PoliToHFI at SemEval-2023 Task 6: Leveraging Entity-Aware and Hierarchical Transformers For Legal Entity Recognition and Court Judgment Prediction

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1401–1411 July 13-14, 2023 ©2023 Association for Computational Linguistics PoliToHFI at SemEval-2023 Task 6: Leveraging Entity-Aware and Hierarchical Transformers For Legal Entity Recognition and Court Judgment Prediction Irene Benedetto1,2 Alkis Koudounas1 Lorenzo Vaiani1 Eliana Pastor1 Elena Baralis1 Luca Cagliero1 1 Politecnico di Torino, {name.surname}@polito.it 2 H-Farm Innovation, {name.surname}@h-farm.com Francesco Tarasconi2 Abstract The use of Natural Language Processing tech- niques in the legal domain has become estab- lished for supporting attorneys and domain ex- perts in content retrieval and decision-making. However, understanding the legal text poses rel- evant challenges in the recognition of domain- specific entities and the adaptation and explana- tion of predictive models. This paper addresses the Legal Entity Name Recognition (L-NER) and Court judgment Prediction (CPJ) and Ex- planation (CJPE) tasks. The L-NER solution explores the use of various transformer-based models, including an entity-aware method at- tending domain-specific entities. The CJPE proposed method relies on hierarchical BERT- based classifiers combined with local input at- tribution explainers. We propose a broad com- parison of eXplainable AI methodologies along with a novel approach based on NER. For the L- NER task, the experimental results remark on the importance of domain-specific pre-training. For CJP our lightweight solution shows perfor- mance in line with existing approaches, and our NER-boosted explanations show promising CJPE results in terms of the conciseness of the prediction explanations. 1 Introduction AI-powered tools for legal AI can analyze vast amounts of documents providing lawyers and judges with relevant insights on case laws (San- sone and Sperlí, 2022). For example, based on the analysis of past court cases legal AI models can predict the outcome of similar cases, thus offer- ing efficient and effective ways of resolving dis- putes (Medvedeva et al., 2022). The legal domain poses relevant challenges due to the specificity of the legal language, the variety of scenarios and application contexts, and the tem- poral evolution of norms and regulations (Chalkidis and Søgaard, 2022; Benedetto et al., 2022). This paper addresses two notable legal AI tasks, i.e., Legal Named Entity Recognition and Court Judgment Prediction with Explanation. Legal Named Entity Recognition The L-NER task aims at annotating portions of legal content with domain-specific entities. Annotations are particularly helpful to support content retrieval and indexing. For example, the identification of relevant legal provisions can help lawyers and judges make informed decisions based on previ- ous cases. Despite NER being a well-known NLP task, the application of existing techniques to le- gal data is challenged by the inherent complex- ity and nuances of the language used in the legal domain (Williams, 2005; Tiersma, 2000). We ad- dress L-NER by exploring the use of established transformer-based models, established for many NLP tasks (Vaswani et al., 2017) and tailored to the legal domain (Chalkidis et al., 2020). Unlike previous studies, we also explore the use of the entity-aware attention mechanism (Yamada et al., 2020), which allows the transformer to also attend to domain-specific entities. The preliminary results achieved …

Content truncated to 500 words