JPH03210659A - Method and device for deciding expression range of compound word - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention]

(Industrial Application Field) The present invention relates to a natural language information processing method and a morphological analysis method for Japanese in an information system, and in particular,
The present invention relates to a compound word expression range determining method and apparatus suitable for determining the range of expressions constituting a compound word.

[Conventional technology]

When a machine processes natural languages such as Japanese,
The first problem is the character input/output problem as a man-machine interface. In the case of Western languages, the number of characters used is about 100, so a European typewriter is sufficient, but in the case of Japanese, at least 2000 characters are used.
Since it is necessary to be able to input and output more than 10,000 character types, the technical difficulty is an order of magnitude greater. There are several ways to input Japanese text, including the multi-stage shift method, full character arrangement method, and code input method. There are Kana-Kanji conversion methods, pattern recognition methods, etc. Input methods other than the Kana-Kanji conversion method and the pattern recognition method are methods suitable for specialized operators, and are difficult for ordinary office workers to easily use. Therefore, the kana-kanji conversion method is now common. However, for special characters, many systems also use a code input method. The code input method is a method in which characters are converted into a multi-digit code and input. For example, JIS Kanji code system, C6
In H.226, characters are represented by four-digit numerical codes, and by inputting this code, the desired Kanji is obtained. The Kana-Kanji conversion method is a method in which when you input a Japanese sentence from the Kana keyboard in Kana or Roman main sentences, the computer automatically converts it into a sentence containing Kanji and Kana, and is one of the research results of natural language processing. . Regarding the problem of improving the efficiency and accuracy of kana-kanji conversion rate, we attempted to use related word information for homophone analysis and word semantic classification information, for example, to facilitate part-of-speech determination in kana-kanji conversion. There is research on proposing a method for separating kana sentences to make it easier to read, and on establishing a syntactic analysis method for full-scale processing of semantic information. FIG. 15 is an explanatory diagram showing the flow of the kana-kanji conversion process. The kana-kanji conversion process first involves a step 1501 of cutting out a series of kana input sentences into processing units for kana-kanji conversion, such as word units or bunsetsu units, and a conversion process of converting the cut-out processing units into corresponding kanji. There is a process 1502. Each process is proceeded by referring to the dictionary 1503. In order to implement such kana-kanji conversion processing for natural language input, it is necessary to analyze the structure of the natural language, and furthermore, in order to perform syntactic analysis, it is necessary to obtain part-of-speech information, etc. . Morphological analysis is a method for obtaining part-of-speech information. A morpheme is a fixed, specifically identifiable (meaningful) segment of a word. For example, in English, rGun (gun) J is one morpheme, and r(i
"un-s (guns)" contains two morphemes, r
"Gun-fighter" contains three morphemes. Morphological analysis is the process of extracting morphemes from a given sentence.
The purpose is to analyze and identify how these words are combined to form words. The purpose of this is to fix the various forms that a word takes according to categories such as number, special name, personal name, etc., and according to case, etc., and also to fix the structure of the word, that is, its base, and its combinations. By extracting the morphemes that contain words, we can analyze how those words are structured. In other words, the main task is to fix a sequence of forms from a sentence given as a character string and extract a sequence of morphemes from these. Regarding the processing of linguistic information mentioned above, the work was supervised by Makoto Nagao,
PP of “Japanese Information Processing” published by Institute of Electronics and Communication Engineers,
1-7.62-75, especially regarding morphological analysis, P
P, 86-113. Conventional morphological analysis methods for Japanese include, for example,
As described in Japanese Patent Application Laid-Open No. 64-48171,
Cut out the smallest unit that becomes a clause from the target sentence,
There is a method that compares and matches the extracted character string with a header string in a dictionary and recognizes the matches as clauses. Furthermore, there is a database system that utilizes a device that implements such a Japanese morphological analysis method. For example, "Natural Language Interface Processing Method Based on the R-Table Hierarchical Model" by Hiroyuki Kinuyo (Information Processing Society of Japan Journal Vol.
, 27 No. 5 May 1986) PP. There are those described in 499-508. In other words, computers and various OAs targeted at unspecified users.
In order to make the device easier to use, it is important to be able to respond to inquiries and give operational instructions in natural language similar to everyday conversation.
