Re: Soundex type search

Path: emory!swrinde!cs.utexas.edu!uunet!mcsun!i2unix!esacs!martin From: martin@esacs.UUCP (Martin Guy) Newsgroups: comp.databases.informix Message-ID: <202@esacs.UUCP> Date: 22 Nov 91 09:20:33 GMT References: <1991Nov20.224415.21672@ddssuprs.uucp@uunet.uu.net> Reply-To: martin@esacs.UUCP (Martin Guy) Organization: Esa Software s.r.l. Rimini (FO), ITALY In article <1991Nov20.224415.21672@ddssuprs.uucp@uunet.uu.net> lpc@ddssuprs.uucp (Luis Caamano) writes: >Does anybody knows of any way to implement sound matching searches >on character fields? That is, to match words that have the same >sound not the spelling? One way that works fairly well for English names is to create a separate column in the table of names, which contains a phonetic key derived from the spelling of the name. You then transform the name you are searching for by the same algorithm, and select matching phonetic keys from the database. To create the phonetic key, you could (for example) ignore all vowels and map groups of similar consonants to the same phonetic value. B and P are similar, for example, as are D, T, and TH. You also need to treat consonant groups intelligently, such as double consonants and absurdities such as GHT. There is a trade-off between the number of distinct vowel groups you define and the closeness of the match that your algorithm will accept. There is a published algorithm called Soundex matching which works something like this, but I don't remember the precise details (sorry). Martin