This paper investigates the validity of syntax analysis for recognition of handwritten Katakana characters. A method of string representation which minimizes the length of the string is proposed and a weighted model for error-correcting parsing with substitution and insertion error is introduced. String representation of this paper has a advantage that some consecutive identical primitives in other methods are denoted only one primitive. Related to introducing a weighted error-correcting grammer, the original grammer is infered from only one sample string which is the shortest and appeares frequently in sample strings. From some experimental results for handwritten Katakana characters, the validity of our method is discussed.