naist-nlp/MultiExpArt

这是一个多语言艺术作品解释数据集，旨在评估大型视觉语言模型在不同语言中生成艺术作品解释的能力。数据集包含了10种语言：英语、日语、德语、法语、意大利语、瑞典语、中文、西班牙语、俄语和荷兰语。通过考虑细微差别和国家特定短语，该数据集不依赖机器翻译来创建多语言扩展数据集，用于评估模型在非英语语言中的生成解释能力，并研究指令微调在丰富英语资源中对其他语言性能的影响。

This is a multilingual artwork explanation dataset designed to evaluate the ability of large-scale vision language models to generate explanations in different languages. The dataset includes 10 languages: English, Japanese, German, French, Italian, Swedish, Chinese, Spanish, Russian, and Dutch. It creates an extended multilingual dataset without relying on machine translation, taking into account nuances and country-specific phrases, to assess the generation explanation abilities of models in non-English languages and to examine the impact of Instruction-Tuning in resource-rich English on performance in other languages.