@inproceedings{hua-xu-2025-less,
    title = "When Less Is More: Logits-Constrained Framework with {R}o{BERT}a for {A}ncient {C}hinese {NER}",
    author = "Hua, Wenjie  and
      Xu, Shenghan",
    editor = "Anderson, Adam  and
      Gordin, Shai  and
      Li, Bin  and
      Liu, Yudong  and
      Passarotti, Marco C.  and
      Sprugnoli, Rachele",
    booktitle = "Proceedings of the Second Workshop on Ancient Language Processing",
    month = may,
    year = "2025",
    address = "The Albuquerque Convention Center, Laguna",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-emnlp/2025.alp-1.25/",
    doi = "10.18653/v1/2025.alp-1.25",
    pages = "192--196",
    ISBN = "979-8-89176-235-0",
    abstract = "This report presents our team{'}s work on ancient Chinese Named Entity Recognition (NER) for EvaHan 20251. We propose a two-stage framework combining GujiRoBERTa with a Logits-Constrained (LC) mechanism. The first stage generates contextual embeddings using GujiRoBERTa, followed by dynamically masked decoding to enforce valid BMES transitions. Experiments on EvaHan 2025 datasets demonstrate the framework{'}s effectiveness. Key findings include the LC framework{'}s superiority over CRFs in high-label scenarios and the detrimental effect of BiLSTM modules. We also establish empirical model selection guidelines based on label complexity and dataset size."
}