JP3442422B2 - Synonym information creation apparatus and method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is to extract synonym information such as synonymous relations and superior-subordinate relations of a word or compound by analyzing natural language sentences in a document registered in a document database. And a document search device that performs document search with high accuracy using the synonym information created by the synonym information creation device.

[0002]

2. Description of the Related Art Conventionally, attempts have been made to improve the accuracy of document retrieval by using synonyms, superordinate terms, and subordinate terms. However, in the conventional document search device using synonym information, all synonym information must be prepared and input manually. Since the synonym information strongly depends on the words used in the document stored in the document database, it is impossible to prepare general information in advance. For this reason, it is necessary to prepare synonym information for each document database to be handled, and it takes a great deal of time and effort to create synonym information for realizing a document retrieval device that accurately retrieves documents.

As one of the methods for creating synonym information,
Attempts have been made to create a thesaurus from a Japanese language dictionary (Information Processing Society of Japan, Natural Language Processing Research Group, 83-1).
6). This document describes a method of extracting the upper-lower relation between words from the word meaning sentence of the Japanese dictionary and automatically constructing a thesaurus by this. But,
As described above, the word required as synonym information depends on the document database to be searched. For this reason, even if a general-purpose thesaurus is created, it is inevitable that the utility value of the search device is low.

An apparatus for automatically extracting synonym information from a sentence by using a method similar to the method for constructing a thesaurus from the word meaning sentence in the above dictionary has been disclosed (JP-A-4-123264). Gazette). In this device, when a syntax pattern expressing a synonymous relationship registered in advance is matched with an input sentence, a synonymous relation is automatically given between corresponding words in the collated sentence. Then, a document is searched using a synonym dictionary that stores the information of the synonymous relationship.

However, it is impossible to extract the correct synonym information completely automatically by the technique of natural language analysis. Therefore, when the synonym information is extracted by the automatic processing, an incorrect synonymous relationship or a word having an upper-lower relationship may be extracted. When a search is performed using a synonym dictionary that stores such synonym information, an incorrect document will be searched. There was a problem in that the trouble of removing the wrong document would rather increase, and there was a practical problem.

[0006]

The method of manually inputting all synonym information requires a large labor cost and is not realistic. In addition, even in the method of creating synonym information from the word meaning sentence of the dictionary, it is difficult to create synonym information useful for users as a general-purpose one. It has been desired to realize a device that makes the creation of a device efficient. in addition,
The synonym information automatically extracted from the natural language sentence contains an error and cannot be automatically used as synonym information.

According to the present invention, synonym information for extracting synonym information candidates from a document written in natural language, correcting / adding based on the candidates, and smoothly creating a synonym dictionary for document retrieval. Provide a producing device.

[0008]

In order to achieve the above object, in the synonym information creating apparatus according to the present invention, synonym information candidates such as synonym relations, words having a superordinate-subordinate relation and compound words are naturally generated. Means for extracting from the sentence, means for displaying the extracted synonym information candidates, means for correcting / adding the displayed synonym information candidates, synonym dictionary for the corrected / added synonym information candidates And a means for covering a matching with a similar expression corresponding to the sentence by inputting a representative example of a natural language sentence expressing a synonymous relationship.

In addition, the synonym information creation device according to the present invention has means for detecting an unregistered word and inputs synonym information for an unregistered word even if a synonym information candidate is not extracted. Encourage.

Further, in order to enable highly accurate extraction of synonym information candidates, there is provided means for registering an exceptional sentence pattern with respect to the extraction rule, and when a certain sentence is matched with the exceptional sentence pattern. , Synonym information is not extracted from the sentence even if it is matched with the corresponding extraction rule.

[0011]

In this way, the user can confirm the corrections and additions after extracting the candidates for the synonym information from the document, so that it is easy to create the synonym dictionary to enable highly accurate document search. Will be able to create.

[0012]

EXAMPLE First, a first example will be described. An embodiment of the present invention will be described with reference to the drawings. FIG. 1 illustrates a device configuration diagram for configuring an embodiment of the present invention.

