@inproceedings{pk-d-2026-thiyaga6851,
    title = "Thiyaga6851 at {S}em{E}val-2026 Task 11: Disentangling Content and Formal Reasoning in Large Language Models using Neuro-Symbolic Mapping",
    author = "Pk, Thiyagarajaa  and
      D., Thenmozhi",
    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-1.276/",
    pages = "2187--2192",
    ISBN = "979-8-89176-414-9",
    abstract = "This paper presents our system for SemEval-2026 Task 11 Subtask 1, which evaluates the formal validity of English syllogisms independently of semantic plausibility. To reduce content effects, we use a hybrid neuro-symbolic pipeline that separates natural-language abstraction from logical inference. The system maps each syllogism into categorical propositions using template rules and a learned parser, followed by explicit role mapping for the major, minor, and middle terms. If the abstraction is structurally complete, an exact Venn-style satisfiability solver checks validity; otherwise, the instance is routed to a learned fallback classifier. Our official submission achieved 71.73{\%} accuracy, a Total Content Effect of 11.84, a Combined Score of 20.19, and a rank of 41st. Development analysis shows that symbolic inference is reliable on well-formed abstractions, while most remaining errors arise from paraphrase, multiword terms, and unstable term alignment."
}