LawShift benchmarks legal judgment prediction under statutory revisions. Covering 31 fine-grained change types, it uses metamorphic testing to examine whether models adapt their judgments to changes in the law.
Publication
Accepted to the NeurIPS 2025 Datasets and Benchmarks Track as a poster and published in the conference proceedings.
Overview
Legal judgment prediction models need to account for statutory revisions rather than rely on a static view of the law. LawShift introduces a Chinese legal benchmark covering 31 fine-grained revision types. Its metamorphic testing framework examines expected relationships between predictions before and after a statute changes.
Evaluation
The paper evaluates five representative legal judgment prediction models, revealing limitations in their response to legal updates and differences in adaptability across model architectures.