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.

sihaochen/propsegment(複製)

データの内容

文を命題に分割する課題と、命題が文書に含意されるかを判定する課題のための英語のデータ。Google Research の PropSegmEnt を、ニュース部分の URL からクローリングで再現したもので、全体の約 96% を復元している。元のデータは命題と含意関係を人手でアノテーションしたもので、segmentation 設定では sentence と、命題の範囲を [M] と [/M] で囲んだ propositions の組が、nli 設定では hypothesis、premise、label の組が入る。

データの概要

項目 値
行数 45,232(2 構成・6 分割の合計)
列数 2〜3(構成による)
容量 13.87 KB(ファイル 3 本)
ファイルの形式 py 1
言語 en
公開 public

構成と分割

構成(config) 分割(split)ごとの行数 列数
nli train 23,835、validation 1,949、test 8,284 3
segmentation train 9,400、validation 686、test 1,078 2

列(構成 nli・分割 train。構成ごとに列が違う)

列 型 先頭の行の値
hypothesis string [M]Coaches at Tremper High School in the Kenosha…
premise string The ACLU said the incidents at the schools are p…
label string n

行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は HYUKI5/dataset-catalog にある。

複製について

sihaochen/propsegment をリビジョン 8454abe0cc508213c931e32fff18f982fc26712f 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

(元のカードに該当の見出しが無いので、ラベルの付け方に触れた段落を抜き出した)

PropSegment(Proposition-level Segmentation and Entailment)は、英語の文を命題に分割し、命題レベルの含意関係を認識するための大規模な人手アノテーション付きデータセットである。含意関係とは、関連する別の文書が各命題を含意するか、矛盾するか、どちらでもないかのことである。

元のデータセットには、人手でアノテーションされた 45k 超の命題(文の中の個々の意味の単位)と、命題と文書の間の 35k 超の含意ラベルが含まれる。

proposition_segmentation.*.jsonl には、文を命題に分割するテキスト分割タスクのデータが含まれる。 出力の命題は、特殊トークン [SEP] で区切って 1 つの文字列に連結される(命題の間に特定の順序はない)。 各命題は [M] と [/M] で囲まれた範囲としてアノテーションされる。

微調整に使うときの注意

  • 調査メモの台帳での区分: A(許諾型のライセンスだけで、疑義の注記が無い)。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
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) cc-by-4.0 / license_use=commercial 確認済 license_use=commercial / [Jev 元データ source] source_annotation_converted(deterministic recast; no model r

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。


PropSegmEnt: A Large-Scale Corpus for Proposition-Level Segmentation and Entailment Recognition

データセットの説明

データセットの概要

これは Google Research の 「PropSegment」データセット を、(ウェブクローリングによって)再現し、加工した版である。

データセットのうち News の部分は URL の形でしか公開されていないため、クローリングによってデータセットを復元した。 全体の約 96% は再現でき、残りの約 4% は URL が無効になっているか、文が編集されていて元のデータセットと対応させられないものである。

PropSegment(Proposition-level Segmentation and Entailment)は、英語の文を命題に分割し、命題レベルの含意関係を認識するための大規模な人手アノテーション付きデータセットである。含意関係とは、関連する別の文書が各命題を含意するか、矛盾するか、どちらでもないかのことである。

元のデータセットには、人手でアノテーションされた 45k 超の命題(文の中の個々の意味の単位)と、命題と文書の間の 35k 超の含意ラベルが含まれる。

詳細はデータセットの論文を参照のこと。

データセットの構造

ここでは seq2seq モデルの入出力用に加工した版を提供する。

proposition_segmentation.*.jsonl には、文を命題に分割するテキスト分割タスクのデータが含まれる。 出力の命題は、特殊トークン [SEP] で区切って 1 つの文字列に連結される(命題の間に特定の順序はない)。 各命題は [M] と [/M] で囲まれた範囲としてアノテーションされる。

{
  "sentence": "This film marks the directorial debut for production designer Robert Stromberg.",
  "propositions": "This film marks the directorial debut for [M]production designer Robert Stromberg.[/M][SEP]This [M]film marks the directorial debut for[/M] production designer [M]Robert Stromberg[/M]."
}

propnli.*.jsonl には、命題と文書の含意タスクの例が含まれる。すなわち、命題と文書が与えられたとき、その命題が文書に対して含意されるか、矛盾するか、中立であるかを予測する。

