MMTIT-Bench：一个面向认知-感知-推理引导的文本图像机器翻译的多语言多场景基准
MMTIT-Bench: A Multilingual and Multi-Scenario Benchmark with Cognition-Perception-Reasoning Guided Text-Image Machine Translation
为视觉语言大模型在图像内文本翻译任务上提供了多语言、多场景的评测基准，并提出将场景认知、文本感知与翻译推理统一进思维链的数据范式，可直接迁移到其他需要视觉定位与语言生成联合推理的 VLLM 任务。
原文依据 We propose Cognition-Perception-Reasoning for Translation (CPR-Trans), a data paradigm that integrates scene cognition, text perception, and translation reasoning
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论端到端的文本图像机器翻译（TIMT）直接跨语言翻译图像中的文本内容，对真实世界的多语言场景理解至关重要。尽管视觉语言大模型（VLLM）已取得进展，但由于评测资源有限，其在多样视觉场景和低资源语言上的鲁棒性仍缺乏充分探索。我们提出 MMTIT-Bench，一个经人工校验的多语言、多场景基准，包含 1,400 张图像，覆盖十四种非英语且非中文的语言，以及文档、场景、网页图像等多种设置，可用于对端到端 TIMT 进行严格评估。除基准评测之外，我们还研究面向推理的数据设计如何改进翻译。尽管近期 VLLM 已开始引入长思维链（CoT）推理，但面向 TIMT 的有效思考范式仍不成熟：现有设计要么以串行方式级联解析与翻译，要么只关注纯语言推理，忽视了 VLLM 中至关重要的视觉认知。我们提出面向翻译的认知-感知-推理（CPR-Trans），这是一种将场景认知、文本感知与翻译推理整合进统一推理过程的数据范式。借助由 VLLM 驱动的数据生成流水线，CPR-Trans 提供了结构化、可解释的监督，使感知与推理相互对齐。在 3B 与 7B 模型上的实验表明，该方法在准确率与可解释性上均取得一致提升。我们将在论文录用后发布 MMTIT-Bench，以推动多语言、多场景 TIMT 研究。
英文原文摘要 1567 字符
End-to-end text-image machine translation (TIMT), which directly translates textual content in images across languages, is crucial for real-world multilingual scene understanding. Despite advances in vision-language large models (VLLMs), robustness across diverse visual scenes and low-resource languages remains underexplored due to limited evaluation resources. We present MMTIT-Bench, a human-verified multilingual and multi-scenario benchmark with 1,400 images spanning fourteen non-English and non-Chinese languages and diverse settings such as documents, scenes, and web images, enabling rigorous assessment of end-to-end TIMT. Beyond benchmarking, we study how reasoning-oriented data design improves translation. Although recent VLLMs have begun to incorporate long Chain-of-Thought (CoT) reasoning, effective thinking paradigms for TIMT are still immature: existing designs either cascade parsing and translation in a sequential manner or focus on language-only reasoning, overlooking the visual cognition central to VLLMs. We propose Cognition-Perception-Reasoning for Translation (CPR-Trans), a data paradigm that integrates scene cognition, text perception, and translation reasoning within a unified reasoning process. Using a VLLM-driven data generation pipeline, CPR-Trans provides structured, interpretable supervision that aligns perception with reasoning. Experiments on 3B and 7B models show consistent gains in accuracy and interpretability. We will release MMTIT-Bench to promote the multilingual and multi-scenario TIMT research upon acceptance.