Coupling Local Context and Global Semantic Prototypes via a Hierarchical Architecture for Rhetorical Roles Labeling

ACL ARR 2025 July Submission1114 Authors

Abstract: Rhetorical Role Labeling (RRL) aims to identify the functional role of each sentence within a document, a task critical for discourse understanding in domains such as law, medicine, and science. While hierarchical models capture local, intra-document dependencies effectively, they struggle to model global, corpus-level regularities. To bridge this gap, we propose two methods that couple local context with global representations in the form of semantic prototypes. Prototype-Based Regularization (PBR) learns soft prototypes through a distance-based auxiliary loss to structure the latent space. Prototype-Conditioned Modulation (PCM) constructs a priori prototypes from the corpus and injects them during both training and inference. We also introduce SCOTUS-Law, the first dataset of U.S. Supreme Court opinions annotated with rhetorical roles at three levels of granularity: category, rhetorical function, and step. Experiments across legal, medical, and scientific benchmarks demonstrate that modeling both local and global perspectives leads to consistent gains over strong baselines, particularly on low-frequency roles, achieving an average gain of $\sim$4 points in Macro-F1.

Paper Type: Long

Research Area: Resources and Evaluation

Research Area Keywords: legal NLP, corpus creation, NLP in resource-constrained settings,

Contribution Types: Approaches to low-resource settings, Data resources

Languages Studied: english

Reassignment Request Area Chair: This is not a resubmission

Reassignment Request Reviewers: This is not a resubmission

A1 Limitations Section: This paper has a limitations section.

A2 Potential Risks: N/A

A2 Elaboration: 9

B Use Or Create Scientific Artifacts: Yes

B1 Cite Creators Of Artifacts: Yes

B1 Elaboration: 5

B2 Discuss The License For Artifacts: Yes

B2 Elaboration: 1

B3 Artifact Use Consistent With Intended Use: Yes

B3 Elaboration: 5

B4 Data Contains Personally Identifying Info Or Offensive Content: No

B4 Elaboration: The data used in this study was collected from CourtListener, an open-access resource. No anonymization was required.

B5 Documentation Of Artifacts: Yes

B5 Elaboration: Appendix C

B6 Statistics For Data: Yes

B6 Elaboration: Appendix C

C Computational Experiments: Yes

C1 Model Size And Budget: Yes

C1 Elaboration: 6.6

C2 Experimental Setup And Hyperparameters: Yes

C2 Elaboration: 5

C3 Descriptive Statistics: Yes

C3 Elaboration: 4

C4 Parameters For Packages: Yes

C4 Elaboration: Appendix B

D Human Subjects Including Annotators: Yes

D1 Instructions Given To Participants: Yes

D1 Elaboration: 4

D2 Recruitment And Payment: N/A

D3 Data Consent: Yes

D3 Elaboration: 4

D4 Ethics Review Board Approval: N/A

D5 Characteristics Of Annotators: Yes

D5 Elaboration: 4

E Ai Assistants In Research Or Writing: Yes

E1 Information About Use Of Ai Assistants: N/A

Author Submission Checklist: yes

Submission Number: 1114