To improve character recognition precision by introducing morpheme analysis that can be applied to an agglutinative language such as Japanese and error-checking an analysis result.

A character recognition processing (maximum likelihood solution is selected based on recognition dictionary collation score and linguistic likelihood) is performed on a plurality of character candidates from an objective character string. Morpheme analysis is performed as a post-processing in a stage where a recognition result is almost decided. It is checked whether a morpheme is incorrectly connected or not with the result as an object. Incorrect connection is checked by detecting disturbance occurred in an arrangement of a character pitch in the morpheme (refer to figure 7) and deciding it by a connection rule based on a grammar attribute such as a part of speech and a character type of a morpheme surface. In a check result, the morpheme which is incorrectly detected is displayed so that it can be identified or evaluation of a recognition score is dropped and the post-processing is performed again.

COPYRIGHT: (C)2004,JPO