JPH08137881A - Device and method for analyzing word relation - Google Patents

Abstract

PURPOSE: To analyze the meaning of words in a set of candidates for meaning or their word through fast and simple logic by analyzing the means of a word and a word string having meaning which are limited to a significant number. CONSTITUTION: A relative information dictionary 100 contains information representing the meaning of a work as bit information by displaying meanings of a slave semantic category by the meanings of a basic semantic category. Then the relative information acquiring means 210 of an analyzing means 200 obtain the meaning of a selected word by referring to the relative information dictionary 100 and using relative information of bit constitution, and recursively repeating logical operation as many times as inputted words, and a relative information output means 220 performs logical operation between relative information on the final word and all words before the final word when the selected word is the final word of the inputted word string, and outputs relative information on the whole set of the inputted words.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a word relation analyzing apparatus and a word relation analyzing method, and a word relation analysis for analyzing the meaning of a combination of input words and the meaning of each word in the combination. The present invention relates to a device and a word relation analysis method.

[0002] More specifically, for a plurality of input words, each word is input based on a combination of input words without specifying what meanings of some or all of the words are used. Also, the present invention relates to a word relation analysis device and a word relation analysis method for analyzing the meaning represented by the entire word set.

[0003]

2. Description of the Related Art Conventionally, in order to obtain the meaning represented by a set of arbitrary words in natural language processing or the meaning of each word at that time, a table is created or a dictionary is created based on the meaning given to each word by a dictionary. , Interpret according to the rules of semantic analysis of natural language.

FIG. 22 shows the configuration of a conventional word analysis system. The system shown in the figure includes a word input unit 10, a morpheme analysis unit 20, an analysis unit 30, a result output unit 40, and a dictionary 5.
0 and a semantic analysis rule 60. The morphological analysis unit 20 divides the word string input by the word input unit 10 into words, and refers to the dictionary 50 for the divided words to add meaning to the words.

The analysis unit 30 applies a semantic analysis rule 60 to the words given the meaning by the morphological analysis unit 20.
Interpret meaning according to. Here, an example will be described in which each input word specifies an address such as a city name.

Address information is one of the important input conditions for identifying a customer in managing / retrieving customer data with a computer. Address analysis is a process that analyzes what a character string entered as an address is, such as a prefecture name, city name, town name, etc., and converts it to the corresponding address code. Ease of handling, dealing with ambiguity, and speeding up processing are required. For example, like directory information
There is a case where address analysis is performed based on the information acquired from the inquirer when the result must be produced while interacting with the inquirer.

The address to be coded is divided into four layers, which are prefectures, cities, groups, towns, towns, and large letters. In the conventional address analysis, there are used (1) a method of delimiting each address hierarchy and (2) a method of inputting the first character of each address hierarchy and confirming it by guidance.

As an example of inputting by dividing each address hierarchy in (1), input using a delimiter symbol for each "city" and "ward", such as "Himeji City △ Shikama Ward △ Ega Kasuga Town" To do. However, “Δ” indicates a division (space).

As an example of inputting the first character of each address hierarchy in (2), first, when the character "ken" is input and displayed, for example, the user inputs "Saitama". next,
When you enter / display the guidance "City", enter "Kasukabe". In this way, the address information is input according to the guidance for each address hierarchy.

As described above, conventionally, the meaning is narrowed down by the inclusion relation of nouns separated by suffixes (prefecture, city), or
The content of the suffix is used to understand the relationship between the meanings of the nouns.

[0011]

However, in the method of inputting by dividing each address hierarchy and the method of inputting the first character of each address hierarchy and confirming with the guidance in the above-mentioned conventional technique, "town / large character", In many cases, a “letter” is composed of multiple words, and the delimiter is often entered incorrectly. Table 1 shows examples of input errors.

There is a restriction on the input order, and in the case of interactive processing, when the names are input in various orders such as town names and city names, or when it is unclear whether the input words are city names or town names. It is difficult to process.

[0013]

[Table 1]

Further, even when inputting in accordance with the guidance, if the address hierarchy is erroneously input, the target address cannot be obtained even if the first character of each address hierarchy is input. Furthermore, conventionally, in applying natural language processing to address analysis, the division of roles between morphological analysis and semantic analysis has become a problem. Addresses are mostly input with place names and suffixes (prefecture, city, ward, town, etc.), so the general grammar of natural sentences cannot be applied. Therefore, the lower order is narrowed down by the inclusion relation of the address (when the town A is included in the city B, it is called the inclusion relation). There are 460,000 addresses (including separate reading) nationwide up to "characters", and almost all combinations of characters exist in kana place names with few characters such as one or two characters. Performing a complete inclusion check using this dictionary in morphological analysis is
There is a problem that the number of solution candidates becomes too large and a practical execution speed cannot be obtained.

Conventionally, in order to match the character strings of words including a suffix, when the suffix is omitted,
There is also the problem that it will not be possible to relieve. Further, in the conventional word relation analysis, since the semantic analysis rule of the natural language that is tabulated or logic is used as the semantic analysis rule, the processing logic becomes complicated, resulting in a great development cost and processing time. There is a problem that enormous work costs are required.

Further, in order to perform the analysis at high speed, it is possible to access all of the dictionary at high speed. For example, if the dictionary is stored in the main memory, high speed access is possible, but the cost is high. is there. Further, when the dictionary is stored in a secondary storage device such as a disk device, there is a problem that high-speed access cannot be expected by reading and processing the dictionary information in the disk device into the main storage device.

The present invention has been made in view of the above points, solves the above-mentioned conventional problems, and does not specify in what meaning (hierarchy) the input arbitrary word set is used. Also, it is an object of the present invention to provide a word relation analysis device and a word relation analysis method capable of analyzing a meaning candidate and the meaning of each word in a set of the words at high speed with a simple logic.

