Cross-lingual performance of the framework

Ascertain the performance of the structured decomposition framework—combining LLM-driven ontology population with SWRL-based reasoning—on non-English languages across rule-governed tasks, accounting for potential differences in language model capabilities.

Background

All experiments in the paper are conducted on English-language datasets spanning legal hearsay determination, scientific method-task application, and clinical trial eligibility. The authors emphasize that they did not evaluate multilingual settings.

Given that LLM capabilities can vary by language, the generalizability of the framework’s improvements and verification benefits to non-English inputs remains explicitly stated as unknown.