JP2000268035A - Semantic dictionary registering device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a semantic dictionary registering device for automatically giving semantic information to an input dictionary entry while utilizing the information of a dictionary entry, to which the semantic information is already given, in a semantic dictionary. SOLUTION: This semantic dictionary registering device is provided with a dictionary data storage part 2 for storing dictionary entries composed of dictionary records, indexes of the dictionary records and semantic information of dictionary records as dictionary data, a synonym dictionary entry semantic information retrieving part 3 for confirming whether the dictionary entry of synonym to the dictionary record of the input dictionary entry and the index of the dictionary record is stored in the dictionary data storage part 2 or not and extracting the semantic information of the synonym dictionary entry when it is stored and an input dictionary entry registering part 10 for applying the semantic information of the synonym dictionary entry as the semantic information of the input dictionary entry when the semantic information of the synonym dictionary entry is extracted by the synonym dictionary entry semantic information retrieving part 3 and registering it into the dictionary data storage part 2.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a semantic dictionary registration device for creating a word dictionary storing semantic information of words, and is applicable to a natural language processing system, an information retrieval system, a machine translation system, and the like. It is.

[0002]

2. Description of the Related Art Conventionally, when constructing a semantic dictionary used in a natural language processing system, a document search system, a machine translation system, or the like, a dictionary entry of a semantic dictionary newly registered by an expert who is familiar with the contents of the semantic dictionary. It is common practice to examine the contents and manually add and register appropriate semantic information to the contents. However, it is difficult to assign this semantic information to the dictionary entry unless the contents of the dictionary entry and the contents of the semantic information to be added are well understood. Therefore, when a dictionary entry is expanded or a user registers a dictionary entry as a user dictionary according to the purpose of use, it requires work by an expert who is familiar with the dictionary entries and the contents of semantic information. Cost and effort.

As one of the semantic dictionary registration methods, a semantic dictionary expert writes a semantic information determination rule for determining semantic information in the mind, and uses the rule to refer to the contents of the dictionary entry. There is a method of adding semantic information. In this method, the expertise of assigning semantic information written in the semantic information decision rule at the time of assigning semantic information can be directly utilized, so that even a non-semantic dictionary expert can efficiently and precisely Provision of information can be expected. However, since the creation of this semantic information determination rule is a manual operation by an expert, it takes a great deal of cost and time to create an accurate assignment flow, and the semantic information determination rule is created by a manual operation. There is a possibility that errors and inconsistencies may be included in the information, which causes a problem in that the accuracy of the assigned semantic information is reduced.

[0004] Also, what meaning a word has,
Automatically checks the tendency of the word to appear in natural language sentences from a large-scale natural language sentence collection (hereinafter referred to as a corpus) and co-occurs with similar words There is known an apparatus for automatically extracting synonym information by recognizing that words have the same meaning (Japanese Patent Laid-Open No. 9-1204).
No. 01). However, the set of synonym words automatically extracted in this way is directly related to the semantic information system used as various conditions in natural language processing systems, document search systems, and machine translation systems. Since there is no relevant relationship, since the words belong to the same group, there is a problem in that the dictionary information to be registered in the dictionary entry cannot be directly determined. Furthermore, in addition to the word notation, various information used in individual systems is usually added to the dictionary entry. However, this device automatically groups a set of synonym words using only the word notation as a clue. Therefore, there is a problem in that it is difficult to add semantic information in consideration of information other than the word notation of the dictionary entry.

[0005]

In the above-described conventional meaning dictionary registering device which is a device for creating a word dictionary storing the meaning information of words, in order to newly register the meaning information in the dictionary entry, the meaning of the meaning dictionary is changed. There is a problem that a great deal of cost and labor are required because the dictionary entry and the contents of the semantic information need to be assigned by a well-understood expert. Also,
In the method of assigning semantic information based on semantic information decision rules written by a semantic dictionary expert, creating an accurate assignment flow requires a great deal of cost and labor, and the semantic information decision rules include errors and inconsistencies. This is a problem in that the accuracy of the assigned semantic information is reduced due to the possibility that the semantic information may be given. Furthermore, in a device for automatically examining the tendency of words that co-occur with the word in the corpus and automatically extracting synonym information, and adding the meaning information of the word, a set of extracted synonym words is used. Is not directly related to the semantic information in the dictionary entry, and does not utilize information other than the word notation in the dictionary entry, so it is difficult to assign semantic information in consideration of information other than the word notation in the dictionary entry That is a problem.

The present invention has been made in view of the above points, and solves the above-mentioned conventional problems. The present invention utilizes the information of a dictionary entry to which semantic information has already been assigned in a semantic dictionary, and uses the information of the semantic information. It is an object of the present invention to provide a semantic dictionary registration device which extracts a dictionary entry in an input dictionary entry and a synonymous semantic dictionary to be assigned, and assigns semantic information assigned to the dictionary entry as semantic information of the input dictionary entry. Aim.

[0007]

SUMMARY OF THE INVENTION The present invention has been made to achieve the above object.
Dictionary data storage means for storing at least a dictionary record, a dictionary entry including the heading of the dictionary record, and semantic information of the dictionary record as dictionary data;
It is checked whether or not the dictionary record of the input dictionary entry and the dictionary entry synonymous with the heading of the dictionary record are stored in the dictionary data storage means. If stored, the semantic information of the synonymous dictionary entry is stored. Synonym dictionary entry semantic information searching means for extracting the synonymous dictionary entry, and when the synonym dictionary entry semantic information searching means extracts semantic information of the synonymous dictionary entry, the synonymous dictionary entry And an input dictionary entry registering means for registering the dictionary information in the dictionary data storage means.

[0008] According to the second aspect of the present invention, in the first aspect,
In the invention described above, at least a semantic information determination condition that is a condition for determining semantic information based on information in a dictionary record and a heading, and a semantic information determination that includes semantic information to be given to a dictionary entry satisfying the semantic information determination condition A semantic information determining rule storing means for storing a rule; wherein the synonym dictionary entry meaning information searching means extracts the meaning of the semantic dictionary determining rule when extracting the semantic information of the synonymous dictionary entry with respect to the input dictionary entry. The determination condition is compared with the dictionary record and / or heading of the input entry, and if there is a meaning determination condition to be matched, the semantic information corresponding to the meaning determination condition is extracted as the semantic information of the synonymous dictionary entry. It is characterized by the following.

According to the third aspect of the present invention, there is provided the second aspect.
In the described invention, the tendency of semantic information given to dictionary records and / or headings of each dictionary entry stored in the dictionary data storage means is automatically analyzed, and the semantic information is stored in the dictionary records and / or headings. A semantic information determination condition that is a condition for determination is automatically extracted, and the semantic information determination rule is stored as a semantic information determination rule including the semantic information determination condition and the semantic information to be assigned to a dictionary entry satisfying the semantic information determination condition. And means for acquiring semantic information determination rules to be registered in the means.

According to a fourth aspect of the present invention, in the first, second or third aspect of the invention, a dictionary of input dictionary entries which must be included when specific semantic information is added to the input dictionary entries. A semantic information determination essential condition that is information in a record and a heading; and a semantic information determination essential rule storage unit that stores a semantic information determination essential rule including the specific semantic information. When extracting the semantic information of the dictionary entry to be synonymous with the input dictionary entry, the semantic determination essential condition of the semantic dictionary determination essential rule is matched with the dictionary record and / or heading of the input entry, and the semantic determination to be matched is performed. It extracts semantic information of synonymous dictionary entries from semantic information corresponding to essential conditions. That.

Further, according to the invention described in claim 5, according to claim 4,
In the described invention, the tendency of semantic information given to dictionary records and / or headings of each dictionary entry stored in the dictionary data storage means is automatically analyzed, and a specific meaning in the dictionary records and / or headings is specified. Automatically extract semantic information determination essential condition, which is a condition that must be included when information is added, and define the semantic information determination essential condition and the semantic information determination essential rule consisting of the specific semantic information as described above. It is characterized by including semantic information determination essential rule acquisition means registered in the semantic information determination essential rule storage means.

According to a sixth aspect of the present invention, in the first, second, third, fourth or fifth aspect of the invention, the dictionary data storage means includes a dictionary record heading of each dictionary entry containing word information. A synonym dictionary entry semantic information search unit that calculates a semantic distance between words that is semantic distance information between word information; When confirming whether a dictionary entry synonymous with a heading is stored in the dictionary data storage means, word information in the heading of the dictionary record of the input dictionary entry,
The inter-word semantic distance between the word information in the heading of the dictionary record of the dictionary entry stored in the dictionary data storage unit is calculated by the semantic distance calculating unit, and the inter-word semantic distance is referred to, It is characterized by extracting the semantic information of the defined dictionary entry.

