analiza aiqbench
This arXiv paper introduces Culture-MMT, a benchmark that tests whether multimodal models can translate social-media posts by combining text, images, and cultural context. Its ablations show that real images help substantially on visually dependent posts (+0.280 on Visual, +0.197 on Hybrid) but barely on symbol-based ones (+0.050), and that random images add nothing. For readers, the takeaway is that visual grounding and cultural-pragmatic adaptation look like separate failure modes, so aggregate translation scores can hide which one a model is failing at.
Beyond Words Alone: Disentangling Distributed Evidence for Visually Grounded Cultural Translation
arXiv:2610.35482v1 Announce Type: cross Abstract: Social media enables cross-lingual communication, yet the meaning of user-generated content (UGC) is often distributed across text, images, and cultural knowledge, making translation challenging. Existing evaluations typically treat images as optional and assume textual information is sufficient, obscuring whether errors stem from missing visual evidence or inadequate cultural adaptation. We introduce \textbf{Culture-MMT}, a diagnostic benchmark for \textbf{distributed-evidence pragmatic reconstruction} in social-media translation. It contains 498 curated, de-identified UGC instances across 20 domains, categorized into Symbol, Visual, and Hybrid by their required evidence. We also develop a human-aligned domain-specific judge, Judger, and a controlled evaluation framework that systematically varies the evidence available to translators. Across 17 model configurations, real images improve performance by +0.280 on Visual and +0.197 on Hybrid, but only +0.050 on Symbol, while random images provide no comparable gains. Human-authored descriptions further show that jointly dependent cases remain more challenging even when visual facts are explicit. These results characterize culturally effective social-media translation as {meaning reconstruction from distributed evidence}, highlighting visual grounding and cultural-pragmatic adaptation as distinct model bottlenecks.
Abstract:Social media enables cross-lingual communication, yet the meaning of user-generated content (UGC) is often distributed across text, images, and cultural knowledge, making translation challenging. Existing evaluations typically treat images as optional and assume textual information is sufficient, obscuring whether errors stem from missing visual evidence or inadequate cultural adaptation. We introduce \textbf{Culture-MMT}, a diagnostic benchmark for \textbf{distributed-evidence pragmatic reconstruction} in social-media translation. It contains 498 curated, de-identified UGC instances across 20 domains, categorized into Symbol, Visual, and Hybrid by their required evidence. We also develop a human-aligned domain-specific judge, Judger, and a controlled evaluation framework that systematically varies the evidence available to translators. Across 17 model configurations, real images improve performance by +0.280 on Visual and +0.197 on Hybrid, but only +0.050 on Symbol, while random images provide no comparable gains. Human-authored descriptions further show that jointly dependent cases remain more challenging even when visual facts are explicit. These results characterize culturally effective social-media translation as {meaning reconstruction from distributed evidence}, highlighting visual grounding and cultural-pragmatic adaptation as distinct model bottlenecks.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2610.35482 [cs.CL]
(or arXiv:2610.35482v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.35482 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 275 cuvinte
Subiecte conexe
Privacy-Preserving Text Sanitization for Distributed Agents Collaboration via Disentangled Representations
arXiv · 17 iun. 2026
Causal Representation Learning for Generalisable Recommendation
arXiv · 27 oct. 2026
MLLMs Hallucinate when Information Distribution Drifts in Synergy Heads
arXiv · 17 sept. 2026
Correct When Paired, Wrong When Split: Decoupling and Editing Modality-Specific Neurons in MLLMs
arXiv · 17 iun. 2026