In response to this strong demand, we are proposing a new interface processing method for natural language query response for database systems. [Problems to be Solved by the Invention] Conventional morphological analysis methods targeting Japanese do not take into account how to determine the range of expressions that constitute a compound word. For example, the Japanese morphological analysis device described in Japanese Patent Application Laid-open No. 64-48171 does not accurately identify the expressions that make up the compound word unless the character string of the expression that makes up the compound word is registered as a dictionary headword. It is impossible to recognize it. For example, when using this Japanese morphological analysis method in a database system, the user must register a large number of headwords. In other words, when considering a database search in Japanese, it is essential to check and recognize that the search conditions or individual words that make up the search target completely match the character strings in the database. Therefore, if a word is incorrectly recognized, character strings cannot be compared and matched during a database search, and the database search cannot be performed. In databases, data corresponding to proper nouns is often composed of compound words. However, when expressing compound words in Japanese, they are not necessarily the same as the compound words stored in the database. It is not necessarily expressed in Rather, compound words are often expressed in various ways, and it has been impossible for conventional techniques to recognize these various expressions of compound words. In addition, in order to be able to recognize various expressions of a compound word with the conventional technology, it is necessary to register a variety of different headwords for the compound word in a dictionary, and it is necessary to construct a dictionary. This results in a huge burden on the user. at the same time,
There was a problem that the storage area of the dictionary was used more than necessary. Furthermore, the above-mentioned problem occurs every time the database to be searched is changed, making it difficult to construct a dictionary for analyzing Japanese. The purpose of the present invention is to solve these problems of the prior art, make it possible to automatically determine the expression range of an input compound word, and furthermore, make it possible to search in a database by inputting a compound word with a high degree of freedom. The object of the present invention is to provide a method and device for determining the range of compound word expression. [Means for Solving the Problems J] In order to achieve the above object, the compound word expression range determination method of the present invention provides (1) a Japanese sentence inputted by an input device,
It is characterized by determining the range of expressions constituting a compound word based on the case particle included in this Japanese sentence and the dictionary, and converting it into an expression that includes the compound word stored in this dictionary. (2) The compound word expression range determination method described in (1) above is characterized in that a case particle table defining character strings corresponding to case particles is used. (3) The compound word expression range determination method described in (2) above is characterized by using a headword division dictionary in which headwords are divided into character strings of minimum units of independent words and stored. (4) In the compound word expression range determination method described in (3) above, the range of expressions constituting the compound word is determined from the input Japanese sentence based on case particles, and the determined expression range is A headword candidate string is generated from the character string, and the range of expressions that make up the compound word is determined by referring to the headword candidate string stored in the headword division dictionary. The method is characterized in that it converts the compound words stored in the dictionary into expressions that include the compound words, and generates a Japanese phrase string for which the a-goku display range has been determined. (5) In the compound word expression range determination method described in (4) above, the continuous range of clauses to which case particles are attached in the input Japanese sentence is determined as the expression range that constitutes the compound word. , from the string in this determined expression range,
It is characterized by generating a headword candidate character string. (6) In the compound word expression range determination method described in (4) above, the character string corresponding to the case particle in the headword candidate character string is deleted, and the headword candidate character string and the headword from which this case particle is deleted are added. The divided headwords stored in the word division dictionary are reconstructed based on the same fixed rules, and the reconstructed headword candidate character strings and the divided headwords are collated and compared, and if they match, this headword is It is characterized by determining and recognizing the range of character strings of the input Japanese sentence that correspond to candidate character strings as the range of expressions that constitute compound words. (7) In the compound word expression range determination method described in (6) above, the structure based on the same fixed rule is to Characterized by rearrangement. (8) In the compound word expression range determination method described in (6) above, the structure based on the same fixed rule is that each character string of the divided independent words in the headword division dictionary is arranged in ascending order of character code. Characterized by rearrangement. (9) In the compound word expression range determination method described in (6) above, the composition based on the same fixed rule is performed by combining character strings of divided independent words of the headword division dictionary. do. The compound word expression range determining device of the present invention is (10)
