Mok, Wai Yin; Mok, Jonathan R.; Mok, Rachel V.
Abstract: We identify seven sentence types for contract law cases and report a sentence classification algorithm that assigns one of the seven types to every sentence of a contract case. Our algorithm relies on spaCy, an industrial-strength Natural Language Processing (NLP) library, to parse and process contract cases. A critical component of our algorithm is a knowledge base that contains various legal abbreviations and corrections that assist tokenization and sentence segmentation. The knowledge base also contains generic sentence fragments that assist classification of sentences other than reference sentences, which can be determined by rules because of their simple sentence structures. To calculate the similarity of a sentence and the sentence fragments in the knowledge base, our algorithm depends on the spaCy’s similarity function, which in turn is based on word vectors, a recent advancement of NLP. Our algorithm yields a classification accuracy rate of 68.67%, demonstrating the validity of this approach.