To shorten an input sentence without using syntax analysis processing even when an input sentence is a long sentence.
This machine translation system 20 includes an erasure knowledge preparation processing part 32 for preparing erasure knowledge 34 related with the elements of a sentence, a long sentence shortening part 60 for shortening an input sentence 36 by using erasure knowledge 34 and a machine translation part 64 for inputting a shortened sentence 62 to be outputted by the long sentence shortening part 60. The erasure knowledge preparation processing part 32 is configured as the learning device of knowledge for a shortened sentence for learning knowledge for sentence shortening from a pair of sentences 30 for learning constituted of a pair of synonymous long and short sentences, and includes a learning data preparing part 50 for preparing learning data 52 by respectively applying predetermined tags to elements configuring the long sentence, which are included in elements configuring the short sentence, and the other elements and an erasure knowledge learning part 54 for capturing erasure knowledge 34 for estimating a tag showing whether or not arbitrary elements can be erased by machine learning using the learning data 52. The long sentence shortening part 60 shortens the input sentence 36 by using erasure knowledge 34.
SHIMOHATA MITSUO
SUMIDA EIICHIRO
Next Patent: BETTING CARD ISSUING DEVICE FOR AUTOMATIC PRINTING