FarsTail

自然语言推理 (NLI)，也称为文本蕴涵，是 NLP 中的一项重要任务，其目标是确定前提 p 和假设 h 之间的推理关系。这是一个三类问题，其中每一对 (p, h) 被分配到以下类别之一：如果假设可以从前提中推断出，则为“ENTAILMENT”，如果假设与前提相矛盾，则为“CONTRADICTION”，以及“NEUTRAL”如果以上都不成立。_x000D_
英语 NLI 有 SNLI、MNLI 和 SciTail 等大型数据集，但波斯语等数据较差的语言的数据集很少。_x000D_
波斯语（波斯语）是一种多中心语言，在伊朗、阿富汗和塔吉克斯坦等国家约有 1.1 亿人使用。在这里，我们展示了第一个用于 NLI 任务的相对大规模的波斯数据集，称为 FarsTail。从 3,539 个多项选择题的集合中生成了总共 10,367 个样本。训练、验证和测试部分分别包括 7,266、1,537 和 1,564 个实例。请参阅手稿以获取更多详细信息。

Natural Language Inference (NLI), also known as textual entailment, is a critical task in Natural Language Processing (NLP) that aims to identify the inferential relationship between a premise p and a hypothesis h. It is a three-class classification problem where each pair (p, h) is assigned to one of the following categories: "ENTAILMENT" if the hypothesis can be inferred from the premise, "CONTRADICTION" if the hypothesis contradicts the premise, and "NEUTRAL" if neither of the above holds. Large-scale datasets for English NLI such as SNLI, MNLI, and SciTail are widely available, but there are very few datasets for low-resource languages like Persian. Persian (Farsi) is a polycentric language spoken by approximately 110 million people across countries including Iran, Afghanistan, and Tajikistan. Here, we present FarsTail, the first relatively large-scale Persian dataset for NLI tasks. A total of 10,367 samples were generated from a collection of 3,539 multiple-choice questions. The training, validation, and test splits contain 7,266, 1,537, and 1,564 instances respectively. Please refer to the manuscript for more details.

搜集汇总

背景与挑战

背景概述

FarsTail是首个用于波斯语自然语言推理任务的相对大规模数据集，包含10,367个样本，分为三类推理关系。该数据集从多项选择题生成，旨在支持波斯语NLP研究，填补了数据稀缺语言的空白。