Further, a further object of the present invention is to make the handling of suffixes of an input word string flexible, and when omitted, interpret the meaning with the omitted information and add the suffix. For example, it is to provide a word relation analysis device and a word relation analysis method capable of performing a meaning interpretation that reflects the suffix.

A further object of the present invention is to provide a word relation analysis device and a word relation analysis method capable of reducing resources without reducing the access speed of a dictionary or the like.

[0020]

FIG. 1 is a block diagram of the first principle of the present invention. The word relation analysis device of the present invention is a word relation analysis device that analyzes the meaning of an input word, the meaning of a set of words, and the meaning of a word is displayed by displaying the meaning of each subordinate meaning category for each meaning of the basic meaning category. Related information dictionary 100 for storing information expressing
And an analysis means 200 for analyzing the meaning of a word having a meaning limited to a significant number and the meaning of a word string made up of words by referring to the related information dictionary 100.

The related information dictionary 100 holds the related information for each word as bit information. Further, the analysis means 200 recursively obtains the logical operation for the number of input words by using the related information bitized by referring to the related information dictionary 100 for the meaning of the selected word. When the selected word is the final word of the input word string, the related information acquisition unit 210 that performs the logical operation of the related information of the final word and all the words before the final word is input. The related information output means 220 which outputs the related information of the whole set of words.

Further, the above-mentioned related information acquisition means 210
When the input word includes a suffix indicating the meaning / content of the meaning category for one of the meaning categories, the masking of the related information of other meaning categories is performed based on the content of the suffix. The mask means 211 for performing is included.

FIG. 2 is a block diagram of the second principle of the present invention. The present invention provides, in a word relation analysis device that analyzes the meaning of an input word, the meaning of a set of words, information representing the meaning of a word by displaying the meaning in a subordinate meaning category for each basic meaning category. The related information dictionary 100 to be stored, the meaning of a word having a meaning limited to a significant number, and the meaning of a word string composed of the words are stored in the related information dictionary 1
It has the 1st analysis means 200 which analyzes with reference to 00.

The related information dictionary 100 holds the related information for each word as coded information. Further, the first analysis unit 200 refers to the related information dictionary for the dependent meaning categories of the words selected in order of the basic meaning categories having the order, and uses the coded related information to use the higher-order basic meanings. A related information acquisition unit that interprets the dependent meaning category within the range of the category, and recursively repeatedly acquires the interpretation of the next basic meaning category within the range of the interpreted dependent meaning category, and a set of input words The related information output means 220 which outputs the related information of the whole is included.

Further, the first analyzing means 200 described above, when the inputted word includes a suffix indicating the meaning / content of the meaning category for one of the meaning categories, the suffix of the meaning category It includes a second analysis unit 212 that determines the basic meaning category based on the content and analyzes the basic meaning category within the range.

Further, according to the present invention, when the words and information are stored in the related information dictionary, the storage order determining means for determining the order of storing in descending order from the highest level to the lower level based on the level of the meaning category of the word. And a first storage means for storing words from the highest rank to a predetermined level among the ranks determined by the storage rank determination means in a first related information dictionary in the first storage means having a high access speed. , Second storage means for storing words of a predetermined level or lower in a second related information dictionary in a second storage means having an access speed slower than that of the first storage means.

The first storage means is a main storage device and the second storage means is a disk device. The word relation analysis method of the present invention is a word relation analysis method for analyzing the meaning of an input word or the meaning of a set of words. The meaning of a word having a meaning limited to and a word string composed of words is analyzed.

The present invention also provides the meaning of the input word,
In a word relation analysis method for analyzing the meaning of a set of words, a related information dictionary that stores information expressing the meaning of words by displaying meanings in subordinate meaning categories for each basic meaning category, The meaning of a word having a meaning limited to a number and a word string composed of the word is analyzed with reference to a related information dictionary.

When performing the above-mentioned analysis, the related information for each word is analyzed by referring to the related information dictionary which is coded information. Also, when performing the above analysis, the subordinate semantic categories of the words selected in order of the basic semantic categories having a hierarchical order are referred to the related information dictionary, and the encoded related information is used to determine the higher-order basic semantic categories. Interprets the subordinate semantic categories within the range of, and then recursively acquires the interpretation of the next basic semantic category within the range of the interpreted subordinate semantic categories, and outputs the related information of the entire set of input words. To do.

Further, when interpreting the basic meaning category in the above, if the input word includes a suffix indicating the meaning and content of the meaning category for one of the meaning categories, the suffix is added. The basic meaning category is determined based on the contents of, and analysis is performed within the range of the basic meaning category.

Further, according to the present invention, the order of storing in descending order from the highest level to the lower level is determined based on the level of the meaning category of the word, and the words from the highest level to a predetermined level in the determined order. Is stored in the first storage means having a high access speed, and words having a predetermined level or less are stored in the second storage means having a slower access speed than the first storage means.

[0032]

FIG. 3 is a diagram for explaining the first principle of the present invention. The present invention expresses the meaning content of another meaning category (hereinafter, subordinate category) depending on the presence or absence of the relationship with each meaning content in a certain meaning category (hereinafter, basic category), and specifies the meaning of each word. Prepare a related information dictionary in advance.

Step 1) The meaning of the word input from the related information dictionary is read as dictionary related information. Step 2) Obtain the semantic content represented by all the subordinate categories for the input first word (that is, obtain the logical sum of the semantic content of each subordinate category), and set the result as B1.

Step 3) The input word number i = 1. Step 4) The input word number i is incremented. Step 5) For each subordinate category of the second word, reflect the content of the result B1 in its semantic content (by taking the logical product of the semantic content of each subordinate category of the word and the result B1),
The semantic content of the second word represented by all the subordinate categories in which the semantic content of the first word is reflected is obtained, and the result B
Set to 2. Similarly, for the next third word, the result B2
The result B3 is obtained by reflecting the meaning contents of the first and second words by a logical operation with and. After that, the input words are recursively processed.

