INFERES

INFERES是首个为欧洲西班牙语设计的自然语言推理（NLI）数据集，由德克萨斯大学奥斯汀分校信息学院的研究团队创建。该数据集包含8055个黄金标准的命题-假设对，涵盖广泛的主题和语言现象，特别关注基于否定的对比和对抗性示例。数据集的创建过程结合了专家语言学家和众包工作者的策略，旨在提供高质量数据并系统评估自动化系统。INFERES不仅用于训练自动化系统，还用于深入理解推理的本质，特别是在否定和指代消解方面的研究。INFERES is the first natural language inference (NLI) dataset designed for European Spanish, developed by a research team from the School of Information at The University of Texas at Austin. This dataset includes 8,055 gold-standard premise-hypothesis pairs, covering a broad spectrum of topics and linguistic phenomena, with special emphasis on negation-based contrasts and adversarial examples. The construction of the dataset integrates strategies involving expert linguists and crowdworkers, with the goal of delivering high-quality data and enabling systematic evaluation of automated systems. INFERES can be utilized not only for training automated systems but also for gaining in-depth insights into the nature of inference, particularly research on negation and coreference resolution.

搜集汇总

背景与挑战

背景概述

INFERES是一个西班牙语自然语言推理（NLI）数据集，专门引入了基于否定结构的对抗性示例，旨在增强模型对否定语义的理解和鲁棒性。该数据集与一篇发表于COLING 2022的学术论文相关联，提供了用于训练和评估的代码及数据资源。