JPS5856071A - Retrieval system by japanese - Google Patents

Abstract

PURPOSE:To eliminate a difference in the representation of enquiry information and to perform desired retrieval effectively by obtaining the clear connection relation between independent words of enquiry information and performing conversion into names, and then retrieving information. CONSTITUTION:The word dividing device 7 of a retrieval name generating device 5 divides enquiry information 1 into words by using a Japanese dictionary 8 and an interword connection inspecting device 9, and those words are stored in a word buffer 10. While attention to independent words of respective words in the buffer 10 and their generation order is paid, phrases having added adjective are extracted and then given respective numbers before being stored in a phrase buffer 12. The contents of this bufer 12 are applied to a connection analyzing device 13, which utilizes a meaning connection inspecting device 15 and a grammatical connection inspecting device 14 to analyze connection relations between the number of the phrases. The results are set in a phrase connection table buffer 16, and a meaning symbol extracting device 17 while extracts words with meaning information from words in the phrases utilizes connection destination phrase information to extract and apply retrieval name information 6 to a retrieving device 2.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention provides a search system in Japanese, particularly a parts database storing a plurality of program modules ("/7tot: i think of air parts)" in Japanese. In order to search by inputting a word, Japanese inquiry information is processed to clarify the relationships between independent words and to absorb differences in expression in the inquiry information. It is related to a search system.

For example, a program module named ``Integration by Runge-Futter method'', a program module named ``Integration by Euler method'', a program module named ``Integration by Poyn method'' are stored. Let us assume a search system that accesses a data e-pace and extracts a desired program module. The search system that works in this case is the first one.
It will look like the one shown in the diagram. In FIG. 1, 1 represents an inquiry information buffer, 2 a search device, 3 a database, and 4 a thesaurus storage buffer.

In the fl'lJl/stem, there is no thesaurus storage buffer 4, and the search device 2 operates to search for a program module a whose name exactly matches the contents of the inquiry information buffer l. In a more advanced system, the thesaurus Kano buffer 4
For example, if we associate the fact that ``integration by the simple Runge-Cook method'' and ``integration by the Poynne method'' are essentially the same, then ``integration by the simple Runge-Futter method'' can be found in the thesaurus. Check the storage buffer 4 and save the data.
The "integral based on the Point method" in base 3 is extracted.

However, in a search system that queries in Japanese, it is difficult to absorb differences in the expression of query information, such as "integration by the Runge-Futter method" and "integration by the Runge-Futter method". desired.

The purpose of the present invention is to solve the above-mentioned problems, and the Japanese language search system of the present invention provides names with added relationships between multiple nouns and noun phrases. A search object database is provided that stores a plurality of search objects, and a search device for accessing the search object database is provided, and the search device is provided with a search object database that stores a plurality of search objects. In a search system that performs a search by specifying a name, a word dividing device that indexes a Japanese dictionary and divides the inquiry information into word units based on nine inquiry information given in Japanese #, and also performs a word connection test; A phrase recognition device that selects an independent word in the words divided by the word dividing device and recognizes a phrase in which an adjunct word is added to the independent word; A grammatical connection test device is used to analyze relationships between kashi-uke and kashi-uke, which adds information on the previous bunsetsu. Tanuke is equipped with a semantic symbol extraction device that refers to information and relates the semantic symbols corresponding to independent words in the phrases across the phrases.
The present invention is characterized in that the given inquiry information is converted into the name of the search object with a clear relationship between independent words, and then the search is performed. This will be explained below with reference to the drawings.

Fig. 2 is a conceptual explanatory diagram of the search system according to the present invention;
The figure shows the configuration of an embodiment of the present invention, and is an explanatory diagram when inquiry information in a form different from that shown in Fig. 3 of Fig. 4 is given.

In FIG. 2, 1-1 and 1-2 represent inquiry information, 2 a search device, 5 a search name creation device, and 6Fi search name information (or buffer).

In the case of the present invention, the search name creation device 5 pays attention to independent words in the inquiry information, whether the inquiry information 1-1 shown in FIG. Moreover, the relationship between the independent words is paid attention to and the same search name information 6 is created. Then, the search device 2 is notified with the search name information 6.

