PURPOSE: To improve the reading precision of the homograph by automatically judging connection strength between semantic attributes, the combination of reading attributes and appearing frequency as to the homograph and a word in the modification relation with the homograph and deciding the most effective homograph.

CONSTITUTION: A semantic attribute correlation table 6 is used for checking strength in the semantic connection relation between the words. In 'the president Taro', for example, there are two words of 'Taro' and 'president'. 'Taro' is the name of a fuel name, and the semantic attribute is the name. The president is the name of a managerial position and the semantic attribute is the managerial position. When the correlation table 6 is retrieved by the 'name' of the semantic attribute 100 of the former word and the 'managerial position' of the semantic attribute 110 of the latter word, connection strength between the semantic attributes becomes '5', and semantic connection strength is detected to be high. The semantic attribute correlation table 6 is given an the connection strength of the semantic attributes, the combination of the reading attributes and appearing frequency are automatically judged. The most effective homograph is decided and a reading error is prevented and the homograph is correctly recognized.