Step 6) When the input word is the last word n, the result B _n is obtained for the last input word n. The result B _n represents the common meaning content regarding the basic categories of all the input words. (In the example of the address, if the basic category is a prefecture name, B _n indicates the prefecture name in which the set of input words (city / town / village name) exists.

Step 7) Result B for each word_nTo
Find satisfying semantic information (Result B _n) And each result B
_i(The logical product operation of i = 1, 2, ..., N-1) is performed).
As a result, the meaning of the set of input words
Also, you can narrow down the candidates for the meaning of each word.
It

FIG. 4 is a diagram for explaining the second principle of the present invention. Step 11) Input the input character string in the order of the hierarchy level from the higher order to the lower order. Step 12) The input character string is divided into words by referring to the related dictionary information and divided into each level.

Step 13) When the division of all the levels of one input character string is completed, the process proceeds to step 14, and if there is a word string that has not been divided, the process proceeds to step 12. Step 14) Analyze the semantic categories of the divided words for each level. At this time, referring to the related information dictionary for the meaning categories of the words selected in the hierarchical order of the basic meaning categories having a hierarchy, the dependent meaning categories are interpreted within the range of the higher basic meaning categories based on the coded related information. Then, the direction relation of the meaning range is acquired.

Step 15) Divide the level for each semantic category. For example, when the information of the input character string is address information, it is divided into a prefecture city / ward level and a town character level. Step 16) If the level divided in Step 15 is equal to or higher than the predetermined level, the process proceeds to Step 17, and if it is a word located below the predetermined level, Step 18
Move to

Step 17) Words higher than a predetermined level are stored in a storage means for high speed access such as a memory of a main storage device. Step 18) Words lower than a predetermined level are stored in a storage device for secondary access, for example, a disk device.

As a result, in the address analysis method using natural language processing, etc., the input character string is divided into levels, and if the divided levels are equal to or higher than a predetermined level, the higher level that is frequently accessed is Is stored in a storage means such as a main storage device that can be accessed at high speed, and other information is stored in a disk device for secondary access.
For example, in the case of address information, if you are required to enter up to the prefectural / city / ward level, for the lower levels of the town letter level, use the already fixed prefecture / ward / ward level word (code) as a key. Since it is accessed, the number of times of access is not so large. That is, in the address word dictionary, the number of words at the prefecture / ward / ward level is smaller than the number of words at a lower level, so the number of access to the dictionary file is small even if the number of input characters increases. Thus, by storing only the higher-level dictionary in a medium that can be accessed at high speed, it is possible to reduce resources without reducing the access speed.

[0042]

Embodiments of the present invention will be described below with reference to the drawings. The embodiment described below will be described by taking an address analysis as an example, in which an address set is represented by a set of input words, and which hierarchy of each address display hierarchy is used for each word.

FIG. 5 is a block diagram of the word relation analyzing apparatus of the first embodiment of the present invention. The word relation analysis device shown in the figure includes a related information dictionary 1, an input / output unit 2 for exchanging data with an external device or a module, a control unit 3 for controlling the whole of the word relation analysis device, and a middle of a logical operation. A temporary storage unit 4 for temporarily storing the result, a related information storage device 5 for storing each word input from the input / output unit 2 and related information as a word set, and a related information storage device 5 controlled by the control unit 3. The logical operation unit 6 performs a logical operation such as a logical product such as a logical product or a logical sum with the data content of each other or with the data content of the related information storage device 5 and outputs the operation result.

FIG. 6 shows the contents of the related information dictionary having the field structure of the first embodiment of the present invention. The example shown in the figure shows the case where the prefecture is the basic category and the subordinate categories are the city / ward level large character / town level character / chome level. Prefectures in the basic category are 4
Corresponding to the bit position of 7 bits, the corresponding bit is "1"
It means that the word has the meaning of the prefecture name of the corresponding bit position.

In the example of the figure, each word is "city level".
The related information, “large-scale / town level” related information, and “character / chome level” related information indicate 47 prefectures 47
It is represented by bit information. As an example, "Tokyo"
Is "00" as bit information of "city level" related information.
11 …………… 01 ”and“ 1100 …………… 00 ”as the bit information of the“ large / town level ”related information, and“ 1100 ……… 00 ”as the bit information of the“ character / chome level ”related information. 0000 ……
……… has 01 ”.

That is, the case corresponding to the 3rd, 4th and 47th bits is "city level" with the name "Tokyo".
There is an address name of, and the prefectures corresponding to the first and second bits are:
It indicates that there is an address name of "Oaza / Machi level" called "Tokyo" (the same applies to the character / chome level).

As described above, in the related information dictionary, whether a word representing each address name exists in any prefecture in “city / ward level”, “town / large character level”, or “letter, chome level”,
It indicates whether or not to do so, and each "city / ward level", "town / large character level", or "character / chome level" corresponds to bit information (47 bits) for the number of prefectures.

The operation of the above-mentioned word relation analyzing apparatus will be described with reference to FIG. FIG. 7 is a flowchart showing the operation of the word relation analysis device of the first exemplary embodiment of the present invention. As a premise of the following processing, it is assumed that the word string input from the input / output unit 2 is divided into words by the morphological analysis function.

Step 101) When the divided words are input to the word relation analysis device, the control unit 3 refers to the related information dictionary 1 for each word and refers to the related information of each word, that is, each word. The meaning candidates are accumulated in the related information storage device 4. (Step 102) The control unit 3 points the first input word and stores the related information of the word in the temporary storage unit 4.

Step 103) The control unit 3 sets the contents of the temporary storage unit 4 as the related information in which the meaning candidate contents of the pointed word are reflected as the meaning candidate contents of the processing word. Transfer to. Step 104) The logical operation unit 6 performs logical operation processing (bit information and bit information indicating the order of words) specific to each application registered in the control unit 3 with respect to the related information of the word pointed to by the control unit 3. And the calculation result is accumulated in the temporary storage unit 4.

