Waystone · Research Preview

Dartmoor Heritage Gazetteer

21,894 place names extracted from 1890s Ordnance Survey six-inch maps using machine learning

This gazetteer was produced by applying the MapReader ML pipeline to NLS-hosted six-inch OS maps of Dartmoor and its surrounds (sheets surveyed 1884–1904, published 1888–1913). Every extracted name has been cross-referenced against four independent datasets to assess its persistence, evolution, and heritage significance.

Key findings

21,894
Place names extracted
5,337
Lost from modern records
5,174
Names changed since 1890s
65%
Validated against GB1900
3,794
Heritage record matches
56%
Etymology resolved

Linguistic landscape

Element-based etymology analysis reveals the layered linguistic history of Dartmoor’s place names. Old English dominates, but Celtic/Brythonic elements persist—particularly on western Dartmoor near the Cornish border—with Norman French concentrated in the manorial lowlands.

Old English (18,298) Norman French (2,109) Celtic/Brythonic (2,050) Latin (241) Old Norse (148) Celtic/Goidelic (168)

Cross-referencing methodology

Interactive viewer

The viewer displays all 21,894 features on the original NLS six-inch tiles. A research sidebar provides faceted filtering by heritage status, GB1900 validation, modern name status, etymology language, and confidence score. Results can be exported as CSV for external analysis.

Launch Viewer Download Data (CSV + GeoJSON)

Viewer password available on request. Data package includes the public gazetteer CSV (34,742 records, 22 fields) and GeoJSON for direct use in GIS tools. 3.5 MB zipped.

Known limitations

This is an ML-extracted dataset, not a manually verified transcription. The confidence score (0.40–1.00) reflects the model’s certainty for each extraction. Approximately 35% of entries lack independent GB1900 confirmation. OCR errors exist, particularly for degraded or ornate typefaces. Cross-reference matches are probabilistic: heritage and modern name matches use spatial proximity combined with text similarity, not manual verification. The “Export filtered CSV” button in the viewer allows researchers to apply their own quality thresholds before analysis.