@inproceedings{jing-etal-2026-fifa,
    title = "{FIFA}: Unified Faithfulness Evaluation Framework for Text-to-Video and Video-to-Text Generation",
    author = "Jing, Liqiang  and
      Lai, Viet Dac  and
      Yoon, Seunghyun  and
      Bui, Trung  and
      Du, Xinya",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.555/",
    pages = "11418--11456",
    ISBN = "979-8-89176-395-1",
    abstract = "Video Multimodal Large Language Models (VideoMLLMs) have achieved remarkable progress in both Video-to-Text and Text-to-Video tasks. However, they often suffer from hallucinations, generating content that contradicts the visual input. Existing evaluation methods are limited to one task (V2T) and also fail to assess hallucinations in open-ended, free-form responses. To address this gap, we propose FIFA, a unified \textbf{F}a\textbf{I}th\textbf{F}ulness ev\textbf{A}luation framework that extracts comprehensive descriptive facts, models their semantic dependencies via a Spatio-Temporal Semantic Dependency Graph, and verifies them using VideoQA models. We further introduce , a tool-based correction framework that revises hallucinated content. Extensive experiments demonstrate that FIFA aligns more closely with human judgment than existing evaluation methods, and that {~} effectively improves factual consistency in both text and video generation."
}