WO2016117920A1

WO2016117920A1 - Knowledge represention expansion method and apparatus

Info

Publication number: WO2016117920A1
Application number: PCT/KR2016/000579
Authority: WO
Inventors: 최기선; 함영균; 서지우
Original assignee: 한국과학기술원
Priority date: 2015-01-20
Filing date: 2016-01-20
Publication date: 2016-07-28

Abstract

A knowledge representation expansion apparatus includes: a predicate-argument structure analyzing unit for extracting a predicate and at least one argument from a text using a meaning representation language; an ontology unit for representing knowledge using a knowledge representation language, which is a structured format understandable by a computer, and for extracting a second predicate corresponding to a first predicate, which is extracted from the predicate-argument structure analyzing unit; and a knowledge representation unit for representing knowledge extracted from the text using the first predicate, when the similarity of the first predicate and the second predicate is equal to or less than a threshold value.

Description

Knowledge expression extension method and device

The present invention relates to a method and apparatus for extending knowledge representation.

Recently, research on question and answer system based on semantic web and big data is active. The semantic web is a semantic web that expresses relationships between information and semantic information (Semanteme) in ontology that can be processed by a computer in a distributed environment such as the Internet. In addition, many studies are being conducted to build an ontology-based knowledge database. Traditionally, however, knowledge is written in natural language, and some studies have shown that more knowledge is contained in unstructured data than in structured databases. Therefore, researches for automatically generating instances of ontology schemas from unstructured data including natural language texts are being conducted to extend the knowledge database.

In particular, the Semantic Web must express the knowledge of the Web in a structured format that can be understood by a computer, that is, Resource Description Framework (RDF) triples. For this purpose, the Semantic Web has properties that can fully describe various attributes of the knowledge elements. Ontology is required. RDF Triple is an international standard governed by the World Wide Web Consortium (W3C). Its knowledge and information are subject (subject), predicate (property) and object (object (literal)). ] In the form of three pairs, where the property corresponds to the predicate of the RDF triple and the relationship between the subject and the object.

DBpedia, the latest technology on the Semantic Web, is a knowledge database built automatically from Wikipedia, the encyclopedia of text. Divipedia uses Divipedia Ontology, originated from Wikipedia's infobox, to express Wikipedia's knowledge. However, while D.B. ontologies may be sufficient to express Wikipedia's summarized knowledge, it is difficult to guarantee that all knowledge in Wikipedia's text can be expressed. Therefore, we need an ontology that can express various attributes of knowledge elements in natural language text, and we need a technology to expand knowledge by automatically building knowledge database based on this.

An object of the present invention is to extend a knowledge expression method and apparatus, and when the knowledge extracted from any text cannot be expressed as a knowledge expression language used in the knowledge expression ontology, a method for extending the knowledge expression using a semantic expression language. will be.

An apparatus for expanding knowledge expression according to an embodiment of the present invention, comprising: a predicate-argument structure analyzer for extracting a predicate and at least one argument from text using a semantic expression language, a knowledge expression language that is a structured format that can be understood by a computer Extracts a second predicate corresponding to the first predicate extracted by the predicate-dissertation structure analysis unit from the ontology unit expressing the knowledge using and the similarity between the first predicate and the second predicate When the reference value is less than or equal to, the first expression includes a knowledge expression unit for representing the knowledge extracted from the text.

The knowledge expression unit may extract the second predicate related to the at least one argument from the ontology unit.

The knowledge expression unit extracts a first domain that is similar to a lexical type assigned to the at least one argument from domains of the knowledge expression language by more than a reference value, and is assigned to the at least one argument among the ranges of the knowledge expression language. The first range similar to the lexical type and the reference value may be extracted, and the first domain and the predicate related to the first range may be extracted as the second predicate.

The knowledge expression unit may generate a string in which information related to any one of the first predicate and the at least one argument is combined, and add the string to the knowledge expression language of the ontology portion.

The knowledge expression language may be a language expressed in a resource description framework (RDF) ternary relationship.

A method according to another embodiment of the present invention extends a knowledge expression, the method comprising: receiving text including at least one sentence, expressing the text as a first predicate and at least one argument based on a semantic expression language And extracting a second predicate corresponding to the first predicate, comparing the similarity between the first predicate and the second predicate, and, if the similarity is equal to or less than a reference value, from the text. Expressing the extracted knowledge using the first predicate.

In the extracting of the second predicate corresponding to the first predicate, the second predicate corresponding to the first predicate may be extracted from the knowledge expression ontology using the vocabulary type assigned to the at least one argument.

