ILCI | 1.3M | ||
| Massive | 115K | ||
| Newly Added | Wiki | 644K | |
| Daily | 139K |
Additionally, we provide augmented back-translation data generated by our intermediate IndicTrans2 models for training purposes. Please refer our paper for more details on the selection of sample proportions and sources.
| English BT data (English Original) | 401.9M |
| Indic BT data (Indic Original) | 400.9M |
Evaluation
IN22 test set is a newly created comprehensive benchmark for evaluating machine translation performance in multi-domain, n-way parallel contexts across 22 Indic languages. It has been created from three distinct subsets, namely IN22-Wiki, IN22-Web and IN22-Conv. The Wikipedia and Web sources subsets offer diverse content spanning news, entertainment, culture, legal, and India-centric topics. IN22-Wiki and IN22-Web have been combined and considered for evaluation purposes and released as IN22-Gen. Meanwhile, IN22-Conv the conversation domain subset is designed to assess translation quality in typical day-to-day conversational-style applications.
| IN22-Gen (IN22-Wiki + IN22-Web) | 1024 sentences | 🤗 ai4bharat/IN22-Gen |
| IN22-Conv | 1503 sentences | 🤗 ai4bharat/IN22-Conv |
LICENSE
The following table lists the licenses associated with the different artifacts released as a part of this work:
| Artifact | LICENSE |
|---|---|
| Existing Mined Corpora (NLLB & Samanantar) | CC0 |
| Existing Seed Corpora (NLLB-Seed, ILCI, MASSIVE) | CC0 |
| Newly Added Mined Corpora (Samanantar++ & Comparable) | CC0 |
| Newly Added Seed Corpora (BPCC-H-Wiki & BPCC-H-Daily) | CC-BY-4.0 |
| Newly Created IN-22 test set (IN22-Gen & IN22-Conv) | CC-BY-4.0 |
| Back-translation data (BPCC-BT) | CC0 |
| Model checkpoints | MIT |
The mined corpora collection (BPCC-Mined), existing seed corpora (NLLB-Seed, ILCI, MASSIVE), Backtranslation data (BPCC-BT), are released under the following licensing scheme:
- We do not own any of the text from which this data has been extracted.
- We license the actual packaging of this data under the Creative Commons CC0 license (“no rights reserved”).
- To the extent possible under law, AI4Bharat has waived all copyright and related or neighboring rights to BPCC-Mined, existing seed corpora (NLLB-Seed, ILCI, MASSIVE) and BPCC-BT.
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},
issn={2835-8856},
year={2023},
url={https://openreview.net/forum?id=vfT4YuzAYA},
note={}
}