公开(公告)号：JP2006350605A

公开(公告)日：2006-12-28

申请号：JP2005174809

申请日：2005-06-15

Applicant: ATR ADVANCED TELECOMM RES INST

Inventor： ANDREW FINCH ,  HWANG YOUNGSOOK ,  SUMIDA EIICHIRO

IPC: G06F17/28 ,  G06F17/30

Abstract: PROBLEM TO BE SOLVED: To provide an apparatus capable of evaluating semantic equivalence of two sentences highly reliably. SOLUTION: This apparatus includes: a POS tagger and stemmer part 96 for tagging each of words in the two sentences with an appropriate POS tag and for outputting a first and second bags of words 98 and 100; and a PER calculation module 102 for calculating an extended PER score for the first bag of words 98 with reference to the second bags of words 100. The extended PER score includes a first group of elements calculated from non-matching words in the first and second bags of words 98 and 100 and a second group of elements calculated from matching words in the first and second bags of words 98 and 100. COPYRIGHT: (C)2007,JPO&INPIT

公开(公告)号：JP2003263433A

公开(公告)日：2003-09-19

申请号：JP2002062624

申请日：2002-03-07

Applicant: ATR ADVANCED TELECOMM RES INST

Inventor： WATANABE TARO ,  IMAMURA KENJI ,  SUMIDA EIICHIRO

IPC: G06F17/28

Abstract: PROBLEM TO BE SOLVED: To provide a method of generating a translation model in a statistical machine translator improving translation performance. SOLUTION: This method of generating the translation model in the statistical machine translator comprises a first step of extracting hierarchical phrase alignment from two-language corpus, a second step of converting the word string of the two-language corpus into a chunk string by dividing the two- language corpus into chunks which are meaningful groups, and a third step of training the translation model with the chunk as a unit on the basis of the two-language corpus converted in the second step. COPYRIGHT: (C)2003,JPO