Text Aphasia Battery (TAB)
收藏资源简介:
文本失语症电池是由斯坦福大学等多家研究机构联合开发的临床基础基准,专为评估语言模型中的失语症样缺陷而设计。该数据集包含561个英语文本样本,其中306个来自失语症患者的AphasiaBank转录资料,255个为AI生成文本,涵盖连接文本、词汇理解等四个子测试维度。数据集通过改编临床验证的快速失语症电池核心组件,采用自动化协议确保评估可靠性。主要应用于计算语言病理学领域,旨在建立标准化文本评估框架以解析人工智能系统的语言退化模式。
The Text Aphasia Battery is a foundational clinical benchmark jointly developed by Stanford University and multiple other research institutions, specifically engineered to assess aphasia-like deficits in language models. This dataset contains 561 English text samples, with 306 derived from AphasiaBank transcriptions of aphasic patients and the remaining 255 being AI-generated texts. It encompasses four sub-test dimensions, including connected text and lexical comprehension, among others. The dataset adapts core components of the clinically validated Quick Aphasia Battery, and utilizes automated protocols to ensure evaluation reliability. Primarily deployed in the field of computational patholinguistics, it aims to establish a standardized text evaluation framework for dissecting language degradation patterns in artificial intelligence systems.

- 1通过斯坦福大学, 加州大学旧金山分校, 天普大学, 圣地亚哥州立大学, 加州大学圣地亚哥分校 · 2025年



