Phonetic matching algorithms

Phonetic matching finds words that sound alike, even if spelled differently. It's used in name deduplication, spell checking, genealogy research, and fuzzy search.

Soundex (1918): oldest algorithm; encodes each name as a letter + 3 digits based on consonant sounds. Still used in US census and genealogy databases. "Smith" = S530, "Smythe" = S530 (same!).
Metaphone: more sophisticated than Soundex; better handles complex English phonics. Handles leading silent letters ("knight" -> NT).
Double Metaphone: generates two codes per name - one for English pronunciation, one for foreign language pronunciation of the same spelling.

When Soundex falls short

Soundex was designed for English names of European origin. It has known failure modes:

Short names: names with fewer than 4 coded characters are padded with zeros, causing false matches. "Lee" and "Li" both encode to L000.
Non-English names: Asian, African, and Arabic names often have phonetic patterns Soundex was not designed for, resulting in poor recall.
Adjacent identical codes: "Jackson" and "Jakson" may not match if the coding rules collapse nearby consonant pairs differently.

For modern applications, consider using edit-distance algorithms like Levenshtein distance or Jaro-Winkler alongside phonetic matching for better recall.

Applications in genealogy and data quality

Phonetic matching is heavily used in:

Genealogy research: historical records (census, birth, immigration) often have names spelled inconsistently by enumerators who recorded names by ear.
Healthcare records: patient name matching across hospital systems, especially for patients who spell their name differently on different visits.
Voter registration: detecting duplicate registrations where names are slightly misspelled.
E-commerce and search: "did you mean?" suggestions and product name disambiguation.