@inproceedings{calderon-reyes-etal-2026-nlp,
    title = "{NLP}-{CIMAT} at {S}em{E}val-2026 Task 9: {LLM}-Based One-Shot and Cross-Lingual Data Augmentation for Polarization Detection",
    author = "Calderon-Reyes, Miriam  and
      Sanchez-Vega, Fernando  and
      Lopez Monroy, Adrian Pastor",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.362/",
    pages = "2886--2893",
    ISBN = "979-8-89176-414-9",
    abstract = "This paper describes our participation in SemEval 2026 Task 9: Multilingual Text Polarization. The task requires estimating polarization levels across languages, where linguistic variability and limited annotated data pose significant challenges. To address data scarcity, we propose a pipeline that combines cross-lingual translation, synthetic data augmentation via LLMs, and domain-specific pre-trained models. Our approach leverages the hypothesis that polarization signals can transfer across languages without substantial loss of semantic alignment, enabling effective data augmentation through translation. Notably, one-shot synthetic example generation emerges as a viable strategy for enriching training data in topic-specific scenarios. Experimental results demonstrate high stability and competitive performance, achieving a macro F1-score of 0.7869 for Spanish and 0.7939 for English on the test set, ranking 21th on the official English leaderboard, while our Spanish results are competitive with top-performing systems, corresponding to 7th place."
}