According to a seventh aspect of the present invention, in the first, second, third, fourth, fifth or sixth aspect, the dictionary data storage means stores the dictionary data in the dictionary record of each dictionary entry. The synonym dictionary includes word information corresponding to record headings and semantic information of the dictionary record, and includes semantic distance calculating means for calculating inter-word semantic distance that is semantic distance information between the word information. The entry semantic information search unit is configured to check whether a dictionary record of the input dictionary entry and a dictionary entry synonymous with a heading of the dictionary record are stored in the dictionary data storage unit. The meaning distance between words between the word information in the word and the word information in the dictionary record of the dictionary entry stored in the dictionary data storage means is defined as the meaning distance. Calculated by the calculation means, with reference to said word between semantic distance, it is characterized in that to extract semantic information of the dictionary entries to be synonymous.

Further, according to the invention described in claim 8, according to claim 6,
Or the invention according to claim 7, further comprising: a thesaurus storage means for storing a thesaurus in which word information having the same meaning and / or synonymous meaning is grouped, wherein the semantic distance calculation means calculates the inter-word semantic distance between the word information. When calculating, it is referred to whether or not both word information belong to the same group in the thesaurus stored in the thesaurus storage means, and calculate the inter-word semantic distance between the word information. It is a feature.

According to the ninth aspect of the present invention, the eighth aspect of the present invention provides
In the described invention, the thesaurus storage means stores conceptual relationship information that is a semantic relationship between groups of word information in the thesaurus and / or a higher-order relationship that is a bundle of a plurality of groups having a common meaning in each group. When the semantic distance calculating means calculates the inter-word semantic distance between the word information, both of the word information belong to the same group in the thesaurus stored in the thesaurus storage means. If they are in another group in addition to whether or not they are included, refer to what kind of conceptual semantic relation information and / or superordinate conceptual relation information exists in both groups, It is characterized in that a meaning distance between words is calculated.

According to a tenth aspect of the present invention, in the invention of the eighth or ninth aspect, the dictionary data storage means includes word information of a first language as word information of a heading of a dictionary record of each dictionary entry. And that the word information of the dictionary record of each dictionary entry includes word information of the second language corresponding to the heading of the dictionary record.

In the invention according to claim 11, in the invention according to claims 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10, the means for storing semantic information determination rules may include: It is characterized by including a means for displaying the meaning information of the synonymous dictionary entries extracted by the meaning dictionary entry meaning information search means.

According to the above configuration, in the semantic dictionary registration device of the present invention, the dictionary data storage means stores a dictionary entry including at least a dictionary record, a heading of the dictionary record, and semantic information of the dictionary record. Store as dictionary data. From now on, semantic information is added to the dictionary entries stored in the dictionary data storage means by the following means.

The synonym dictionary entry semantic information search means includes:
It is checked whether or not the dictionary record of the input dictionary entry and the dictionary entry synonymous with the heading of the dictionary record are stored in the dictionary data storage means. If stored, the semantic information of the synonymous dictionary entry is determined. Can be extracted. The input dictionary entry registering means, when the synonym dictionary entry meaning information searching means and the meaning information of the synonymous dictionary entry are extracted by the synonym dictionary entry meaning information searching means, as the meaning information of the input dictionary entry, Attach semantic information of synonymous dictionary entries,
It can be registered in the dictionary data storage means. As a result, for the input dictionary entry to which no semantic information has been added, the semantic information that reflects the result of adding the semantic information that has been added to the dictionary entry with reference to the dictionary entry to which the semantic information has been added in advance. Can be provided.

Further, the semantic information determination rule storage means includes at least a semantic information determination condition that is a condition for determining semantic information based on the information in the dictionary record and the index, and a dictionary entry that satisfies the semantic information determination condition. Can be stored. The synonym dictionary entry semantic information search means, when extracting the semantic information of a dictionary entry synonymous with the input dictionary entry, compares the semantic determination condition of the semantic dictionary determination rule with the dictionary record and / or heading of the input entry. If there is a semantic determination condition to be collated and collated, semantic information corresponding to the semantic determining condition can be extracted as semantic information of a synonymous dictionary entry. As a result, the semantic information of the input dictionary entry is determined by utilizing not only the dictionary entry to which the semantic information has been added in advance but also the semantic information determination rule for determining the semantic information using the information of the input dictionary entry. If a condition considered particularly effective in determining semantic information is described in the semantic information determination rule, semantic information can be efficiently and accurately added to the input dictionary entry.

Further, the semantic information determination rule acquiring means automatically analyzes the tendency of the semantic information given to the dictionary record and / or the index of each dictionary entry stored in the dictionary data storage means, and And / or automatically extracting a semantic information determination condition that is a condition for determining semantic information in the heading, and determining the semantic information including the semantic information determination condition and the semantic information to be assigned to a dictionary entry satisfying the semantic information determination condition. It can be registered as a rule in the semantic information determination rule storage means. Thereby, as a semantic information determination rule used when determining the semantic information of the input dictionary entry, the semantic information in the dictionary data storage means reflects the tendency of the dictionary entries to which the semantic information has been added. Since an effective condition is automatically obtained, semantic information can be efficiently and accurately added to the input dictionary entry without manual creation of a semantic information determination rule.

Further, the semantic information determination essential rule storage means stores the dictionary information of the input dictionary entry and the semantic information determination information which is included in the index, which must be included when specific semantic information is given to the input dictionary entry. An essential condition and a semantic information determination essential rule composed of the specific semantic information can be stored. The synonym dictionary entry semantic information searching means, when extracting the semantic information of the dictionary entry to be synonymous with the input dictionary entry, the semantic dictionary essential condition, the semantic essential condition, the dictionary record of the input entry and / or the heading And the semantic information of the synonymous dictionary entry can be extracted from the semantic information corresponding to the meaning determination essential condition to be compared. As a result, it is possible to prevent the semantic information that should not be given from the characteristics of the input dictionary entry from being accidentally given due to the search result of the synonym dictionary entry or the determination result of the semantic information determination rule. Semantic information that should not be added can be excluded in advance when searching for a synonym dictionary entry or determining a semantic information determination rule, so that semantic information can be efficiently added.

Further, the semantic information determination essential rule acquiring means automatically analyzes the tendency of the semantic information given to the dictionary record and / or heading of each dictionary entry stored in the dictionary data storage means, and Automatically extract the semantic information determination essential condition, which is a condition that must be included when specific semantic information is given in the record and / or heading, and the semantic information determination essential condition and the specific semantic information Can be registered in the semantic information determination essential rule storage means. Thereby, as a semantic information determination essential rule utilized when determining the semantic information of the input dictionary entry, the semantic information in the dictionary data storage means reflects the tendency of the essential condition for the attribute value of the dictionary entry to which the semantic information has been added. Effective conditions for determining information are automatically obtained, so even if there is no manual work to create rules for determining semantic information, it is necessary to eliminate semantic information that must not be given efficiently and accurately. Semantic information can be added to the input dictionary entry.

In the dictionary data storing means, the heading of the dictionary record of each dictionary entry may include word information. The semantic distance calculation means can calculate the inter-word semantic distance that is the semantic distance information between the word information. The synonym dictionary entry semantic information searching means, when confirming whether or not a dictionary record of the input dictionary entry and a dictionary entry synonymous with the heading of the dictionary record are stored in the dictionary data storage means, The inter-word semantic distance between the word information in the heading of the dictionary record and the word information in the heading of the dictionary record of the dictionary entry stored in the dictionary data storage means is calculated by the semantic distance calculating means, By referring to the inter-word semantic distance, semantic information of synonymous dictionary entries can be extracted. Thus, when searching for the input dictionary entry and the dictionary entry in the synonymous dictionary data storage means, not only information on whether or not the word information in the headings of both dictionary entries match, but also match. Even in the case where there is no word information, information on how close the two word information are in meaning can be used, so that a similar dictionary entry can be more accurately searched and semantic information can be added.

Further, the dictionary data storage means can include word information corresponding to the heading of the dictionary record and the semantic information of the dictionary record in the dictionary record of each dictionary entry. The semantic distance calculation means can calculate the inter-word semantic distance that is the semantic distance information between the word information. The synonym dictionary entry semantic information searching means, when confirming whether the dictionary record of the input dictionary entry and the dictionary entry synonymous with the heading of the dictionary record are stored in the dictionary data storage means, The inter-word semantic distance between the word information in the dictionary record and the word information in the dictionary record of the dictionary entry stored in the dictionary data storage unit is calculated by the semantic distance calculation unit. Referring to the semantic distance,
It is possible to extract the meaning information of synonymous dictionary entries. Thus, when searching for a dictionary entry in the dictionary data storage means that is similar to the input dictionary entry, not only information on whether or not the word information in both dictionary entries match, However, since it is possible to utilize information on how close the word information is in meaning, both meanings can be more accurately searched for synonymous dictionary entries and semantic information can be added.

