Learning the visualness of text using large vision-language models (EMNLP 2023)