FIG. 3 shows the configuration of an embodiment of the present invention. Code 1 in the figure,
2, 3, 5.6 correspond to Figure 1 or Figure 2 @
F is a word dividing device which divides the contents of the inquiry information buffer 1 into words using a Japanese dictionary 8 and a word connection testing device 9, and stores the words in a word buffer IOK. Note that the word buffer 10 stores inquiry information divided into t,
A word code of a word, a meaning symbol of the word, grammatical information of the word, semantic information of the word, etc. are stored as one word.

Reference numeral 11 is a clause recognition device, which pays attention to the independent words of each word on the word puffer 10 and the order in which they occur, and attaches #1 to one independent word! (auxiliary verb or particle) is extracted, and a =D number is assigned to each clause. The result is stored in the illustrated clause buffer 12. In the case shown, the phrase with ID number ``1'' that is ``by the Runge-Kutster method'' and the phrase with iQ number ``2'' that is ``integrate'' are extracted. The analysis device uses the illustrated grammatical connection testing device 14 and the semantic connection testing device 15 based on the contents of the phrase buffer 12 to analyze the phrase (jl) number “1”) and the phrase (j
l) number "2") to analyze the relationship.

As a result, the bunsetsu modification is set in the table buffer 16, and the previous bunsetsu information iIy is assigned to the modification.

Reference numeral 17 is a semantic symbol extracting device, in which the bunsetsu dependencies extract words in the bunsetsu that have semantic information for each bunsetsu set in the table buffer 16, and the above-mentioned bunsetsu dependencies extract previous bunsetsu information. Search name information 6 using iD'
Extract.

Each device will be explained in more detail below.

The word dividing device divides the contents of the inquiry information buffer 1 into, for example, "yomi". That is, even if a kanji code is included, the Japanese dictionary 8 is indexed using the kanji code and then decomposed into kana codes. Then, the query information is summarized into words using the maximum likelihood evaluation method based on the Japanese dictionary 8, and the inquiry information is divided into words. Also, the connection test between the obtained words is performed using the word connection test device 9. Test.

Regarding the above-mentioned Japanese dictionary 8 and its index mode, please refer to Japanese Patent Application No. 172472 (1972) filed by the present applicant "Word Search Method Based on Reading and Writing" and Japanese Patent Application No. 1983 (1972).
What is disclosed in No. 72474 "Kanji-kana conversion device" can be used. Regarding the above-mentioned maximum likelihood evaluation method, the method disclosed in Japanese Patent Application No. 54-922'F1 "Kana-Kanji Conversion Device" can be used, and regarding the word connection test, the method disclosed in Japanese Patent Application No. 54-92270 "Kana-Kanji Conversion Device" can be used. It is possible to use the one disclosed in "Word Connection Testing Apparatus".

As described above, the word segmentation device 7 uses the word buffer 10
At the top, set the words into which the inquiry information is divided. It should be noted that the grammatical information and semantic information regarding the rate IFK can be considered to be stored together with the words in the Japanese dictionary 8, and the grammatical information is used in the illustrated grammatical connection testing device 14, and the semantic information is is used in the illustrated semantic connection testing device 15. Furthermore, for "K" shown in the diagram, "M" is used alone.
There are two types: ``law'' which means ETHOD and ``law'' which has a meaning when combined with other nouns, such as when used as ``Runge-Hutter method'', but the above semantic information distinguishes between these. You can think of it as a thing.

The phrase recognition device 11 (as described above, groups phrases including independent words and assigns ID numbers to them), uses grammatical information based on the words on the word buffer lO and the order of occurrence of the words. Summarize into phrases.

The modification analysis device 13 uses the grammatical connection testing device 14 and the semantic connection testing device 15 for each word in the phrase buffer 12. The grammatical connection testing device 14 has a table that indicates whether or not a grammatical connection is possible. For example, as shown in the figure, a noun is allowed to be followed by a case particle, a s-pronounced noun is allowed to be followed by a s-s-bending ending, a case particle is allowed to be followed by a s-s-bending noun, and so on. It is described in From this, the end of the phrase in ID number r l J is ``by'' (
case particle) is the beginning of the clause with ID number ``2'', ``integral'' (
It can be seen that it is allowed to be followed by a (subject noun). Further, the semantic connection testing device 15 tests the connection relationship between the above-mentioned "Runge Kutta" and "Ho". Similar to the grammatical connection testing device 14, it is equipped with a table indicating whether or not semantic connections are possible.
Kutta' (semantic information rloOJ) is prohibited from being followed by 'law' (semantic information '300J) which has the meaning of METHOD, and the above 'r method' has the meaning, information rloOJ).
A “law” with 200J, the “Runge-Hutter method”
Check that it is organized as follows.

