Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

facebook/mlqa(複製)

データの内容

英語、アラビア語、ドイツ語、スペイン語、ヒンディー語、ベトナム語、簡体字中国語の 7 言語にわたる、抽出型の質問応答のデータ。1 行が 1 問で、context に段落、question に質問、answers に段落中の答えの文字列と開始位置が入る。事例は平均して 4 言語の間で並列になっており、言語横断の評価に使える。

データの概要

項目 値
行数 747,713(61 構成・116 分割の合計)
列数 4
容量 44.49 KB(ファイル 3 本)
ファイルの形式 py 1
言語 en, de, es, ar, zh, vi, hi
公開 public

構成と分割

構成(config) 分割(split)ごとの行数 列数
mlqa-translate-train.ar train 78,058、validation 9,512 4
mlqa-translate-train.de train 80,069、validation 9,927 4
mlqa-translate-train.vi train 84,816、validation 10,356 4
mlqa-translate-train.zh train 76,285、validation 9,568 4
mlqa-translate-train.es train 81,810、validation 10,123 4
mlqa-translate-train.hi train 82,451、validation 10,253 4
mlqa-translate-test.ar test 5,335 4
mlqa-translate-test.de test 4,517 4
mlqa-translate-test.vi test 5,495 4
mlqa-translate-test.zh test 5,137 4
mlqa-translate-test.es test 5,253 4
mlqa-translate-test.hi test 4,918 4
mlqa.ar.ar test 5,335、validation 517 4
mlqa.ar.de test 1,649、validation 207 4
mlqa.ar.vi test 2,047、validation 163 4
mlqa.ar.zh test 1,912、validation 188 4
mlqa.ar.en test 5,335、validation 517 4
mlqa.ar.es test 1,978、validation 161 4
mlqa.ar.hi test 1,831、validation 186 4
mlqa.de.ar test 1,649、validation 207 4

(ほか 41 構成。全体は HF のデータセットビューアで見られる)

列(構成 mlqa-translate-train.ar。どの構成も同じ列)

列 型
context string
question string
answers list<?>
id string

行数と列は元のカードの YAML(dataset_info)の記載。データセットビューアでは読めなかった。全データセットの一覧は HYUKI5/dataset-catalog にある。

複製について

facebook/mlqa をリビジョン 397ed406c1a7902140303e7faf60fff35b58d285 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。

出典

論文

アノテーションの方法

収集の根拠

More Information Needed

初期のデータ収集と正規化

More Information Needed

元の言語の作成者は誰か?

More Information Needed

アノテーションの過程

More Information Needed

データセットの作成者

More Information Needed

微調整に使うときの注意

  • 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
  • Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
  • 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。

調査メモの記述

メモ 節 ライセンス(記載) 確認 注記
multilingual_cjk.md 1. 大規模多言語データセット(XNLI, PAWS-X, MASSIVE, SIB-200, Belebele ほか) (L53) cc-by-sa-3.0 確認済 確認印なし(各ファイルの凡例により確認済扱い) / 原文注記: 抽出型 QA。script同梱(mlqa.py)。分類への変換は「解答可能か」程度

元のカード(日本語訳)

以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。


データセットカード: "mlqa"

目次

  • データセットの説明
    • データセットの概要
    • 対応するタスクとリーダーボード
    • 言語
  • データセットの構造
    • データのインスタンス
    • データのフィールド
    • データの分割
  • データセットの作成
    • 収集の根拠
    • 元データ
    • アノテーション
    • 個人情報と機微な情報
  • データの利用上の注意
    • データセットの社会的影響
    • バイアスに関する議論
    • その他の既知の制約
  • 追加情報
    • データセットの作成者
    • ライセンス情報
    • 引用情報
    • 貢献

データセットの説明

データセットの概要

MLQA(MultiLingual Question Answering)は、言語横断の抽出型質問応答の性能を評価するためのベンチマークデータセットである。
MLQA は、SQuAD 形式の 5K を超える抽出型 QA の事例(英語では 12K)から成り、英語、アラビア語、ドイツ語、スペイン語、ヒンディー語、ベトナム語、簡体字中国語の 7 言語にわたる。
MLQA は並列性が高く、QA の事例は平均して 4 つの異なる言語の間で並列になっている。

