Natural language processing method for converting a first natural language into a second natural language using data structures

• US 6,126,306 A
• Filed: 09/10/1992
• Issued: 10/03/2000
• Est. Priority Date: 09/11/1991
• Status: Expired due to Fees

• Current Assignee: Shimon Ando
• Original Assignee: Shimon Ando
• Inventors: Ando, Shimon
• Primary Examiner(s): Hayes, Gail O.
• Application Number:
  US07/943,401
• Time in Patent Office: 2,945 Days
• Field of Search: 364/419.02, 364/419.08
• US Class Current: 708/605
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/55 Rule-based translation

Abstract

A method which includes performing a structure analysis on a natural sentence inputted by making use of a word dictionary DIC-WD and a configuration dictionary DIC-KT and converting letter series KNJ of the inputted natural sentence into a language structure information series IMF-LSL. The natural sentence inputted in the form of the language structure information series IMI-LSL is subjected in such a manner to application of meaning analysis grammar IMI-GRM to cause a single or a plurality of meaning frames IMF-FRM to be read out from a meaning frame dictionary DIC-IMI in accordance with commands of the meaning analysis grammar IMI-GRM. When a plurality of meaning frames IMI-FRM are read out a meaning frame which defines an abstract meaning expressed by the inputted natural sentence is synthesized by case coupling and/or logic coupling the meaning frames IMI-FRM. Words WD, particles JO and symbols KI are inserted into the meaning frames IMI-FRM read out or the meaning frame IMI-FRM synthesized to thereby determine and produce data sentence DT-S correctly expressing the meaning of the inputted natural sentence in a computer whereby the language structure information series IMF-LSL is converted into the data sentence DT-S in the form of data structure PSMW with a multi layered case-logic language structure.

Claims

1. A method of storing natural language in a computer and generating further natural language based on the stored natural language by the computer comprising the steps of:

preparing a word dictionary which stores language structure information defining individual function of letter series representing words;

preparing a configuration dictionary which stores language structure information defining mutual connecting relations of letter series representing particles and symbols;

preparing a meaning frame dictionary which stores meaning frames defining abstract meaning structures corresponding to letter series representing words;

preparing a meaning analysis grammar which commands mutual case coupling relations and mutual logical coupling relations between words, particles, symbols and the meaning frames corresponding to combinations of the language structure information and further commands insertion of the words, the particles and the symbols into the meaning frames;

performing a structure analysis on a natural sentence inputted by making use of the word dictionary and the configuration dictionary;

converting the letter series of the inputted natural sentence into a language structure information series;

subjecting the inputted natural sentence in the form of the language structure information series to the meaning analysis in such a manner that through application of the meaning analysis grammar to the language structure information series a single or a plurality of meaning frames are read out from the meaning frame dictionary in accordance with commands of the meaning analysis grammar;

synthesizing, when a plurality of meaning frames are read out, a meaning frame which defines an abstract meaning expressed by the inputted natural sentence by case coupling and/or logic coupling the meaning frames; and

inserting words, particles and symbols into the meaning frames read out or the meaning frame synthesized to thereby determine and produce data sentence correctly expressing the meaning of the inputted natural sentence in the computer, whereby the language structure information series is converted into the data sentence in the form of data structure with a multi layered case-logic language structure.

2. A method according to claim 1, wherein the data structure includes at least, a first element which stores words, a second element which stores particles, a third element which stores symbols, a fourth element which stores the number of objective data structure to be connected by the case combination, a fifth element which stores the type of case combination, a sixth element which stores the number of objective data structure to be connected by the logical combination, and a seventh element which stores the type of logical combination;

the case logic structure, which determines the entire framework of the abstract meaning expressed by the natural sentence which has been input, is formed by storing the type of case combination between words expressed by the natural language inputted in the fifth element representing collection in the data structure which expresses the number of objective data structure to be connected by case combination in the fourth element of objective data structure to be connected by logical combination in the sixth element and type of logical combination in the seventh element; and

storing the words, particles, and symbols of the natural sentence inputted, in the first element, element and third element in the case logical structure, to determine the meaning of the natural sentence inputted, whereby the meaning of the input natural sentence is accurately expressed in the computer, and natural language processing is easily performed by the computer.

3. A method according to claim 2, wherein the data structure further comprises an eighth element which stores the number of the data structure to be connected by case combination and an ninth element which stores the number of the data structure to be connected by logic combination.