Further, the thesaurus storage means can store a thesaurus in which word information having the same meaning and / or synonymous meaning is grouped. Meaning distance calculation means,
When calculating the inter-word semantic distance between the word information, it is determined whether or not both word information belong to the same group in the thesaurus stored in the thesaurus storage means. Can be calculated. Thereby, when searching the input dictionary entry and the dictionary entry in the synonymous dictionary data storage means, the word information in both dictionary entries and / or the semantic closeness of the word information in the heading of the dictionary entry are calculated. In doing so, it can be based on the information of the group of words in the thesaurus, and the word information included in the same group can be confirmed, so the basis of the criteria for calculating the semantic closeness is clear At the same time, semantic information reflecting the characteristics of the field can be given by selecting a thesaurus to be used for a specific field or selecting a thesaurus automatically extracted from the specific field corpus.

Further, the thesaurus storage means stores conceptual relation information that is a semantic relation between groups of word information in the thesaurus and / or superordinate concepts that are a relation bundled by a plurality of groups having a common meaning in each group. Relationship information can be included. When calculating the inter-word semantic distance between the word information, the semantic distance calculation means includes, in addition to whether or not both word information are in the same group in the thesaurus stored in the thesaurus storage means, If they belong to another group, also refer to what concept semantic relationship information and / or superordinate concept relationship information exists in both groups,
The inter-word meaning distance between the word information can be calculated. Thereby, when searching the input dictionary entry and the dictionary entry in the synonymous dictionary data storage means, the word information in both dictionary entries and / or the semantic closeness of the word information in the heading of the dictionary entry are calculated. In doing so, not only does the word information of both groups belong to the same group in the thesaurus, but also based on the relationship information between the two groups that contain each other's word information even if they do not belong to the same group. Since it is possible to calculate semantic closeness, it is possible to search for similar dictionary entries more accurately and to add semantic information, and to set related information between groups as a thesaurus to be used for specific fields. Field, or select a thesaurus to which information on the relationship between groups is automatically extracted from a specific field corpus. To reflect the, it is possible to impart a meaning information.

The dictionary data storage means includes word information of the first language as the word information of the heading of the dictionary record of each dictionary entry, and stores the heading of the dictionary record as the word information of the dictionary record of each dictionary entry. May be included in the second language. Thus, when a word in the first language and a corresponding word in the second language are included in a dictionary entry in a dictionary for machine translation, etc., the closeness of the word in the two languages is determined. They can be calculated separately, and more accurate semantic information can be given by making use of the semantic characteristics of these two languages, and even if the inter-word semantic distance cannot be calculated from words in one language, another Since similar dictionary entries can be searched based on the semantic distance between words of the language, when the number of dictionary entries in the dictionary data storage unit is small or when semantic information with few dictionary entries is given In such a case, the semantic information for the input dictionary entry can be more appropriately given.

Further, the semantic information determination rule storage means can display the meaning information of the synonymous dictionary entries extracted by the synonym dictionary entry meaning information searching means. As a result, the user can confirm the contents of the semantic information registered in the input dictionary entry, and can make appropriate corrections and the like.

[0030]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. In the present embodiment, it is assumed that the semantic dictionary is a syntactic and semantic dictionary for a machine translation system, and the first natural language is Japanese and the second natural language is English.

FIG. 1 shows a basic block configuration of a meaning dictionary registration device according to an embodiment of the present invention. The meaning dictionary registration device shown in FIG. 1 includes an input dictionary entry reading unit 1, a dictionary data storage unit 2, a synonym dictionary entry meaning information search unit 3, a meaning information determination rule storage unit 4, a meaning information determination rule acquisition unit 5, a meaning. It comprises an information determination essential rule storage unit 6, a semantic information determination essential rule acquisition unit 7, a semantic distance calculation unit 8, a thesaurus storage unit 9, and an input dictionary entry registration unit 10.

The input dictionary entry reading unit 1 inputs an input dictionary entry which is made up of an input dictionary record and a heading of the input dictionary record and to which no semantic information is added. It should be noted that the heading of the input dictionary record contains Japanese word information that is the first natural language, and the input dictionary record contains the second natural language corresponding to the Japanese word information. Contains English word information.

The dictionary data storage unit 2 stores a dictionary record,
A plurality of dictionary entries each including a heading of a dictionary record and semantic information already assigned to the dictionary record are stored. In each dictionary entry, the word information of the first natural language, Japanese, is included in the heading of the dictionary record, and the second word corresponding to the Japanese word information is included in the dictionary record. Contains word information of English, which is a natural language.

The synonym dictionary entry meaning information search unit 3
For the input dictionary entries read by the input dictionary entry reading unit 1 and to which no semantic information is assigned, search the dictionary entries stored in the dictionary data storage unit 2 for the input dictionary entries and the similar dictionary entries. Then, the semantic information assigned to the dictionary entry having the same meaning is assigned as the semantic information of the input dictionary entry. The synonym dictionary entry semantic information search unit 3 includes a dictionary data search unit 31, a semantic information determination essential rule matching unit 32, a semantic information determination rule matching unit 33, and a synonym dictionary entry determining unit 34.
Consists of

The dictionary data search unit 31 of the synonym dictionary entry meaning information search unit 3 searches the dictionary data storage unit 2 for a set of dictionary entries that satisfy the condition for the dictionary entry specified by the semantic information determination essential rule matching unit 32. I do. The semantic information determination essential rule matching unit 32 includes: an input dictionary entry read by the input dictionary entry reading unit 1;
The rule of the semantic information determination essential rule stored in the semantic information determination required rule storage unit 6 is checked against the semantic information determination essential condition of each rule, and a set of semantic information determination essential rules that do not contradict the information in the input dictionary entry. The dictionary data search unit 31 uses the extracted semantic information candidate, which is the semantic information specified by the extracted semantic information determination essential rule, as a key, and the information in the dictionary entry has the semantic information as a key. The dictionary data storage unit 2 searches for a dictionary entry candidate that is a set of dictionary entries.

The semantic information determination rule matching unit 33 includes a semantic information candidate obtained by the input dictionary entry and the semantic information determination essential rule matching unit 32 and a semantic information determination rule storage unit 4.
Is compared with the contents of each rule in the semantic information determination rules stored in, and a set of semantic information determination rules from which semantic information candidates are selected without being inconsistent with the information in the input dictionary entry is extracted. Based on the rule certainty assigned to the semantic information determination rule including the semantic information, the semantic information certainty, which is the certainty of each semantic information, is calculated for each type of the semantic information in the table. The synonym dictionary entry determining unit 34 determines the input dictionary entry and each dictionary entry in the synonym dictionary entry candidates obtained by the semantic information determination essential rule matching unit 32,
How closely the contents of both dictionary entries match
The two dictionary entries and the word information included in the headings of the two dictionary entries are input to the semantic distance calculation unit 8, and the similarity between the two dictionary entries is determined based on the result of how much semantically calculated as a result. The degree is calculated, and the semantic dictionary entry is determined in consideration of the semantic information certainty obtained for each semantic information obtained by the semantic information determination rule matching unit 33, and the semantic information of the synonym dictionary entry is extracted. .

The semantic information determination rule storage unit 4 stores a semantic information determination condition which is a condition for determining semantic information based on the information in the dictionary record and the index, and a semantic information to be assigned to a dictionary entry satisfying the semantic information determination condition. Are stored in plural rules. It should be noted that each of the semantic information determination rules is given a rule certainty indicating the reliability of the rule.

The semantic information decision rule acquiring unit 5 comprises a dictionary entry analyzing unit 51, a semantic information decision rule extracting unit 52, and a semantic information decision rule registration unit 53.

The dictionary entry analysis unit 51 of the semantic information determination rule acquiring unit 5 is configured to perform a correlation between the type of the semantic information with respect to the dictionary record of each dictionary entry stored in the dictionary data storage unit 2 and the information in the headings of the dictionary record. Is automatically analyzed. Based on the analysis result obtained by the dictionary entry analysis unit 51, the semantic information determination rule extraction unit 52 converts a semantic information determination condition, which is a condition for determining semantic information in a dictionary record and a headline, into a statistical condition of the condition. Automatically extract with accurate information. Semantic information determination rule registration unit 53
Is based on the semantic information determining condition obtained by the semantic information determining rule extracting unit 52 and the semantic information to be given to the dictionary entry satisfying the semantic information determining condition, as a semantic information determining rule storage unit 4 Register with.

The semantic information determination essential rule storage unit 6 stores the dictionary information of the input dictionary entry which must be included when specific semantic information is given to the input dictionary entry, and the semantic information determination essential information which is information in the index. A condition and a semantic information determination essential rule including the specific semantic information are stored.

