CODAL: Contradiction Detection in the Duties of Actors in Legal Documents

Contradiction detection in textual data is a crucial task for verifying the validity of information stored in textual form. Existing studies in this field can be categorized into three approaches, traditional methods, neural networkbased methods, and bias-resistant methods. Identifying contradictions in legal documents prevents assigning duties to unqualified or multiple actors. However, to the best of our knowledge, contradiction detection in legal documents has not yet been explored. In this paper, we propose CODAL, a method which first leverages existing legal documents to learn and predict actors based on their duties, and second applies the acquired knowledge to detect contradictions in new legal documents. Experimental results demonstrate that CODAL achieves 68 % accuracy in assigning the correct actor to a new duty. effectively identifies contradictions in legal assignments, highlighting cases where duties are mistakenly assigned to unqualified actors. Additionally, the obtained results can also be used to determine the topics allocated to an actor.