摘要 |
PURPOSE:To provide the method for recognizing an undefined word in a Japanese sentence in which no manual operation is required and an undefined word is recognized with high accuracy without dispersion of recognition. CONSTITUTION:A word string candidate generating processing means 6 generates a candidate of a word string for each paragraph as to a character string in a Japanese sentence. A defect sentence extract processing means 7 extracts a defective paragraph from each word string candidate. A link word extract processing means 8 sets a tentative setting position of a linked word, retrieves a connection part of speech pattern dictionary 4 to extract a link word. A link word representative index extract processing means 9 places priority onto link words based on a weight index of a part of speech and extracts an index of a link word from a representative index dictionary 5 as to the part of speech of the link word. A link word verification processing means 10 generates a Japanese sentence character string in which a link word representative index is set to verify it that the sentence is not a defective paragraph. A link word recognition processing means 11 recognizes a link word candidate with highest priority among link word candidates which are verified not to form a defective paragraph as an undefined word. |