The semantic information determination essential rule acquisition unit 7 includes a dictionary entry analysis unit 71, a semantic information determination essential rule extraction unit 72, and a semantic information determination essential rule registration unit 73.

The dictionary entry analysis section 71 of the semantic information determination essential rule acquisition section 7 automatically determines the tendency of the semantic information given to the dictionary records and headings of each dictionary entry stored in the dictionary data storage means 2. analyse.
Based on the analysis result obtained by the dictionary entry analysis unit 71, the semantic information determination essential rule extraction unit 72 specifies a condition that must be included when specific semantic information is given in the dictionary record and the heading. Is automatically extracted. Based on the semantic information determination essential condition obtained by the semantic information determination essential rule extraction unit 72 and the specific semantic information, the semantic information determination required rule storage unit 73 stores Register in 6.

The semantic distance calculation unit 8 includes a word information pair input unit 8
1, a thesaurus search unit 82 and an inter-word distance calculation unit 83

Word information pair input unit 81 of semantic distance calculation unit 8
Is a pair of the word information in the heading of the input dictionary entry and the word information in the heading of the dictionary entry stored in the dictionary data storage unit 2, or the word information in the input dictionary entry and the dictionary data storage A pair of word information in the dictionary entry stored in the section 2 is input. The thesaurus search unit 82 searches for a thesaurus stored in the thesaurus storage unit 9 using the pair of word information input by the word information pair input unit 81 as a key, and includes the same meaning and / or the same including individual word information. The word information having the meaning of the synonym is extracted from the type of the group and the semantic relation information between the groups including the individual word information. The inter-word distance calculation unit 83
Based on the type of the group including the pair of word information obtained by the thesaurus search unit 82 and the semantic relation information between the groups, it is determined whether the two word information match or not. If they are not in the same group in the same thesaurus, and if they are not in the same group in the same thesaurus, there is a semantic relationship between the two groups,
Calculate the semantic distance between both word information.

The thesaurus storage unit 9 stores a thesaurus composed of word category information in which word information having the same meaning and / or synonymous meaning is grouped and conceptual relation information which is a semantic relation between each group of word information. are doing.

When the synonym dictionary entry meaning information search unit 3 extracts the meaning information of the synonymous dictionary entry, the input dictionary entry registration unit 10 determines the meaning of the synonymous dictionary entry as the meaning information of the input dictionary entry. The semantic information is added and registered in the dictionary data storage 2.

FIG. 2 shows an example of dictionary entries of dictionary data stored in the dictionary data storage unit according to one embodiment of the present invention. The dictionary data includes, for example, a Japanese-English syntactic translation dictionary used in a Japanese-English machine translation system. In the Japanese-English syntactic translation dictionary, Japanese verb expressions are registered as headings as dictionary entries, The dictionary entry is given the particle pattern of the Japanese verb. For example, for the verb "Acquire,""S" (subject, noun) "ga" (particle) "O1" (direct purpose Words, nouns) "O" (particles) "
For the verb “suppress”, use “S” (subject, noun)
The particle pattern "ga" (particle) "O1" (direct object, noun) "wo" (particle), and the verb "attach" is "S" (subject, noun) "ga" (particle) " The particle pattern "O1" (direct object, noun) "wo" (particle) "O2" (indirect object, noun) "ni" (particle) is "S" (subject) for the verb "integrate" , Noun) "ga" (particle) "O1" (direct object, noun) "wo" (particle) "O2"
(Indirect object, noun) The particle pattern "ni" (particle) is "S" (subject,
Noun) "ga" (particle) "O1" (direct object, noun)
A particle pattern of "wo" (particle), "O2" (indirect object, noun) "ni" (particle) "is written in a Japanese slot, and the verb is used as a headword. It is written to search for English translations and to search for semantic information of the verb.

Further, a selection condition indicating the meaning of the case element of the verb is described. For example, for the verb “acquire”, the meaning of the case element shown in the particle pattern (S is O1) is translated in English as “S acquire O1,” and the semantic information is “S gets O1. In the case of "buy", it indicates that S has the meaning of "subject" and O1 has no restriction. For the verb “suppress”, the meaning of the case element indicated in the particle pattern (S is O1) is translated into English as “S control O1”, and the semantic information is “S is the attribute of O1. In the case of “change”, it indicates that there is a constraint that S has the meaning of “subject” and O1 has the meaning of “price”. As for the verb “attach”, the meaning of the case element shown in the particle pattern (S is O1 to O2) is “S attach O1 to O
When translated in English as “2”, and the semantic information is “S combines O1 with O2”, it indicates that there is no restriction.

The verb "integrate" is translated into English as "S combine O1 into O2" for the meaning of the case element shown in the particle pattern (S is O1 to O2), and the semantic information is obtained. Is "S joins O1 and O2", it means that S has the meaning of "subject" and that there are no restrictions on O1 and O2.

As for the verb “directly connected”, the meaning of the case element indicated in the particle pattern (S is O1 to O2) is translated into English as “S link O1 directly to O2”. Information says "S works with O1 and O2"
Indicates that S has the meaning of "subject" and that O1 and O2 have no restrictions.

Next, the processing procedure of the semantic dictionary registration device having the basic block configuration shown in FIG. 1 will be described. Here, FIG.
The procedure of this processing will be described with reference to the input dictionary entry shown in FIG.

First, the input dictionary entry reading unit 1
Inputs a dictionary entry which is made up of an input dictionary record and a heading of the input dictionary record and has no semantic information. This input dictionary entry is used to assign semantic information by the semantic dictionary registration device. In the example of the input dictionary entry in FIG. 3, the same type of information as the dictionary data of the Japanese-English syntactic translation dictionary used in the Japanese-English machine translation system in FIG. 2 is stored, and the Japanese verb notation is registered in the heading. And the input dictionary entry contains
The Japanese verb particle pattern is attached. More specifically, for example, the verb “obtain” in FIG. 3 is composed of the verb “S” (subject, noun) “ga” (particle) “O1” (direct object, noun) “o” (particle) ” As a selection condition indicating the meaning of the case element of the verb of the verb, the meaning of the case element shown in the particle pattern (S is O1) is expressed in English as “S get O1”. In the case of translation, it indicates that there is a constraint that S has the meaning of "subject" and O1 has the meaning of "concrete".
However, the input dictionary entry of FIG. 3 does not have the semantic information provided in the dictionary entry of FIG. 2 and is in a state of “?” Indicating unknown.

The input dictionary entry to which the input semantic information is not assigned is searched by the synonym dictionary entry semantic information search unit 3 from the dictionary entries stored in the dictionary data storage unit 2 and the input dictionary entry. Of the dictionary entry, and the semantic information assigned to the dictionary entry having the same meaning is added as the semantic information of the input dictionary entry.

In the synonym dictionary entry semantic information search unit 3, first, the semantic information determination essential rule matching unit 32 checks the input dictionary entry and the semantic information determination essential rules stored in the semantic information determination essential rule storage unit 6. Of the rules in the input dictionary entry, and a set of semantic information mandatory rules that do not contradict the information in the input dictionary entry is specified. The extracted individual semantic information mandatory rules are specified. The dictionary data search unit 31 searches the dictionary data storage unit 2 for a synonymous dictionary entry candidate that is a set of dictionary entries having meaning information whose key is the information in the dictionary entry, using the semantic information candidate that is the semantic information as a key. I do.

The semantic information mandatory rule stored in the semantic information essential rule storage unit 6 is used by the semantic information essential rule acquisition unit 7 in the embodiment of the semantic dictionary registration apparatus shown in FIG. The dictionary entry information to which the semantic information stored in the data storage unit has been added is analyzed, and a semantic information determination essential rule is automatically created and stored in the semantic information determination essential rule storage unit 6. The procedure for acquiring the semantic information determination essential rule will be described with reference to the basic block diagram of FIG.

First, the dictionary entry analysis unit 71
The tendency of the semantic information given to the dictionary record and the index of each dictionary entry stored in the dictionary data storage means 2 is automatically analyzed. For example, in the dictionary data shown in FIG. 2, as shown in FIG. 4, each dictionary entry is classified according to the type of semantic information, and processed so that essential conditions of dictionary entries having the same semantic information can be easily extracted.

Next, the semantic information determination essential rule extraction unit 72
Accordingly, based on the analysis result obtained by the dictionary entry analysis unit 71, the semantic information determination essential condition, which is a condition that must be included when specific semantic information is given in the dictionary record and the heading, is included. Automatically extract. For example, from the analysis result obtained by the dictionary entry analysis unit 71 shown in FIG. 4, a method of extracting the same feature possessed by all dictionary entries having the same semantic information as a semantic information determination essential condition can be considered. Specifically, looking at the information in the Japanese slots, if the semantic information is “S operates in combination with O1 and O2”, all the Japanese slots have “S is O1
To "O2", and if the semantic information is "S will get or buy O1" or "S will change the attribute of O1", then all Japanese slots will have "S ", These sets shown in bold in FIG. 4 are automatically extracted as one of the semantic information determination essential conditions.