The present invention is characterized in that it includes a program that executes the compound word expression range determination method described in (4) above, and is added to a Japanese language processing system. Furthermore, (11) In a database system that performs operations based on operation instructions input in Japanese text, as a natural language interface, it is possible to analyze Japanese sentences representing operation instructions input into a computer and provide operation instructions to the computer. Features. [Operation] In the present invention, when detecting a heading candidate character string from an input Japanese sentence, the compound word expression range determining device:
Continuous character strings in phrases to which particles equal to the case particle string in the case particle table are attached are determined as heading candidate character strings. However, non-consecutive character strings in clauses to which particles equal to the case particle string in the case particle table are attached are determined at this point as the range of character strings constituting one clause. Next, when determining the range of strings that make up a compound word, delete the string that matches the case particle string in the case particle table from the heading candidate string, and then delete this case particle. The heading candidate character strings and the divided headwords of the headword divided dictionary are rearranged using the same rule. After sorting, both strings are compared. If they match, the range of character strings in the input Japanese sentence that correspond to this character string is determined as the range of character strings that constitute the compound word. At the same time, a regular word in the database corresponding to this character string is obtained from the headword division dictionary. The headword division dictionary that is referred to when determining the range of a clause consists of divided headword strings in which headwords are divided into the smallest unit of independent words, and regular word characters that are words in the database that correspond to this headword string. columns and other dictionary contents. [Embodiments] Examples of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a functional block diagram showing an embodiment of a compound word expression range determining device according to the present invention. This method realizes the compound word expression range determining method of the present invention, and will be explained below in terms of its functional blocks. A heading character string candidate estimation unit 1 generates a heading candidate character string 12 from an input Japanese character string 11 by referring to a case particle table 2 storing case particles, and a heading word division dictionary 4
, and generates a Japanese phrase string 13 for which the range of compound seasonal expressions has been determined. As is clear from the figure, the heading character string candidate estimating unit 1 and the bunsetsu range determining unit 3 represent processing, and the case particle table 2 and the heading word division dictionary 4 are files (also called tables). With this configuration, the following processing operations are performed. First, from the input Japanese character string 11, the range of consecutive clitic clauses with case particles is estimated to be a character string that may constitute a compound word, and heading candidates that can be character strings that constitute a compound word are estimated. Detect character string 12. Using this heading candidate character string 12 as input information, from the heading candidate character string 12,
Delete character strings corresponding to case particles and create headword division dictionary 4
Sort using the same rules as the divided headwords. Thereafter, the character strings are compared with the divided headwords, and if they match, the range of character strings that make up the compound word is determined as a range that can become character strings that make up the compound word. When detecting the heading candidate character string 12, reference is made to the case particle table 2 that defines character strings that can be case particles. In addition, when determining the range of a clause, use a divided heading string that is a headword divided by the smallest independent word, a regular word string that is a word in the database that corresponds to this heading string, and other words. Refer to the headword division dictionary 4 that defines the dictionary contents. That is, when the user inputs the Japanese character string 11, the heading character string candidate estimating unit l uses the case particle table 2.
, the range of character strings constituting the compound word, that is, the range of character strings that can be the header strings of the headword division dictionary 4, is estimated, a header candidate character string 12 is generated, and this is held in a memory (not shown). . Using the retained heading candidate character strings 12 as input information, the clause range determination unit 3 determines the range of character strings forming the compound word while referring to the headword division dictionary 4. and,
A Japanese phrase string 13 for which the range of complex phrase expression has been determined is generated and held in a memory (not shown). In this way, according to this embodiment, each functional block is realized by program logic. Therefore, each functional block can be implemented as an LSI, and the processing speed can be increased as a compound word expression range determination device. FIG. 2 is a block diagram showing the overall hardware configuration of the compound word expression range determination device in FIG. 1. An input/output device 5 that inputs and displays the Japanese character string 11 in FIG. 1, a processor 6 that executes the processing in FIG. 1 based on the program, and a case particle table 2 and various programs in FIG. It consists of a storage device 7 that stores. Furthermore, the storage device 7 includes a working area 8 which is a memory for executing each process of the processor 6, a storage area 10 for a heading character string candidate estimation program, a case particle table storage area 20, a clause range determination program storage area 30, and a heading word storage area 10. It is composed of a storage section of a divided dictionary storage area 40. Each program stored in the storage device 7 is executed by the processor 6
It is executed in In the execution, the input/output device 5 is used as necessary. Figure 3 shows the PAD (Proble+s Analyst) of the headword character string candidate estimation program in Figure 2.
