Our submission on Task 2 on Multilingual Claim Normalization, achieving achieved top-three positions in 15 of the 20 languages.
Claim normalization is the process of converting informal social media posts into concise, self-contained, and verifiable statements. Training was executed on three platforms: Kaggle kernels equipped with two T4 GPUs, Google Colab Pro sessions with a single T4 GPU, and an on-premise server hosting an A100, which provided training of different models of different sizes.
If you use our work, please cite our paper https://ceur-ws.org/Vol-4038/paper_79.pdf:
@misc{almada2025claimnorm,
title={AKCIT-FN at CheckThat! 2025: Switching Fine-Tuned SLMs and LLM Prompting for Multilingual Claim Normalization},
author={Fabrycio Leite Nakano Almada and Kauan Divino Pouso Mariano and Maykon Adriell Dutra and Victor Emanuel da Silva Monteiro and Juliana Resplande Sant'Anna Gomes and Arlindo Rodrigues Galvão Filho and Anderson da Silva Soares},
year={2025},
eprint={2509.11496},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2509.11496},
}This work has been fully funded by the project "Computational Techniques for Multimodal Data Security and Privacy" supported by the Advanced Knowledge Center in Immersive Technologies (AKCIT), with financial resources from the PPI IoT/Manufatura 4.0 / PPI HardwareBR of the MCTI grant number 057/2023, signed with EMBRAPII.
This work has been fully funded by the project Computational Techniques for Multimodal Data Security and Privacy supported by Advanced Knowledge Center in Immersive Technologies (AKCIT), with financial resources from the PPI IoT/Manufatura 4.0 / PPI HardwareBR of the MCTI grant number 057/2023, signed with EMBRAPII.