FedericaFabri1/ITALIC-declarative-statements

Description

This dataset contains affirmative sentences derived from the multiple-choice questions of the ITALIC dataset.

Each affirmative sentence is created through a natural merging of the original question and its corresponding answer option. The reformulation preserves the original semantic content while transforming the interrogative multiple-choice structure into declarative statements.

About this dataset

Author of this derived version: Federica Fabri.

This dataset was created for academic purposes, as part of a thesis project on the evaluation of factual knowledge in Italian language models

Dataset Structure

This dataset is built only from the "culture and commonsense" macro-category of ITALIC (excluding the "language capability" macro-category), consistently with the goal of probing factual and cultural knowledge rather than purely linguistic skills.

The 5,964 selected unique questions are expanded into their individual answer options — one correct assertion plus its incorrect "distractor" assertions — for a total of 21,589 rows. The number of options per question varies: 4 questions have 2 options, 2,262 have 3, 3,690 have 4, and 7 have 5 options. A single residual anomaly (one question with only one option) is present and is programmatically excluded during the inference/analysis phase.

Fields

• id (int64): question identifier, repeated across all rows of the same question; used as the key to group the correct assertion with its distractors.
• question (string): the original question text.
• category (string): the cultural sub-category of the item (e.g. literature, art_history, geography).
• answer (string): the single answer option handled in that row (correct or distractor).
• statement (string): the declarative assertion obtained by merging question and answer; this is the string fed to the models for scoring.
• label (int64): 1 if the statement is the correct assertion (gold target), 0 if it is an incorrect distractor.

Derivation Notice

This dataset is derived from the ITALIC dataset (Crisp-Unimib): https://huggingface.co/datasets/Crisp-Unimib/ITALIC

All credit for the original content goes to the original authors. This version only modifies the linguistic structure (question + answer option → affirmative sentence) without altering the underlying meaning.

The original factual content comes from publicly available Italian public competition tests; for primary provenance see the ITALIC documentation.

License

This dataset is released under the MIT License, the same license as the original ITALIC dataset.

Citation

This derived dataset:

Fabri, F. (2026). Italic Affirmative Sentences: a declarative-statement dataset derived from ITALIC [Data set]. Hugging Face. https://huggingface.co/datasets/FedericaFabri1/italic_affirmative_sentences

Original ITALIC benchmark:

@inproceedings{seveso2025italic,
  title     = {ITALIC: An Italian Culture-Aware Natural Language Benchmark},
  author    = {Seveso, Andrea and Potert{\`\i}, Daniele and Federici, Edoardo and Mezzanzanica, Mario and Mercorio, Fabio},
  booktitle = {Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)},
  pages     = {1469--1478},
  year      = {2025}
}