sD iagram) It is possible to construct a compound word by focusing on the case particle and using the range of continuous separation in case particle-enclosed clauses for the Japanese character string input by the user. It is presumed that there is a
This figure shows the processing up to determining candidates for headwords that can become compound word clauses. This process will be explained below with reference to the PAD diagram. First, the system waits for input of a Japanese character string for bunsetsu division from the user (step 101), and then substitutes the Japanese character string input by the user into a character string variable INP. As a result, the bunsetsu-intelligible Japanese character string is assigned to the variable INP (step 102) 6. Note that INP simply indicates a character string variable and is not an abbreviation with a specific meaning. Next, the following processing is performed from the first clause to the last clause in the variable INP (step 103). With reference to the case particle table 2 in FIG. and the clause immediately following it are connected with a period (,) and substituted into INP (step 105). next,
Immediately after a clause that has been concatenated or immediately after a clause that has not been concatenated at all, it is separated by a comma (1) and substituted into INP (step 106). This indicates that the character string in the range of consecutive periods is the heading halo type character string 12 (see FIG. 1). The segmented Japanese character string that waits for input in step lot is an example of a Japanese character string. Even if the input Japanese character string is written in solid text rather than in separate parts, it can be easily divided into clauses by morphological analysis. In this way, by the processing from step 101 to step 106, from the segmented Japanese character strings shown in FIG. 7(a), FIG. 8(a), and FIG. 9(a), which will be described later, , Fig. 7(b), Fig. 8(b), and Fig. 9(
Generate the heading candidate character string shown in b). A method for determining the range of phrase character strings using the heading candidate character strings generated in this manner will be described below. FIG. 4 is an explanatory diagram showing the contents of the case particle table in FIG. 1. It consists of case particle character strings 21 such as ``haj'', ``ga'', ``wo'', ``ni'', ``de'', and ``naru''. This is referred to in step 104 in FIG. FIG. 5 is a PAD diagram of the phrase range determination processing program in FIG. 2. This figure shows the processing up to determining the range of character strings that will become one clause, using the heading candidate character string 12 in FIG. 1 as input information. This process will be explained below with reference to the PAD diagram. While a period exists in the INF in which the heading candidate character string is stored, the following processing is performed (step 301).
. From the last character in the variable INP, a character string in a continuous range with periods is cut out and substituted into the character string variable INP(2) (step 302). Next, with reference to the case particle table, INPC corresponding to the case particle string 2I in FIG.
2) Delete the character string inside step 303), INF(
2) Sort the character codes in descending order for each period character string in step 304), and delete the period from the character string in INP(2) (step 305). next,
"O" is assigned to an integer variable FLG (flag) (step 306). First, the headword division dictionary 4 (see FIG. 1) is moved record by record, and the following processing is performed until the last record is reached. Take out the division headword 41 of the record and assign it to the character string variable INP (3) (Step 308), and set the variable IN
Within the range of consecutive character strings with periods in P(3), the character strings in division units are sorted in descending order of character code (step 309), and periods are deleted from the character string in INF(3) (step 309). 310). Next, the variable INP(2
) and the character string in variable INF(3) are equal or not is determined in step 311), and if they are equal, the following processing is performed. Integer Noo4 is assigned to integer variable FLG (step 312). Next, I corresponding to the character string INF(2)
A regular word, which is a word in the database corresponding to the headword, is acquired from the headword division dictionary and associated with the character string in the NF (step 313). Next, the variable INP corresponding to the character string of variable INF (2)
The case particle character string immediately before the period in the character string inside is deleted (step 314). A period is deleted from the character string in variable INP corresponding to variable INP(2) (step 315). Next, it is determined whether the character string immediately before the beginning of the character string in the variable INP corresponding to the character string in the variable INF (2) is a period (step 316), and if it is a period, the character string in the INF Change the period to a comma (step 317), then change the label (LABEL) (2
) (step 318). Next, it is determined whether the integer FLG is rQJ (step 319), and if it is "0", the following processing is performed. Step 3
20), and the process moves to label (LABEL) (1) (step 321). Through steps 301 to 321, FIG. 7(b), which will be described later.