The knowledge expression ontology uses a knowledge expression language that expresses knowledge in a ternary relation of a subject, predicate, and object, and extracting a second predicate corresponding to the first predicate. A predicate kit that is similar to the lexical type assigned to the at least one item among the subjects of the knowledge expression language or more than the reference value, and is similar to the lexical type assigned to the at least one item among the objects of the knowledge expression language. Can be extracted with the second predicate.

The expressing using the first predicate may generate a string in which information related to any one of the first predicate and the at least one argument is combined, and express the knowledge extracted from the text using the string.

The method may further include adding the character string to a knowledge expression language of the knowledge representation ontology.

An apparatus according to another embodiment of the present invention extends a knowledge expression, the method comprising: interpreting a predicate-argument structure of text, matching the predicate-argument structure of the text with a ternary relation of the knowledge expression language, and Adding the first predicate extracted from the predicate-dissertation structure of the text as a predicate of the knowledge expression language based on a matching similarity.

The adding of the knowledge expression language as a predicate may include extracting a second predicate matching the first predicate of the predicate-non-serial structure of the text from the ternary relation of the knowledge expression language, the first predicate and the second predicate. Comparing the similarity of the predicate, and if the similarity is less than the reference value, adding the first predicate to the knowledge expression language.

The method may further include expressing the text in a ternary relationship using the first predicate.

Matching the ternary relation of the knowledge expression language may match the predicate-nonserial structure of the text to the ternary relation based on the similarity between the domains and the range of the ternary relations extracted from the predicate-terminal structure of the text. can do.

According to an embodiment of the present invention, when the knowledge extracted from a text cannot be expressed as the knowledge expression language used in the knowledge expression ontology, the knowledge expression may be extended using the semantic expression language. That is, according to the embodiment of the present invention can solve the problem that the knowledge representation ontology does not have sufficient coverage when building the knowledge database from the web text.

According to an embodiment of the present invention, the knowledge database can be expanded quickly and easily by expressing knowledge included in unstructured data such as natural language as a knowledge expression language in a computer understandable format based on sentence semantic predicate-dissertation structure.

According to an embodiment of the present invention, the "relationship" ontology of the knowledge database can be expanded to increase knowledge expression power and can be applied to CGC (Collaboratively Generated Content) oriented knowledge forms and interpretations.

1 is an illustration of a semantic expression language according to an embodiment of the present invention.

2 is a block diagram of an apparatus for expanding knowledge representation according to an embodiment of the present invention.

FIG. 3 is an exemplary diagram illustrating a result of analyzing a predicate-dissertation structure according to an embodiment of the present invention. FIG.

4 is an exemplary diagram illustrating a ternary relation knowledge expression structure according to an embodiment of the present invention.

5 is a flowchart of a method of expanding an expression of knowledge according to an embodiment of the present invention.

6 is a flowchart illustrating a method of extending knowledge representation according to an embodiment of the present invention.

7 is a diagram illustrating a result of analyzing a predicate-dissertation structure of an example sentence according to an embodiment of the present invention.

8 is a diagram illustrating a ternary relation knowledge expression structure of an example sentence according to an embodiment of the present invention.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is said to "include" a certain component, it means that it can further include other components, without excluding other components unless specifically stated otherwise.

The knowledge database stores structured information in the knowledge expression language. Ontology represents knowledge in a structured format that can be understood by a computer. The knowledge expression language may vary, but may be, for example, an RDF triple. RDF triples represent a knowledge and information in the ternary relation of a subject (Subject (resource)), predicate (Predicate (property)), and object ((Object (literal)), where a predicate or property is a predicate. , Represents the relationship / property between the entity at the subject (object) and the entity or value at the object (object).

Since the ontology is limited to structured information, it is difficult to express knowledge extracted from an unstructured knowledge source. In particular, we examined whether sufficient knowledge can be extracted from the text by calculating the ontology of Divipedia, which is the center of linked data, and when expressing new knowledge using unstructured text as a knowledge source. It can be seen that this is limited.

The following describes how to extend knowledge expression based on semantic expression language. That is, when the knowledge extracted from the text cannot be expressed in the current knowledge expression language, a method of extending the knowledge expression by creating a new ontology instance will be described.

Referring to Figure 1, it will be described taking the following query example as an example. "This" in the query is "interferon".

