English-Vietnamese Parallel Corpora

This repository contains English-Vietnamese parallel corpora collected from publicly available sources/projects for the VLSP 2020 English-Vietnamese News Translation Shared Task. It includes:

• basic: parallel corpus from basic sentences, crawled from english-vietnamese websites and from Tatoeba corpus
• EVBcorpus: parallel corpus of law and news texts, compiled by Dr. Quoc Hung Ngo. Source: EVBcorpus.
• wiki-alt: multilingual wiki text from Asian Language Treebank project.
• ted-like: one-speaker talks similar to TED talks, extracted from many multilingual sites, including TED, Khan Academy, Amara...
• openSub: parallel corpus of movie subtitles, extracted from the (https://www.opensubtitles.org/) website by OPUS project
• indomain-news: parallel corpus of news domains, extracted and sentence-aligned from various newspapers. Besides, there are development and test sets in the same domain (news) which are used for the VLSP 2020 evaluation campaign.
• iwslt15: parallel corpus extracted from TED talks, served for the IWSLT15 evaluation campaign. It also includes development, test and official test sets used for the IWSLT15 evaluation campaign. NOTE: ted-like corpus contains all the data in this iwslt15 corpus, including development and test sets. So, the systems trained on data containing ted-like might overfit and perform unfairly on IWSLT15 English-Vietnamese benchmarks (e.g. paperwithcode)
• mono-vi: a monolingual corpus extracted and filtered from Vietnamese popular newspapers. It should be used for Back-Translation approaches.

Statistics of Training Data

Here are the statistics of those data:

Dataset	Number of sentences	Notes
basic	8,818	basic, conversational sentences
EVBcorpus	45,308	law and news texts
wiki-alt	20,000	wiki texts
ted-like	546,190	one-speaker talks, educational domain
openSub	3,505,276	movie subtitles, noisy
indomain-news	20,000	news, in-domain data for VLSP 2020
iwslt15	133,137	ted talks, fully overlapped to ted-like
mono-vi	2,000,000	monolingual corpus of news domain
Total	4,145,592	excluding iwslt15 and mono-vi
Total Good	640,316	excluding iwslt15, mono-vi and openSub

Statistics of Development/Test Data

At the end of the VLSP 2020 evaluation campaign, the task organizers published the official test set VLSP20-official. Here are the statistics of those validations and test sets:

Evaluation Campaign	Dataset	Number of sentences	Notes
	dev	7,462	TED talks, development/validation set
IWSLT15 (ilwslt15)	test	6,322	TED talks, public test set
	IWSLT15.official_testdev	500	TED talks, private/official test set
VLSP2020 (indomain-news)	dev	1,007	news domain, development/validation set
	tst	1,220	news domain, public test set
VLSP2020 (VLSP20-official)	offi_test	789	news domain, private/official test set

Among those corpora, the openSub corpus is the largest but the most noisy one. We trained a strong NMT system on Total data (including openSub) and the same sytem on Total Good (without openSub) and the latter outperforms the former on VLSP 2020 test sets.

Cite

Please cite our paper if you use the data in your research:

@inproceedings{vlsp2020-mt,
title     = {{Goals, Challenges and Findings of the VLSP 2020 English-Vietnamese News Translation Shared Task}},
author    = {Thanh-Le Ha and Van-Khanh Tran and Kim-Anh Nguyen},
booktitle = {{Proceedings of the 7th International Workshop on Vietnamese Language and Speech Processing - VLSP 2020}},
year      = {2020}
}

Acknowledgment

• TED and WIT3: https://wit3.fbk.eu/. M. Cettolo, C. Girardi, and M. Federico. 2012. "WIT3: Web Inventory of Transcribed and Translated Talks". In Proc. of EAMT, pp. 261-268, Trento, Italy.
• WIKI-ALT: Asian Language Treebank project. Hammam Riza, Michael Purwoadi, Gunarso, Teduh Uliniansyah, Aw Ai Ti, Sharifah Mahani Aljunied, Luong Chi Mai, Vu Tat Thang, Nguyen Phuong Thai, Vichet Chea, Rapid Sun, Sethserey Sam, Sopheap Seng, Khin Mar Soe, Khin Thandar Nwet, Masao Utiyama, Chenchen Ding. 2016. "Introduction of the Asian Language Treebank". Oriental COCOSDA.
• Tatoeba : J. Tiedemann. 2012. "Parallel Data, Tools and Interfaces in OPUS". In Proceedings of the 8th International Conference on Language Resources and Evaluation (LREC 2012).
• QED: A. Abdelali, F. Guzman, H. Sajjad and S. Vogel. 2014 "The AMARA Corpus: Building parallel language resources for the educational domain". The Proceedings of the 9th International Conference on Language Resources and Evaluation (LREC'14). Reykjavik, Iceland, 2014. Pp. 1856-1862. Isbn. 978-2-9517408-8-4.
• OpenSubtitles: P. Lison and J. Tiedemann. 2016. "OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles". In Proceedings of the 10th International Conference on Language Resources and Evaluation (LREC 2016)
• VINBDI: VinGroup Big Data Institute for the data collection and official test translation fundings.