Regarding the semantic connection test, the present applicant previously filed a patent application 1973-1 oza364! For the modification analysis device, the one disclosed in Japanese Patent Application No. 56-102634 "Japanese Language Analysis System" can be used.

The dependency analysis device 13 examines the dependency between clauses as described above, and sets the dependency relationship in the table buffer 16. Next, the semantic symbol extraction device 1γ extracts the semantic information RUNGE KUTTA <BY MEANS OF> about the word in the clause with the BD number “1” in the buffer 16 and which has semantic information, and the iD number “2”. That r in the clause 1NTI
GRALJ is extracted and the Japanese arrangement relation is converted to a predetermined arrangement relation to obtain the search name report 6. Since the processing is as described above, the inquiry information 1-2 shown in Figure 2 is Even if , is input, the search name information at the 60th stage will be exactly the same.

FIG. 4 shows an explanatory diagram illustrating a mode when information such as "sort the list in ascending order" is input as inquiry information.

Reference numerals 1, 2, 3, 6, 16 in the figure. 1 corresponds to FIG. 3, and the area surrounded by a chain line corresponds to parts 1 to 16 shown in FIG.

In the case of the above inquiry information 1-3, there are three independent words. Then, the clauses with ID numbers "1" and "2" are shown as information ID/, respectively.
3” Semantic symbol extraction device 1
Based on this, r 5ORTiN[l+ <
BY > 1NcuBi < 0RDERJ and rsORTiNG <0BTECT> 2 with Li5TJ
The search device 2 extracts two pieces of information, and notifies the search device 2 of the two pieces of information in a form that is combined with an AND.

As described above, according to the present invention, differences in expression in inquiry information are absorbed, making it possible to supply desired searcher information to a search device.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram explaining a general aspect of a search system, FIG. 2 is a conceptual diagram of a search system according to the present invention, FIG. 3 is a configuration of an embodiment of the present invention, and FIG. This is an explanatory diagram when inquiry information in a format different from that shown in the figure is given. In the figure, 1 is the inquiry information, 2 is the search device, 3 is the data pace, 5 is the searcher creation device, 6 is the search name information,)
is a word segmentation device, 8 is a Japanese dictionary, 9 is a word connection test device, IO is a word no (tsufa), 11 is a clause recognition device, 1
2 is a phrase) (tsufa, 13 is a dependency is an analysis, a device, 14
is a grammatical connection testing device, 15 is a semantic connection testing device, 1
Reference numeral 6 represents a table buffer for bunsetsu relations, and reference numeral 1) represents a semantic symbol extraction device. Patent applicant Fujitsu Limited

Claims (1)

[Claims] (Relationships between multiple nouns or noun phrases) The receiver has a search object database that stores a plurality of search objects that are given names with added relationships, and one search object. In a search system that is equipped with a search device that accesses a database of objects, and performs a search by instructing the search device to specify the name of the object to be searched, the system searches for the target object based on inquiry information given in Japanese. A word division device that indexes a Japanese dictionary and divides the matching information into word units, and also performs a word connection test;
A phrase recognition device that adds adjunct words to K and recognizes nine phrases, and at least a grammatical connection verification device for each phrase obtained by the phrase recognition device; The 夛Uke is obtained by the analysis device, which gives information on the preceding bunsetsu, and the ｛Uke is obtained by the analysis device.The conjugation uke between the nine clauses corresponds to the independent word in the bunsetsu while referring to the information. Equipped with a semantic symbol extraction device that relates semantic symbols across clauses, the query information given above is converted into the name of the search object with clear relationships between independent words, and then searched. A search system in Japanese, which is characterized in that it performs the following functions.