From the heading candidate character string ``What is the purchase price of skirts handled by the office in Tokyo?'' shown in Figure 7(e), the Japanese phrase string [the office in Tokyo sells] ,J is generated as the purchasing unit price of ,handled skirts. That is, FIG. 7(C) and FIG. 7(d) show the process of generating FIG. 7(e) from FIG. 7(b). Similarly, in steps 301 to 321, the eighth
The heading candidate character string shown in Figure (b) “6 offices in Tokyo,
From ``What is the cumulative sales result of skirts?'', the composite seasonal expression range determined Japanese phrase string ``What is the cumulative sales result of skirts at the Tokyo office'' shown in FIG. 8(e) is generated. Similarly, the composite seasonal expression range from the heading candidate character string "The business office is in Tokyo, Skirt's. Cumulative sales results are" shown in FIG. 9(b) to be described later in FIG. 9(e). A determined Japanese phrase string ``What are the cumulative sales results of skirts at the Tokyo office?'' is generated. In this embodiment, in step 309, the character strings in division units are sorted in ascending order of character code, that is,
They are arranged in ascending order of character code. The character strings in division units may be sorted in descending order of character codes, or in a combination of character strings in division units. FIG. 6 is an explanatory diagram showing an example of the contents of a headword division dictionary to be referred to by the bunsetsu range determination unit in FIG. 1. Divided headword 41. It has information on regular words 42 and parts of speech 43, which are words in the database. Note that it is easily possible to have information other than headwords. FIG. 7 is an explanatory diagram showing a first processing example of the Japanese language processing system in FIG. 1. The bunsetsu division Japanese character string input by the user is
It shows how the complex phrase expression range is converted into a Japanese phrase string. The phrase segmented Japanese character string input by the user shown in Figure 7(a) [ΔPurchase price of ΔSkirt sold by ΔSales Office in Tokyo is 8] is the header character in Figure 1. The column candidate estimation unit 1 generates the heading candidate character string [Tokyo office] shown in FIG. 7(b). The purchase price of skirts handled is converted to ``. Furthermore, this heading candidate character string is processed by the clause range determination unit 3 in FIG. Word clause sequence [The purchase price of skirts handled by the Tokyo office is”
is converted to In other words, the heading candidate character string [Purchasing price of skirts handled by a sales office in Tokyo] shown in Figure 7(b) is equivalent to , of the skirt,
The purchase price is '', and in Figure 7 (d), ``Skirts handled by the Tokyo sales office.'' The purchasing unit price is determined by J, and the R Tokyo office in Figure 7 (e)
The purchase price of skirts handled is converted to ``. FIG. 8 is an explanatory diagram showing a second processing example of the Japanese language processing system in FIG. 1. The bunsetsu division Japanese character string input by the user is
It shows how the compound seasonal expression range is converted into a Japanese phrase string. The phrase segmented Japanese character string input by the user shown in FIG. 1, the heading candidate character string [
What are the cumulative sales results for skirts at the sales office in Tokyo?
is converted to Furthermore, this heading candidate character string is
The phrase range determining unit 3 in the figure determines the range of phrases in FIG. 8(c) and (
Through step d), the compound phrase expression range is determined to be the Japanese phrase string ``What is the cumulative sales record of skirts at the Tokyo office'' shown in FIG. 8(e). In other words, the heading candidate character string shown in Figure 8(b) [Total sales results for skirts in Tokyo office] is as follows:
Figure 8(c) [Tokyo, sales office]. The cumulative sales results for skirts are J, and Figure 8 (d)
What are the cumulative sales results for skirts at the sales office in Tokyo?
After that, it is converted to ``The unit purchase price of the skirt at the Tokyo office'' in Figure 7(e). FIG. 9 is an explanatory diagram showing a third processing example of the Japanese language processing system in FIG. 1. The bunsetsu division Japanese character string input by the user is
It shows how the complex phrase expression range is converted into a Japanese phrase string. The segmented Japanese character string input by the user shown in Figure 9(a) "The business office is ΔTokyo ΔSkirt's ΔSales ΔCumulative ΔAchievements are J" The heading character string candidate estimating unit 1 converts the heading candidate character string F business office shown in FIG. Furthermore, this heading candidate character string is processed by the phrase range determination unit 3 in FIG. The phrase string is converted to "Tokyo office's skirts. Cumulative sales results are". In other words, the heading candidate character string shown in FIG. 9(b) "What are the cumulative sales results for skirts at the office in Tokyo?"
