TyDi-AS2

Dataset Summary

TyDi-AS2 and Xtr-TyDi-AS2 are multilingual Answer Sentence Selection (AS2) datasets comprising 8 diverse languages, proposed in our paper accepted at ACL 2023 (Findings): Cross-Lingual Knowledge Distillation for Answer Sentence Selection in Low-Resource Languages. Both the datasets were created from TyDi-QA, a multilingual question-answering dataset. TyDi-AS2 was created by converting the QA instances in TyDi-QA to AS2 instances (see Dataset Creation for details). Xtr-TyDi-AS2 was created by translating the non-English TyDi-AS2 instances to English and vise versa. For translations, we used Amazon Translate.

Languages

TyDi-AS2 (original)

• bn: Bengali
• en: English
• fi: Finnish
• id: Indonesian
• ja: Japanese
• ko: Korean
• ru: Russian
• sw: Swahili

File location: jsonl/original/

For non-English sets, we also have English-translated samples used for the cross-lingual knowledge distillation (CLKD) experiments in our paper.

File location: jsonl/x-to-en/

Xtr-TyDi-AS2 (translationese)

Xtr-TyDi-AS2 (X-translated TyDi-AS2) dataset consists of non-English AS2 instances translated from the English set of TyDi-AS2.

• bn: Bengali
• fi: Finnish
• id: Indonesian
• ja: Japanese
• ko: Korean
• ru: Russian
• sw: Swahili

File location: jsonl/en-to-x/

Dataset Structure

Data Instances

This is an example instance from the English training split of TyDi-AS2 dataset.

{
  "Question": "When was the Argentine Basketball Federation formed?",
  "Title": "History of the Argentina national basketball team",
  "Sentence": "The Argentina national basketball team represents Argentina in basketball international competitions, and is controlled by the Argentine Basketball Federation.",
  "Label": 0
}

For English-translated TyDi-AS2 dataset and Xtr-TyDi-AS2 dataset, the translated instances in JSONL files are listed in the same order of the original (native) instances in the original TyDi-AS2 dataset.

For example, the 2nd instance in jsonl/x-to-en/en_from_bn-train.jsonl (English-translated from Bengali) corresponds to the 2nd instance in jsonl/original/bn-train.jsonl (Bengali).

Similarly, the 2nd instance in jsonl/en-to-x/bn_from_en-train.jsonl (Bengali-translated from English) corresponds to the 2nd instance in jsonl/original/en-train.jsonl (English).

Data Fields

Each instance (a QA pair) consists of the following fields:

• Question: Question to be answered (str)
• Title: Document title (str)
• Sentence: Answer sentence in the document (str)
• Label: Label that indicates the answer sentence correctly answers the question (int, 1: correct, 0: incorrect)

Data Splits

		#Questions				#Sentences
	train	dev	test		train	dev	test
Bengali (bn)	7,978	2,056	316		1,376,432	351,186	37,465
English (en)	6,730	1,686	918		1,643,702	420,899	249,513
Finnish (fi)	10,859	2,731	1,870		1,567,695	408,205	298,093
Indonesian (id)	9,310	2,339	1,355		960,270	236,076	97,057
Japanese (ja)	11,848	2,981	1,504		3,183,037	822,654	444,106
Korean (ko)	7,354	1,943	1,389		1,558,191	392,361	199,043
Russian (ru)	9,187	2,294	1,395		3,190,650	820,668	367,595
Swahili (sw)	8,350	2,850	1,896		1,048,303	269,894	74,775

See our paper for more details about the statistics of the datasets.

Dataset Creation

Source Data

The source of TyDi-AS2 dataset is TyDi QA, which is a question answering dataset.

Annotations

Annotation process

TyDi QA is a QA dataset spanning questions from 11 typologically diverse languages. Each instance comprises a human-generated question, a single Wikipedia document as context, and one or more spans from the document containing the answer. To convert each instance into AS2 instances, we split the context document into sentences and heuristically identify the correct asnwer sentences using the annotated answer spans. To split documents, we use multiple different sentence tokenizers for the diverse languages and omit languages for which we could not find a suitable sentence tokenizer:

1. bltk for Bengali
2. blingfire for Swahili, Indonesian, and Korean
3. pysdb for English and Russian
4. nltk for Finnish
5. Konoha for Japanese

Who are the annotators?

Shivanshu Gupta converted TyDi QA to TyDi-AS2. Yoshitomo Matsubara translated non-English samples to English and vice versa for Xtr-TyDi-AS2 dataset Since sentence tokenization and identifying answer sentences can introduce errors, we conducted a manual validation of the AS2 datasets. For each language, we randomly selected 50 instances and verified the accuracy of the answer sentences through manual inspection. Our findings revealed that the answer sentences were accurate in 98% of the cases.

Additional Information

Dataset Curators

Shivanshu Gupta (@shivanshu)

Licensing Information

CDLA-Permissive-2.0

Citation Information

@inproceedings{gupta2023cross-lingual,
  title={{Cross-Lingual Knowledge Distillation for Answer Sentence Selection in Low-Resource Languages}},
  author={Gupta, Shivanshu and Matsubara, Yoshitomo and Chadha, Ankit and Moschitti, Alessandro},
  booktitle={Findings of the Association for Computational Linguistics: ACL 2023},
  pages={14078--14092},
  year={2023}
}

Contributions

• Shivanshu Gupta
• Yoshitomo Matsubara
• Ankit Chadha
• Alessandro Moschitti