対応するタスクとリーダーボード

More Information Needed

言語

MLQA は、英語、アラビア語、ドイツ語、スペイン語、ヒンディー語、ベトナム語、簡体字中国語の 7 言語の QA の事例を含む。

データセットの構造

データのインスタンス

mlqa-translate-test.ar

  • ダウンロードしたデータセットのファイルのサイズ: 10.08 MB
  • 生成されたデータセットのサイズ: 5.48 MB
  • ディスクの使用量の合計: 15.56 MB

'test' の例は次のとおり。


mlqa-translate-test.de

  • ダウンロードしたデータセットのファイルのサイズ: 10.08 MB
  • 生成されたデータセットのサイズ: 3.88 MB
  • ディスクの使用量の合計: 13.96 MB

'test' の例は次のとおり。


mlqa-translate-test.es

  • ダウンロードしたデータセットのファイルのサイズ: 10.08 MB
  • 生成されたデータセットのサイズ: 3.92 MB
  • ディスクの使用量の合計: 13.99 MB

'test' の例は次のとおり。


mlqa-translate-test.hi

  • ダウンロードしたデータセットのファイルのサイズ: 10.08 MB
  • 生成されたデータセットのサイズ: 4.61 MB
  • ディスクの使用量の合計: 14.68 MB

'test' の例は次のとおり。


mlqa-translate-test.vi

  • ダウンロードしたデータセットのファイルのサイズ: 10.08 MB
  • 生成されたデータセットのサイズ: 6.00 MB
  • ディスクの使用量の合計: 16.07 MB

'test' の例は次のとおり。


データのフィールド

データのフィールドは、すべてのスプリットで同じである。

mlqa-translate-test.ar

  • context: string 型の特徴量。
  • question: string 型の特徴量。
  • answers: 次を含む辞書型の特徴量:
    • answer_start: int32 型の特徴量。
    • text: string 型の特徴量。
  • id: string 型の特徴量。

mlqa-translate-test.de

  • context: string 型の特徴量。
  • question: string 型の特徴量。
  • answers: 次を含む辞書型の特徴量:
    • answer_start: int32 型の特徴量。
    • text: string 型の特徴量。
  • id: string 型の特徴量。

mlqa-translate-test.es

  • context: string 型の特徴量。
  • question: string 型の特徴量。
  • answers: 次を含む辞書型の特徴量:
    • answer_start: int32 型の特徴量。
    • text: string 型の特徴量。
  • id: string 型の特徴量。

mlqa-translate-test.hi

  • context: string 型の特徴量。
  • question: string 型の特徴量。
  • answers: 次を含む辞書型の特徴量:
    • answer_start: int32 型の特徴量。
    • text: string 型の特徴量。
  • id: string 型の特徴量。

mlqa-translate-test.vi

  • context: string 型の特徴量。
  • question: string 型の特徴量。
  • answers: 次を含む辞書型の特徴量:
    • answer_start: int32 型の特徴量。
    • text: string 型の特徴量。
  • id: string 型の特徴量。

データの分割

name test
mlqa-translate-test.ar 5335
mlqa-translate-test.de 4517
mlqa-translate-test.es 5253
mlqa-translate-test.hi 4918
mlqa-translate-test.vi 5495

データセットの作成

収集の根拠

More Information Needed

元データ

初期のデータ収集と正規化

More Information Needed

元の言語の作成者は誰か?

More Information Needed

アノテーション

アノテーションの過程

More Information Needed

アノテーターは誰か?

More Information Needed

個人情報と機微な情報

More Information Needed

データの利用上の注意

データセットの社会的影響

More Information Needed

バイアスに関する議論

More Information Needed

その他の既知の制約

More Information Needed

追加情報

データセットの作成者

More Information Needed

ライセンス情報

More Information Needed

引用情報

@article{lewis2019mlqa,
  title = {MLQA: Evaluating Cross-lingual Extractive Question Answering},
  author = {Lewis, Patrick and Oguz, Barlas and Rinott, Ruty and Riedel, Sebastian and Schwenk, Holger},
  journal = {arXiv preprint arXiv:1910.07475},
  year = 2019,
  eid = {arXiv: 1910.07475}
}

