@inproceedings{tetakali-tetakali-2026-silkpeak,
    title = "{S}ilk{P}eak at {S}em{E}val-2026 Task 6: When Politicians Dodge {---} Unmasking Evasion in Political Interviews through Joint Multi-Task Transformer Learning",
    author = "Tetakali, Amruth  and
      Tetakali, Lavanya",
    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-1.253/",
    pages = "2020--2025",
    ISBN = "979-8-89176-414-9",
    abstract = "This paper describes a system for SemEval-2026 Task 6 (CLARITY), which focuses on recognizing evasive communication in political interviews. The approach treats the one subtask{---}determining the clarity level of an answer {---}as a single joint multi-task problem. A DeBERTa-v3-Large encoder is shared across both tasks, processing the question and answer as a single concatenated sequence. By updating independent linear classification heads simultaneously, the model allows the fine-grained learning signals from the evasion taxonomy to directly inform the broader clarity-level decisions, and vice versa. On the official evaluation set, this joint discriminative system achieves a 0.76 macro F1 score on Task 1. This approach significantly outperforms standard single-task baseline models, hierarchical bi-encoding architectures, and generative large language models like LoRA-tuned LLaMA-3-8B."
}