Query: This is a glycoprotein to an animal infected with a virus, the cells produced. It acts as a deterrent to the infection and proliferation of viruses. It is mass-produced with the development of genetic engineering and is used to treat viral diseases such as type B infection and herpes (herpes).

Answer: Interferon

The ontology of the knowledge database allows this (interferon) to express the type of "glycoprotein" as structured information (RDF). However, in unstructured queries, predicates such as "infected", "generating", "retarding", "acting", "produced", "used in therapy" are important information, It is difficult to express them.

The present invention enhances the expressive power of knowledge using a semantic expression language. Here, the semantic expression language is a language for expressing the meaning of a sentence based on a relationship between a predicate (Property / Predicate) and an argument (Argument). Predicate-argument structure refers to the relationship of arguments that a predicate requires in constructing a sentence. The number of arguments depends on the predicates. A predicate can require one essential argument to create a clause or sentence, and a predicate can require two or three arguments.

The semantic expression language can describe the causes, consequences, opinions, behaviors, and conditions for a particular entity that is difficult to express in the DIBIDI ontology. For example, the predicate-discussion structure may be extracted using FrameNet, but is not limited thereto. Framenet is a language resource constructed by annotating how vocabulary is used in sentences in the form of semantic-frames.

Referring to FIG. 1, a query statement may be expressed as a graph of a framenet structure of an RDF structure. As such, the query statement can be expressed in a predicate-discussion structure. For example, "infected" can be expressed as "Influence_of_event_on_cognizer" in Framenet, "create" can be expressed as "Creating" in Framenet, and "inhibiting" It may be expressed as "Intercepting" of the framenet, and "treat" may be expressed as "Cure" of the framenet.

FIG. 2 is a block diagram of an apparatus for expanding knowledge representation according to an embodiment of the present invention, FIG. 3 is an exemplary view illustrating a result of analyzing a predicate-nonsense structure according to an embodiment of the present invention, and FIG. An exemplary diagram illustrating a ternary relation knowledge expression structure according to an embodiment of the present invention.

Referring to FIG. 2, the knowledge expression expanding apparatus (hereinafter referred to as “device”) 100 may include a text input unit 110, a predicate-dissertation structure analysis unit 130, a knowledge expression ontology unit 150, and a knowledge expression unit ( 170).

The text input unit 110 receives text including at least one sentence.

The predicate-argument structure interpreter 130 divides the text into a predicate and at least one argument based on the semantic expression language. A semantic expression language specifies at least one argument that must be present in any word of a sentence (eg, a word corresponding to a predicate), and expresses the meaning of the sentence using a predicate-dissertation structure. Referring to FIG. 3, the predicate-dissertation structure interpreter 130 finds a predicate (predicate.L) in the text, and finds at least one argument (item 1 to n) corresponding to the predicate. At this time, the predicate-argument structure analyzer 130 may output lexical types T.1 to T.n of each argument. For example, the semantic expression language may be FrameNet. In the case of using the framenet to analyze the predicate-dissertation structure, the predicate-dissertation structure analyzer 130 identifies the frame target in the sentence and finds the frame element. Here, the frame object corresponds to the predicate of the sentence, and the frame element corresponds to the argument related to the predicate. The predicate-argument structure analysis unit 130 may output an annotation text on the framenet analysis result.

The knowledge representation ontology unit 150 expresses knowledge in a structured format that can be understood by a computer. To this end, the knowledge representation ontology unit 150 describes the attributes of the knowledge elements using the knowledge expression language. For example, the knowledge expression language may be a resource description framework (RDF), and knowledge is expressed as an RDF triple, that is, a ternary relationship <S, P, O>. The knowledge expression ontology unit 150 expresses the text in a predefined ternary relationship. Referring to FIG. 4, the knowledge expression language may be RDF, and may be expressed as <Domain (D), Predicate (Predikit), Range (Range, R)>. Here, the domain D is a class of the domain related to the predicate, and corresponds to the class of the subject in the ternary relationship. The scope R is the class of the scope related to the predicate, which corresponds to the class of the object in the ternary relationship. For example, Divipedia Ontology can be read from the sentence ("Cheol was born in 1944 in Korea") from <People: "Pole", dbo: birthPlace, Place: "South Korea"> and <People: "Pole", dbo We can extract: birthDay, time: "1944"> in a ternary relation of knowledge expressions.