4. A method according to claim 1, wherein a minimum meaning unit including at least six cases of Case A an agent case, Case T a time case, Case S a space case, Case O an object case, Case P a predicate case and Case X an auxiliary case defined by the data structure, which includes a first element which stores words, a second element which stores particles, a third element which stores symbols, a fourth element which stores data commanding prohibition of outputting the stored word in a natural sentence, a fifth element which stores number of object data structure in which the same word is to be inserted, a sixth element which stores data defining the content of the word to be stored, a seventh element which stores number of object data structure to be connected by case combination, an eighth element which stores a type of the case combination, a ninth element which stores number of object data structure to be connected by logic combination and a tenth element which stores a type of logic combination;

whereby more complicated meaning structures are constructed by connecting single or multiple minimum meaning units by case combination or by logic combination, to form the meaning frames which express an abstract meaning.

5. A method according to claim 4, wherein the data structure further comprises an eleventh element which stores the number of the data structure to be connected by case combination and a twelfth element which stores the number of the data structure to be connected by logic combination.

6. A method according to claim 1, wherein the data structure includes first data structure and the second data structure, and the first data structure includes at least a first element which stores words, a second element which stores particles, a third element which stores symbols, a fourth element which stores the data commanding prohibition of outputting of the stored word in a natural sentence, a fifth element which stores number of the first data structure in which the same word is to be inserted, a sixth element which stores the data defining the content of the word to be stored, a seventh element which stores the number of the first data structure or the number of the second data structure to be connected by case combination, an eighth element which stores a type of case combination, a ninth element which stores the number of data structure to be connected by logic combination, and a tenth element which stores a type of the logic combination;

the second data structure includes at least a eleventh element which stores particles, a twelfth element which stores symbols, a thirteenth element which stores the number of the first data structure connected as Case A (agent case), a fourteenth element which stores the number of data structure MW connected as Case T (time case), a fifteenth element which stores the number of the first data structure connected as Case S (space case), a sixteenth element which stores the number of the first data structure connected as Case O (object case), a seventeenth element which stores number of data structure connected as Case P (predicate case), and an eighteenth element which stores number of the first data structure connected as Case X (auxiliary case).

7. A method according to claim 1, wherein when words and particles are inserted into the meaning frame which is read from the meaning frame dictionary, or inserted into the synthesized meaning frame, and when the arrangement in the language structure information contains word+particle in the language structure information series, then data structure, in which the same particle is set, is searched for by tracing a searching path in the meaning frame which is set according to the designated order of priority, and the word and the particle are respectively inserted into first element and second element of the searched for data structure.

8. A method according to claim 7, wherein particles in the meaning frame which was called up from the meaning frame dictionary or in the synthesized meaning frame are set to permit alternation whereby input natural sentences having a variety of expressions are stored in the form of the data structure.

9. A method according to claim 7, wherein a plurality of case particles designated in the meaning frame are stored in a third element of the data structure for the meaning frame via the coordinates in a case particle table which stores a group of case particles.

10. A method according to claim 1, wherein, when word is inserted into the meaning frame which was read out from the meaning frame dictionary or into the synthesized meaning frame, data structure, in which word has not yet been inserted into the element, is searched for by tracing a search path in the meaning frame which is set up according to the designated order of priority and then the word is inserted into the element in the searched for data structure.

11. A method according to claim 1, wherein when words and particles are inserted into the meaning frame which is read out from the meaning frame dictionary or inserted into the synthesized meaning frames a predetermined range in the language structure information series defined by starting point and ending point is designated in advance in which range there exists the word possibly inserted in the meaning frame, whereby words not related to the insertion into the meaning frame are eliminated and only the words related to the meaning frame are correctly inserted.

12. A method according to claim 11, wherein the word+particle in the predetermined range containing possible insertable word are inserted starting from the word at the ending point ending to the word at the starting point in such a manner that data structure, in which the same particle is set, is searched for by tracing a searching path in the meaning frame which is set according to the designated order of priority, and the word and the particle are respectively inserted into a first element and a second element of the searched for data structure and the remaining words in the predetermined range are further inserted starting from the word at the starting point ending to the word at the ending point in such a manner that data structure, in which word has not yet been inserted into the element, is searched for by tracing a search path in the meaning frame which is set up according to the designated order of priority and then the word is inserted into the element in the searched for data structure.