Next, the semantic information determination essential rule registration unit 73
Thus, the semantic information determination essential condition for the semantic information obtained by the semantic information determination essential rule extraction unit 72 is registered in the semantic information determination essential rule storage unit 6 as a semantic information determination essential rule. For example, a semantic information determination essential rule as shown in FIG. 5 is created from the semantic information determination essential conditions shown in bold in FIG. 4 and registered in the semantic information determination essential rule storage unit 6.

In the embodiment described here, the semantic information determination essential rule is automatically extracted from the dictionary data stored in the dictionary data storage means 2. For example, the semantic information determination essential rule is manually created. However, it is also possible to adopt a form in which essential conditions that are imagined by humans when they are given to the mind are utilized in the present semantic dictionary registration device. In addition, a form in which both the semantic information determination essential rule automatically extracted from the dictionary data and the manually created semantic information determination essential rule are utilized as the semantic information determination essential rule is also possible.

Using, as an example, the semantic information determination essential rule shown in FIG. 5 created by the processing of the semantic information determination essential rule acquisition unit 7 as described above, the semantic information determination essential rule for the input dictionary entry shown in FIG. The processing of the matching unit 32 will be described.

First, the input dictionary entry is checked against the semantic information decision essential condition of each rule in the semantic information decision essential rule, and a set of semantic information decision essential rules that are consistent with the information in the input dictionary entry is extracted. The semantic information candidates that are the semantic information specified by the extracted semantic information determination essential rules are extracted. For example, for the input dictionary entry of FIG.
Since “1” is “1”, the semantic information of “S receives and buys O1” and “S changes the attribute of O1” that are consistent with those in the input dictionary entry are extracted as semantic information candidates.

Next, using the semantic information candidate as a key, the dictionary data search unit 31 stores a synonymous dictionary entry candidate which is a set of dictionary entries having meaning information whose key is information in the dictionary entry. Search from 2. For example, for the input dictionary entry of FIG. 3, two types of semantic information candidates, “S receives and buys O1” and “S changes the attribute of O1,” are extracted.
6, the dictionary entry shown in FIG. 6 having this semantic information is extracted as a synonymous dictionary entry candidate.

Using the semantic information candidate and the synonym dictionary entry candidate obtained by the semantic information determination essential rule collating unit 32, the semantic information determining rule collating unit 33 extracts the information in the input dictionary entry and the semantic information candidate. The contents of each of the semantic information determination rules stored in the semantic information determination rule storage unit 4 are compared with each other, and the semantic information accuracy, which is the accuracy of each semantic information in the semantic information candidate, is calculated.

The semantic information determination rule stored in the semantic information determination rule storage unit 4 is stored in the dictionary data storage unit 5 by the semantic information determination rule acquisition unit 5 in one embodiment of the semantic dictionary registration device shown in FIG. 2 is analyzed to automatically create a semantic information determination rule and store it in the semantic information determination rule storage unit 4. The procedure for acquiring the semantic information determination rule will be described with reference to the basic block diagram of FIG.

First, the dictionary entry analysis unit 51
The tendency of the semantic information given to the dictionary record and the index of each dictionary entry stored in the dictionary data storage means 2 is automatically analyzed. For example, in the dictionary data shown in FIG. 2, as in the case of the dictionary entry analysis unit 71 of the semantic information determination essential rule acquisition unit 7, as shown in FIG. Processing is performed so that dictionary entry conditions having semantic information can be easily extracted.

Next, the semantic information determination rule extraction unit 52 uses the analysis result obtained by the dictionary entry analysis unit 51 to determine the semantic information in the dictionary record and the heading. Is automatically extracted together with the statistical accuracy information of the condition. For example, based on the analysis result obtained by the dictionary entry analysis unit 51 shown in FIG. 4, if the information of the Japanese slot and the slot of the semantic constraint is used as the semantic information determination condition, as shown in FIG. The frequency of appearance for each feature in the dictionary entry is tabulated for each type of information, and the features in the dictionary entry are used as semantic information determination conditions. Make rules as information. Note that the example shown in FIG. 7 shows a case in which the accuracy information is the accuracy information for each semantic information. In the analysis result of FIG. 4, the semantic information “S receives O1 and buys” is a feature value “ "S", "O1", "S:
The appearance frequencies of “subject” and “O1: no semantic constraint” are each “1”, and the characteristic values “S”, “O1”, and “S” for the semantic information “S changes the attribute of O1” :
The appearance frequencies for the "subject" and "O1: price" are each "1", and the characteristic values "S", "O1", and "O" for the semantic information "S operates O1 in combination with O2".
The appearance frequency for “2” is “3” and “S: subject” respectively.
Is "2", the appearance frequency for "S: no semantic constraint" is "1", and the appearance frequency for "O1: no semantic constraint" and "O2: no semantic constraint" is "3", respectively. Is shown.

Next, the semantic information decision rule registration unit 53 registers the semantic information decision condition for the semantic information obtained by the semantic information decision rule extraction unit 52 as a semantic information decision rule in the semantic information decision rule storage unit 4. . For example, a semantic information determination rule as shown in FIG. 7 created by the semantic information determination rule extraction unit 52 is registered in the semantic information determination rule storage unit 4.

In the embodiment described here, the semantic information determination rule is automatically extracted from the dictionary data stored in the dictionary data storage means 2. For example, the semantic information determination rule is created manually. It is also conceivable that the semantic dictionary registration device utilizes a selection condition that is imagined by a human when he or she gives it. Further, a form in which both the semantic information determination rule automatically extracted from the dictionary data and the manually created semantic information determination rule are used as the semantic information determination rule is also conceivable.

Using, as an example, the semantic information determination rule shown in FIG. 7 created by the processing of the semantic information determination rule acquisition unit 5 as described above, the semantic information determination rule matching unit 33 for the input dictionary entry shown in FIG. Will be described.

First, the semantic information candidates obtained by the input dictionary entry and semantic information determination essential rule collation unit 32 are collated with the contents of each rule in the semantic information determination rule, and are consistent with the information in the input dictionary entry. A set of semantic information determination rules from which semantic information candidates are selected is extracted. For example,
The input dictionary entry shown in FIG. 3 and "S is O1" shown in FIG.
When the semantic information candidates of “get and buy” and “S changes the attribute of O1” are compared with the semantic information determination rule of FIG. 7, the parts shown in bold in FIG. The candidate is extracted as a semantic information determination rule to be selected.

Next, for each type of semantic information in the semantic information candidates, the semantic information certainty, which is the certainty for each semantic information, is calculated based on the rule certainty information given to the semantic information determination rule including the semantic information. For example, two types of semantic information candidates "S gets O1 and buys it" and "S changes the attribute of O1"
For example, when the ratio of the sum of the accuracy information of the portions shown in bold in FIG. 7 to the total sum is calculated as the semantic information accuracy, as shown in FIG. 8, “S receives O1 and buys it. "And" S change the attribute of O1 ", the sum of the accuracy information is 3, and the semantic information accuracy is both 0.5.

Using the synonym dictionary entry candidate obtained by the semantic information determination essential rule matching unit 32, the synonym information accuracy obtained by the semantic information determination rule matching unit 33, and the input dictionary entry, In the dictionary entry determination unit 34,
First, the word information in the input dictionary entry and the word information in each dictionary entry of the synonymous dictionary entry candidates are input to the semantic distance calculation unit 8, and how much the two word information calculated as a result is semantically close to each other. Based on the result of the above, the similarity of each dictionary entry of the similar dictionary entry candidate with respect to the input dictionary entry is calculated. For example, from the input dictionary entry shown in FIG. 3 and the synonym dictionary entry candidates shown in FIG. 6, Japanese words in the heading slots of both dictionary entries and English words in the English slot are input to the semantic distance calculation unit 8. In this case, the word pairs shown in FIG. 9 are input to the semantic distance calculation unit 8, and the inter-word semantic distance of each word pair is calculated. In the example of FIG. 9, the similarity of meaning is
Only verb words that are expected to be effective in determining thesaurus entries are calculated as input words.

The processing procedure of the semantic distance calculator 8 will be described with reference to the basic block diagram of FIG.

First, the word information pair input unit 81 of the semantic distance calculation unit 8 inputs a pair of word information in an input dictionary entry and word information in each dictionary entry of a synonym dictionary entry candidate. For example, in the example of FIG. 9, two types of Japanese word information pairs and two types of English word pairs are input.

