Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates data selection and augmentation techniques to improve cross-domain transfer in regulatory compliance detection, framed as an NLI task. They compare random sampling, cross-entropy difference, importance weighting, and embedding-based retrieval for selecting relevant data from a source domain to augment a target domain. Results show that targeted data selection significantly reduces negative transfer, enhancing the generalizability of compliance detection models across different regulations.
Stop training compliance models on everything: carefully selected data from other regulations dramatically boosts generalization.
Automating the detection of regulatory compliance remains a challenging task due to the complexity and variability of legal texts. Models trained on one regulation often fail to generalise to others. This limitation underscores the need for principled methods to improve cross-domain transfer. We study data selection as a strategy to mitigate negative transfer in compliance detection framed as a natural language inference (NLI) task. Specifically, we evaluate four approaches for selecting augmentation data from a larger source domain: random sampling, Moore-Lewis's cross-entropy difference, importance weighting, and embedding-based retrieval. We systematically vary the proportion of selected data to analyse its effect on crossdomain adaptation. Our findings demonstrate that targeted data selection substantially reduces negative transfer, offering a practical path toward scalable and reliable compliance automation across heterogeneous regulations.