Input means 101 such as a keyboard including a pointing device such as a mouse, output means 103 such as a CRT or bitmap display, storage means 104 such as a semiconductor memory, a magnetic disk or an optical disk, a sentence analysis and a synonym information extraction process. The processing means 102 for performing

FIG. 2 is a functional block diagram. A document data storage unit 205 that stores document data, a sentence analysis unit 201 that performs morphological analysis and syntax analysis for each sentence, and an unnecessary word dictionary 206 that registers words that do not need to be extracted as original words or synonyms of synonym information. , An extraction rule storage unit 208 for storing a syntactic pattern for extracting synonym information,
According to the information stored in the extraction rule storage unit 208 and the unnecessary word dictionary, the syntactic structure input from the sentence analysis unit 201 is analyzed to extract synonym candidate, and the synonym candidate storage unit 209.
Displaying the candidates stored in the synonym candidate extraction unit 202 and the synonym candidate storage unit 209 on the display, adding / correcting the synonym information, and storing the result in the synonym dictionary 207. -Correction unit 203, extraction rule input unit 204 for inputting extraction rules to be stored in extraction rule storage unit 208, synonym candidate storage unit 20 for storing synonym candidates
9, a synonym dictionary 207 that stores the confirmed synonyms, and a control unit 210 that controls the whole.

The document data storage unit 205 in FIG.
The unnecessary word dictionary 206, the extraction rule storage unit 208, the synonym dictionary 207, and the synonym candidate storage unit 209 correspond to the storage unit 104 in FIG. Further, the sentence analysis unit 201, the synonym candidate extraction unit 202, the display / correction unit 203, the extraction rule storage unit 208, and the control unit 210 are the processing means 10 in FIG.
It is executed in 2.

The processing of the sentence analysis unit 201 may use the techniques of morphological analysis and syntactic analysis used in commercially available machine translation systems and the like. FIG. 3 illustrates a processing flow of the control unit 210. The control unit 210 is an element module that controls the entire apparatus. The control unit 210 waits for an input from the keyboard or mouse, and when an input is made, performs processing according to the input code. That is,
When the “extract” command is input, the extraction process of FIG. 4A is started. When the "rule input" command is input, the rule input process of FIG. 4B is started. When the "display / correction" command is input, the display / correction process of FIG. 4C is started. Further, when the "end" command is input, the entire processing is ended.

When the extraction process is started, the document data storage unit 2
Sentence analysis unit 201 for the sentences stored in 05
Performs parsing by invoking. Next, the synonym candidate extraction unit 202 is activated with the obtained syntactic analysis result as an input, and synonym candidate extraction is performed.

FIG. 5 shows the flow of processing of the synonym candidate extraction section. The synonym candidate extraction unit 202 checks whether or not the input syntax pattern matches the extraction rule stored in the extraction rule storage unit 208.
The three sets of (original language, synonym candidate, relationship) are stored in the synonym candidate storage unit 209. Note that if the original language or synonym candidates are registered in the unnecessary word dictionary 206, or if the retrieved pair of original language and synonym is already stored in the synonym candidate storage unit 209, the synonym candidate storage is performed. It is not stored in the unit 209.

FIG. 6 shows an example of the format of the unnecessary word dictionary 206 and its stored contents. The unnecessary word dictionary 206 is
This is a dictionary that registers words that cannot be keywords for search. As shown, it is composed of a headline and a part of speech. In the illustrated example, for example, “thing” or “thing”
It stores the content indicating that nouns such as are unnecessary as keywords.

FIG. 7 shows an example of the format of the extraction rule storage unit 208 and the contents stored therein. Extraction rule storage unit 2
08 is a syntax pattern for extracting synonym information,
The relationship between the words extracted by the syntactic pattern is stored. The rule 1 in the illustrated example is "A1 is A2
Is a kind of. The syntax pattern for the sentence
The relation expressed by the syntax pattern indicates that it is a superordinate relation. Here, "A1" and "A2" have special meanings, and respectively represent an original word and a synonym (including a high-order word and a low-order word) for the original word.

Further, "ga", "dk", and "no" are
Represents a case in syntax. FIG. 8 illustrates an example in which the extraction process is performed on a certain sentence. Sentence analysis is performed on the input sentence in step 302 shown in FIG. Next, in step 303, the synonym candidate extraction unit 202 is activated.

The synonym candidate extraction unit 202 collates with the extraction rule stored in the extraction rule storage unit 208 according to the processing flow shown in FIG. In the example shown in FIG.
Rule 2 matches. The words to be matched with “A1” and “A2” in the extraction rule are “CD” and “negotiable deposit”, respectively. Therefore, the triplet stored in the synonym candidate storage unit is (CD, negotiable deposit, synonym).