13. A method according to claim 1, wherein the data sentence includes a question data sentence which was converted from a natural sentence which was input as a question sentence, and a text data sentence converted from a natural sentence which was input as a text sentence, a base point for starting search in the question data sentence in the form of data structure, and a base point for starting search in the text data sentence in the form of data structure are provided, individual search paths are set up from the search start base point for the question data sentence, and from the search start base point for the text data sentence, the respective search paths are divided into a plurality of search sections defining as a search section starting point at a data structure at the search starting base point or a data structure representing the case of a primary sentence in the search path and defining as a search section ending point at a data structure of which connected upper level data structure is a primary sentence when a data structure to be connected in the upper level is designated in a first element-MW of the data structure at the search section starting point or at a data structure at which no data structures to be connected upper level and to right side via a second element are designated, the respective divided search sections for the question data sentence and the text data sentence are traced along the respective search paths if a word, which exists in the divided search section of the question data sentence, also exists in the divided search section of the text data sentence which corresponds to the divided search section of the question data sentence, the divided search section of the text data sentence is assigned an evaluation point based on the case of the data structure in which the word exists, and on the position of the word in language structure, then the evaluation points for all the divided search sections are totalled, and the conformity of pattern-matching between the question data sentence and the text data sentence is evaluated on the basis of the total number of evaluation points.

14. A method according to claim 1, wherein the data sentence includes a question data sentence [QDT-S]] converted from a natural sentence which was input as a question sentence and a text data sentence [TDT-S]] converted from a set of natural sentences which was input as a text sentence, a search path established in the question data sentence [QDT-S]] by designating the case selection order in the primary sentence, as well as the selection order of data structure to be connected in the data structure, is traced to discover the words WD which have been inserted into a first elements of the data structure, the discovered words are arranged in order of discovery as searched-for words [RWD, then existence of searching words in the set of the text data sentences]], which are similar to the searched-for word is checked according to the discovery order, if a searching word exists, a preliminary evaluation is carried out to check the conformity between the type of case in the primary sentence in the question data sentence to which the searched-for word is connected via a case combination, and the type of case in the primary sentence in the text data sentence to which the searching word SWD is connected via case combination, after passing the above preliminary evaluation, the primary sentence of the question data sentence is determined to be the search start base point for the question data sentence;

and the primary sentence in the text data sentence is determined to be the search start base point for the text data sentence, pattern-matching evaluation is performed for all the text data sentences which have passed the preliminary evaluation in such a manner that a base point for starting search in the question data sentence in the form of data structure, and a base point for starting search in the text data sentence in the form of data structure are provided, individual search paths are set up from the search start base point for the question data sentence, and from the search start base point for the text data sentence, the respective search paths are divided into a plurality of search sections defining as a search section starting point at a data structure at the search starting base point or a data structure representing the case of the primary sentence in the search path and defining as a search section ending point at a data structure of which connected upper level data structure is a primary sentence when a data structure is be connected in upper level to designated in a first element of the data structure at the search section starting point or at a data structure at which no data structures to be connected upper level and to right side via a second element are designated, the respective divided search sections for the question data sentence and the text data sentence are traced along the respective search paths if a word, which exists in the divided search section of the question data sentence, also exists in the divided search section of the text data sentence which corresponds to the divided search section of the question data sentence, the divided search section of the text data sentence is assigned an evaluation point based on the case of the data structure in which the word exists, and on the position of the word in language structures then the evaluation points for all the divided search sections are totalled, and then the text data sentences which have passed the preliminary evaluation are then ranked according to the evaluation points which represent the conformity of the pattern-matching.

15. A method according to claim 14, wherein an answer sentence is prepared based on the text data sentence which has the highest number of evaluation points.

16. A method according to claim 1, wherein when outputting a series of letters of a natural language while tracing the produced data sentence in the form of data structure along an output path established by designating the case selection order in primary sentences and the selection order of data structure to be connected in the data structure, the output order of the series of letters of words, particles and symbols in the data structure is designated, whereby a multiplicity of natural languages having a variety of word orders are produced based on the data sentence stored.

17. A method according to claim 16, wherein further preparing an inflective suffix particle table which contains inflective suffix particles defined by two coordinates, and also a tense negative suffix particle table which stores the tense negative particles and the tense-negative suffix particles and the two coordinates corresponding to various expressions including past, present, affirmative, negative and polite expressions, and when there is an inflective suffix or inflective tense negative suffix particle between two expressive and non-inflective words or tense negative particles, coordinate which is stored in a first element of the data structure in which the preceding word exists or coordinate which is determined from the tense negative suffix particle table by using a second element of the data structure in which the tense negative particle exists, is obtained, and further a coordinate which is stored in the first element of the data structure in which the following word exists or a coordinate which is determined from the tense negative suffix particle table by using the second element of the data structure in which the tense negative particle exists. then the inflective suffix particle or the tense negative suffix particle is determined based on the obtained two coordinates by using the inflective suffix particle table whereby a natural sentence is generated.