In Figure 9(c), [Skirt's business office is in Tokyo]
The cumulative sales result is J, and after going through Figure 9(d), [The sales office is in Tokyo, the cumulative sales result of skirts is]
As mentioned above, according to this embodiment, the compound word is converted into the division headword 41 by the independent word of the minimum unit.
(See Figure 6), there is no need to register other expressions of this compound word in the dictionary, it is possible to reduce storage space, and it is possible for the user to accept other expressions of the compound word. This increases the degree of freedom in expressing the Japanese input by the user, and at the same time reduces the burden on the user who has to think about how to express the Japanese manually. Further, in this embodiment, it is also possible to register the heading candidate character string 12 in the dictionary 4 as a new heading word. Next, an application example of the compound word expression range determining device according to the present invention will be explained. FIG. 10 is a functional block diagram showing the configuration of an embodiment of a database search system according to the present invention. Using the Japanese phrase string 13 whose compound word expression range has been determined, obtained by the compound word expression range determination method shown in FIG.
It performs database searches. That is, the first
A database search system in Japanese is constructed using the compound word expression range determination device shown in the figure. A heading character string candidate estimating unit 1 generates a heading candidate character string 12 by referring to a case particle table 2 from a Japanese character string 11, and a composite character string candidate estimating unit 1 generates a heading candidate character string 12 by referring to a headword division dictionary 4. A compound word expression range determination device having a phrase range determining unit 3 that generates a Japanese phrase string 13 with a determined phrase expression range is configured to perform a semantic understanding process that converts the Japanese phrase string 13 with a determined compound phrase expression range into an internal expression format 52. Internal expression generation unit 5
1 and an SQL generation unit 53 that generates 5QL 54 from this internal representation format 52. Then, it is configured by adding it to a processing section consisting of a database search section 55 that searches a database 56 based on 5QL54. Furthermore, SQL is structured Query.
It is an abbreviation for LanguageJ, and is a language for database searches. Below, the meaning understanding/internal expression generation unit 51 and the SQL generation unit 5
3 and the processing operation by the database search unit 55 will be explained. The meaning understanding/internal expression generation unit 51 uses the complex phrase expression range determined Japanese phrase string 13 as input information, understands the meaning for information retrieval, generates an internal expression format 52, and stores this in a memory (not shown). Hold. The retained internal representation format 52 is used as input information. 5QL54, which is a database search language, is generated by the SQL generation unit 53 and held in a memory (not shown). The database search unit 55 searches the database 56 using the held 5QL 54 as input information. FIG. 1+ is an explanatory diagram showing an example of the Japanese phrase string in which the complex phrase expression range has been determined in FIG. 10. The example of the Japanese clause sequence in which the range of complex phrase expression in FIG. 7(e) has been determined is Japanese sentence 13 for operation instructions for information retrieval.
An example is shown in which the value is set to 1. In other words, the phrase-separated Japanese character string input by the user shown in Figure 7 (a) [Δ unit price of Δ skirts handled by the Δ sales office in Tokyo is] is shown in Figure 1O. The heading character string candidate estimating unit 1 and phrase range determining unit 3 generate the Japanese phrase string ``Skirts, which is handled by the Tokyo office.'' The unit price is converted to ``. The Japanese sentence 131 converted in this way is used for operating instructions for information retrieval. FIG. 12 is an explanatory diagram showing an example of the internal representation format in FIG. 1O. Using the Japanese sentence 131 representing the operation instruction for information retrieval shown in No. 11 as input information, the semantic understanding/internal expression generation unit 51 shown in FIG. 1O extracted search conditions and search targets for information retrieval. This is an example of the internal representation format 521,
This is a search logical expression composed of column names and column values of the database 56 in FIG. 1O. Note that the generation of the internal expression format 52 by the semantic understanding/internal expression generation unit 51 is described in detail in Hiroyuki Kinuyo [Natural Language Interface Processing Method J Based on Table Hierarchical Model (Transactions of Information Processing Society of Japan Vol. 27 No. 5 May 1986), PP, 499-508. FIG. 13 is an explanatory diagram showing an example of the SQL in FIG. 10. A 5QL 541 is generated using the internal representation format 521 formed by the search logical formula in FIG. 12 as input information. This is an example of 5QL54 generated by the SQL generation unit 53 in FIG. 10. Note that the processing method of the SQL generation unit 53 in FIG. 1O is a conversion from a formal language to a formal language, which can be easily realized using conventional technology, and a description thereof will be omitted. FIG. 14 is an example of data stored in the database in [ ]O. A business office table 1410 and a product table 1420 are stored. The office table 1410 includes office threads 1411 such as "Tokyo office" and "Osaka office" and the location 14 of each office.