FIG. 9 shows an example of the format of the synonym candidate storage unit and its stored contents. The synonym candidate storage unit stores at least a triplet (original language, synonym, relation), but in the illustrated example, the identifier of the rule used for matching is also used as information at the time of matching. Is stored.

The display / correction unit 203 is shown in FIGS.
The process flow of is illustrated. Also, an example of display and input is shown in FIG.
It is shown in FIG. In FIG. 13B, an area 1001 displays an original word, an area 1002 displays an upper word, and an area 1003 displays a synonym.
The area 004 is an area for displaying a lower word. Also, 1
Buttons for inputting commands to the display / correction unit are arranged in the area 005. Although FIG. 13 does not display the types such as the upper word, the lower word, and the synonyms, these types may be displayed.

After being activated, the display / correction unit 203 displays a list of the original word lists stored in the synonym candidate storage unit 209 (FIG. 13A). By selecting an original word from the displayed list display, synonym information corresponding to the original word is displayed (FIG. 13B).

When the synonym information is displayed, FIG.
According to the menu displayed in (b), it waits for any of the "correction" command, "next" command, "store" command, "related word" command, and "end" command to be input.

In the case of the "correct" command, step 90 in FIG.
In step 6, the correction process is executed. The correction process is a subroutine after step 915 in FIGS. 11 and 12.
In the case of the "next" command, synonym information is displayed for the original language that has not been confirmed and corrected.

In the case of the "store" command, the currently displayed synonym information is stored in the synonym dictionary and deleted from the synonym candidate storage unit 209. In the case of the “related word” command, all words related to the word included in the synonym information currently being confirmed and modified (words existing in other synonym information) are stored in the synonym candidate storage unit 209 and the synonym dictionary 207. A search is performed and a list of these words is displayed (FIG. 13 (d)). FIG.
In (d), “→” represents a relationship from a lower word to a higher word, and “=” represents a synonymous relationship.

In the case of the "end" command, the display of the synonym information is erased, and the processing of the display / correction unit ends. In the case of the "correction" command, step 906 performs the processing of step 915 and thereafter. Here, individual words displayed as synonym information are to be processed. Therefore, the word to be corrected is managed by a variable called "correction pointer", and the word pointed by the pointer is highlighted.

The input code input by the user to move the correction target corresponds to a predetermined "next word" command, "previous word" command, "next relation" command, and "prerelation" command. In the case of code, after changing the value of the correction pointer corresponding to each code, the position of highlighting is changed in step 916.

The "next-relationship" here means a synonym field if the correction pointer points to a word in the higher-order field, and a lower order if it points to a word in the synonym field. This is an instruction to change the word pointed by the correction pointer to the field of the upper word if it points to the field of the lower word in the field of the word.

If the correction pointer points to the last word of the lower word, or if the "next word" command is input,
It moves cyclically such that the correction pointer moves to the first word of the broader terms.

When the "correct word" command is input, the process shown in FIG.
In step 927 of step 2, the input for correcting the word is accepted. That is, as shown in FIG. 13C, the cursor is displayed to accept the correction input.

When the "add word" command is input, processing for adding a word is performed in step 929.
For example, when the position of the correction pointer immediately before in FIG. 13 is “deposit”, as shown in FIG. 14F, when the position of the previous pointer is “negotiable deposit”, as shown in FIG. Allocates an area and displays the cursor.

When the "delete" command is input, the word pointed by the correction pointer is deleted. Note that FIG. 13 (e)
Indicates that there is no synonym candidate for the target relationship (lower term in the figure). In this case,
When the "word addition" command is input, the cursor is displayed at the same position and the word input is accepted.

In the display / correction unit 203 of this embodiment, the original word list is displayed first, and the user selects the original word to be corrected from the list.
It is also possible to modify so as to display the synonym information about the first original word stored in 9.

FIG. 15 shows an example of the format of the synonym dictionary 207 and its stored contents. The synonym dictionary is composed of original word headings, relationships, and synonyms. In the example of FIG. 15, for example, the word “deposit” is a superordinate word to the original word “CD”, and “negotiable deposit”. Is a synonym, indicating that there is no word corresponding to a subordinate word.

