Automatic Detection of Contradictions in Legal Texts: A Computational Linguistic Approach

Contradictory occurrences are often found in different documents and processes, automating this using intelligent techniques such as natural language model will not only save a lot of time but also help with the process of solving problematic and deceptive issues. In particular, identifying contradictory statements in legal proceedings is largely manual in nature. Laws and their interpretations, legal arguments and agreements are typically expressed in writing, leading to the production of vast corpora of legal text. The focus of the present study is contradictions occurring in legal texts. Consequently, we developed a model that identifies and delineates contradictory statement within written legal texts and documents. Building upon this resource, the entire work develops and implements a supervised classification framework based on transfer learning with BERT, fine-tuned for the sentence-pair task of identifying contradictions. The proposed methodology conceptualizes contradiction detection as a binary classification problem, where the system predicts whether two legal sentences are contradictory, relying on both surface-level lexical cues and deeper argumentative patterns. On the LTCDD, a specialized corpus designed to capture contradiction phenomena specific to legal discourse, results demonstrate that transformer-based models can effectively detect contradictions, with accuracy ranging from 0.8367 to 0.8688.