PURPOSE: To perform high-precision word recognition efficiently in a short time by storing candidated characters, indicated by candidated categories, and dictionary words in register and performing collation, and then finding the degree of coincidence.

CONSTITUTION: An input word supplied in the form of a string of (n) characters is supplied to a recognition part 11 to find (r) candidated categories. Those candidated categories are stored in the candidated character register 21 of a word collation part 13 through an editing part 12. Words read out of a word dictionary 14 are stored in a word dictionary register 23. Respective characters stored in the register 23 are supplied to a comparator 24, which compares in coincidence them with the candidated categories, read out of the register 21 successively, corresponding to word character positions. Information on coincidence detection is supplied to a coincidence degree calculation part 26 to calculate the degree of coincidence. Then, (x) words having higher degrees of coincidence are supplied to the editing part 12 as expected words.