Step 105) The current related information of the word next to the word pointed by the control unit 3 and the contents of the temporary storage unit 4 are transferred to the logical operation unit 6, and the result of the logical product operation is again obtained. Transfer to the temporary storage unit 4. (Step 106) The control unit 3 advances the position of the pointed word by one.

Step 107) The control unit 3 checks whether or not the word currently pointed to is the last word in the word set. If it is the last word, the process proceeds to step 108, and if it is not the last word, , And shifts to step 103. Step 108) The logical operation unit 6 sequentially calls the contents of the temporary storage unit 4 (corresponding to the related information of the last word) and the related information of each word from the related information storage device 5, and the logical operation unit 6 calculates the logical product. After taking, the logical product result is overwritten in the related information storage area of each word in the related information storage device 5.

Step 109) The control unit 3 sequentially calls the related information of all the words stored in the related information storage device 5 by the control unit 3, and the temporary storage unit 4 is used as a buffer for storing the intermediate results. The arithmetic unit 6 calculates the logical product and stores the logical sum result in the related information storage area as a set of words in the related information storage device 5.

Thus, the input / output unit 2 outputs the related information candidates represented by the word string from the related information storage device 5 and the related information of each word at that time. FIG. 8 is a flowchart showing the processing unique to the application program according to the first embodiment of this invention.

(Step 201) The control section 3 sequentially selects words to be processed from the beginning of the input word string. (Step 202) The logical operation unit 6 takes the logical product of the dictionary related information of the current word and the result B of the previous word. At this time, when the current word is the first word, it is considered that the initial value of the result of the previous word is "ALL 1", and therefore the processing result is the same value as when nothing is done.

Step 203) The control section 3 judges whether or not the word currently pointed to by the pointer has a suffix, and branches the processing conditionally. If there is a suffix, the process proceeds to step 204, and if the word has no suffix, the process proceeds to step 205.

Step 204) If the word has a suffix, mask processing is performed based on the content of the suffix so that there is no relevant information corresponding to the current word. For example,
Masking is performed by setting bit information other than "city level" to "ALL0" based on the presence of suffixes "shi" and "ku". Hereinafter, the result of this masking will be described as the result A. That is, when the dictionary-related information is “1010,1111,1110” and the word has a suffix of “city level”, the word is “large letter / town level”, “letter / town level”. , "Character / Chome level" does not have any meaning, and is masked to be "1010,000,000".

Step 205) The logical sum of the contents of each block of the related information of the word pointed by the pointer is calculated. That is, the logical sum of 4 bits of "city level", 4 bits of "large character / town level" and 4 bits of "character / chome level" is obtained, and the result is explained as a result B. For example, if the logical sum of “1010”, “1111”, and “1110” is taken, the result B = “1111”. When the mask processing is performed as described above and "10,100,000", the result B is "1010".

Step 206) The control unit 3 judges whether the current word is the last word or not by referring to the pointer, and if it is the last word, the procedure goes to Step 207. If it is not the final word, the pointer is incremented and the process proceeds to step 201.

(Step 207) The logical operation unit 6 obtains the logical product of the result B of the final word and the result A of all the input words, and sets the result as the output result of each word. (Step 208) The logical operation unit 6 takes the logical product of the respective results B of all the input words and sets the result as the output result as a combination of words.

Hereinafter, the above-mentioned processing will be specifically described by showing each data flow of FIGS. 9 to 13. 9 to 13
Shows a data flow of word association analysis of the present invention. Figure 9
13 to FIG. 13, for simplification, it is assumed that there are originally 47 bits for 47 prefectures for all levels of “city level”, “large character / town level”, and “character / chome level”. This is an example shown with 4 bits (for 4 prefectures).

Hereinafter, the case where the input words are the word 1 "minatoku", the word 2 "roppongi", and the word 3 "tokyo" will be described as an example. In the data flow of FIG. 9, for example, referring to the related information dictionary 1 of FIG. 6 with these words, the address name word “Minato” exists in “Hokkaido, Iwate” at the city / ward level, and is in the large or small town. At the level, the related information described by the bit information indicating that it exists in "Hokkaido, Aomori, Iwate, ..., Okinawa" is obtained. The control unit 3 stores the related information in the related information storage device 5 (FIG. 7: step 102). Here, as shown in FIG. 9, the word pointer is set to the word 1 (first word), and the processing unique to the application in this embodiment is performed. When the suffix meaning is added and input, or when the control unit 3 determines that the suffix is a suffix (FIG. 8:
In step 203), the suffix "ku" exists such as word 1 "minatoku", and mask processing is performed for other levels so that only the "city-level" information in the dictionary-related information is valid. Then, the processing of dropping the information of "large character / town level" and "character / chome level" is performed (FIG. 8: step 204).

In other words, the "large character / town level" and the "character / chome level" portions are "0000" and "0000",
In the related information storage device 5, 1010,000,000 is stored.

Next, in the data flow of FIG.
The logical sum of the information at each level of “city / ward level”, “large character / town level”, and “character / chome level” is calculated. In this example, it becomes 1010, 1010, 1010, and the related information result in which the bit is in the prefecture in which the address name word may exist is temporarily stored in the temporary storage unit 4 (FIG. 7: step 104, FIG. 8: Step 20
5).

The current contents of the temporary storage unit 4 correspond to the result B of step 205 in FIG. 8, and as shown in FIG. 11, the contents of the temporary storage unit 4 "1010, 1010, 1010" and the next word 2 The logical product result "0000,0010,0010" with "Roppongi" is stored in the related information storage device 5 via the temporary storage unit 4 (FIG. 7: step 105, FIG. 8: step 202).

This accumulated information identifies candidates that should be the meanings of the word on the current word pointer (2) after checking the relationship of the word (1) before the current word pointer (2). The indicated bit is the result of only the related information, and corresponds to the result B of step 205 in the flowchart of FIG.

