Datasets:

ArXiv:
License:
IN22-Conv / README.md
pranjalchitale's picture
Update TMLR citation.
189b3d9
metadata
language:
  - as
  - bn
  - brx
  - doi
  - en
  - gom
  - gu
  - hi
  - kn
  - ks
  - mai
  - ml
  - mr
  - mni
  - ne
  - or
  - pa
  - sa
  - sat
  - sd
  - ta
  - te
  - ur
language_details: >-
  asm_Beng, ben_Beng, brx_Deva, doi_Deva, eng_Latn, gom_Deva, guj_Gujr,
  hin_Deva, kan_Knda, kas_Arab, mai_Deva, mal_Mlym, mar_Deva, mni_Mtei, 
  npi_Deva, ory_Orya, pan_Guru, san_Deva, sat_Olck, snd_Deva, tam_Taml, 
  tel_Telu, urd_Arab
license: cc-by-4.0
language_creators:
  - expert-generated
multilinguality:
  - multilingual
  - translation
pretty_name: in22-conv
size_categories:
  - 1K<n<10K
task_categories:
  - translation

IN22-Conv

IN-22 is a newly created comprehensive benchmark for evaluating machine translation performance in multi-domain, n-way parallel contexts across 22 Indic languages. IN22-Conv is the conversation domain subset of IN22. It is designed to assess translation quality in typical day-to-day conversational-style applications. The evaluation subset consists of 1503 sentences translated across 22 Indic languages enabling evaluation of MT systems across 506 directions.

Currently, we use it for sentence-level evaluation of MT systems but it can be repurposed for document translation evaluation as well.

Here is the domain distribution of our IN22-Conv evaluation subset.

domain count
hobbies 120
daily_dialogue 117
government 116
geography 114
sports 100
entertainment 97
history 97
legal 96
arts 95
college_life 94
tourism 91
school_life 87
insurance 82
culture 73
healthcare 67
banking 57
total 1503

Please refer to the Appendix E: Dataset Card of the preprint on detailed description of dataset curation, annotation and quality control process.

Dataset Structure

Dataset Fields

  • id: Row number for the data entry, starting at 1.
  • doc_id: Unique identifier of the conversation.
  • sent_id: Unique identifier of the sentence order in each conversation.
  • topic: The specific topic of the conversation within the domain.
  • domain: The domain of the conversation.
  • prompt: The prompt provided to annotators to simulate the conversation.
  • scenario: The scenario or context in which the conversation takes place.
  • speaker: The speaker identifier in the conversation.
  • turn: The turn within the conversation.

Data Instances

A sample from the gen split for the English language (eng_Latn config) is provided below. All configurations have the same structure, and all sentences are aligned across configurations and splits.

{
   "id": 1,
   "doc_id": 0,
   "sent_id": 1,
   "topic": "Festivities",
   "domain": "culture",
   "prompt": "14th April a holiday",
   "scenario": "Historical importance",
   "speaker": 1,
   "turn": 1,
   "sentence": "Mom, let's go for a movie tomorrow."
}

When using a hyphenated pairing or using the all function, data will be presented as follows:

{
   "id": 1,
   "doc_id": 0,
   "sent_id": 1,
   "topic": "Festivities",
   "domain": "culture",
   "prompt": "14th April a holiday",
   "scenario": "Historical importance",
   "speaker": 1,
   "turn": 1,
   "sentence_eng_Latn": "Mom, let's go for a movie tomorrow.",
   "sentence_hin_Deva": "माँ, चलो कल एक फिल्म देखने चलते हैं।"
}

Sample Conversation

Speaker Turn
Speaker 1 Mom, let's go for a movie tomorrow. I don't have to go to school. It is a holiday.
Speaker 2 Oh, tomorrow is the 14th of April right? Your dad will also have the day off from work. We can make a movie plan!
Speaker 1 That's a good news! Why is it a holiday though? Are all schools, colleges and offices closed tomorrow?
Speaker 2 It is Ambedkar Jayanti tomorrow! This day is celebrated annually to mark the birth of Dr. B. R Ambedkar. Have you heard of him?
Speaker 1 I think I have seen him in my History and Civics book. Is he related to our Constitution?
Speaker 2 Absolutely! He is known as the father of the Indian Constitution. He was a civil rights activist who played a major role in formulating the Constitution. He played a crucial part in shaping the vibrant democratic structure that India prides itself upon.
...

Usage Instructions

from datasets import load_dataset

# download and load all the pairs
dataset = load_dataset("ai4bharat/IN22-Conv", "all")

# download and load specific pairs
dataset = load_dataset("ai4bharat/IN22-Conv", "eng_Latn-hin_Deva")

Languages Covered

Assamese (asm_Beng) Kashmiri (Arabic) (kas_Arab) Punjabi (pan_Guru)
Bengali (ben_Beng) Kashmiri (Devanagari) (kas_Deva) Sanskrit (san_Deva)
Bodo (brx_Deva) Maithili (mai_Deva) Santali (sat_Olck)
Dogri (doi_Deva) Malayalam (mal_Mlym) Sindhi (Arabic) (snd_Arab)
English (eng_Latn) Marathi (mar_Deva) Sindhi (Devanagari) (snd_Deva)
Konkani (gom_Deva) Manipuri (Bengali) (mni_Beng) Tamil (tam_Taml)
Gujarati (guj_Gujr) Manipuri (Meitei) (mni_Mtei) Telugu (tel_Telu)
Hindi (hin_Deva) Nepali (npi_Deva) Urdu (urd_Arab)
Kannada (kan_Knda) Odia (ory_Orya)

Citation

If you consider using our work then please cite using:

@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={}
}