Project Information
- Funding/Details: ARIS project (Collaborator)
- Duration: 1. 10. 2024 - 1. 10. 2027
- Website: https://www.cjvt.si/llm4dh/
Description
The Large Language Models for Digital Humanities (LLM4DH) project aims to enhance Slovene research capacity by applying state-of-the-art language technologies and artificial intelligence to digital humanities (DH). By addressing the shortcomings of large language models (LLMs) in less-resourced languages, the project implements knowledge injection, speech technologies, and multimodal models to support fields such as linguistics, contemporary history, and law.
Under the leadership of Prof. Matija Marolt (multimedia lead at UL FRI), our research group leads Task T4.1: Named Entity Graphs. This task investigates networks of people, organizations, and places across historical and contemporary corpora. Current Named Entity Recognition (NER) and Linking (NEL) methods often fail to adapt to morphologically rich and less-resourced languages like Slovenian, leaving them underrepresented in historical databases.
To overcome this, we adapt NER and NEL methodologies to build comprehensive interaction graphs for historical entities. Using pretrained, language-agnostic reasoning models like ULTRA, the task supports the analysis and visualization of relationship networks, sentiments, and narratives in historical newspapers, parliamentary transcripts, and literary texts. This work provides a crucial open-access foundation for diachronic analysis and critical discourse in Slovene digital humanities.
Publications
- Dobranic, F., Munda, T., Pejic, O., Gorjanc, V., Smajdek, U., Bordon, D., Lenardic, J., Konovsek, T., Tekavcic, K. P. d. M., Bohak, C., & Fiser, D. (2026). Approaches to Analysing Historical Newspapers Using LLMs. CoRR, abs/2603.25051.
- Munda, T., Dobranić, F., Šmajdek, U., Pejić, O., Bohak, C., Gorjanc, V., & Fišer, D. (2026). Benchmarking LLMs for aspect-based sentiment classification in Slovene historical periodicals. In LREC 2026 : Shaping multilingual, multimodal AI for the social sciences and humanities (LLMs4SSH) @ LREC 2026 : workshop proceedings (pp. 103-113).
- Dobranić, F., Šmajdek, U., Pejić, O., Bohak, C., Gorjanc, V., Munda, T., & Fišer, D. (2026). Thematic landscapes of the past : analysing Slovene historical periodicals with topic modelling. In LREC 2026 : first workshop on creating interoperable corpora of historical newspapers : workshop proceedings (pp. 92-99).
- Šmajdek, U. & Bohak, C. (2025). NERVIS : an interactive system for graph-based exploration and editing of named entities. CoRR, abs/2510.04971, pp. 147-156.