Next, the word pointer is advanced to word 2 (FIG. 7: step 106). Since word 2 has no suffix, the masking process is not performed and the process proceeds as it is. After that, this operation is repeated recursively, and when the word pointer points to the word 3 (that is, the last word to be processed) as shown in FIG. 12, the recursive processing is terminated (see FIG. 7: Step 107), the contents of the temporary storage unit 4 at the end of the process [corresponding to the related information of the word 3 (final word) “0010”] and the current related information of the word subjected to each process. By taking the logical product, the related information of each word as shown in the output result is obtained (FIG. 7: step 108, FIG. 8: step 207).

In the following, regarding the result of the final word, "city level", "town level", and "letter level" are all the same bit string, and all related information storage devices 1 on the head side where the processing is completed. The result of the logical product with the information stored in is shown. The output result (related information of word 1 is obtained by ANDing the related information “0010,0010,0010” of word 3 and the information “1010,000,0000” stored in the related information storage device of word 1. ) Becomes “0010, 0000, 0000”.

Related information of word 3 "0010,001"
The output result (related information of word 2) is “0000,0010” by logically ANDing the information “0000,0010,0010” stored in the related information storage device of word 2 with “0,0010”. , 0010 ”.

Related information of word 3 "0010,001"
The output result (related information of word 3) is obtained by logically ANDing 0,0010 ”and the information“ 010,000,000,000 ”stored in the related information storage device of word 3. , 0000 ".

Finally, as shown in FIG. 13, it is possible to take the logical product of the related information of all the words to obtain the related information as a word set and output it (FIG. 7: Step 109, FIG. 8: Step 208). This result shows that the word 1 “Minatoku” and the word 2 are displayed in the prefecture indicated by the 3rd bit of the word set related information “0010”
It indicates that both “Roppongi” and word 3 “Tokyo” exist, and related information of each word is, for example, word 1 “Minatok” is at “city level” and word 2 “Roppongi” is “ The word is "city / city level" or "city / city level" or "letter / chome level".
It indicates whether it exists in the address level of "town level" or "character / chome level".

That is, in the output result of FIG. 13, word 1
The output result of "Minatoku" is "010,000,000,000", and there is a bit in the "city level", and in other "large character / town level" or "character / chome level",
Since there are no bits, it means that the hierarchy (meaning) of "Minatoku" is not "large character / town level" or "character / chome level".

In the output result of FIG. 13, word 2
The output result of "Roppongi" is "0000,0010,0010", and there are bits at "Large / town level" and "Character / Chome level", so the level of "Roppongi" is "Large / town". It indicates that it is “level” or “character / chome level”.

Further, in the output result of FIG. 13, the output result of the word 3 "Tokyo" is "010,000,000000", and the "city level" has a bit. "Level" or "character / chome level",
Since there are no bits, it means that the hierarchy (meaning) of "Tokyo" is at "city level", not "large character / town level" or "character / chome level".

As described above, according to the present embodiment, like the address information, the related information dictionary is generated by limiting the number of significant meanings that are set in advance, and the prefecture name to which each input word belongs (47) and the number of levels to which the word belongs (city level) are combined to analyze the meaning of the input word. Therefore, "Minato"
Means the ward level when the prefecture name is “Tokyo”, and means the town level when the prefecture name is “XX prefecture” and “XX prefecture”. In this case, there are bits at multiple locations in the bit information. The meaning of a word is the hierarchy of prefectures, municipalities,
It consists of multiple categories that belong to which prefecture,
The input words have different meanings depending on the combination.

For example, ““ Minato ”,“ Tokyo ”,
The meaning (hierarchy) of "Minato" differs depending on whether it is entered in the "Roppongi" group and when it is entered in the "Minato", "Aomori", "Hachinohe" group. The bit information of each word is referred to and the judgment is made from the result of the logical operation.

Next, the case where there is an inclusion relation between semantic categories will be described. When performing inclusion check by word division, it is necessary to bring the address code into the word dictionary.
However, many address codes correspond to one word. For example, "Ioi" is 7 in cities and towns all over the country.
0 exists. So, to prevent the dictionary from becoming huge,
The address to which a certain word belongs is configured to have a bit string at the prefecture level. For example, the bit string indicates that "Ioi" exists in "Hyogo Prefecture", "Tokushima Prefecture" at the prefecture / city / district level, and "Hokkaido", "Yamagata Prefecture", "Kanagawa Prefecture" at the town / large character level. It has the following flags.

[0078]

[Table 2]

By having the address code as a flag, the amount of dictionary can be increased by 18 bytes per word (47 bits * 3). In addition, such a dictionary structure has the following advantages. The inclusion check can be easily realized simply by bit operation. Since the dictionary has information up to prefectures, the address table required for address candidate extraction can be divided into prefecture units. Therefore, the number of data to be searched is reduced and the table access becomes faster.

In the above example, the semantic categories are described as two categories, ie, a hierarchy of prefectures, cities, towns and villages, and which prefecture they belong to, but the present invention is not limited to this example, and there are n types of categories. Good. [Second Embodiment] Next, as a second embodiment of the present invention,
In order to efficiently store the dictionary contents in a device that can be accessed at high speed, high-level information that is frequently accessed is stored in the main storage device, and low-level information is stored in the disk device.

This uses the address word dictionary in order to deal with the replacement of the hierarchical order of the address word dictionary, the lack of hierarchical levels at the address level, etc. The address word dictionary includes prefectures, city / ward, large letters, towns, and The dictionary becomes very large because all words up to the letters and the cue must be stored. This makes it costly to store all of the dictionary contents in a device that can be accessed at high speed. Therefore, in the present invention,
This is dealt with by storing high-level words with high access frequency and a small number of words in the main storage device and other low-level words in the disk device.

