groundedness scores
收藏资源简介:
该数据集名为groundedness scores,由爱丁堡大学和剑桥大学的研究团队创建,旨在量化不同语言中词汇类别的语义内容性。数据集包含30种语言的groundedness值,通过多语言多模态语言模型计算得出,用于衡量词汇在图像描述中的语义信息量。数据集的创建过程基于图像与语言模型的 surprisal 差异,量化了词汇在不同语境中的语义内容性。该数据集主要应用于语言类型学研究,旨在解决词汇类别在不同语言中的语义功能问题,特别是功能词与词汇词之间的语义差异。
This dataset, named groundedness scores, was developed by a research team from the University of Edinburgh and the University of Cambridge. Its core objective is to quantify the semantic contentfulness of lexical categories across diverse languages. The dataset includes groundedness values for 30 languages, which are computed using multilingual multimodal language models to measure the semantic informativeness of vocabulary in image descriptions. The construction of this dataset is based on the surprisal discrepancy between image and language models, which quantifies the semantic contentfulness of words across different contexts. Primarily, this dataset is utilized in linguistic typology research, aiming to resolve the issue of semantic functions of lexical categories across different languages, especially the semantic disparities between function words and content words.