The knowledge expression unit 170 converts the predicate-dissertation structure of the text into the format of the knowledge expression ontology unit 150. The knowledge expression unit 170 compares the similarity of the knowledge expressions and determines whether the knowledge interpreted by the predicate-dissertation structure analysis unit 130 can be expressed in the format of the knowledge expression ontology unit 150. When the knowledge interpreted by the predicate-dissertation structure analysis unit 130 can be sufficiently represented in the format of the knowledge expression ontology unit 150, the knowledge expression unit 170 is the knowledge expression ontology unit 150 in the format of knowledge. Extract If the knowledge interpreted by the predicate-argument structure analysis unit 130 is not sufficiently represented in the format of the knowledge expression ontology unit 150, the knowledge expression unit 170 is interpreted by the predicate-argument structure analysis unit 130. Express knowledge using knowledge. Therefore, the knowledge expression unit 170 extracts knowledge from the text based on the semantic expression language when it is difficult to properly express the meaning of the text in a predefined ternary relationship. In addition, the knowledge expression unit 170 may transmit the attribute (corresponding to the ontology instance and the predicate) generated using the semantic expression language to the knowledge expression ontology unit 150. The knowledge expression ontology unit 150 may add information (ontology instances) generated using the semantic expression language to the knowledge expression language.

As such, the knowledge expression extension apparatus 100 may extend the knowledge expression of the knowledge expression ontology using the semantic expression language.

Referring to FIG. 5, the device 100 receives text including at least one sentence (S110).

The apparatus 100 expresses the text as a predicate and at least one argument based on the semantic expression language (S120). The apparatus 100 searches for predicates (predicates.L) and predicates (items 1 to n) in the text as shown in FIG. 3. In this case, the device 100 may output the lexical types T.1 to T.n of each argument.

The apparatus 100 extracts a predicate (predicate.K) corresponding to a predicate (predicate.L) extracted as a semantic expression language from the knowledge expression ontology (S130). The device 100 matches the predicate-nonserial structure of the text into a ternary relationship of the knowledge expression language. When the arguments corresponding to the domain D and the range R of the ternary relation knowledge expression are secured according to the result of the predicate-dissertation structure analysis, the device 100 is assigned to the domain D and the range R as shown in FIG. 4. You can extract the corresponding predicate (predicate.K). The device 100 may find a domain D and a range R that are the same or similar to the lexical type of the argument.

The apparatus 100 determines the similarity between the predicate (predicate.L) extracted as the semantic expression language and the predicate (predicate.K) of the knowledge expression language (S140). In this case, the apparatus 100 may determine the similarity between the predicate (predicate.L) extracted as the semantic expression language and the string combining the lexical type of the argument and the predicate (predicate.K) of the knowledge expression language.

Methods of determining similarity include: 1) similarity at the string level (2), similarity in word semantics (measurement of similarity using the concept hierarchy using language resources), and 3) measurement of word similarity based on corpus. There is a way. 1) In order to measure the similarity at the string level, there is a method of calculating the number of edits that a string takes to convert to a target string, and traditionally such as Levenshtein Distance. . 2) The similarity in word semantics is calculated by measuring the similarity between words in a hierarchical structure using a semantic lexical database such as WordNet. Traditionally, the method of measuring the minimum distance between nodes in a WordNet hierarchy, such as path similarity, the method of measuring the minimum distance and maximum depth between nodes, such as Leacock & Chodorow similarity, and the Wu & Palmer similarity Similarly, there is a method of utilizing the depth of a node and the distance from the minimum upper node between nodes. In the corpus-based word similarity measurement of 3), each word in the corpus is calculated to have a specific vector value in the dimensional space, thereby measuring the similarity between words in the similar vector space. Recently, an approach using word embedding has been used.

In a similar case, the device 100 extracts knowledge from text using a knowledge expression language already stored (S150). Since the knowledge interpreted in the semantic expression language can be sufficiently represented in the format of the knowledge expression ontology, the apparatus 100 expresses the knowledge of the text in the format of the knowledge expression language. That is, since the apparatus 100 is similar to the predicate (predicate.L) extracted as the semantic expression language more than the reference value of the predicate (predicate.K) of the knowledge expression language, the format of the knowledge expression language does not need to be expanded. Judges that the input text can be represented sufficiently. Knowledge may be expressed as <a vocabulary corresponding to a domain (D), a predicate.K, a vocabulary corresponding to a range (R)>.

If not, the apparatus 100 generates a predicate including a predicate (predicate.L) extracted as a semantic expression language (S160).

