Answers are grounded in this paper and cite their sources.
Interpreting S-101 Electronic Navigational Charts requires connecting chart objects and encoded attributes with Feature Catalogue (FC) definitions and Data Classification and Encoding Guide (DCEG) guidance. This study evaluates a read-only access layer that supplies this information to large language models. Five access conditions were compared using 65 public S-101 trial and test cells and two researcher-designed 90-question sets. Claude Opus 4.7, GPT-5.5, and Gemini 3.1 Pro each completed three runs under a uniform protocol, producing 8100 responses from predefined evidence inputs. Structured access to chart instances and FC/DCEG information achieved higher feature-identification and lexical catalogue-grounding scores than raw-record or generic retrieval across all three models. The principal access contrasts survived multiple-testing correction. Geometry-based re-scoring lowered nearby-feature scores while retaining the advantage over generic retrieval, highlighting the importance of spatial evidence selection. Adding structured source identifiers enabled source inspection, with no significant improvement in nearby-feature identification. These findings identify structured chart and catalogue access as a useful design basis for S-101 interpretation assistants. Operational benefits require evaluation with chart producers and validators; the proposed assistant complements formal validation.
No takes yet. Share an insight, caveat, or question.
Cho et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: