Margins2Models

The initiative addresses a pressing global challenge: the underrepresentation of Hungarian and other low-resource languages and their visual heritage in artificial intelligence systems.

The project aims to develop a generative AI platform capable of culturally sensitive and linguistically accurate image and video generation. Current models are predominantly trained on English-language, global datasets, resulting in distorted or incomplete representations of Hungarian visual culture. The initiative seeks to address this gap and promote the preservation of cultural diversity in the digital space.

The initiator and leader of the research is Dr. Brigitta Iványi-Bitter, PhD, head of the Heritage in Motion Lab operating within the MOME Center for Research. The platform will be built on richly annotated, multilingual datasets derived from museum and archival collections, enabling culturally grounded model training. The project places strong emphasis on the ethical use of data and the accurate representation of linguistic and cultural contexts.

Coordinated by MOME, the project is implemented in international collaboration with leading partners including the University of Lisbon (INESC-ID), the University of California, Los Angeles (UCLA), and the Holon Institute of Technology (Israel). The consortium brings together expertise from artificial intelligence, language technology, design, and cultural heritage studies.

A key focus of the project is the development of multilingual generative models, alongside the creation of an evaluation framework (benchmark system) capable of measuring cultural accuracy and bias in image-generation systems. This represents a significant innovation, offering guidance for both national and international AI development efforts.

In addition, the research will establish an intercultural ethical framework to support the responsible use of cultural heritage data and foster the development of culturally reliable generative AI systems. Expected outcomes include a functional AI prototype, a multilingual dataset, and educational and professional guidelines.

In the long term, the initiative aims to position Hungary not only as a user but as a proactive contributor to the future of artificial intelligence, demonstrating how cultural heritage can be integrated into emerging technologies in an authentic and sustainable way.