FIG. 14 shows the prefecture of the second embodiment of the present invention.
An example of a city / ward level related information dictionary is shown. The figure shows a table in which connection suffixes and address codes are described for address name words at each prefecture / city / ward level. In this example, "Tokyo" indicates that the suffix "tou" is connected and the address code is "130000", and "Fuchiyu" is connected to the suffix "city" and the address code is It is shown that there are two locations, "130206" and "340302".

FIG. 15 shows an example of a town character level related information dictionary according to the second embodiment of the present invention. This figure is a table in which address codes for address name words at each town character level are described. In the present embodiment, both "Midori" and "Midoricho" have the address code "130203001".
Corresponding to "2", "Honmachi" is the address code "13020"
It is shown that there are two locations of "63030" and "3403020018".

FIG. 16 is a block diagram of the word relation analyzing apparatus of the second embodiment of the present invention. In the figure, the same parts as those in FIG. 5 are designated by the same reference numerals. The word analysis device shown in the figure includes a related information dictionary 1, an external device or an input / output unit 2 for exchanging data with a module, a control unit 3 for controlling the entire device for realizing a word relation analysis method, and a logical operation process. It is controlled by the temporary storage unit 4 for temporarily storing the result, each word input from the input / output unit 2, and the related information storage device 5 for storing related information as a word set, and the control unit 3, and is hierarchically coded. Further, the inclusion check unit 7 checks the inclusion relationship between the data contents of the related information storage device 5 or between the data contents of the partial storage unit 4 and the data contents of the related information storage device 5.

FIG. 17 is a flow chart showing the word relation analysis method of the second embodiment of the present invention. Step 301) First, a set of words input from the input / output unit 2 is divided for each word by referring to the prefectural city / ward level word dictionary of the related information dictionary 1, and the related information of each word is acquired. .

Step 302) The related information for each divided word (word division solution) is stored in the temporary storage unit 4. Step 303) The inclusion checking unit 7 causes the temporary storage unit 4
Mutual inclusion check is performed for each of the word division solutions with respect to the content stored in, and the related information storage device 5 obtains the word string from the beginning to the word position where the inclusion is established and the address code indicated by the word string. Transfer to.

Step 304) The correspondence between the word string for which the inclusion combined in Step 303 is established and its related information is set as one record, and the control unit 3 sets a record pointer in the first record. (Step 305) The control unit 3 divides the input word string subsequent to the word string of the record pointed to by the town character level word dictionary, and stores only the associated address code in the related information. Regarding the record which is overwritten in the device 5 and the inclusion is not established, the related information storage device 5
Delete the relevant record stored in.

Step 306) The position of the record pointed to by the control unit 3 is advanced by one. (Step 307) It is checked whether or not the record pointed to by the control unit 3 is the last record in the related information storage device 5, and if it is the last record, step 308.
If it is not the final record, go to step 30.
5, and the process is repeated until the record pointed to by the control unit 3 becomes the final record.

By such a method, it is possible to output the candidate of the related information represented by the set of words and the related information of each word at that time from the related information storage device 5 through the input / output unit 2. Next, an example of using the analyzed information as described above will be shown.

FIG. 18 shows an application example of the second embodiment of the present invention. First, as input conditions, words up to a predetermined level in the address hierarchy are always input as address information from the input / output unit 2. For example, it is defined as "Must input word at city / ward level". Also, the inversion of the solution at the top of the hierarchical level of the address is entered in the content. For example, there should be no reversal between prefecture level and city level, such as "Fuchu City Tokyo".

The dictionary structure is as shown in FIG.
The dictionary having the address code in the address table is divided into words up to a predetermined level in the address hierarchy (words up to the prefectural city level) and other words, and FIGS.
The prefecture city / ward address dictionary and the town character address dictionary are created as shown in.

In FIG. 18, a character string that matches a word in the prefecture / city / ward address dictionary 11 is searched from the front of the input input address character string according to the above input condition, and word division is performed. At that time, candidates are enumerated as city level from the front of the input address to a certain part. After that, the remaining input address character strings are divided into words from each of the candidates using the town address dictionary and the address code is compared with the city / ward level address code of the candidate. , To leave it as the final candidate. After that, in an actual application, narrowing down processing (intention understanding) is performed in order to specify how the candidate is shown to the user, and then output.

For example, in the input of "Tsushima", the following candidates can be considered in the prefecture / city / ward address dictionary. 1 Tsu / Shima → Tsu City + Shima 2 Tsu | Shi / Ma → Tsu City + Ma 3 Tsushima → Tsushima City Next, search for “Shima” using the street address dictionary 12, but under “Tsu City” Since there is neither a "sima" nor a "ma", in this case, it is decided as 3 "Tsushima City".

With the above dictionary construction, an address hierarchy having a relatively small capacity (for example, "prefectural city / ward level" which is a higher hierarchy)
By storing the word dictionary (prefecture city / ward address dictionary) up to a predetermined level in (1) in a high-speed device, efficient analysis can be speeded up. The input condition (1) is a natural input condition with practically no restrictions.

A specific example will be described below. FIG. 19 is a diagram showing a specific example of the second exemplary embodiment of the present invention. In the figure,
In the dictionary generation unit 8, the address information input from the input / output unit 2 is coded, sorted into a higher level and a lower level, and the higher-level address code is registered in the prefectural city / district address dictionary of the main storage device 9. , Lower-level addresses are registered in the street address dictionary of the disk device 11. "11, Honmachi, Fuchu-shi, Tokyo""1-1-1, Roppongi, Minato-ku, Tokyo""20 Midori-cho, Musashino-shi, Tokyo""40 Honcho, Mishima-shi, Shizuoka""2 Shimanouchi, Chuo-ku, Osaka-shi, Osaka" was entered To do. Here, the following address code is acquired with reference to the related information dictionary 1. "130000/130206/13020300
12 "" 130000/136000/13600011
11 "" 130000/131000/13020300
12 "" 100000/11090/111009008
00 ”“ 7000000/700100/70011001
00 ”The address codes 1 to 3 above are stored in the temporary storage unit 4.
Since these address codes are entered from the higher order to the lower order, the first address code is "130000 /
130206 ", is" 130000/13600
"0", "130000/131000", "1"
"00000/11090" is "700000/7"
"00100" is registered in the memory of the main storage device 9 as a prefecture / city / ward address dictionary.

