task_categories:
- sentence-similarity
language:
- si
- ta
- en
Dataset summary
This is a gold-standard benchmark dataset for document alignment, between Sinhala-English-Tamil languages. Data had been crawled from the following news websites.
| News Source | url |
|---|---|
| Army | https://www.army.lk/ |
| Hiru | http://www.hirunews.lk |
| ITN | https://www.newsfirst.lk |
| Newsfirst | https://www.itnnews.lk |
The aligned documents have been manually annotated.
Dataset
The folder structure for each news source is as follows.
army
|--Sinhala
|--English
|--Tamil
|--armynews_english_sinhala.txt
|--armynews_english_tamil.txt
|--armynews_sinhala_tamil.txt
Sinhala/English/Tamil - contain the crawled data for the respective news source army_news_english_sinhala.txt - contains the annotated aligned documents between English and Sinhala languages. armynews_english_tamil.txt - contains the annotated aligned documents between English and Tamil languages. armynews_sinhala_tamil.txt - contains the annotated aligned documents between Sinhala and Tamil languages.
Citation Information
@article{fernando2022exploiting,
title={Exploiting bilingual lexicons to improve multilingual embedding-based document and sentence alignment for low-resource languages},
author={Fernando, Aloka and Ranathunga, Surangika and Sachintha, Dilan and Piyarathna, Lakmali and Rajitha, Charith},
journal={Knowledge and Information Systems},
pages={1--42},
year={2022},
publisher={Springer}
}