摘要 |
PROBLEM TO BE SOLVED: To accurately detect an error place of a morphological analysis result for a text based on grammar of a new field for which correct answer data of morphological analysis is not provided, without preparing the correct answer data of the new field.SOLUTION: A tabulation section 22 compiles statistics related to morphemes of an object text and stores the statistics in an object text statistics DB 32. An error candidate extraction section 23 extracts all morphemes n-gram from a morphologically analyzed object text as an error candidate. A feature generation section 24 acquires the statistics being a feature of each of the error candidates from the object text statistics DB 32 to generate a feature vector. An error extraction section 25, using information of a comparison text statistics DB 33 in which statistics are stored related to morphemes of a comparison text of a field which is different from the feature vector and the object text, extracts, as an error place, an error candidate which contains a morpheme whose difference between the object text and the comparison text in statistics related to the morpheme is large and whose appearance frequency is small in the object text statistics DB 32, and which contains a morpheme given prescribed part-of-speech information. |