Regulating LLM training under EU copyright law and its linguistic implications for smaller languages such as Estonian

Abstract

The rapid advancement of artificial intelligence (AI) and text and data mining (TDM) technologies has become a crucial factor in supporting linguistic diversity of smaller languages, such as Estonian. However, the growing reliance on large-scale data processing raises significant legal and ethical challenges, particularly concerning the lawful use of copyrighted materials to train AI systems. While the European Union has introduced regulatory instruments such as the AI Act and the Copyright Directive to address these concerns, the intersection of AI development, TDM, and intellectual property rights remains insufficiently resolved. This research aims to explore how AI-driven text and data mining can both comply with European copyright law and actively contribute to the preservation and wider use of the Estonian language. The authors investigate this research topic from the perspective of the Estonian language, enmeshed within Estonian cultural norms and parameters. The analysis examines the legal and economic appropriateness of enabling the training of generative AI-based large language models (LLMs) in the Estonian language as a means of promoting the language and assuring its appropriate use in digital environments, thereby contributing to its long-term relevance. Methodologically, the study combines a comparative analysis of EU, US, and Asian legal frameworks with a theoretical evaluation of existing models for balancing innovation and rights protection. The article seeks to provide concrete recommendations for developing AI and data mining practices that enhance linguistic innovation while safeguarding creators’ rights, thereby contributing to Estonia’s digital and cultural resilience within the EU’s evolving AI ecosystem.