Diagnosing the limits of large language models for structured scientific evidence extraction
A schema-grounded, tiered evaluation framework for structured scientific evidence extraction that increases task difficulty from single-entity identification to relational tuples and corpus-level statistical tasks and provides a reusable query suite and diagnostic evaluation framework to support future fine-grained ben...