Contributions to automatic legal document summarization: Judgements from the Portuguese Supreme Court

• Autor(a) principal: Dias, Margarida Rebelo
• Data de Publicação: 2024
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Texto Completo: http://hdl.handle.net/10071/33681
• topic: Automatic text summarization; Legal document summarization; Extractive summarization; Abstractive summarization; European Portuguese; Sumarização de texto automática; Sumarização de documentos jurídicos; Sumarização extrativa; Sumarização abstrativa; Português Europeu

As information continues to grow in an exponential way, overtaking humans capacity to reach all of it, it is crucial to develop strategies to minimize the time spent on reading and comprehending information. In the legal field, the process of summarization has been used for this purpose, however, it is still done manually by legal experts. This dissertation focuses on testing different summarization models in order to understand their efficacy in automating the summarization process, specifically for Portuguese legal documents from the Portuguese Supreme Court of Justice. Automatic summarization models have been developed in a variety of areas. Conversely, the legal field brings some constraints because of the length of the documents and the particular vocabulary used in them. We implemented three different models: a sentence-level model, a summary-level model, and a hybrid approach to evaluate the generation of summaries using both extractive and abstractive summarization methods. For each experiment, we used two different input texts: the original documents and specific sections from the original documents. For the evaluation process, we use the ROUGE and BERTscore metrics, where we compare the generated summaries with the reference summaries available for each document. The analysis of the results made us conclude that the extractive models are effective at reducing document length, particularly with the summary-level approach, and that abstractive techniques can improve summary fluency. Furthermore, it was confirmed that the use of a summary-level approach has a significant effect on the summarization of Portuguese legal documents.