Arabic Dataset for Commonsense Validation 

用于常识验证的阿拉伯语数据集基于常识验证和解释 (ComVE) 任务我们翻译了用于常识验证的原始英文数据集。提供的数据集中的每个示例都由 2 个句子组成：{s1, s2} 和一个指示哪一个无效的标签。这对句子过于相似，单词之间存在细微差别，无法识别哪个句子没有意义。

据我们所知，没有公开的阿拉伯语数据集可用于常识验证。提供的数据集有 12k 行，由三个文件组成：训练、验证和测试文件。每行由两个句子和不合理句子的标签组成。

An Arabic commonsense validation dataset: We translated the original English commonsense validation dataset based on the Commonsense Validation and Explanation (ComVE) task. Each example in the provided dataset consists of two sentences: {s1, s2} and a label indicating which one is nonsensical. The two sentences are highly similar, with only minor lexical differences, making it difficult to identify the nonsensical one.
To the best of our knowledge, no publicly available Arabic datasets for commonsense validation exist. The provided dataset contains 12,000 rows and is composed of three files: training, validation, and test files. Each row includes two sentences and a label marking the nonsensical sentence.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于常识验证的阿拉伯语资源，通过翻译英文原始数据构建，包含12k行训练、验证和测试样本，每行提供两个相似句子及其无意义标签。它由约旦科技大学于2020年发布，旨在填补阿拉伯语常识验证数据的空白。