貢献

このデータセットを追加してくれた @patrickvonplaten、@M-Salti、@lewtun、@thomwolf、@mariamabarham、@lhoestq に感謝する。

原文(英語)

Dataset Card for "mlqa"

Dataset Summary

MLQA (MultiLingual Question Answering) is a benchmark dataset for evaluating cross-lingual question answering performance.
MLQA consists of over 5K extractive QA instances (12K in English) in SQuAD format in seven languages - English, Arabic,
German, Spanish, Hindi, Vietnamese and Simplified Chinese. MLQA is highly parallel, with QA instances parallel between
4 different languages on average.

Supported Tasks and Leaderboards

More Information Needed

Languages

MLQA contains QA instances in 7 languages, English, Arabic, German, Spanish, Hindi, Vietnamese and Simplified Chinese.

Dataset Structure

Data Instances

mlqa-translate-test.ar

  • Size of downloaded dataset files: 10.08 MB
  • Size of the generated dataset: 5.48 MB
  • Total amount of disk used: 15.56 MB

An example of 'test' looks as follows.


mlqa-translate-test.de

  • Size of downloaded dataset files: 10.08 MB
  • Size of the generated dataset: 3.88 MB
  • Total amount of disk used: 13.96 MB

An example of 'test' looks as follows.


mlqa-translate-test.es

  • Size of downloaded dataset files: 10.08 MB
  • Size of the generated dataset: 3.92 MB
  • Total amount of disk used: 13.99 MB

An example of 'test' looks as follows.


mlqa-translate-test.hi

  • Size of downloaded dataset files: 10.08 MB
  • Size of the generated dataset: 4.61 MB
  • Total amount of disk used: 14.68 MB

An example of 'test' looks as follows.


mlqa-translate-test.vi

  • Size of downloaded dataset files: 10.08 MB
  • Size of the generated dataset: 6.00 MB
  • Total amount of disk used: 16.07 MB

An example of 'test' looks as follows.


Data Fields

The data fields are the same among all splits.

mlqa-translate-test.ar

  • context: a string feature.
  • question: a string feature.
  • answers: a dictionary feature containing:
    • answer_start: a int32 feature.
    • text: a string feature.
  • id: a string feature.

mlqa-translate-test.de

  • context: a string feature.
  • question: a string feature.
  • answers: a dictionary feature containing:
    • answer_start: a int32 feature.
    • text: a string feature.
  • id: a string feature.

mlqa-translate-test.es

  • context: a string feature.
  • question: a string feature.
  • answers: a dictionary feature containing:
    • answer_start: a int32 feature.
    • text: a string feature.
  • id: a string feature.

mlqa-translate-test.hi

  • context: a string feature.
  • question: a string feature.
  • answers: a dictionary feature containing:
    • answer_start: a int32 feature.
    • text: a string feature.
  • id: a string feature.

mlqa-translate-test.vi

  • context: a string feature.
  • question: a string feature.
  • answers: a dictionary feature containing:
    • answer_start: a int32 feature.
    • text: a string feature.
  • id: a string feature.

Data Splits

name test
mlqa-translate-test.ar 5335
mlqa-translate-test.de 4517
mlqa-translate-test.es 5253
mlqa-translate-test.hi 4918
mlqa-translate-test.vi 5495

Dataset Creation

Curation Rationale

More Information Needed

Source Data

Initial Data Collection and Normalization

More Information Needed

Who are the source language producers?

More Information Needed

Annotations

Annotation process

More Information Needed

Who are the annotators?

More Information Needed

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

More Information Needed

Citation Information

@article{lewis2019mlqa,
  title = {MLQA: Evaluating Cross-lingual Extractive Question Answering},
  author = {Lewis, Patrick and Oguz, Barlas and Rinott, Ruty and Riedel, Sebastian and Schwenk, Holger},
  journal = {arXiv preprint arXiv:1910.07475},
  year = 2019,
  eid = {arXiv: 1910.07475}
}

Contributions

Thanks to @patrickvonplaten, @M-Salti, @lewtun, @thomwolf, @mariamabarham, @lhoestq for adding this dataset.

Downloads last month
76

Paper for HYUKI5/facebook__mlqa