Legal Question Answering Based on Logical Reasoning in Large Pretrained Language Models
Joseph Dimos, Eva Fourel · Frontiers in artificial intelligence and applications · 2024
Legal Question Answering (LQA) systems are facing a wide range of challenges. Deep learning-based methods need to be improved to better perform on the underlying logical reasoning tasks. We propose a hybrid neurosymbolic framework to achieve this, by enhancing the pretraining of BERT based models with LogicQA and the integration of PyReason.