prosa-text/nusa-translation

Dataset Description

NusaTranslation is a dataset for machine translation tasks, featuring curated pairs of translations, totaling around 300,000 instances, from Indonesian to three distinct languages:

• Balinese (ban)
• Buginese (bug)
• Minangkabau (min)

How to use the data

To access the data you can use the Hugging Face Python datasets library. To load NusaTranslation, simply call datasets.load_dataset() as shown on the snippet below:

import datasets

min_dataset = datasets.load_dataset("prosa-text/nusa-translation", name="min")
ban_dataset = datasets.load_dataset("prosa-text/nusa-translation", name="ban")
bug_dataset = datasets.load_dataset("prosa-text/nusa-translation", name="bug")

Data Fields

Every instance contains the following fields:

• text: original text in Indonesian.
• translated: the translated text.
• language: the language code (min, ban, bug)

Example:

{
  "text": "Lagu tersebut di bawah naungan perusahaan musik Indonesia, Trinity Optima Production. Video musik lagu tersebut dirilis pada 19 April bersamaan dengan rilisnya lagu tersebut dalam versi bahasa Inggris. Video musik diunggah di kanal YouTube pribadinya dan telah ditonton lebih dari dua juta kali. Lalu pada 3 Juni 2019, ia merilis lagu \"I Love You 3000\" yang ia tulis sendiri.",
  "translated": "Lagu tasabuik di bawah naungan parusahaan musik Indonesia, Trinity Optima Production. Video musik lagu tasabuik dirilis pado 19 April basamaan jo rilisnyo lagu tasabuik dalam versi bahasa Inggris. Video musik diunggah di kanal YouTube pribadinyo jo lah ditonton labiah dari duo juta kali. Lalu pado 3 Juni 2019, inyo marilis lagu \"I Love You 3000\" nan inyo tulis surang.",
  "language": "min"
}

Data Instances & Splits

The data is splitted into three data splits, i.e., training, validation, and test.

Language	Data Split	Num Sample
Balinese	Training	126972
	Validation	3000
	Test	10000
Buginese	Training	128472
	Validation	3000
	Test	10000
Minangkabau	Training	126972
	Validation	3000
	Test	10000

Data Analysis

lang	data	total_characters	average_characters_per_row	variance_characters	total_words	average_words_per_row	variance_words	total_tokens	average_tokens_per_row	variance_tokens
Balinese	all	49891009	356.44	3168.84	7130696	50.94	48.71	13606931	97.21	287.61
Buginese	all	50037752	353.69	3259.37	7179251	50.75	52.63	14798634	104.60	354.50
Minangkabau	all	48748663	348.27	2956.63	7086576	50.63	44.72	13212897	94.40	253.39

Balinese: The dataset has the highest average number of characters and words per row Highest average number of characters and words per row: This indicates the tendency to write long sentences, possibly reflecting a detailed or elaborate writing style.

Buginese: The dataset has the highest average number of tokens per row and the highest total number of characters and tokens Highest average tokens per row: This indicates that Buginese sentences typically have more words and potentially smaller word breaks, leading to a higher token count. This could relate to a more concise or direct writing style compared to other 2 languages. Highest total characters and tokens: This suggests that Buginese sentences in the dataset tend to be longer than those in other languages, possibly due to more complex sentence structures.

Minangkabau: The dataset has the lowest total number of characters and tokens and the lowest average number of words per row. Lowest total characters and tokens: Having the least overall content suggests that Minangkabau sentences might be shorter but use denser word choices or have many common character combinations. Lowest average words per row: This implies that Minangkabau sentences might use longer words and less unique character combinations compared to other languages.

Annotation process

Human translation is carried out by determining the boundaries of the rules in the translation process. We instructed the annotators to retain the meaning of the text and to keep entities, such as persons, organizations, locations, and time with no target language translation the same. Specifically, we instructed them to: (1) maintain the sentence's sentiment polarity; (2) preserve entities; and (3) maintain the complete information content of the original text.

This translation method achieved a total of 140,000 sentences for each language. The details after being converted to words are 7,128,771 words for Balinese, 7,116,103 words for Buginese, and 7,085,176 words for Minangkabau.

Who are the annotators?

We conduct corpus construction through human annotation by expert annotators. All expert annotators are native speakers of each target language who have gone through a selection process. In the process of developing data in a local language, a competent and experienced team in the required local language is certainly needed. Annotators play a crucial role in compiling high quality local language data. Therefore, strict qualifications are required for the candidate annotators who will be recruited.

The qualifications include educational background and experience related to language. Annotator candidates must have good knowledge of the language and the sentence structure of the local language they are proficient in. Good writing skills are also a plus, considering that the annotator's tasks include creating dialogues and paragraphs. Additionally, annotators are expected to have resilience in working with a large amount of data, so commitment from annotators is also required.

The recruitment process has successfully gathered a total of 255 annotators candidates for 3 different languages. There are 110 candidates, or approximately 43%, who were eligible to participate in the annotation process. There are 46 candidates for the Balinese language, 40 candidates for the Buginese language, and 24 candidates for the Minangkabau language. Out of that number, only 86 people persevered until the annotation process was completed, while the rest withdrew from the project midway through.

The following is the distribution of dialect diversity from the annotators.

Language	Dialects
Balinese	Badung, Bali, Bali Aga, Bangli, Buleleng, Dataran,
	Denpasar, Gianyar, Karangasem, Klungkung, Singaraja,
	Tabanan.
Buginese	Barru, Bone, Bugis, Bulukumba, Magai Io, Makassar,
	Maros, Pangkep, Pinrang, Sengkang, Sidenreng Rappang,
	Sinjai, Soppeng, Wajo.
Minangkabau	Agam, Bukittinggi, Minangkabau, Padang, Padang Panjang,
	Pariaman, Pasaman, Payakumbuh, Sijunjung, Tanah Datar.

Personal and Sensitive Information

In the process of defining topics, there are several topics that have the potential to cause opinion bias among annotators. These topics are usually related to emotions, for instance liking or disliking something. It should be understood that this is the annotator's subjectivity and has nothing to do with the organization's values.

Additional Information

Licensing Information

The dataset is released under the terms of CC-BY-SA 4.0. By using this dataset, you are also bound to the respective Terms of Use and License of the dataset. For commercial use in small businesses and startups, please contact us (business@prosa.ai) for permission to use the datasets by informing company profile and propose of usage.

Citation Information

@article{purwarianti2023nusatranslation,
  title={NusaTranslation: Dialogue Summarization and Generation for Underrepresented and Extremely Low-Resource Languages},
  author={Purwarianti, Ayu and Adhista, Dea and Baptiso, Agung and Mahfuzh, Miftahul and Yusrina Sabila and Cahyawijaya, Samuel and Aji, Alham Fikri},
  journal={arXiv preprint arXiv:(coming soon)},
  url={https://huggingface.co/datasets/prosa-text/nusa-translation},
  year={2023}
}

Acknowledgement

This research work is funded and supported by The Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH and FAIR Forward - Artificial Intelligence for all. We thank Direktorat Jenderal Pendidikan Tinggi, Riset, dan Teknologi Kementerian Pendidikan, Kebudayaan, Riset, dan Teknologi (Ditjen DIKTI) for providing the computing resources for this project.

Contact Us

If you have any question please contact our support team at business@prosa.ai.