ABSTRACTION-ERC/subCat-human

SubCat: A Dataset of Subordinate Categories in Human Mind and LLMs for the Italian Language

A psycholinguistic italian dataset released with the paper How Humans and LLMs Organize Conceptual Knowledge: Exploring Subordinate Categories in Italian. It contains a list of subordiante categories, or exemplars, for 187 concrete words or, basic-level categories.

Dataset Creation

The dataset was created to study how Italian L1 speakers generate exemplars for common object categories. The stimuli consisted of 187 basic-level concrete categories (e.g., dog, table) organized under 12 superordinate semantic categories (e.g., animals, furniture).

An exemplar generation task was administered to 365 Italian L1 speakers. Participants were presented with a list of 15-16 categories and asked to produce as many exemplars as possible for each concept at their own pace.

The final human dataset, after cleaning and standardization, consists of 24,659 exemplars.

Data Processing

Raw data underwent a post-processing step to correct common typos and misspellings. This was done to ensure the consistency and accuracy of the final dataset. The corrected exemplars were then standardized to a common format.

Dataset Structure

The dataset contains the aggregated results of the human experiment. For row in the dataset contains a unique subordinate exemplars and related statistics. The dataset contains the following columns:

1. category: the super-ordinate category
2. concept: the basic-level category
3. exemplar: the generated/produced sub-ordinate level exemplar/concept
4. exemplar_string: a sanitized version of the exemplar
5. availability: a metric which represents how readily the exemplar is produced as a member of its associated category
6. count: the number of occurrences of the exemplar across participants
7. min_rank: the minimum rank of exemplar's occurrence
8. max_rank: the highest rank of exemplar's occurrence
9. mean_rank: the average rank of exemplar's occurrence
10. first_occur: the ratio of exemplar occurring at first rank, divided by the total number of exemplar's occurrence
11. dominance: the proportion of participants who produce the exemplar given its associated category
12. abs_freq_corpus: only for LLM's generated exemplars, the number of exemplar's occurrences in the italian corpus ItTenTen

Citation

If you find this dataset is useful in your own work, please consider citing it as follows:

@inproceedings{pedrotti-etal-2025-humans,
    title = "How Humans and {LLM}s Organize Conceptual Knowledge: Exploring Subordinate Categories in {I}talian",
    author = "Pedrotti, Andrea  and
      Rambelli, Giulia  and
      Villani, Caterina  and
      Bolognesi, Marianna",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-long.224/",
    doi = "10.18653/v1/2025.acl-long.224",
    pages = "4464--4482",
    ISBN = "979-8-89176-251-0",
}