Corpus-based Name Standardization

Variation in the spelling of names has various origins, many of which many are difficult to describe by rule. We present a method that uses both rules and a similarity measure of a probabilistic nature, and which can make use of existing onomastic corpora. Rules first convert an unknown name to a semiphonemic form. Then a selection is made of possible candidates in the onomastic corpus. For this set, the similarity to the unknown name is computed and a decision procedure chooses the best candidate. If no specific onomastic corpus is available, the method provides a tool for a clustering of similar names. The method is demonstrated on a corpus of 49.193 first names from 18th century parish registers, in the availability of a Dutch corpus with 22.579 variants of 4.482 base forms of first names.