FIG. 16 shows the flow of processing of the extraction rule input unit 204. Further, FIG. 18 shows the display and an input example. The extraction rule input unit 204 stores and manages the input syntax patterns and relationships in the syntax pattern temporary storage and the relationship temporary storage, respectively.

In the extraction rule input section 204, the commands ("input" command, "relational input" command, "store") entered by the user are input.
Command, or "end" command).

In the case of "input" instruction, step 1 of FIG.
The processing of 204 and below is performed (FIG. 18A). That is,
After receiving the user's input (FIG. 18B) and performing sentence analysis on the input sentence, the syntax pattern of the analysis result is stored in the syntax pattern temporary storage. However, when the character string "A1" meaning the original word or the character string "A2" meaning the synonym does not exist in the input sentence, the warning is displayed (FIG. 18 (d)).

When any of "upper", "synonymous", and "lower" is input as the "relationship input", it is stored in the relational temporary storage and displayed as shown in FIG. 18 (c). When the "store" command is input, and when each data is stored in the syntax pattern temporary storage or the relational temporary storage, the data is stored in the extraction rule storage unit 208. If no data is set in the temporary storage, see FIG.
A warning is displayed as in (e).

FIG. 17 illustrates an example in which the sentence input by the extraction rule input unit 204 is converted into a rule. In the input sentence, the character strings “A1” and “A2” mean the original language and the synonyms, respectively.

As described above, in the present embodiment, a device is provided which can easily confirm / correct / add information of a synonym candidate extracted from the sentence in the document stored in the document data storage unit 205. It becomes possible to do.

Next, the second embodiment will be described. In the second embodiment, processing of unregistered words is performed by the synonym candidate extraction unit 202.
Add to. Generally, in document retrieval, words that are important keywords are often not words prepared for general purposes. For example, when considering the search for newspaper articles, proper nouns are important as keywords, and when specified in articles on new product launch, product names are important as keywords. It is impossible to prepare such words in advance. Therefore, these words are treated as unregistered words in morphological analysis and syntactic analysis.

For these unregistered words, it is important to enter synonym information. FIG. 19 illustrates a processing flow of the synonym candidate extraction unit 202 according to the second embodiment. First, it is determined whether or not there is an unregistered word in the sentence input in step 1501, and if it does not exist,
In step 1506, the processing illustrated in FIG. 5 is performed.

Next, when an unregistered word is used as the original language, it is determined whether or not to match the extraction rule (step 1502).
When the original word or the synonym candidate does not exist in the unnecessary word dictionary, it is stored as a synonym candidate storage unit 209 in a triplet (original word, synonym candidate, relation) ( Steps 1503, 150
4, 1505).

If the extraction rule is not matched, in step 1507 (original word, <null>, <nu>
11>) are stored in the synonym candidate storage unit. The other processing units are exactly the same as those in the first embodiment, and detailed description thereof will be omitted.

As described above, in the second embodiment, since the unregistered word is also stored in the synonym candidate storage unit 209, it is displayed / displayed.
The correction unit 303 can perform the same processing as in the first embodiment.

It is also possible to replace unregistered word extraction with important word extraction. For the important word extraction, the existing process disclosed conventionally (The Institute of Electronics, Information and Communication Engineers, DI, Vol. J74-DI, No. 8) may be used. That is, the document data storage 2
The sentence stored in 05 is analyzed to determine the important word,
Create an important word table.

In the synonym candidate extraction unit 202, when there is an important word included in the important word table in the input sentence,
It can be modified to perform the same processing as in FIG.

Next, a third embodiment will be described. In the first embodiment, the synonym information is extracted based on the explicit extraction rule, but it is difficult to express the syntactic pattern of the extraction rule only explicitly. Therefore, the third embodiment is modified so that two types of rules can be described as negative rules. One is the type that expresses the exception of the syntax pattern, and the other is
One is a type that means that it is not taken out as a synonymous relationship when it is matched with the syntactic pattern.

FIG. 20 illustrates an example of storage for exception rules. Rule 15 is a description example of the exception rule. "N
The "ull-synonymous" relationship means that, when matching a syntactic pattern, it matches the corresponding rule (rule 2), but invalidates the matching. That is, "an electronic dictionary is an electronic dictionary. The sentence "not a dictionary" means "A
This is a rule that means that "1 is A2" is matched with the syntactic pattern of the sentence, but "electronic dictionary" and "electronic dictionary" are not taken out as synonyms.

