The Application of Text Entailment Techniques in COLIEE 2020

Juliano Rabelo, Mi-Young Kim, Randy Goebel

Abstract: We develop a method to identify entailment relationships in the texts of case law documents, in the context of two tasks of the Competition on Legal Information Extraction and Entailment (COLIEE 2020). The first task consists in, given a 1) base case, 2) a text fragment from that base case, and 3) the list of paragraphs from a noticed case, identify which of those paragraphs entail the given text fragment. For that problem, we apply a combination of transformer-based and textual similarity techniques. The second task consists in identifying entailment relationships between yes/no questions extracted from the Japanese law bar exam and paragraphs extracted from the Japanese statute legislation. Our approach is to first define a set of classes of textual entailment that cover our analysis of the statute law data, and then implement a pre-processing step to construct the statute law-specific entailment types before exploiting them with deep learning.