Explainable Natural Language Inference in the Legal Domain via Text Generation

Natural language inference (NLI) in the legal domain is task of predicting entailment between premise, i.e. law, and hypothesis, which a statement regarding issue. Current state-of-the-art approaches to NLI with pre-trained models do not perform well domain, presumably due discrepancy level abstraction premise hypothesis convoluted nature language. Some difficulties specific are that 1) tend be extensive length; 2) comprises multiple rules, only one rules related hypothesis. Thus small fractions statements relevant for determining entailment, while rest noise, and; 3) often abstract written terms, whereas concrete case tends more ordinary vocabulary. These problems accentuated by scarcity such data high cost.