The Romanian language in the age of large models
How well current models handle Romanian, where they get it systematically wrong, what is missing from the training data and who should produce the missing datasets.
Romania's National Culture Day (Ziua Culturii Naționale) falls right in the middle of the festival. We treat it as a working theme, not as an excuse for a gala evening.
Large language models are trained overwhelmingly on English. Romanian is a low-resource language compared with what the models consume: less clean data, fewer evaluation sets, fewer tools. The consequences show in the quality of the answers, in the cost of processing and in who can afford to build.
This is not a heritage problem, it is an infrastructure problem. And 15 January is the right day to put it on the table in front of the people who can do something about it.
How well current models handle Romanian, where they get it systematically wrong, what is missing from the training data and who should produce the missing datasets.
Without Romanian evaluation sets we cannot tell whether a system works here. What exists, what is open, and what universities and companies should build together.
Digitising collections and what can be done next: searching old texts, handwriting recognition, linking collections across institutions, public access.
What university researchers already do with automated tools: corpus analysis, critical editions, history, linguistics. What they lack to go further.
What it means for a Romanian author, illustrator or composer to have their work in the training data. What European law says and what comes next.
What can be done intelligently with the heritage of Iași: the Palace of Culture (Palatul Culturii), the university collections, the city archives. What we will not do: generated poetry presented as a party trick.
The morning of 15 January, on the main stage. It is the only part of the festival designed to speak to people outside the industry as well: teachers, librarians, museum curators, researchers, cultural journalists, students of literature and history.
Working on one of these themes at a university, a cultural institution or a company? Write to us at [email protected], with the subject line “AI and Romanian culture”. The programme is being put together until December 2026.
The Smart City Romania network, run by ARSC
Everything you need to know about Romania's fastest-growing industry