@inproceedings{nuthakki-etal-2026-team,
    title = "Team Evaluators at {S}em{E}val-2026 Task 6: Instruction-Tuned {LLM}s for Clarity and Evasion Classification in Political Interviews",
    author = "Nuthakki, Siva  and
      Pulagam, Sanjay  and
      Woona, Sai",
    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-1.426/",
    pages = "3442--3445",
    ISBN = "979-8-89176-414-9",
    abstract = "This work is part of the SemEval-2026 CLARITY shared task (Task 6), which focuses on detecting clarity and evasion in political question{--}answer pairs from interviews and debates. The competition includes two subtasks: clarity-level classification (Clear Reply, Ambiguous,Clear Non-Reply) and evasion-level classification, which identifies one of nine fine-grained evasion techniques. The dataset consists of annotated question{--}answer pairs with hierarchical labels for both clarity and evasion, enabling comprehensive evaluation of nuanced discoursephenomena. We fine-tune open-source large language models using Low-Rank Adaptation (LoRA) and supervised fine-tuning (SFT), employing structured prompts that jointly encode the question and answer to capture discoursecues. Models are evaluated using Macro F1, the official metric of the shared task. Our system achieves a Macro F1 of 0.83 on Subtask 1 (5th place) and 0.54 on Subtask 2 (9th place), demonstrating that parameter-efficient fine-tuning of LLMs is effective for modeling strategic ambiguity in political discourse."
}