Stanisław Giziński, University of Warsaw, MI²DataLab, and Bartosz Pieliński, Asst. Prof., Faculty of Political Science and International Studies, University of Warsaw, MI²DataLab
This presentation focuses on the innovative application of Large Language Models (LLMs) to Institutional Grammar (IG) coding. It will provide a brief overview of LLMs, highlighting their potential as a solution for IG coding challenges. This presentation will argue for integrating LLMs in this domain by showing the current landscape of IG research, marked by diverse implementation strategies and a lack of standardized IG-code legal text corpus. Specifically, the utilization of the few-shot approach will be explored. In this method, LLMs are conditioned with a minimal set of IG-coded examples sufficient for the automated coding of legal documents. This approach will be evaluated using a selection of international conventions related to cultural heritage, climate change, and wildlife protection.
