Party Extraction from Legal Contract Using Contextualized Span Representations of Parties

Proceedings of Recent Advances in Natural Language Processing, pages 1085–1094 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_116 1085 Party Extraction from Legal Contract Using Contextualized Span Representations of Parties Sanjeepan Sivapiran, Charangan Vasantharajan, Uthayasanker Thayasivam Department of Computer Science and Engineering, University of Moratuwa Colombo, Sri Lanka {sanjeepan.18, charangan.18, rtuthaya}@cse.mrt.ac.lk Abstract Extracting legal entities from legal documents, particularly legal parties in contract documents, poses a significant challenge for legal assis- tive software. Many existing party extraction systems tend to generate numerous false pos- itives due to the complex structure of the le- gal text. In this study, we present a novel and accurate method for extracting parties from le- gal contract documents by leveraging contex- tual span representations. To facilitate our ap- proach, we have curated a large-scale dataset comprising 1000 contract documents with party annotations. Our method incorporates sev- eral enhancements to the SQuAD 2.0 question- answering system, specifically tailored to han- dle the intricate nature of the legal text. These enhancements include modifications to the ac- tivation function, an increased number of en- coder layers, and the addition of normalization and dropout layers stacked on top of the output encoder layer. Baseline experiments reveal that our model, fine-tuned on our dataset, outper- forms the current state-of-the-art model. Fur- thermore, we explore various combinations of the aforementioned techniques to further en- hance the accuracy of our method. By employ- ing a hybrid approach that combines 24 encoder layers with normalization and dropout layers, we achieve the best results, exhibiting an exact match score of 0.942 (+6.2% improvement). 1 Introduction Extracting legal entities from legal documents is an essential challenge for legal assistive software (Leivaditi et al., 2020). Its goal is to extract struc- tured information — “what are the legal attributes (agreement date, party, license, etc) that are in- volved” — from unstructured text. A contract doc- ument is a legally binding agreement between two or more parties. It outlines the terms and condi- tions of the relationship and sets forth the rights and obligations of each party (Chalkidis et al., 2017). Here, the party is a person or entity who takes part in a legal transaction, for example, a person with an immediate interest in an agreement or deed, or a plaintiff or a defendant in a lawsuit. A “third party” is a person who is a stranger to a transaction, contract, or proceeding. Extracting parties’ information from legal con- tracts can provide numerous benefits to legal assis- tant software such as Concord1, ContractWorks2, and HelloSign3. First and foremost, this can aid legal professionals in identifying parties involved in similar cases, trends, and patterns in legal dis- putes, and assist in more efficient legal research. Secondly, the extraction of parties allows reviewing multiple documents efficiently with the quick doc- ument search feature, and it leads to focusing on relevant information and documents through assis- tant software. Moreover, the manual extraction of parties is prone to human errors, which can lead to inaccuracies and inconsistencies (Hendrycks et al., 2021). By automating …

Content truncated to 500 words