12. “Skirt”, [Product name of slacks J etc. 141
3. Purchase unit price 1414 and sales unit price 14 corresponding to each product
15. It is composed of cumulative sales results 1416. In addition, the product table 1420 includes model numbers 1422 corresponding to each product name 1421 such as [skirts] and [slacks].
, manufacturing factory name 1423, and inventory quantity 1424. The database search system shown in FIG.
Based on the input of the Japanese sentence 131 shown in Figure 1, "Tokyo Sales Office" is entered from the sales office thread 1411 in the sales office table 1410, "skirt" is entered from the product name 1413, and purchase unit price 1
414 to [2500]. As described above, according to the database search system shown in FIG. 10, in addition to the processing by the compound word expression range determination method shown in FIG. 1, the following processing is possible. When performing a database search in Japanese, it is essential that the input question sentence be a natural language sentence with a high degree of freedom at the level of daily conversation. In other words, in databases, there are many cases where data corresponding to proper nouns is composed of compound words.
When a compound word is expressed in Japanese, it is not necessarily expressed in the same way as the compound word stored in a database, and the compound word is often expressed in a variety of ways. With conventional technology, it was impossible to recognize the various expressions of compound words and to match them with the compound words in the database, but according to the database search system using this Japanese language processing system, Make it possible to recognize diverse expressions. Therefore, when the user inputs a search sentence, there is no need to match it exactly to a compound word in the database, and the user can perform a search by inputting Japanese with a high degree of freedom. As described above, according to this embodiment, by dividing a compound word into the smallest units of independent words and registering them as headwords in the dictionary, there is no need to register other expressions of the compound word in the dictionary, and the storage area is reduced. I can do it. In particular, since the scale of a dictionary used for information retrieval depends on the scale of the information medium, the storage area for the dictionary must be reduced as much as possible. Furthermore, although it is very time-consuming to register all the expressions of a compound word, this embodiment can save this time and effort. Furthermore, users who input Japanese character strings do not need to be aware of the exact compound word, and can input Japanese character strings with free expression, which reduces the burden on the user. [Effect of the invention J According to the present invention, it is possible to automatically determine the expression range of an input compound word, and furthermore, it is possible to search in a database by inputting Japanese character strings with a high degree of freedom. I can do it.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram showing an embodiment of the compound word expression range determining device according to the present invention, and FIG. 2 is a block diagram showing the overall hardware configuration of the compound word expression range determining device in FIG. 1. , Figure 3 is a PAD diagram of the headword character string candidate estimation program in Figure 2, Figure 4 is an explanatory diagram showing the contents of the case particle table in Figure 1, and Figure 5 is the clause range determination process in Figure 2. PAD of the program, Figure 6 is an explanatory diagram showing an example of the contents of the headword division dictionary in Figure 1, Figure 7 is the first part of the Japanese language processing system in Figure 1.