The address code "13020300
12 ”,“ 13600100111 ”, and“ 130
2030012 ”, is“ 1100900800 ”,
Registers "7001100100" as a town character address dictionary in the disk device 11. In this way, by generating an address dictionary from the input address data, for example,
Taking the address input when inquiring about a telephone number as an example, the number of input characters is actually 10 to 16 in most cases. Also,
There may be a long character string of 30 to 40 characters, and as in the above example, the coded address data is, for example,
When the telephone number is guided by dividing the prefecture into the city and the town and dividing into the dictionary in the main storage device 9 and the dictionary in the disk device 11, the part related to the prefecture and city is the main access speed is high. The storage device 9 is accessed, and the disk device 11 is accessed for the part related to the town letter. In other words, by constructing the word dictionary of the relatively small size in the upper hierarchy with a memory having a short access time, the cost increase is small, and even if the number of characters is large, a storage device having a relatively long access time such as a disk device is used. It is possible to reduce the number of accesses to the word and reduce the word analysis time.

The present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the claims, and are applicable to information services such as editing of address information and inquiries of place names and addresses. Widely applicable.

[0098]

As described above, according to the word relation analyzing apparatus and the word relation analyzing method of the present invention, a combination of words is represented in an arbitrary word string based on bitwise related information of each word. The meaning candidate and the meaning of the word can be realized by simply performing a logical operation without designating information as to what meaning each word uses.

The address table is about 40 including characters.
The word dictionary has 160,000 words. Inquiry example 10
Analysis of 000 shows that the average number of word divisions is 2
As shown in FIG. 20, 99% falls within 8 cases. Since the average number of input characters is eight, about 200 solution candidates are output without inclusion check, so that a filtering effect of about 1/100 can be obtained.

Therefore, according to the present invention, it is possible to perform the analysis processing of word relations at high speed with a simple logic. In addition, FIG. 21 is a diagram for explaining the effect of the second embodiment of the present invention. In the figure, under the condition that the same small amount of memory (high-speed medium) can be used for the address language analysis and the high-speed address analysis, the vertical axis indicates the number of dictionary file accesses and the horizontal axis indicates the input. Indicates the number of characters. The dotted line a shows the number of dictionary file accesses when the address language analysis is performed, and the solid line b shows the file access number when the high speed address analysis is performed. a: Address language analysis ... Register only the following key index in the memory: Register the address dictionary in the file of the disk device: b: High-speed address analysis ... Register only the prefecture / city / ward address dictionary in the memory (corresponding to the upper basic category): Town The character address dictionary is the simulation result when it is registered in the file of the disk device.

The longer the input character string corresponding to the word string to be analyzed, the more the dictionary file access frequency increases in the address language analysis. However, in high-speed address analysis, the number of dictionary file accesses does not increase so much even if the input character string becomes long, and the number of dictionary file accesses within the range of the number of input characters for address input can be ignored. It is possible to analyze the address with.

As described above, according to the present invention, in an arbitrary word string, based on the hierarchically encoded related information of each word, the candidate of the meaning represented by the combination of words and the meaning of each word at that time are represented. Even if is input without specifying what meaning each word is used for, it is possible to perform analysis with a small amount of resources, at high speed, and with a simple hierarchical inclusion check.

[Brief description of drawings]

FIG. 1 is a first principle configuration diagram of the present invention.

FIG. 2 is a second principle configuration diagram of the present invention.

FIG. 3 is a diagram illustrating the first principle of the present invention.

FIG. 4 is a diagram illustrating a second principle of the present invention.

FIG. 5 is a block diagram of a word relation analysis device according to the first embodiment of this invention.

FIG. 6 is a diagram showing an example of a related information dictionary according to the first embodiment of this invention.

FIG. 7 is a flowchart showing an operation of the word relation analysis device of the first exemplary embodiment of the present invention.

FIG. 8 is a flowchart showing processing unique to an application program according to the first embodiment of this invention.

FIG. 9 is a data flow (No. 1) of the word relation analysis according to the first embodiment of this invention.

FIG. 10 is a data flow (No. 2) of the word relation analysis according to the first embodiment of this invention.

FIG. 11 is a data flow (No. 3) of the word relation analysis according to the first embodiment of this invention.

FIG. 12 is a data flow (No. 4) of the word relation analysis according to the first embodiment of this invention.

FIG. 13 is a data flow (No. 5) of the word relation analysis according to the first embodiment of this invention.

FIG. 14 is a diagram showing an example of a prefecture / ward / ward-level address dictionary according to the second embodiment of the present invention.

FIG. 15 is a diagram showing an example of a street character level address dictionary according to a second embodiment of the present invention.

FIG. 16 is a block diagram of a word relation analysis device of a second exemplary embodiment of the present invention.

FIG. 17 is a flowchart showing a word relation analysis method according to the second embodiment of the present invention.

FIG. 18 is a diagram showing a usage example of the second exemplary embodiment of the present invention.

FIG. 19 is a diagram showing a specific example of the second exemplary embodiment of the present invention.

FIG. 20 is a diagram for explaining the first effect of the present invention.

FIG. 21 is a diagram for explaining the second effect of the present invention.

FIG. 22 is a block diagram of a conventional word analysis system.