Next, the thesaurus search unit 82 searches for a thesaurus stored in the thesaurus storage unit 9 using the word information pair input by the word information pair input unit 81 as a key.
The type of the word information group including the individual word information and having the same meaning and / or a similar meaning, and the semantic relationship information between the groups including the individual word information are extracted.

The thesaurus stored in the thesaurus storage unit 9 searched by the thesaurus search unit 82 includes word category information in which word information having the same meaning and / or synonymous meaning is grouped, and word group information between each group of word information. Consisting of conceptual relationship information that is a semantic relationship and / or superordinate conceptual relationship information that is a relationship bundled by a plurality of groups having a common meaning in each group.

For example, in the example of the input word pair shown in FIG. 9, a Japanese word pair and an English word pair are input, so that two types of thesaurus for Japanese and English are stored in the thesaurus storage unit 9. It will be stored.

FIG. 10 shows an example of a Japanese thesaurus stored in the thesaurus storage 9 according to one embodiment of the present invention. The thesaurus for Japanese is, for example, a semantic category in which word information having the same meaning and / or a synonymous meaning is grouped, a semantic category name indicating the semantic category, and conceptual relationship information that is a semantic relationship between the semantic categories, for example, , A semantic category semantically encompasses another semantic category,
That is, it is composed of superordinate concept relation information indicating a relation between semantic categories when a certain semantic category is a superordinate concept of another semantic category.

In the thesaurus for Japanese in FIG. 10, "<
The words in “>” indicate the meaning category name, and the words in “[]” described immediately below indicate the list of words included in the group of the meaning category. Contains a word "...", indicating that the display of the list of words included in this word category is omitted, and a line above the word category is connected to another semantic category. Indicates that the upper semantic category is a superordinate concept of the lower semantic category.If there is "...." immediately below "[]", the semantic category indicates a plurality of different semantic categories. Although it is a superordinate concept of the category, it indicates that it is omitted in the figure.

More specifically, a part of the contents of the thesaurus for Japanese in FIG. 10 will be described. In the semantic category of <acquisition>, the words "acquire" and "acquire" are both included. The superordinate concept of the semantic category of <> is a semantic category of <transaction>, and the superordinate concept of the semantic category of <transaction> is a semantic category of <act>.

The processing procedure of the thesaurus retrieving unit 82 is described by using a Japanese word pair as a key among the input word pairs shown in FIG.
The case of searching for the Japanese thesaurus in FIG. 10 will be described as an example. First, "get" as an input word pair
When the Japanese thesaurus shown in FIG. 10 is searched using “and” as the key, both “to get” and “to get” are included in the same semantic category of <get>.
As a search result of the thesaurus search unit 82, for example, both “<acquire>” and “acquire” meaning categories of “acquire” are output and “0” is output as the inter-category distance. Next, using “acquire” and “suppress” as keys as input word pairs, FIG.
When a Japanese thesaurus of 0 is searched, “acquire” is included in the semantic category of “acquire”, and “suppress” is included in the semantic category of <regulation>, and up to <act>, which is a common superordinate concept between the two categories. Since <acquisition> needs to be traced back to the upper concept with two levels of <acquisition> and four levels of <regulation>, the search result of the thesaurus retrieval unit 82 may be, for example, "<acquisition>" However, “<Regulation>” is output as the semantic category of “suppress”, and “6”, which is the sum of the number of steps up to the common superordinate concept, is output as the inter-category distance.

FIG. 11 shows an example of an English thesaurus stored in the thesaurus storage unit 9 according to an embodiment of the present invention. The thesaurus for English is, for example, a semantic category in which word information having the same meaning and / or a synonymous meaning is grouped, conceptual relation information that is a semantic relation between the semantic categories,
For example, when the semantic category semantically includes another semantic category, that is, upper semantic relation information indicating a relation between semantic categories when a certain semantic category is a superordinate concept of another semantic category. .

For example, in the English thesaurus of FIG. 11, words in "[]" indicate a list of words included in the same semantic category. When "..." is included in "[]", it indicates that the display of the list of words included in this word category is omitted. When a word category is connected to another semantic category by a line, it indicates that the upper semantic category is a superordinate concept of the lower semantic category. When there is "...", it means that the semantic category is a superordinate concept of a plurality of different semantic categories, but is omitted in the figure.

To explain a part of the contents of the English thesaurus in FIG. 11, "control" and "operate" belong to the same semantic category.
The superordinate concept of the semantic category of "rate" is a semantic category containing "manipulate", and the superordinate concept of the semantic category of "manipulate" is a semantic category containing "handle" and "palm". The superordinate concept of the semantic category of “handle” and “palm” is a semantic category including “touch”, which indicates that the semantic category of “touch” has no superordinate concept.

The processing procedure of the thesaurus retrieving unit 82 will be described by taking as an example a case where the English thesaurus shown in FIG. 11 is retrieved by using an English word pair as a key among the input word pairs shown in FIG. First, “get” and “acqui
Searching for the English thesaurus in Fig. 11 using "re" as a key,
Since both “get” and “acquire” are included in the same semantic category, as the search results of the thesaurus search unit 82, for example, both “[get, acquire]” as the semantic categories of “get” and “acquire” However, “0” is output as the distance between categories. Next, when the English thesaurus of FIG. 11 is searched using “get” and “control” as input word pairs, there is no common upper-level semantic category for “get” and “control”. For example, “[get, acquire]” is output as the semantic category of “get” and “[control, operate]” is output as the semantic category of “control” as the search result of the search unit 82, and “infinity” is set as the inter-category distance. "Large" is output.

In the embodiment described above, a Japanese thesaurus and an English thesaurus are used as the thesauri stored in the thesaurus storage unit 9. However, as the Japanese thesaurus, a conventional thesaurus disclosed conventionally is used. Japanese thesaurus (for example, “Japanese vocabulary system”, Iwanami Shoten,
1997) can be used, and as an English thesaurus, an existing English thesaurus previously disclosed (for example, “Word Net-An Electronic Lexical Dat
abase ", MIT Press, 1998). Of course, it is also possible to use a thesaurus in one or the other language instead of using both Japanese and English thesauruses. Further, in the embodiment described here, a completed thesaurus is used as the thesaurus. However, an existing thesaurus creation device disclosed in the related art (for example, Japanese Unexamined Patent Application Publication No. 9-120) is used.
U.S. Pat. No. 401) may utilize a corpus-oriented thesaurus automatically extracted from the corpus.

Based on the type of group including the pair of word information obtained by the thesaurus search unit 82 and the semantic relationship information between groups, the inter-word distance calculation unit 83 determines whether the two word information match. No, if they do not match, they are in the same group in the thesaurus, or if they are not in the same group in the thesaurus, because there is a semantic relationship between both groups, Calculate the semantic distance between both word information.

For the word-to-word distance calculation unit 83, for example,
The Japanese and English word pairs shown in FIG.
, And a procedure for specifically calculating the inter-word distance from the obtained result will be described. In the example in which the word pair in FIG. 9 is input, the number of relationships between the semantic categories to which both words belong is calculated as the distance between categories from the thesaurus search unit 82, but the distance between word calculation unit 83 Can be calculated according to, for example, an inter-word distance calculation rule as shown in FIG. Specifically, "0" indicates that the input word pair matches, "0.2" indicates that the input word pair does not match but both input words fall into the same semantic category and the inter-category distance is "0", and the inter-category distance is "0". “1” is “0.4”, the distance between categories is “2” to “4” is “0.6”, and the distance between categories is “5” or more is “0.8”. The case where the distance is “infinity” can be calculated as “1”.

When the inter-word distance of the specific word pair in FIG. 9 is calculated using the inter-word distance calculation rule of FIG. 12, first,
From the Japanese word pair “get” and “get”, the distance between categories is “0” as the output of the thesaurus search unit 82.
Is obtained, and since "acquire" and "acquire" are not the same word, "0.
2 "is output. Further, from the Japanese word pair “acquire” and “suppress”, the result of the inter-category distance “6” is obtained as an output of the thesaurus search unit 82.
Since “acquire” and “suppress” are not the same word, “0.8” is output as the inter-word distance. The English word pair “get” and “acquire” can be used in a thesaurus search unit 8
As the output of 2, the result that the inter-category distance is “0” is obtained, and “get” and “acquire” are not the same word, so “0.2” is output as the inter-word distance. From the English word pair “get” and “control”, the distance between categories is “infinity” as an output of the thesaurus search unit 82.
Since "get" and "control" are not the same word, "1" is output as the inter-word distance.

As a result, the inter-word distance is given by the semantic distance calculation unit 8 to each word pair in FIG. 9, for example, as shown in FIG. It is sent to the decision unit 34.

