Reasoning as the Engine Driving Legal Arguments

Legal decisions often include reasoning sentences, which explain how evidence and legal rules support findings of fact. These sentences serve multiple purposes: providing explanations for rulings, identifying methods of evidence assessment, and facilitating the classification of cases and reasoning patterns. To determine which sentences qualify as reasoning sentences, lawyers consider factors such as relevance to the issue of fact, credibility of evidence, conflict between evidence, and the comparison of probative value. Machine learning (ML) models can be trained to identify reasoning sentences with reasonable accuracy, despite the complexity of the task. However, it is important to consider the nature of ML errors and the cost of misclassification in the context of specific use cases. Practical error analysis shows that the majority of misclassifications involve confusion with other sentence types, such as evidence or finding sentences. The wording of sentences can also make their classification ambiguous, even for legal experts. Despite the challenges, ML models can still provide useful predictions and assist in the extraction and analysis of legal reasoning.