Skip to main content

OCR and fuzzy place-name searching

OCR can turn a real place name into nonsense, especially in old type, damaged scans and te reo Māori material. Search should therefore generate candidates rather than rely on exact spelling alone.

Build variants systematically​

For a target name test:

  • exact current form
  • no macrons
  • common historical spelling
  • first distinctive word/root
  • likely OCR substitutions such as rn/m, I/l/1, c/e, broken hyphens

Papers Past supports exact phrase, all/any words, Boolean operators and filters. Its search help explicitly recommends historical variants/wildcards.

Fuzzy match tables​

For a large extracted list, normalise a comparison field (lowercase, stripped punctuation and optional macron-insensitive form) while keeping name_as_source unchanged.

Use string similarity only to rank Gazetteer candidates. A high fuzzy score does not establish geographic identity when many places share related names.

Verify on the page image​

OCR is the search index. The original scan is the evidence. Check the actual glyphs before publishing a spelling.

For local document collections see Local AI and Find ingoa wāhi.

Last reviewed: 26 August 2026

Last updated on