@inproceedings{cafagna-etal-2023-hl,
    title = "{HL} Dataset: Visually-grounded Description of Scenes, Actions and Rationales",
    author = "Cafagna, Michele  and
      van Deemter, Kees  and
      Gatt, Albert",
    editor = "Keet, C. Maria  and
      Lee, Hung-Yi  and
      Zarrie{\ss}, Sina",
    booktitle = "Proceedings of the 16th International Natural Language Generation Conference",
    month = sep,
    year = "2023",
    address = "Prague, Czechia",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/jlcl-multiple-ingestion/2023.inlg-main.21/",
    doi = "10.18653/v1/2023.inlg-main.21",
    pages = "293--312",
    abstract = "Current captioning datasets focus on object-centric captions, describing the visible objects in the image, often ending up stating the obvious (for humans), e.g. {\textquotedblleft}people eating food in a park{\textquotedblright}. Although these datasets are useful to evaluate the ability of Vision {\&} Language models to recognize and describe visual content, they do not support controlled experiments involving model testing or fine-tuning, with more high-level captions, which humans find easy and natural to produce. For example, people often describe images based on the type of scene they depict ({\textquotedblleft}people at a holiday resort{\textquotedblright}) and the actions they perform ({\textquotedblleft}people having a picnic{\textquotedblright}). Such concepts are based on personal experience and contribute to forming common sense assumptions. We present the High-Level Dataset, a dataset extending 14997 images from the COCO dataset, aligned with a new set of 134,973 human-annotated (high-level) captions collected along three axes: scenes, actions and rationales. We further extend this dataset with confidence scores collected from an independent set of readers, as well as a set of narrative captions generated synthetically, by combining each of the three axes. We describe this dataset and analyse it extensively. We also present baseline results for the High-Level Captioning task."
}