8 is an explanatory diagram showing a second processing example of the Japanese language processing system in FIG. 1, and FIG. 9 is an explanatory diagram showing a third processing example of the Japanese language processing system in FIG. 1. FIG. 10 is a functional block diagram showing the configuration of an embodiment of a database search system according to the present invention, and FIG. 11 is an explanation showing an example of a Japanese phrase string whose complex phrase expression range has been determined in FIG. 10. Figure 12 is an explanatory diagram showing an example of the internal expression format in Figure 1O, Figure 13 is an explanatory diagram showing an example of SQL in Figure 1O, and Figure 14 is the data stored in the database in Figure 10. Example, 15th
The figure is an explanatory diagram showing the flow of kana-kanji conversion processing. l: Headline character string candidate estimation unit, 2: Case particle table, 3
: clause range determination unit, 4: headword division dictionary, 5: input/output device, 6: processor, 7: storage device, 8: working area, 9:, 10: storage area for headline character string candidate estimation program, 11: Japanese character string, 12: Heading candidate character string, 13: Japanese phrase string with complex phrase expression range determined,
20: Case particle table storage area, 21: Case particle character string. 30: Clause range determination program storage area, 40: Divided headword dictionary storage area, 41: Divided headwords, 42:
Regular language, 43: Part of speech, 51: Meaning understanding/internal expression generation unit, 52: Internal expression format, 53: SQL generation unit, 54: S
QL, 55: Database search section, 56: Database, 131: Japanese text. 521: Internal representation format, 541: SQL, 1410:
Sales office table, 1411: Sales office thread, 1412: Address, 1413: Product name, 1414: Purchase unit price, 1415
: Sales unit price, 1416: Cumulative sales results, 1420: Product table, 1421: Product name, 1422: Model number, 142
3: Manufacturing factory name. 1424: Stock quantity, +501: Cutting process, 1502
: Conversion process, 1503: Dictionary. Fig. Fig. 4 Fig. Fig. 1 (Part 1) 19 Fig. (Part 2) Fig. 6 Fig. Fig. (a) (e) (el The cumulative sales of skirts at the Tokyo office are shown in Figures (a) and (e) The cumulative sales of skirts at the Tokyo office are shown in Figure 1. Figure 501 502 503

Claims (1)

[Claims] 1. It has an input device for inputting Japanese sentences and a Japanese electronic dictionary, and morphologically analyzes the Japanese sentences inputted by the input device into a string of words. Among them, in a Japanese language processing system that recognizes words that match character strings containing compound words stored in the dictionary as clauses, the Japanese sentence inputted by the input device is recognized as a case particle included in the Japanese sentence. A method for determining a compound word expression range, comprising determining the range of expressions constituting a compound word based on the dictionary, and converting the range into an expression that includes the compound word stored in the dictionary. 2. The compound word expression range determining method according to claim 1, characterized in that a case particle table defining character strings corresponding to the case particles is used. 3. The compound word expression range determining method according to claim 2, which uses a headword division dictionary in which headwords are divided into character strings of minimum units of independent words and stored. . 4. In the compound word expression range determination method according to claim 3, from the input Japanese sentence, based on the case particle,
Determine the range of expressions constituting the compound word, generate a headword candidate character string from the character strings in the determined expression range, and further store the headword candidate character string in the headword division dictionary. determine the range of expressions constituting the compound word by referring to the character string, convert it into an expression that includes the compound word stored in the dictionary, and generate a Japanese phrase string with the compound phrase expression range determined. A compound word expression range determination method characterized by: 5. In the compound word expression range determination method according to claim 4, a range of consecutive clauses to which the case particle is attached in the input Japanese sentence is determined as an expression range constituting the compound word; A method for determining a compound word expression range, characterized in that the headword candidate character string is generated from a character string within the determined expression range. 6. In the compound word expression range determination method according to claim 4, a character string corresponding to a case particle of the headword candidate character string is deleted, and the headword candidate character string from which the case particle is deleted is divided into the headword candidate character string. The divided headwords stored in the dictionary are reconstructed based on the same fixed rules, and the reconstructed headword candidate character string and the divided headwords are collated and compared, and if they match, the headword candidate characters are A compound word expression range determination method, characterized in that the range of character strings of the input Japanese sentence corresponding to the string is determined and recognized as the range of expressions constituting the compound word. 7. In the compound word expression range determination method according to claim 6, the configuration based on the same fixed rule is to rearrange the divided independent words of the headword division dictionary in descending order of character codes in units of character strings. A compound word expression range determination method characterized by the following. 8. In the compound word expression range determination method according to claim 6, the configuration based on the same fixed rule is to rearrange the divided independent words of the headword division dictionary in ascending order of character strings in units of character strings. A compound word expression range determination method characterized by the following. 9. The compound word expression range determining method according to claim 6, wherein the composition based on the same fixed rule is performed by combining character strings of divided independent words of the headword division dictionary. A method for determining the range of word expressions. 10. A compound word expression range determination device, comprising a program for executing the compound word expression range determination method according to claim 4, and added to the Japanese language processing system according to claim 1. 11. In a database system that performs operations based on operation instructions inputted in Japanese sentences, the database system is characterized in that, as a natural language interface, the Japanese sentences representing the operation instructions inputted into the computer are analyzed and the operation instructions are given to the computer. The compound word expression range determining device according to claim 10.