A Spelling Checker Based on Affix Classes

Abstract

A technique is described for encoding dictionary entries in terms of a mapping over equivalence classes. The empirically determined classes are based on the property of affixability. The method is demonstrated in a spelling checker although the syntactic and semantic regularities in the classes also have potential relevance to work in natural language processing and descriptive linguistics. In a spelling checker, the primary advantage of the method is an increased accuracy and reduced dictionary size as compared to more traditional “heuristic” checkers that do incomplete affix analysis and use larger dictionaries (e. g., the UNIX** spell program). Various factors in reducing the search for a legal bound stem are discussed, and an affix stripping procedure is presented. The method is compared with other spelling checkers.

UNIX is a trademark of AT&T Bell Laboratories