Evaluate robustness beyond Japanese news and small controlled collections

Evaluate the robustness of jaROTE on much larger collections containing thousands of documents and in domains outside Japanese news.

Background

The experiments assess jaROTE primarily on two Japanese news corpora and a limited retrieval dataset constructed from 97 Nikkei articles. The evaluation therefore does not establish whether the rule-based pipeline remains reliable at substantially larger scale or in domains whose temporal-expression patterns differ from Japanese news. Robustness under these broader conditions is explicitly left unresolved.