This type of rule is valid only for the corresponding rule. Rule 16 is a description example of the invalidation rule. The relationship "not-synonymous" means that the combination of words does not have a synonymous relationship when matching a syntactic pattern. In this example, "A
1 is different from A2. It means that no synonym relationship is set between words expressed as "."

When matching with this rule, each data is stored in the invalid word relation table (consisting of three groups of (original word, synonym, relation)). Then, the words matched with this set are not extracted as synonym candidates thereafter.

The process for the exception rule is added to the synonym candidate extraction unit 202. The process flow is shown in FIG.
In this embodiment, when the input sentence is matched with the invalidation rule and the exception rule, the information is used in the subsequent extraction processing. However, when the input sentence is matched with the exception rule or the invalidation rule, the synonym candidate storage unit 20
It is easy to scan 9 to find corresponding synonym information, and modify them so as to delete them from the synonym candidate storage unit 209. Furthermore, by newly providing a field for storing a flag indicating that deletion should be performed for the triple of (original language, synonym, relation), it is possible to interactively confirm deletion without deleting all at once. As described above, it is easy to change the display / correction unit.

Next, a fourth embodiment will be described. In the fourth embodiment, a process of applying the synonym information extracted by the synonym candidate extraction unit to other original words is added. FIG. 22 illustrates a processing flow of the synonym candidate extraction unit 202 according to the present exemplary embodiment. The difference from FIG. 19 which illustrates the processing flow of the synonym candidate extraction unit of the second embodiment is that steps 1808 and 1
809, and other steps (1801 to 180)
7) is exactly the same as steps (1501 to 1507) in FIG.

If the input sentence does not match the extraction rule, in step 1808, the matching degree of the character string between the unregistered word to be processed and all the original words stored in the synonym candidate storage unit 209 or the synonym dictionary 207. To calculate. If the matching degree exceeds a predetermined threshold value and there is an original word, in step 1809, the synonym information (upper word, lower word, synonym) given to the original word is read and becomes a processing target. The synonym candidate storage unit 209 stores the synonym information of the unregistered word.

For the calculation of the degree of coincidence, for example, the following formula is used. S (A, B) = C (A, B) / Max (A, B) (1) In the above formula, A and B are words for measuring the degree of coincidence, and M
ax (A, B) is the character string length on the side of the long character string in the word A or B, and C (A, B) is the character string length in which A and B match. For example, A = “J3100”, B = “J310”
When 0ZD "is set, the calculation can be performed as follows.

S (“J3100”, “J3100ZD”) = 5/7 =
0.7 (2) If this value is higher than a predetermined threshold value, step 1809 is performed. For example, the synonym candidate storage unit 20
When the information ("J3100", "personal computer", upper level) is stored in 9, the new word "J3100"
Assuming that 3100ZD ”is input, synonym information (“ personal computer ”) of the original word (“ J3100 ”) with a high degree of matching of the character strings stored in the synonym candidate storage unit 209
Read the upper level, and from now on (“J3100ZD”,
“Personal computer, upper” is newly added as a synonym candidate storage unit 209.
To store.

Needless to say, the equation (1) can be modified in various ways as a measure for measuring the matching of character strings. The first to fourth embodiments have a means for inputting the extraction rule. However, it is possible to store the extraction rules in advance and configure the device without the extraction rule input unit 204. That is, by preparing the extraction rules that can be prepared for general use, the user's labor can be reduced.

Further, a simpler structure can be achieved by adopting a structure that does not have the unnecessary word dictionary 206. Further, in the described embodiment, the synonym candidate storage unit 2
09 and the synonym dictionary 207 are separately configured, but they can be configured as one. In this case, it is necessary to be able to add information as to whether or not confirmation / correction / addition is completed for each original word and synonym information.

In the above embodiment, the synonym dictionary 207 is 1
However, assuming that a plurality of users use a common dictionary, a plurality of synonym dictionaries 207 may be provided depending on the users.

[0063]

According to the present invention, it is possible to provide an apparatus for interactively constructing a synonym dictionary while taking out information of synonym candidates from a document stored as document data and performing confirmation / correction / addition. . With this device, it becomes possible to easily create a highly accurate synonym dictionary that cannot be obtained by automatic extraction.

[Brief description of drawings]

FIG. 1 is a diagram showing a device configuration.

FIG. 2 is a diagram showing a functional configuration.