The apparatus 100 extracts knowledge from the text using the generated predicate (S170). That is, if the device 100 can express the text in the ternary relation existing in the knowledge expression ontology, the input device expresses the input text based on the stored knowledge expression ontology, and if the text cannot be expressed in the knowledge expression ontology, the input text is predicate-determined. Expressed in extended ternary relation using structure predicates. Knowledge is: vocabulary corresponding to domain (D), predicate.L, vocabulary corresponding to range (R)> or vocabulary corresponding to domain (D), predicate.L + vocabulary type corresponding to range (R), Vocabulary corresponding to the range (R)>.

The device 100 adds the generated predicate to the knowledge expression ontology (S180). The generated predicate is added as a new knowledge representation instance.

In the following, we will explain how to extract knowledge from the example sentence ("Cheol was born in 1944 in Korea").

FIG. 6 is a flowchart illustrating a knowledge expression extension method according to an embodiment of the present invention. FIG. 7 is a view illustrating a result of analyzing a predicate-dissertation structure of an example sentence according to an embodiment of the present invention. Is a diagram illustrating a ternary relation knowledge expression structure of an example sentence according to an embodiment of the present invention.

Referring to FIG. 6, the device 100 receives text (“Br. Was born in 1944 in Korea.”) (S210).

The apparatus 100 classifies text into predicates and arguments based on the semantic expression language as shown in FIG. 7. If the argument for the predicate ("born") is "Who", "when" or "where", then the strings corresponding to the argument are "Abstract", "Korea", and "1944". When using a framenet, the frame target is "born" and the frame predicate class is "being_born". The frame arguments for the frame predicate class ("being_born") are defined as "Child", "Place", and "Time", so the frame argument-string pairs are Child-Joe, Place-Korea, and Time-1944. The vocabulary type for the argument is also determined, the vocabulary type of "Child" is "people", the vocabulary type of "Place" is "place", and the vocabulary type of "Time" is "time ( time) ".

The apparatus 100 compares the domain of the dispute with the ternary relation, and extracts a dispute that matches the domain of the ternary relation among the disputes (S230). The device 100 may find a domain of ternary relation similar to the lexical type of the arguments. The device 100 finds the domain / range related to the argument in order to convert the predicate-claim structure into a ternary relationship, which may first make a non-domain similarity measure. The device 100 may determine that "people" of the lexical type of the argument is similar to "people" which is a domain of ternary relation.

The device 100 compares the range of the argument and the ternary relation, and extracts a dispute that matches the range of the ternary relation among the arguments (S240). The device 100 may determine that "time" of the lexical type of the argument is similar to "Time" which is a range of ternary relations.

Since the apparatus 100 extracts the subject (domain) and the object (range) required by the ternary relation knowledge expression, the apparatus 100 extracts a predicate (predikit) related to the subject (domain) and the object (range) (S250). Referring to FIG. 8, the predicate (fredikit) related to the domain "people" and the range "Time" is "birthday".

The apparatus 100 measures the similarity between the predicate ("being_born") of the semantic expression language and the predicate ("birthday") of the ternary relation (S260). At this time, the device 100 combines the predicate "being_born" with "time" which is a lexical type / related range of the related argument / related argument to generate a combined string ("being_bornTime"), and "being_bornTime" and "birthday". "Can be compared.

If the predicates are similar, the device 100 expresses the knowledge extracted from the text using the predicate (“birthday”) of the ternary relationship (S270). The knowledge extracted from the text can be <Bill, birthday, 1994>, and "Bail" and "1994" can be URIs linked.

If the predicates are not similar, the apparatus 100 expresses the knowledge extracted from the text using the predicate "being_born" of the semantic expression language (S280). That is, since the device 100 currently defined in the knowledge expression language ("birthday") does not sufficiently express the meaning of the sentence, the apparatus 100 uses the predicate of the semantic expression language instead of the predicate of the ternary relation. Herein, the newly generated predicate may be a string including "being_born", for example, "being_bornTime". The knowledge extracted from the text is expressed in an extended ternary relationship, and may be, for example, <Atract, being_born, 1994> or <Atract, being_bornTime, 1994>. "Withdrawal" and "1994" can be URIs linked.

The device 100 stores the new predicate as a predicate related to the domain "people" and the range "Time". Here, the new predicate is a string including "being_born", for example, may be "being_bornTime".

The predicate currently defined in the knowledge expression language ("birthday") contains time information similar to "1944", but "1944" is the birth year, not "birthday", so that it can express insufficient knowledge. Thus, the device 100 may replace "being_born" or more specifically "being_bornTime" with a predicate instead of "birthday."