[Explanation of symbols]

 1 Related information dictionary 2 Input / output unit 3 Control unit 4 Temporary storage unit 5 Related information storage device 6 Logical operation unit 7 Encapsulation check unit 8 Dictionary generation unit 9 Main storage device 10 Disk device 11 Prefectural city / district address dictionary 12 Town address Dictionary 100 Related information dictionary 200 Analysis means 210 Related information acquisition means 211 Masking means 212 Second analysis means 220 Related information output means

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location 8420-5L 15/38 M

Claims (19)

[Claims] 1. A word relation analysis device that analyzes the meaning of an input word and the meaning of a set of said words, and expresses the meaning of said word by displaying the meaning of each dependent meaning category for each meaning of a basic meaning category. Associated information dictionary for storing the stored information, and analysis means for analyzing the meaning of a word having a meaning limited to a significant number and the meaning of a word string consisting of the word by referring to the related information dictionary. And a word relation analysis device. 2. The related information dictionary holds related information for each word as bit information.
The described word relation analysis device. 3. The analysis means recursively repeats a logical operation for the number of input words by using the related information in which the meaning of the selected word is bitized by referring to the related information dictionary. When the selected word is the final word of the input word string, the related information acquisition means to be acquired and the related information of the final word and all the words before the final word are logically operated and input. 2. The word relation analysis device according to claim 1, further comprising: related information output means for outputting related information of the entire set of the selected words. 4. The related information acquiring means, based on the content of the suffix, when the input word includes a suffix indicating the meaning / content of the meaning category for one of the meaning categories. 4. The word relation analysis apparatus according to claim 3, further comprising masking means for masking related information of another semantic category. 5. A word relation analysis device for analyzing the meaning of an input word and the meaning of a set of said words, and displaying the meaning of each word by displaying the meaning in a subordinate meaning category for each basic meaning category. And a first analysis means for analyzing the meaning of a word having a meaning limited to a significant number and the meaning of a word string consisting of the word, by referring to the related information dictionary. A word relation analysis device characterized by the above. 6. The word relation analysis apparatus according to claim 5, wherein the related information dictionary holds the related information for each word as coded information. 7. The first analyzing means refers to the related information dictionary for subordinate meaning categories of words selected in a hierarchical order of basic meaning categories having an order, and refers to the related information dictionary to obtain higher ranks by using the related information. The related information acquisition means that interprets the dependent meaning category within the range of the basic meaning category of, and repeatedly recursively acquires the interpretation of the next basic meaning category within the range of the interpreted dependent meaning category, 6. The word relation analysis device according to claim 5, further comprising: related information output means for outputting related information of the entire set of words. 8. The first analysis means, when the input word includes, for one of the meaning categories, a suffix indicating the meaning / content of the meaning category, the content of the suffix is added. 8. The word relation analysis apparatus according to claim 7, further comprising a second analysis means for determining a basic meaning category based on the basic meaning category and analyzing the basic meaning category within the range. 9. A storage order determining means for determining a storage order in descending order from the highest level to a lower level based on the level of the meaning category of the word, and the highest order among the orders determined by the storage order determining means. Firstly, words from the upper level to a predetermined level are stored in a first related information dictionary in a first storage means having a high access speed.
Storage means, and second storage means for storing words having a predetermined level or lower in a second related information dictionary in a second storage means having an access speed slower than that of the first storage means. Item 5
The described word relation analysis device. 10. The first storage means is a main storage device, and the second storage means is a disk device.
The described word relation analysis device. 11. A word analysis method for analyzing the meaning of an input word or the meaning of a set of said words, using relevant information for displaying the meaning of a word by a plurality of meaning categories, and limiting the number to a significant number. A word relation analysis method, characterized by analyzing the meaning of a word having a meaning and a word string composed of the word. 12. The related information of each word is held as bit information, the logical operation of the bit information between the words is performed, the operation result of the logical operation is held for each word, and the bit of all words is held. The word relation analysis method according to claim 11, wherein a logical operation is recursively performed on the information. 13. The word constituted by the bit information of the related information is given to each word of the word string, and the bit information of the first word is based on a certain semantic category and the logic of the bit information of another semantic category is given. Performs an arithmetic operation, sequentially with the result of the logical operation of the first word, and recursively repeats the logical operation of the bit information of the next word, and the logical operation of the logical operation result of the bit information of the last word and the bit information of all words. 12. The word relationship analysis method according to claim 11, wherein the meaning of each word in the word string is specified. 14. The meaning category of the input word is 1
12. The word relation analysis method according to claim 11, wherein when a suffix is included for one, mask processing of related information of another semantic category is performed based on the content of the suffix. 15. A word relation analysis method for analyzing the meaning of an input word or the meaning of a set of said words, wherein information representing the meaning of a word is displayed for each basic meaning category by displaying the meaning in a subordinate meaning category. And a related information dictionary stored in, and analyzing the meaning of a word having a meaning limited to a significant number and the meaning of a word string consisting of the word by referring to the related information dictionary. Method. 16. The word relation analysis method according to claim 15, wherein the related information for each word is analyzed with reference to the related information dictionary which is coded information. 17. When performing an analysis, the subordinate semantic categories of words selected in order of basic semantic categories having an order are referred to the related information dictionary, and higher-order basic meanings are obtained by using the encoded related information. Interpret the dependent semantic category in the range of the category, recursively iteratively obtain the interpretation of the next basic semantic category in the range of the interpreted dependent semantic category, and obtain the related information of the entire set of input words. The word relation analysis method according to claim 15, which outputs the word relation. 18. When interpreting the basic meaning category, if one of the meaning categories includes a suffix indicating the meaning / content of the meaning category in the input word, the suffix of the meaning category is input. 18. The word relation analysis method according to claim 17, wherein the basic meaning category is determined based on the content, and analysis is performed within the range of the basic meaning category. 19. The order of storing in descending order from the highest level to the lower level is determined based on the level of the meaning category of the word, and the access speed of words from the highest level to a predetermined level in the determined order. 12. The word relation analysis method according to claim 11, wherein the word relation analysis method stores the words in the first storage means having a high speed, and stores the words having a level equal to or lower than the predetermined level in the second storage means having an access speed slower than that of the first storage means.