{
  "hypothesis": "[M]The Departed is[/M] a 2006 feature film [M]directed by Martin Scorsese.[/M]",
  "premise": "The Departed is a 2006 American crime thriller film directed by Martin Scorsese and written by William Monahan. It starred Leonardo DiCaprio, Matt Damon, Jack Nicholson, and Mark Wahlberg, with Martin Sheen, Ray Winstone, Vera Farmiga, and Alec Baldwin in supporting roles. It is a remake of the Hong Kong film Infernal Affairs (2002).\nThe Departed won the Oscar for Best Picture at the 79th Academy Awards. Scorsese received the Oscar for Best Director, Thelma Schoonmaker the Oscar for Best Editing and William Monahan the Oscar for Best Adapted Screenplay.",
  "label": "e"
}

引用

@inproceedings{chen2023propsegment,
    title = "{PropSegmEnt}: A Large-Scale Corpus for Proposition-Level Segmentation and Entailment Recognition",
    author = "Chen, Sihao and Buthpitiya, Senaka and Fabrikant, Alex and Roth, Dan  and Schuster, Tal",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
    year = "2023",
}
原文(英語)

PropSegmEnt: A Large-Scale Corpus for Proposition-Level Segmentation and Entailment Recognition

Dataset Summary

This is a reproduced (i.e. after web-crawling) and processed version of the "PropSegment" dataset from Google Research.

Since the News portion of the dataset is released only via urls, we reconstruct the dataset by crawling. Overall, ~96% of the dataset can be reproduced, and the rest ~4% either have url no longer valid, or sentences that have been edited (i.e. cannot be aligned with the orignial dataset).

PropSegment (Proposition-level Segmentation and Entailment) is a large-scale, human annotated dataset for segmenting English text into propositions, and recognizing proposition-level entailment relations --- whether a different, related document entails each proposition, contradicts it, or neither.

The original dataset features >45k human annotated propositions, i.e. individual semantic units within sentences, as well as >35k entailment labels between propositions and documents.

Check out more details in the dataset paper.

Dataset Structure

Here we provide processed versions of the dataset for seq2seq model inputs/outputs.

proposition_segmentation.*.jsonl contains data for the text segmentation task, i.e. split a sentence into propositions. The output propositions are concatenated as one string (with no particular order between them) by a special token [SEP]. Each proposition is annotated as spans enclosed by [M] and [/M].

{
  "sentence": "This film marks the directorial debut for production designer Robert Stromberg.",
  "propositions": "This film marks the directorial debut for [M]production designer Robert Stromberg.[/M][SEP]This [M]film marks the directorial debut for[/M] production designer [M]Robert Stromberg[/M]."
}

propnli.*.jsonl contains examples for the proposition-to-document entailment task, i.e. Given a proposition and a document, predict whether the proposition can be entailed/contradicted, or neutral with respect to the document.

{
  "hypothesis": "[M]The Departed is[/M] a 2006 feature film [M]directed by Martin Scorsese.[/M]",
  "premise": "The Departed is a 2006 American crime thriller film directed by Martin Scorsese and written by William Monahan. It starred Leonardo DiCaprio, Matt Damon, Jack Nicholson, and Mark Wahlberg, with Martin Sheen, Ray Winstone, Vera Farmiga, and Alec Baldwin in supporting roles. It is a remake of the Hong Kong film Infernal Affairs (2002).\nThe Departed won the Oscar for Best Picture at the 79th Academy Awards. Scorsese received the Oscar for Best Director, Thelma Schoonmaker the Oscar for Best Editing and William Monahan the Oscar for Best Adapted Screenplay.",
  "label": "e"
}

Citation

@inproceedings{chen2023propsegment,
    title = "{PropSegmEnt}: A Large-Scale Corpus for Proposition-Level Segmentation and Entailment Recognition",
    author = "Chen, Sihao and Buthpitiya, Senaka and Fabrikant, Alex and Roth, Dan  and Schuster, Tal",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
    year = "2023",
}
Downloads last month
61

Paper for HYUKI5/sihaochen__propsegment