Dataset Card for "IndicParaphrase"

Dataset Summary

IndicParaphrase is the paraphrasing dataset released as part of IndicNLG Suite. Each input is paired with up to 5 references. We create this dataset in eleven languages including as, bn, gu, hi, kn, ml, mr, or, pa, ta, te. The total size of the dataset is 5.57M.

Supported Tasks and Leaderboards

Tasks: Paraphrase generation

Leaderboards: Currently there is no Leaderboard for this dataset.

Languages

• Assamese (as)
• Bengali (bn)
• Gujarati (gu)
• Kannada (kn)
• Hindi (hi)
• Malayalam (ml)
• Marathi (mr)
• Oriya (or)
• Punjabi (pa)
• Tamil (ta)
• Telugu (te)

Dataset Structure

Data Instances

One example from the hi dataset is given below in JSON format.

{
  'id': '1',
 'input': 'निजी क्षेत्र में प्रदेश की 75 प्रतिशत नौकरियां हरियाणा के युवाओं के लिए आरक्षित की जाएगी।',
 'references': ['प्रदेश के युवाओं को निजी उद्योगों में 75 प्रतिशत आरक्षण देंगे।',
'युवाओं के लिए हरियाणा की सभी प्राइवेट नौकरियों में 75 प्रतिशत आरक्षण लागू किया जाएगा।',
'निजी क्षेत्र में 75 प्रतिशत आरक्षित लागू कर प्रदेश के युवाओं का रोजगार सुनिश्चत किया जाएगा।',
'प्राईवेट कम्पनियों में हरियाणा के नौजवानों को 75 प्रतिशत नौकरियां में आरक्षित की जाएगी।',
'प्रदेश की प्राइवेट फैक्टरियों में 75 फीसदी रोजगार हरियाणा के युवाओं के लिए आरक्षित किए जाएंगे।'],
 'target': 'प्रदेश के युवाओं को निजी उद्योगों में 75 प्रतिशत आरक्षण देंगे।'
}

Data Fields

• id (string): Unique identifier.
• pivot (string): English sentence used as the pivot
• input (string): Input sentence
• references (list of strings): Paraphrases of input, ordered according to the least n-gram overlap
• target (string): The first reference (most dissimilar paraphrase)

Data Splits

We first select 10K instances each for the validation and test and put remaining in the training dataset. Assamese (as), due to its low-resource nature, could only be split into validation and test sets with 4,420 examples each. Individual dataset with train-dev-test example counts are given below:

Language	ISO 639-1 Code	Train	Dev	Test
Assamese	as	-	4,420	4,420
Bengali	bn	890,445	10,000	10,000
Gujarati	gu	379,202	10,000	10,000
Hindi	hi	929,507	10,000	10,000
Kannada	kn	522,148	10,000	10,000
Malayalam	ml	761,933	10,000	10,000
Marathi	mr	406,003	10,000	10,000
Oriya	or	105,970	10,000	10,000
Punjabi	pa	266,704	10,000	10,000
Tamil	ta	497,798	10,000	10,000
Telugu	te	596,283	10,000	10,000

Dataset Creation

Curation Rationale

[More information needed]

Source Data

Samanantar dataset

Initial Data Collection and Normalization

Detailed in the paper

Who are the source language producers?

Detailed in the paper

Annotations

[More information needed]

Annotation process

[More information needed]

Who are the annotators?

[More information needed]

Personal and Sensitive Information

[More information needed]

Considerations for Using the Data

Social Impact of Dataset

[More information needed]

Discussion of Biases

[More information needed]

Other Known Limitations

[More information needed]

Additional Information

Dataset Curators

[More information needed]

Licensing Information

Contents of this repository are restricted to only non-commercial research purposes under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). Copyright of the dataset contents belongs to the original copyright holders.

Citation Information

If you use any of the datasets, models or code modules, please cite the following paper:

@inproceedings{Kumar2022IndicNLGSM,
  title={IndicNLG Suite: Multilingual Datasets for Diverse NLG Tasks in Indic Languages},
  author={Aman Kumar and Himani Shrotriya and Prachi Sahu and Raj Dabre and Ratish Puduppully and Anoop Kunchukuttan and Amogh Mishra and Mitesh M. Khapra and Pratyush Kumar},
  year={2022},
  url = "https://arxiv.org/abs/2203.05437"
}  

Contributions