arXiv:2512.14926v2 Announce Type: replace-cross Abstract: Focusing on low-resource languages is an essential step toward democratizing generative AI. In this work, we contribute to reducing the multimodal NLP resource gap for Romanian. We translate the widely known Flickr30K dataset into Romanian and further extend it for visual question answering by leveraging open-source LLMs. We demonstrate the usefulness of our datasets by fine-tuning open-source VLMs on Romanian visual question answering. We select VLMs from three widely used model families: LLaMA 3.2, LLaVA 1.6, and Qwen2. For fine-tunin
This work directly addresses the existing void in high-quality multimodal NLP resources for low-resource languages like Romanian, leveraging recent advances in open-source LLMs.
Democratizing generative AI by extending its capabilities to low-resource languages reduces digital inequalities and expands the global addressable market for AI applications, fostering greater innovation outside major language blocs.
The availability of translated and extended datasets for Romanian VLMs enables the development of more diverse and linguistically inclusive AI systems, moving beyond English-centric models.
- · Romanian AI developers
- · Generative AI LMMs
- · Low-resource language communities
- · Multilingual AI platforms
- · Developers solely focused on high-resource languages
Increased accessibility and utility of multimodal AI for Romanian speakers and businesses.
Potential for other low-resource language communities to replicate this methodology, further diversifying the AI landscape.
Broadened global competition in AI development as more nations build out domestic language capabilities.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI