Automatic summarization of legal documentsMiščič, Andrej (Avtor) Žitnik, Slavko (Mentor) automatic text summarizationextractive summarizationabstractive summarizationlegal documentsnatural language processingThe adoption of modern natural language processing is crucial for the legal industry to process large amounts of text data and provide efficient services. Legal research is one the most impacted areas, allowing legal practitioners to find relevant legislation and case law faster. Intending to provide summaries of long legal documents, we tackle the task of automatic summarization of Slovene judicial decisions. We propose GloBerta-Sum, an extractive approach based on recently introduced Slovene pretrained language models. It exploits the structure of judicial decisions to deal with their length and is trained on proposed soft labels to mitigate the effect of a high sentence compression ratio. We additionally combine GloBerta-Sum with an abstractive model to form a hybrid system capable of producing summaries in a paraphrasing manner. We evaluate our approaches using automatic metrics and human evaluation. Results show that our approaches match the relevance of human written summaries, albeit producing a bit less coherent summaries containing more redundant information. Nevertheless, we believe our work highlights the potential of using the proposed methodology to equip legal documents with summaries that allow legal practitioners to quickly assess their relevance.20222022-11-14 07:40:01Magistrsko delo/naloga142575VisID: 34301COBISS_ID: 130546947sl