somosnlp-hackathon-2026/Onexe-QA-Dataset
收藏资源简介:
该数据集专门设计用于评估大型语言模型(LLMs)在加那利西班牙语语境下的方言、语言和文化理解能力。它包含4,683个基于加那利语言学院官方词典的评估问题,每个记录使用41种不同的问题模板随机但可重复地生成。主要目标是衡量模型在不同疑问结构下识别加那利地区术语的能力,并评估其对词汇和语法变化的鲁棒性,而无需直接访问答案。
This dataset has been designed specifically for evaluating the dialectal, linguistic, and cultural understanding of Large Language Models (LLMs) within the context of Canarian Spanish. It contains 4,683 evaluation questions based on the official lexicon of the Academy of Canarian Language (Academia Canaria de la Lengua - ACL). Each record presents a linguistic query phrased using one of 41 distinct question templates generated randomly yet reproducibly. The main goal is to measure a models ability to identify Canarian regional terms under diverse interrogative structures and evaluate its robustness to lexical and grammatical variations without having direct access to the answers in this subset.




