Agent-Specific Deontic Modality Detection in Legal Language

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11563 - 11579 December 7-11, 2022 ©2022 Association for Computational Linguistics Agent-Specific Deontic Modality Detection in Legal Language Abhilasha Sancheti†‡, Aparna Garimella‡, Balaji Vasan Srinivasan‡, Rachel Rudinger† †University of Maryland, College Park ‡Adobe Research {sancheti, rudinger}@umd.edu {garimell, balsrini}@adobe.com Abstract Legal documents are typically long and written in legalese, which makes it particularly diffi- cult for laypeople to understand their rights and duties. While natural language understand- ing technologies can be valuable in supporting such understanding in the legal domain, the limited availability of datasets annotated for de- ontic modalities in the legal domain, due to the cost of hiring experts and privacy issues, is a bottleneck. To this end, we introduce, LEXDE- MOD, a corpus of English contracts annotated with deontic modality expressed with respect to a contracting party or agent along with the modal triggers. We benchmark this dataset on two tasks: (i) agent-specific multi-label deontic modality classification, and (ii) agent-specific deontic modality and trigger span detection us- ing Transformer-based (Vaswani et al., 2017) language models. Transfer learning exper- iments show that the linguistic diversity of modal expressions in LEXDEMOD general- izes reasonably from lease to employment and rental agreements. A small case study indicates that a model trained on LEXDEMOD can de- tect red flags with high recall. We believe our work offers a new research direction for deontic modality detection in the legal domain1. 1 Introduction A contract is a legal document executed by two or more parties. To sign a contract (e.g., lease agree- ments, terms of services, privacy policies, EULA, etc.), it is important for these parties to precisely understand their obligations, entitlements, prohi- bitions, and permissions as described in the con- tract. However, for a layperson, understanding contracts can be difficult due to their length and the complexity of legalese used. Therefore, a layper- son often signs agreements without even reading them (Cole, 2015; Obar and Oeldorf-Hirsch, 2020). 1The code and data are available at https://github. com/abhilashasancheti/LexDeMod Figure 1: Sample contract3 indicating the terminologies used to refer to the elements of a contract. ‘shall’ trig- gers obligation for Lessee and entitlement for Lessor. Contracting party or agent is referred to via an “alias” (such as Lessor or Lessee) throughout the contract. Having a system which can provide an “at a glance” summary of obligations, entitlements, prohibitions, and permissions to a contracting party (henceforth, “agent”2), will be of great help not only to the agents but also to legal professionals for contract review. While existing language processing and understand- ing systems can be used for legal understanding, limited availability of annotated datasets in the le- gal domain due to the cost of hiring experts and privacy issues is a bottleneck. Furthermore, the highly specialized lexical and syntactic features of legalese make it difficult to directly apply systems trained on data from other linguistic domains (e.g., news) to the legal domain. For an “at a glance” summary of contracts, we first need to …

Content truncated to 500 words