AI for Science Conference 2025: Highlights from Our Team

From 22 to 26 September 2025, the Faculty of Computer and Information Science at the University of Ljubljana hosted the 28th Discovery Science Conference and a series of co-located events under the joint name AI for Science. The conference brought together researchers and practitioners working at the intersection of artificial intelligence, data science, and big data analysis, exploring their applications in scientific and industrial contexts.

This year, our Centre of Excellence in Artificial Intelligence for Digital Humanities (AI4DH) supported the dedicated AI and Digital Humanities track, an interdisciplinary forum focused on the opportunities and challenges of applying AI to humanities research.

Keynote Highlights

We had the privilege of hearing from Dr Marieke van Erp, leader of the Digital Humanities Research Lab at the KNAW Humanities Cluster. Her keynote talk, “Reflections on Reflection in Humanities and AI,” highlighted three key insights:

Human oversight remains important: historical data often contains information that computers cannot easily interpret. For example, an entry such as 34.737, 2, 8 (a value for coffee in the Dutch East India Company archives) makes perfect sense to historians but not to machines.

Data quality is crucial: even with powerful large language models, the principle of “garbage in, garbage out” still applies. Analysis is only as strong as the underlying data.

Interdisciplinary collaboration is essential: computer scientists aim to simplify, while humanities scholars often introduce complexity. To capture phenomena such as the history of odours in a database, both perspectives are necessary.

Her core message: AI is not only a technical field but also a humanities discipline, as it ultimately reflects our attempts to understand human cognition and intelligence.

Contributions from Our Team

Our AI4DH team actively contributed to the programme, with several members presenting their latest research.

• Copyright and Large Language Models
  Dr Simon Krek and Dr Maja Bogataj Jančič examined the Slovenian legal framework for training LLMs. While Slovenia provides a strong legal basis for using copyright materials in training large language models, institutions remain hesitant to fully leverage its potential in practice.
  
• AI for Language Games
  Dr Špela Arhar Holdt, Tjaša Arčon, and Dr Iztok Kosem explored how LLMs can generate “distractors” (plausible but incorrect answers) for CJVT Games, a Slovenian word game platform. Their findings show that LLMs perform remarkably well in generating quality data for language games.
  
• Evaluating LLMs in Value Annotation
  Dr Ajda Pretnar Žagar presented her study on applying LLMs to value annotation tasks. While useful for exploratory analysis, LLMs often fail to provide reliable or interpretable results in nuanced or multilingual settings.
  
• Grammatical Analysis of Multilingual Corpora
  Dr Kaja Dobrovoljc and collaborators introduced an innovative approach that integrates Universal Dependencies data into LLMs to enhance grammatical analysis across multiple languages.
  
• Grammar Error Correction in Slovenian
  Dr Marko Robnik Šikonja and dr Špela Arhar Holdt together with co-authors presented efforts to build a synthetic dataset for grammar error correction in Slovenian.
  
• Folkloristic Motifs and LLMs
  Tjaša Arčon, Dr Polona Tratnik and Dr Marko Robnik Šikonja showcased a novel methodology for automatically detecting motifs in folktales, using the Cinderella tale as a case study. Their approach demonstrates how LLMs can uncover complex narrative patterns, enabling large-scale comparative folklore analysis.
  
• From Datasets to Data Spaces
  Dr Slavko Žitnik discussed the evolution of data spaces, from European data strategies to sectoral implementations in health, mobility, and culture. Special attention was given to the Language Data Space (LDS), a European initiative for large-scale, multilingual language data sharing that is poised to advance both AI research and digital humanities.
  

Looking Ahead

A book of abstracts will soon be available on the conference website. Be sure to visit our photo gallery to see our team in action. We are proud to have contributed to such an inspiring event and look forward to continuing our work at the crossroads of AI and digital humanities.