| 摘要 |
PURPOSE: To reduce the number of times of retrieval of a dictionary and the number of candidate words obtained from the dictionary and to improve the efficiency of character recognition post-processing by sampling a candidate word group by using a 'mixing method'. CONSTITUTION: Assuming that it is, for example, L=2 in the 'mixing method', the number of keys whose length is smaller than two generated by aαcandidate character conducting type dictionary consultationβis reduced to one several tenth compared with the case that it is L=5. In the case of an ordinary Japanese language dictionary, most words are ones whose length is two or less than two in the number of words, and moreover, the number of candidate words (whose length is three or exceeds three in the number of words) obtained byαdictionary conducting type dictionary consultation' is reduced to several on the average by applying constraint with respect to a leading L character. Therefore, the handling time of dictionary retrieval and the sampling of candidate word can be reduced compared with the case that a conventional system is used individually, respectively. For example, the number of times of dictionary retrieval by the 'mixing method'=the number of retrieval target keys 90 whose length is less than two is eight, and the number of candidate words 902 obtained by the 'dictionary conducting type dictionary consultation' is two, which makes only ten.
|
