Type to search. Arrow keys to move, Enter to open, Esc to close.

All research
SearchState as of: 2026-10

Accent-insensitive search for a bilingual static site

How a serverless search finds Hungarian words typed without accents, with typos and in inflected forms, and what that costs in index size.

Research question: How can a serverless search find Hungarian words typed without accents, with typos and in inflected forms?

Project requirement

A bilingual static portfolio with search and no external service. Hungarian visitors often type without accents, and Hungarian words appear with suffixes in the text, for example "modellekkel".

Method and validation

  • One index per language generated at build time, searched in the browser with MiniSearch.
  • The same normalization at indexing and query time: Unicode NFD decomposition, removal of diacritic marks, lower case.
  • Prefix matching with a fuzzy tolerance of 0.2, every query word required, title and tags weighted higher.
  • Checked on the built Hungarian index (2026-10-10, 43 entries): "feltöltés" and "feltoltes" return the same 5 results; "nyelvi modell" and the misspelled "nyelvi modelek" both return 4, led by the small language model research; the fragment "kepoptim" also matches.
  • Measured on 2026-10-10, with 43 index entries; index size: English 34.7 KB (10.8 KB gzipped), Hungarian 38.8 KB (12.1 KB gzipped).

Findings

  • Removing diacritics on both sides gives accent-insensitive search without a Hungarian dictionary.
  • Prefix matching covers most suffixes, because Hungarian suffixes follow the word stem ("modell" matches "modellekkel").
  • The fuzzy tolerance handles common typos such as a missing double consonant.
  • At this content size, loading the whole index on first focus is cheap; a split index is not needed.

Decision

One index per language, loaded lazily, with NFD normalization, prefix matching, fuzzy 0.2 and all words required. No stemming.

Limits and open questions

  • Words whose stem changes, for example "tó" and "tavak", are not matched by prefix; that would need a Hungarian stemmer.
  • Removing accents can merge different words, for example "kor" and "kór"; this was not examined separately.
  • The index grows with the content; much larger content would need a split index or another tool.

Sources