FIG. 3 is a flowchart showing a processing flow of a control unit.

FIG. 4 is a flow diagram showing extraction processing, rule input processing, display / correction processing.

FIG. 5 is a flowchart showing a processing flow of a synonym candidate extraction unit.

FIG. 6 is a diagram showing an example of an unnecessary word dictionary.

FIG. 7 is a diagram showing an example of an extraction rule storage unit.

FIG. 8 is a diagram showing an example of extraction processing.

FIG. 9 is a diagram showing an example of a synonym candidate storage unit.

FIG. 10 is a flowchart showing a processing flow of a display / correction unit.

FIG. 11 is a flowchart showing the flow of correction processing execution.

FIG. 12 is a flowchart showing the flow of correction processing execution.

FIG. 13 is a diagram showing a display example of a display / correction unit.

FIG. 14 is a diagram showing a display example of a display / correction unit.

FIG. 15 is a diagram showing an example of a synonym dictionary.

FIG. 16 is a diagram showing an example of an extraction rule input unit.

FIG. 17 is a diagram showing an analysis example of an extraction rule input unit.

FIG. 18 is a diagram showing a display example of an extraction rule input unit.

FIG. 19 is a flowchart showing a processing flow of a synonym candidate extraction unit of the second embodiment.

FIG. 20 is a flowchart showing an example of storage of exception rules and invalidation rules.

FIG. 21 is a flowchart showing a processing flow of a synonym candidate extraction unit of the third embodiment.

FIG. 22 is a flowchart showing a processing flow of a synonym candidate extraction unit of the fourth embodiment.

─────────────────────────────────────────────────── ─── Continuation of the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/30 320 G06F 17/28 JISST file (JOIS)

Claims (4)

(57) [Claims] 1. A document data storage unit for storing document data, a syntactic analysis unit for parsing each sentence in a document in the document data storage unit, and a pair of a syntactic pattern and a synonym relation is extracted. Extraction rule storage means for storing as a rule; synonym candidate extraction means for extracting synonym candidates by matching the syntactic pattern obtained by the syntactic analysis means with the extraction rule; and within the document in the document data storage means An unregistered word detecting means for detecting an unregistered word in each sentence, and a synonym candidate for storing the synonym candidate obtained by the synonym candidate extracting means and the unregistered word detected by the unregistered word detecting means Storage means; and correction means for correcting / confirming / adding to the synonym candidates stored in the synonym candidate storage means, wherein the synonym candidate storage means includes: The one that matches the extraction rule is stored as a synonym candidate.
And has no synonyms that do not match the above extraction rules
A synonym information creation device for storing as a synonym candidate . 2. A document data storage unit for storing document data, a syntactic analysis unit for parsing each sentence in a document in the document data storage unit, and a pair of a syntactic pattern and a synonym relation is extracted. Extraction rule storage means for storing as a rule, exception rule storage means for storing an exception of the extraction rule as an exception rule, invalidation rule storage means for storing a pattern of words that do not form a synonymous relationship as an invalidation rule, and The syntactic pattern obtained by the syntactic analysis means, the synonym candidate extracting means for collating the extraction rule, the exception rule and the invalidation rule to extract synonym candidates, and the synonym obtained by the synonym candidate extracting means. A synonym candidate storing means for storing a candidate, and a correcting means for correcting / confirming / adding the synonym candidate stored by the synonym candidate storing means, Broadcast creation device. 3. A synonym information creating method using a computer, wherein syntactic analysis is performed on each sentence in document data stored in advance to obtain a syntax pattern, and the syntactic pattern is stored in advance. extraction rule and the collation to extract synonyms candidate from the syntax pattern, detects the unregistered word in each sentence of the document data, those that have been matched to the extraction rule of the unregistered word and the synonyms candidate storing as a synonym candidate, one that does not match with the extraction rule of the unregistered word,
A synonym information creating method for storing synonym candidates that do not have synonyms and correcting / confirming / adding the synonym candidates. 4. A method for creating synonym information using a computer, wherein syntactic analysis is performed on each sentence in document data stored in advance to obtain a syntactic pattern. Extraction rules that represent the correspondence between patterns and synonyms, exception rules that represent exceptions to the extraction rules, and invalidation rules that represent patterns of words that do not form a synonym relationship are extracted and stored as synonym candidates. Then, the synonym information creating method of correcting / confirming / adding the synonym candidate.