README.md

metadata

annotations_creators: []
language_creators:
  - crowdsourced
  - expert-generated
  - machine-generated
  - found
  - other
languages:
  - asm-IN
  - ben-IN
  - brx-IN
  - guj-IN
  - hin-IN
  - kan-IN
  - kas-IN
  - kok-IN
  - mai-IN
  - mal-IN
  - mar-IN
  - mni-IN
  - nep-IN
  - ori-IN
  - pan-IN
  - san-IN
  - sid-IN
  - tam-IN
  - tel-IN
  - urd-IN
licenses:
  - cc-by-nc-4.0
multilinguality:
  - multilingual
pretty_name: Aksharantar
size_categories: []
source_datasets:
  - original
task_categories:
  - text-generation
task_ids: []

Dataset Card for Aksharantar

Table of Contents

• Table of Contents
• Dataset Description
  • Dataset Summary
  • Supported Tasks and Leaderboards
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Social Impact of Dataset
  • Discussion of Biases
  • Other Known Limitations
• Additional Information
  • Dataset Curators
  • Licensing Information
  • Citation Information
  • Contributions

Dataset Description

• Homepage: https://indicnlp.ai4bharat.org/indic-xlit/
• Repository: https://github.com/AI4Bharat/IndicXlit/
• Paper:
• Leaderboard:
• Point of Contact:

Dataset Summary

Aksharantar is the largest publicly available transliteration dataset for 20 Indic languages. The corpus has 26M Indic language-English transliteration pairs.

Supported Tasks and Leaderboards

[More Information Needed]

Languages

Assamese (asm)	Hindi (hin)	Maithili (mai)	Marathi (mar)	Punjabi (pan)	Tamil (tam)
Bengali (ben)	Kannada (kan)	Malayalam (mal)	Nepali (nep)	Sanskrit (san)	Telugu (tel)
Bodo(brx)	Kashmiri (kas)	Manipuri (mni)	Oriya (ori)	Sindhi (snd)	Urdu (urd)
Gujarati (guj)	Konkani (kok)

Dataset Structure

Data Instances

A random sample from Hindi (hin) Train dataset.

{
'unique_identifier': 'hin1241393', 
'native word': 'स्वाभिमानिक', 
'english word': 'swabhimanik', 
'source': 'IndicCorp', 
'score': -0.1028788579
}

Data Fields

• unique_identifier (string): 3-letter language code followed by a unique number in each set (Train, Test, Val).

• native word (string): A word in Indic language.

• english word (string): Transliteration of native word in English (Romanised word).

• source (string): Source of the data.

• score (num): Character level log probability of indic word given roman word by IndicXlit (model). Pairs with average threshold of the 0.35 are considered.

  For created data sources, depending on the destination/sampling method of a pair in a language, it will be one of:

  • Dakshina Dataset
  • IndicCorp
  • Samanantar
  • Wikidata
  • Existing sources
  • Named Entities Indian (AK-NEI)
  • Named Entities Foreign (AK-NEF)
  • Data from Uniform Sampling method. (Ak-Uni)
  • Data from Most Frequent words sampling method. (Ak-Freq)

Data Splits

Subset	asm-en	ben-en	brx-en	guj-en	hin-en	kan-en	kas-en	kok-en	mai-en	mal-en	mni-en	mar-en	nep-en	ori-en	pan-en	san-en	sid-en	tam-en	tel-en	urd-en
Training	179K	1231K	36K	1143K	1299K	2907K	47K	613K	283K	4101K	10K	1453K	2397K	346K	515K	1813K	60K	3231K	2430K	699K
Validation	4K	11K	3K	12K	6K	7K	4K	4K	4K	8K	3K	8K	3K	3K	9K	3K	8K	9K	8K	12K
Test	5531	5009	4136	7768	5693	6396	7707	5093	5512	6911	4925	6573	4133	4256	4316	5334	-	4682	4567	4463

Dataset Creation

Information in the paper.

Curation Rationale

[More Information Needed]

Source Data

Initial Data Collection and Normalization

Information in the paper.

Who are the source language producers?

[More Information Needed]

Annotations

Information in the paper.

Annotation process

Information in the paper.

Who are the annotators?

Information in the paper.

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

This data is 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 the mined data under the Creative Commons CC0 license (“no rights reserved”), and the Aksharantar benchmark and all manually transliterated data under the Creative Commons CC-BY license (“no rights reserved”).
• To the extent possible under law, AI4Bharat has waived all copyright and related or neighboring rights to Aksharantar.
• This work is published from: India.

Creative Commons Attribution-NonCommercial 4.0 International.

Citation Information

Contributions