multi_nli_mismatch

  • 説明:

Multi-Genre Natural Language Inference (MultiNLI) コーパスは、テキスト含意情報で注釈が付けられた 433,000 のセンテンス ペアのクラウドソース コレクションです。このコーパスは SNLI コーパスをモデルにしていますが、さまざまなジャンルの話し言葉と書き言葉をカバーし、独特のクロスジャンル一般化評価をサポートしているという点で異なります。このコーパスは、コペンハーゲンの EMNLP で開催された RepEval 2017 ワークショップの共有タスクの基礎として機能しました。

スプリット
'train' 392,702
'validation' 10,000
  • 機能構造:
FeaturesDict({
   
'hypothesis': Text(shape=(), dtype=string),
   
'label': Text(shape=(), dtype=string),
   
'premise': Text(shape=(), dtype=string),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
仮説文章ストリング
ラベル文章ストリング
前提文章ストリング
  • 引用
@InProceedings{N18-1101,
  author
= "Williams, Adina
            and Nangia, Nikita
            and Bowman, Samuel"
,
  title
= "A Broad-Coverage Challenge Corpus for
           Sentence Understanding through Inference"
,
  booktitle
= "Proceedings of the 2018 Conference of
               the North American Chapter of the
               Association for Computational Linguistics:
               Human Language Technologies, Volume 1 (Long
               Papers)"
,
  year
= "2018",
  publisher
= "Association for Computational Linguistics",
  pages
= "1112--1122",
  location
= "New Orleans, Louisiana",
  url
= "http://aclweb.org/anthology/N18-1101"
}