The synonym dictionary entry determination unit 34 also takes into account the inter-word distance obtained by the semantic distance calculation unit 8 and the semantic information accuracy, which is the accuracy of each semantic information obtained by the semantic information determination rule matching unit 33. To determine the thesaurus entry.

As a method of calculating the degree of synonym of each dictionary entry in the synonym dictionary entry candidate for the input dictionary entry using the inter-word distance and the degree of semantic information accuracy, for example, the semantic information for the semantic information of each dictionary entry is used. The accuracy is Pm, and the semantic distance calculation unit 8 in each dictionary entry
If the number of words for which the inter-word distance has been calculated is N, the inter-word distance for the i-th word in the words for which the inter-word distance has been calculated is Dwi (i = 1 to N). The similarity P of the entry can be calculated by, for example, the following formula.

[0093]

(Equation 1)

Using this equation, the semantic degree P between the input dictionary entry of FIG. 3 and each dictionary entry in the synonym dictionary entry candidates of FIG. As shown in FIG. 13, the calculation is performed using the inter-word distance and the semantic information accuracy of FIG. First, input dictionary entries and headings are "acquired"
8, the semantic information accuracy Pm is “0.5” from FIG. 8, the number of words N calculated for the inter-word distance is “2”, and the two types of inter-word distance are Japanese words. Distance Dw1
Is "0.2" and the English word-to-word distance Dw2 is "0.2", so that the value of the similarity P is "0.3". The similarity degree P between the input dictionary entry and the dictionary entry whose heading is “suppress” is as follows: the semantic information accuracy Pm is “0, 5” from FIG. 8, the number of words N for which the inter-word distance is calculated is “2”,
For the two types of inter-word distances, the Japanese inter-word distance Dw1 is "0.
8 ”, since the English word distance Dw2 is“ 1 ”,
As the value of the similarity P, “−0, 4” is obtained. The degree of synonym of each dictionary entry and the input dictionary entry obtained individually is given to each dictionary entry in the synonym dictionary entry candidates as shown in FIG.

Based on the result of the calculation of the degree of synonym, a synonym dictionary entry is determined from the synonym dictionary entry candidates. For example, from the input dictionary entries of FIG. 3 and the synonym dictionary entry candidates of FIG. 6, among the synonym dictionary entry candidates of FIG. Is determined as a synonym dictionary entry (FIG. 15).

Finally, the input dictionary entry registration unit 10 gives the semantic information of the individually determined synonym dictionary entries as the semantic information of the input dictionary entry. The dictionary entry is registered in the dictionary data storage unit 2 as one entry in the dictionary data. For example, the semantic information “S receives O1 or buys” is extracted from the synonym dictionary entry of FIG. 15, and it is added as the semantic information of the input dictionary entry of FIG. The input dictionary entry to which the semantic information is added is registered as a new dictionary entry in the dictionary data in the dictionary data storage unit 2 as shown in FIG.

In the description of the embodiment shown in FIG. 1, one entry similar to the input dictionary entry is selected from the dictionary data, and the semantic information assigned to the similar dictionary entry is selected. Although assigned as an input dictionary entry, similar dictionary entries to the input dictionary entry are considered for each type of semantic information, a plurality of entries including the same semantic information are selected as similar dictionary entries, and the semantic information is assigned to the input dictionary entry. It is also possible to add appropriate semantic information as the semantic information of the input dictionary entry by referring to the information of the dictionary data without determining a specific similar dictionary entry.

The input dictionary entry registration unit 10 is provided with a display function, and the input dictionary entry to be newly registered is presented to the user together with the semantic information, so that the user can confirm it and register it in the dictionary data storage unit 2. Is also possible. Further, the user can confirm the display contents, make appropriate changes, and register them in the dictionary data storage unit 2.

The present invention is not limited to the above embodiment, but can be variously modified or applied within the scope of the claims.

In the semantic dictionary registration device according to the present invention shown in FIG.
The synonym dictionary entry semantic information search unit 3, semantic information determining rule obtaining unit 5, semantic information determining essential rule obtaining unit 7, semantic distance calculating unit 8, and input dictionary entry registering unit 10 include a computer and its peripheral devices. Each of the dictionary data storage unit 2, the semantic information determination rule storage unit 4, the semantic information determination essential rule storage unit 6, and the thesaurus storage unit 9 can be realized by a program executed by the computer. It can be realized using a storage device connected directly to the computer or via a communication line. in this case,
The program executed by the computer can be distributed via a predetermined storage medium or a communication line.

[0101]

According to the semantic dictionary registration device of the present invention, an input dictionary entry is read, and this input dictionary entry is compared with dictionary data in a dictionary data storage section composed of input dictionary entries to which semantic information has already been added. Then, in the dictionary data, a dictionary entry and a synonymous dictionary entry are searched, and semantic information assigned to the synonymous dictionary entry is assigned as semantic information of the input dictionary entry. As a result, for the input dictionary entry to which no semantic information has been added, the semantic information that reflects the result of adding the semantic information that has been added to the dictionary entry with reference to the dictionary entry to which the semantic information has been added in advance. Can be provided.

Further, a semantic information determination rule and / or a semantic information determination essential rule which describes a condition and / or an essential condition for determining specific semantic information in the information of the dictionary entry are stored, and a synonym for the input dictionary entry is stored. When extracting the semantic information of the dictionary entry to be compared, the semantic dictionary determination rule and / or the semantic information determination essential rule are compared with the contents of the input entry, and the dictionary of the dictionary data storage unit having the condition to be matched and the semantic information of the essential condition Dictionary entries in the data can be extracted as synonymous dictionary entries. As a result, a dictionary entry that is effectively and accurately synonymous can be selected, and semantic information can be added to the input dictionary entry.

The semantic information determination rule and / or the semantic information determination essential rule automatically analyze the tendency of the semantic information assigned to each dictionary entry stored in the dictionary data storage means, and perform the semantic information determination rule and / or the semantic information determination rule. Semantic information determination required rules can be automatically extracted. Thereby, as a semantic information determination rule utilized when determining the semantic information of the input dictionary entry and / or a semantic information determination essential rule, in determining the semantic information reflecting the assignment tendency of the dictionary entry to which the semantic information has been assigned. Since the effective conditions are automatically obtained, the semantic information can be efficiently and accurately added to the input dictionary entry even if there is no manual creation of the semantic information determination rule and / or the semantic information determination essential rule.

In the dictionary data storage means, each dictionary entry can include word information. The semantic distance calculating means can calculate the inter-word semantic distance which is the semantic distance information between the word information. Synonym dictionary entry semantic information searching means, when checking whether a dictionary entry synonymous with the input dictionary entry is stored in the dictionary data storage means, word information of the input dictionary entry,
The inter-word semantic distance between the word information of the dictionary entry in the dictionary data storage means is calculated, and by referring to the inter-word semantic distance, the semantic information of the synonymous dictionary entry can be extracted. This makes it possible to search not only for input dictionary entries and similar dictionary entries, but also for information on how close the word information of both dictionary entries is in meaning, as well as whether the word information in both dictionary entries match. Since it is possible, similar dictionary entries can be searched more accurately and semantic information can be added.

Further, the thesaurus storage means can store a thesaurus obtained by grouping word information having the same meaning and / or synonymous meaning and / or conceptual relationship information which is a semantic relationship between the groups. Meaning distance calculation means,
When calculating the inter-word semantic distance between word information, whether or not both word information are in the same group in the thesaurus, and if there is another group, what kind of semantic relationship information exists between groups The inter-word semantic distance between the pieces of word information can be calculated with reference to FIG. Thereby, when calculating the semantic closeness between the input dictionary entry and the word information, it can be based on the information of the group of the words in the thesaurus and / or the relation information between the groups. Clarify the grounds for the criteria of the standard, and add semantic information that reflects the characteristics of the field by selecting a thesaurus for use in a specific field as a thesaurus to be used or selecting a thesaurus automatically extracted from a specific field corpus Can be.

Further, the dictionary data storage means includes the first language word information as the word information of the heading of the dictionary record of each dictionary entry, and stores the heading of the dictionary record as the word information of the dictionary record of each dictionary entry. May be included in the second language. As a result, in a dictionary for machine translation or the like, more accurate semantic information can be provided by making use of the characteristics of the meanings of the two languages, and even if the inter-word semantic distance cannot be calculated from words in one language. A dictionary entry with a similar meaning can be searched based on the semantic distance between words of words in different languages, so if the number of dictionary entries in the dictionary data storage unit is small or if there is a small number of dictionary entries, meaning information is added. In such a case, the semantic information for the input dictionary entry can be given more appropriately.

Further, the semantic information determination rule storage means can display the meaning information of the synonymous dictionary entries extracted by the synonym dictionary entry meaning information searching means. As a result, the user can confirm the contents of the semantic information registered in the input dictionary entry, and can make appropriate corrections and the like.

