Automatisk summering av lagligt bindande kontrakt med hjälp av LLM

The precise and technical language used in legal contexts is essential to ensure that the law is interpreted and enforced appropriately, but it also makes the language of the law inaccessible. Violating the terms of a contract, even unintentionally, can have severe consequences. But retaining legal counsel is costly, and reading through a contract is timeconsuming regardless of legal knowledge. This study explores the effectiveness of using Large Language Models (LLMs) to generate summaries of legally binding agreements (LBA) and whether it can address these concerns. By evaluating the quality of summaries of terms of service (TOS) agreements generated by LLMs using ROUGE and BERTScore metrics with gold standard reference summaries, it was determined that the summaries had a high word similarity and a notable semantic likeness. Furthermore, qualitative analysis revealed that errors in summaries generated by LLMs were relatively rare, indicating that LLM can generate accurate summaries of LBAs, and that this process could be a significant improvement and provide value for organizations and individuals.