公开(公告)号：JPH08147299A

公开(公告)日：1996-06-07

申请号：JP28308794

申请日：1994-11-17

Applicant: IBM JAPAN

Inventor： NASUKAWA TETSUYA

IPC: G06F17/21 ,  G06F17/27 ,  G06F17/28

Abstract: PURPOSE: To enable syntax analysis with a certain degree of accuracy even for any sentence by reanalyzing a sentence, which can not be analyzed, by utilizing the word train of a sentence, for which the syntax analysis is enabled, in the same context in syntax analystic processing depending on grammatical knowledge. CONSTITUTION: A morpheme analysis block 104 divides a sentence applied from an input block 102 and analyzes the part of speech or inflection of each word while referring to a dictionary and a syntax analysis block 106 performs processing for providing information to consist of tree structure based on the output information from the mo pheme analysis block 104. Further, a context analysis block 108 holds the context information of plural entier inputted snetences and performs processing to provide the exact analyzed result as much as possible by applying context information to the sentence showing the unsuitable analyzed result outputted from the syntax analysis block 106, namely, to the sentence (grammatically unsuitable sentence) not to be analyzed by the conventional syntax analysis depending on the grammatical knowledge. Thus, the accuracy of the natural language processing system is improved and the syntax analysis is enabled with a certain degree of accuracy.

公开(公告)号：JPH11184855A

公开(公告)日：1999-07-09

申请号：JP35438697

申请日：1997-12-24

Applicant: IBM JAPAN

Inventor： NASUKAWA TETSUYA ,  TAKEDA KOICHI ,  WATANABE HIDEO

IPC: G06F17/27 ,  G06F17/28

Abstract: PROBLEM TO BE SOLVED: To improve the selection accuracy of translated words in a machine translation mode without lowering the processing efficiency by using plural types of dictionaries including a context dictionary when a word that is not defined in a compound word dictionary is translated in a sentence. SOLUTION: Every input sentence is taken out at its head (110), and the compound words corresponding to a word string composing a single input sentence are retrieved from a compound word dictionary. When each of wards which are not corresponding to the compound words is translated (120), its translation is decided by a context dictionary and the translated word is obtained based on the translation (130). The translated word undergoes the translation result registering processing. Then it's checked whether an object word is stored in a translation result recording buffer as a header. If the object word is stored in the buffer, it's checked whether or not its translated word is stored in the buffer and all words which are not corresponding to the compound words are processed (140, 160). Then all words are translated (170). When it's decided that a full sentence is translated, the sentence is registered (180 to 195) after undergoing the retranslation effect evaluation processing.