[Brief description of the drawings]

FIG. 1 is a basic block configuration diagram of a meaning dictionary registration device according to an embodiment of the present invention.

FIG. 2 is a diagram showing dictionary entries of dictionary data according to an embodiment of the present invention.

FIG. 3 is a diagram showing an input dictionary entry according to an embodiment of the present invention.

FIG. 4 is a diagram illustrating dictionary entries in which dictionary data according to an embodiment of the present invention is classified according to semantic information;

FIG. 5 is a diagram showing entries of semantic information determination essential rules according to the embodiment of the present invention.

FIG. 6 is a diagram showing synonym dictionary entry candidates according to the embodiment of the present invention.

FIG. 7 is a diagram showing entries of a semantic information determination rule according to an embodiment of the present invention.

FIG. 8 is a diagram illustrating semantic information accuracy for a semantic information candidate according to an embodiment of the present invention.

FIG. 9 is a diagram illustrating word pairs input to a semantic distance calculation unit according to an embodiment of the present invention.

FIG. 10 is a diagram showing a Japanese thesaurus according to an embodiment of the present invention.

FIG. 11 is a diagram showing an English thesaurus according to one embodiment of the present invention.

FIG. 12 is a diagram showing a rule for calculating the distance between words according to an embodiment of the present invention.

FIG. 13 is a diagram illustrating a word-to-word distance for a word pair input to a semantic distance calculation unit according to an embodiment of the present invention;

FIG. 14 is a diagram showing a similarity degree for a similarity dictionary entry candidate according to an embodiment of the present invention.

FIG. 15 is a diagram showing a synonym dictionary entry according to an embodiment of the present invention.

FIG. 16 is a diagram illustrating an input dictionary entry to which semantic information is added according to an embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input dictionary entry reading part 2 Dictionary data storage part 3 Synonym dictionary entry semantic information search part 31 Dictionary data search part 32 Semantic information determination essential rule matching part 33 Semantic information determination rule matching part 34 Synonym dictionary entry determining part 4 Semantic information Decision rule storage unit 5 semantic information decision rule acquisition unit 51 dictionary entry analysis unit 52 semantic information decision rule extraction unit 53 semantic information decision rule registration unit 6 semantic information decision essential rule storage unit 7 semantic information decision essential rule acquisition unit 71 dictionary entry analysis Part 72 semantic information determination essential rule extraction part 73 semantic information determination essential rule registration part 8 semantic distance calculation part 81 word information pair input part 82 thesaurus search part 83 inter-word distance calculation part 9 thesaurus storage part 10 input dictionary entry registration part

Claims (11)

[Claims] 1. A dictionary data storage means for storing, as dictionary data, a dictionary entry including at least a dictionary record, a heading of the dictionary record, and semantic information of the dictionary record, a dictionary record of an input dictionary entry, and the dictionary record A dictionary entry meaning information search for checking whether or not a dictionary entry synonymous with the heading of the dictionary entry is stored in the dictionary data storage means, and extracting the meaning information of the synonymous dictionary entry if it is stored Means, when the meaning information of the synonymous dictionary entry is extracted by the synonym dictionary entry meaning information searching means, the meaning information of the synonymous dictionary entry is given as the meaning information of the input dictionary entry; Input dictionary entry registration means for registering in a data storage means, a semantic dictionary registration characterized by comprising: Location. 2. A semantic information determination condition comprising at least a semantic information determination condition for determining semantic information based on information in a dictionary record and a heading, and semantic information to be assigned to a dictionary entry satisfying the semantic information determination condition. Means for storing semantic information determining rules for storing rules, wherein the synonym dictionary entry meaning information searching means extracts the meaning of the semantic dictionary determining rules when extracting the meaning information of the synonymous dictionary entries for the input dictionary entries. The determination condition is compared with the dictionary record and / or heading of the input entry, and if there is a meaning determination condition to be matched, the semantic information corresponding to the meaning determination condition is extracted as the semantic information of the synonymous dictionary entry. The semantic dictionary registration device according to claim 1, wherein: 3. A method for automatically analyzing the tendency of semantic information given to a dictionary record and / or a heading of each dictionary entry stored in the dictionary data storage means, and converting the semantic information in the dictionary record and / or the heading. A semantic information determination condition that is a condition for determination is automatically extracted, and the semantic information determination rule is stored as a semantic information determination rule including the semantic information determination condition and the semantic information to be assigned to a dictionary entry satisfying the semantic information determination condition. 3. A semantic dictionary registration apparatus according to claim 2, further comprising means for acquiring semantic information determination rules registered in said means. 4. An essential condition for determining semantic information, which is information in a dictionary record and a heading of an input dictionary entry that must be included when specific semantic information is given to an input dictionary entry, and the specific semantic information The semantic information determination essential rule storing means for storing a semantic information determination essential rule comprising: the synonym dictionary entry semantic information searching means, when extracting the semantic information of the synonymous dictionary entry for the input dictionary entry, The meaning determination essential condition of the semantic dictionary determination essential rule is compared with the dictionary record and / or heading of the input entry, and the semantic information of the synonymous dictionary entry is extracted from the semantic information corresponding to the semantic essential requirement to be matched. 4. The semantic dictionary registration device according to claim 1, wherein the registration is performed. 5. A tendency of semantic information assigned to a dictionary record and / or a heading of each dictionary entry stored in the dictionary data storage means, and automatically analyzes a specific meaning in the dictionary record and / or the heading. Automatically extract semantic information determination essential condition, which is a condition that must be included when information is added, and define the semantic information determination essential condition and the semantic information determination essential rule consisting of the specific semantic information as described above. 5. The semantic dictionary registration device according to claim 4, further comprising a semantic information determination essential rule acquisition unit registered in the semantic information determination essential rule storage unit. 6. The dictionary data storage means, wherein a heading of a dictionary record of each dictionary entry includes word information, and a semantic distance calculation for calculating an inter-word semantic distance, which is semantic distance information between the word information. Means for retrieving whether or not a dictionary record of the input dictionary entry and a dictionary entry synonymous with a heading of the dictionary record are stored in the dictionary data storage means. Calculating the meaning distance between words between the word information in the heading of the dictionary record of the input dictionary entry and the word information in the heading of the dictionary record of the dictionary entry stored in the dictionary data storage means. And calculating means for extracting meaning information of synonymous dictionary entries by referring to the meaning distance between words. 3, 4 or 5
Description dictionary registration device. 7. The dictionary data storage means includes, in a dictionary record of each dictionary entry, word information corresponding to a heading of the dictionary record and semantic information of the dictionary record. A synonym dictionary entry semantic information search unit, the synonym dictionary entry semantic information search unit includes a dictionary record of an input dictionary entry and a dictionary entry synonymous with a heading of the dictionary record. When confirming whether or not the dictionary information is stored in the dictionary data storage means, the word information in the dictionary record of the input dictionary entry and the word information in the dictionary record of the dictionary entry stored in the dictionary data storage means The meaning distance between words is calculated by the meaning distance calculation means, and the meaning of the dictionary entry to be synonymous is calculated with reference to the meaning distance between words. The semantic dictionary registration device according to claim 1, 2, 3, 4, 5, or 6, wherein the taste information is extracted. 8. A thesaurus storing means for storing a thesaurus in which word information having the same meaning and / or synonymous meaning is grouped, wherein said semantic distance calculating means calculates an inter-word semantic distance between word information. In this case, it is characterized by referring to whether or not both word information belong to the same group in the thesaurus stored in the thesaurus storage means, and calculating the inter-word semantic distance between the word information. The semantic dictionary registration device according to claim 6 or 7, wherein: 9. The thesaurus storage means stores conceptual relation information, which is a semantic relation between groups of word information in the thesaurus, and / or higher-order relations, which are bundled in a plurality of groups having a common meaning in each group. When the semantic distance calculating means calculates the inter-word semantic distance between the word information, both the word information belong to the same group in the thesaurus stored in the thesaurus storage means. If they are in another group in addition to whether or not they are included, refer to what kind of conceptual semantic relation information and / or superordinate conceptual relation information exists in both groups, 9. The semantic dictionary registration device according to claim 8, wherein a semantic distance between words is calculated. 10. The dictionary data storage means includes word information of a first language as word information of a heading of a dictionary record of each dictionary entry, and stores word information of the dictionary record as word information of a dictionary record of each dictionary entry. The semantic dictionary registration device according to claim 8 or 9, wherein word information of a second language corresponding to the headline is included. 11. The semantic information determination rule storage means,
6. The apparatus according to claim 1, further comprising means for displaying semantic information of synonymous dictionary entries extracted by said synonym dictionary entry meaning information searching means.
A semantic dictionary registration device according to 6, 7, 8, 9 or 10.