@inproceedings{buyukbas-etal-2026-sabancigroup4,
    title = "Sabancigroup4 at {S}em{E}val-2026 Task 5: Uncertainty-Aware Semantic Plausibility Scoring via {GNLL} Regression and {LLM} Rationales",
    author = {B{\"u}y{\"u}kba{\c{s}}, Salih  and
      Benli, Doruk  and
      Kara, Osman  and
      Kek{\"u}ll{\"u}o{\u{g}}lu, Dilara},
    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-1.144/",
    pages = "1056--1062",
    ISBN = "979-8-89176-414-9",
    abstract = "SemEval-2026 Task 5 is a shared task on rating the plausibility of an ambiguous homonym in a predetermined context. The dataset of this task consists of a precontext {\&} sentence {\&} ending combinations for each homonym, and the plausibility of the sample is manually rated by 5 annotators. The task of participating teams was to automatically predict the plausibility with respect to the mean rate given by the annotators. Unlike traditional models that rely on single-label selection, this task frames disambiguation as a probabilistic distribution over multiple plausible meanings. To this end, we propose an uncertainty-aware training strategy using GNLL regression, and semantic context enrichment through POS tags and LLM rationales. Our system exhibits competitive performance, achieving 90{\%} accuracy within standard deviation and 81{\%} Spearman correlation, and placing us in the ninth place in the leaderboard."
}