TypedCSIP: Typed Counterfactual Pretraining for Chinese Legislative Conflict Classification
2026-05-25 • Computation and Language
Computation and Language
AI summaryⓘ
The authors present TypedCSIP, a method to better identify conflicts in legal provision pairs and classify the type of legal inconsistency. They use expert-written edits as examples of how the provisions could change to avoid conflict during training but only look at the original pairs when testing. Their approach improves classification accuracy modestly compared to previous models and works well even on new, unseen data. They also show their method focuses on conflict classification and doesn't help for related legal retrieval tasks. The authors provide their code and experiment records for others to review.
conflict classificationcounterfactual pretraininglegal doctrineLCR-CN benchmarkmacro-F1 scorecross-backbone replicationpretrained encoderlegal provision pairslegal inconsistency typesexpert revision
Authors
Yao Liu
Abstract
TypedCSIP is a typed counterfactual pretraining method for the conflict-classification task of the LCR-CN benchmark (Zhao et al., 2026): given a (superior, subordinate) provision pair, predict whether the pair conflicts and which of four legal-doctrine types (Responsibility, Condition, Sanction, Definition) describes the inconsistency. We exploit LCR-CN's expert-written minimal revisions as training-time counterfactual supervision; at test time the classifier reads only the original pair. Stage 1 pretrains a shared encoder with a typed Counterfactual Selective Intervention Pretraining objective on (superior, subordinate, expert-revised) triplets, treating the expert revision as a counterfactual that the typed factor head must classify as carrying no conflict evidence. Stage 2 transfers the encoder to a five-way classification head. The confirmatory test was registered on the Open Science Framework before observing v6 measurements: 18 seeds, locked rule requiring mean per-seed difference at least 0.8 pp with both seed-bootstrap and Student-t 95% lower bounds above zero. On the 696-record test split, the v2 variant improves macro-F1 over the strongest single-model baseline by +0.916 pp on chinese-roberta-wwm-ext and +1.288 pp on the SAILER cross-backbone replication; both cells pass the rule. A cold-start stratified result on the 244 Unseen-gB records keeps the gain positive on both backbones. A cross-task diagnostic shows the Stage-2 encoder is classification-specialized and does not transfer to LCR-CN's superior-law retrieval task, so we scope the contribution to conflict classification. We release code, 72 pre-registered prediction files, matched-seed and MLM-control auxiliaries, and the OSF pre-registration record.