The Impact of AI on Low-resource languages and their visual heritage
The research examines how well AI-generated visual content can reflect the visual heritage of cultures that are underrepresented in Machine learning-ready digital databases. The rapid advancement of AI-driven creative tools predominantly benefits high-resource languages, such as English, Spanish, Chinese, leading to linguistic and cultural homogenization. Low-resource languages like Hungarian remain underrepresented in AI datasets, limiting their presence in global digital platforms. This exclusion marginalizes Hungary’s visual heritage, resulting in misrepresentation of its visual heritage. Without dedicated AI development, the public knowledge about the visual heritage of low-resource languages faces long-term erosion.
Partner organizations:
- The Jan Evangelista Purkyně University in Ústí nad Labem, Design Faculty, Czechia
- Institute of Philosophy, Slovak Academy of Sciences, v.v.i., Bratislava, Slovak Republic
- University of Silesia in Katowice, Faculty of Humanities, Centre for Critical Technology Studies, Katowice, Poland
- Fameplay.ai – Czech Video Center, Prague, Czechia
- Budapest University of Technology and Economics, Faculty of Electrical Engineering and Informatics, Department of Telecommunications and AI, Hungary
- Budapest University of Technology and Economics, Faculty of Economy and Social Sciences, Department of Sociology and Communication, Hungary
- ELTE BTK Digital Humanities, Budapest, Hungary
Project Phases
Scope of Research
This research initiative, led by MOME, aims to address the underrepresentation of low-resource languages, such as Hungarian, and their associated visual cultures in artificial intelligence systems. The project defines the status quo of the problematic AI systems and develops AI tools tailored to our low-resource visual heritage, enhancing the visibility of Hungarian cultural heritage. Based on a transparent legal framework it will create machine-learning-compatible datasets from public collections to improve AI model performance.
Development project
To preserve Hungarian and other low-resource language related visual heritages in an accurate way, this project will develop AI-compatible datasets from digital archives. The development project will enhance AI models to support low-resource languages by fine-tuning open-source frameworks. The lab develops text-to-video applications that generate AI-powered visual content inspired by Hungarian and other low-resurce language related art histories, and design histories. We Foster cross-sector collaboration with art and film museums, archives, and cultural institutions to ethically source and utilize heritage data.
Student participation in the project
There are several courses attached to this research and development project, in which students can take a look under the hood. Their visual research and experimentation informs the research projects. We encourage students to apply for an internship at the Heritage in Motion lab. (Regarding the internship opportunity write to: ivanyi-bitter.brigitta@mome.hu)
Result
The project’s outcome will enhance the reliability of the digital ecosystem. The AI tools developed through this initiative will make it possible to accurately integrate visual heritage associated with Hungarian—and other similar languages—into newly generated AI content. This will be a valuable resource for artists and designers in the creative industry, as well as for academic researchers, educators, and teachers, allowing them to illustrate their presentations with precise, AI-generated images. We also hope that our project will help speed up the digitization of both public and private collections related to the Hungarian language. Moreover, the research and development process we establish could serve as a model for other countries and communities with similarly low digital resource visual heritage.
Scope of Research
Development project
Student participation in the project
Result