As such, the apparatus 100 may automatically extend the limited expressive power of the knowledge expression language using the semantic expression language, and thereby, may construct a knowledge expression language capable of extracting more accurate knowledge.

On the other hand, the device 100 may determine that "place" of the lexical type of the argument is similar to "Place" which is the range of the ternary relationship. The predicate (fredkit) associated with the domain "people" and the scope "Place" is "birthplace". The apparatus 100 may extract knowledge by using "birthplace" as it is or by using a predicate extended to "being_bornPlace".

The device 100 may extend knowledge representation power of ontology-based knowledge database as well as Divpedia. The apparatus 100 may be ontology in a format in which a classification of a word of a sentence is designated, such as a framenet, and may be extended to a semantic expression language in which arguments related to a word are designated.

As described above, according to the exemplary embodiment of the present invention, when the knowledge extracted from any text cannot be expressed as the knowledge expression language used in the knowledge expression ontology, the knowledge expression may be extended using the semantic expression language. That is, according to the embodiment of the present invention can solve the problem that the knowledge representation ontology does not have sufficient coverage when building the knowledge database from the web text.

The knowledge expression expansion apparatus 100 may store instructions for performing the knowledge expression expansion method described with reference to FIGS. 1 to 8, or may be stored in a memory or a memory for temporarily storing the instructions by loading the instructions from the storage device. And a processor for processing the knowledge representation extension method of the present invention by executing instructions, or loaded instructions. Instructions for performing the knowledge expression extension method described with reference to FIGS. 1 to 8 are implemented as a program that can be processed by a processor.

The embodiments of the present invention described above are not only implemented through the apparatus and the method, but may be implemented through a program for realizing a function corresponding to the configuration of the embodiments of the present invention or a recording medium on which the program is recorded.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims

As a knowledge expression expansion device,

A predicate-argument structure interpreter that extracts a predicate and at least one argument from text using a semantic expression language;

An ontology branch that expresses knowledge using a knowledge expression language, which is a structure that the computer can understand, and

Extracting a second predicate corresponding to the first predicate extracted by the predicate-non-serial structure analysis unit from the ontology unit and using the first predicate when the similarity between the first predicate and the second predicate is equal to or less than a reference value Knowledge expression unit for expressing knowledge extracted from the text

Knowledge expression expansion device comprising a.
In claim 1,

The knowledge expression unit

And a knowledge expression extension device for extracting the second predicate related to the at least one argument from the ontology unit.
In claim 2,

The knowledge expression unit

Extracting a first domain that is similar to the lexical type assigned to the at least one argument from the domains of the knowledge expression language, and having a reference value that is similar to or greater than the reference value; And a first range similar to the above, and extracting the first domain and a predicate related to the first range as the second predicate.
In claim 3,

The knowledge expression unit

And generating a character string combining information related to any one of the first predicate and the at least one argument, and adding the character string to the knowledge expression language of the ontology part.
In claim 1,

The knowledge expression language is a knowledge expression extension device that is a language expressed in a resource description framework (RDF) ternary relationship.
As a device extends knowledge representation,

Receiving text including at least one sentence,

Expressing the text as a first predicate and at least one argument based on a semantic expression language,

Extracting a second predicate corresponding to the first predicate from the knowledge expression ontology;

Comparing the similarity between the first predicate and the second predicate, and

Expressing the knowledge extracted from the text using the first predicate when the similarity is equal to or less than a reference value

Knowledge expression expansion method comprising a.
In claim 6,

Extracting a second predicate corresponding to the first predicate

And extracting the second predicate corresponding to the first predicate from the knowledge representation ontology using the lexical type given to the at least one argument.
In claim 6,

The knowledge expression ontology uses a knowledge expression language that expresses knowledge in a ternary relation of a subject, a predicate, and an object.

Extracting a second predicate corresponding to the first predicate

A predicate kit that is similar to the lexical type assigned to the at least one item among the subjects of the knowledge expression language or more than the reference value, and is similar to the lexical type assigned to the at least one item among the objects of the knowledge expression language. Knowledge expression extension method for extracting the expression as the second predicate
In claim 6,

Expressing using the first predicate

A knowledge expression extension method for generating a string in which information related to any one of the first predicate and the at least one argument is combined, and expressing knowledge extracted from the text using the string.
In claim 9,

Adding the string to a knowledge expression language of the knowledge representation ontology

Knowledge expression extension method further including
As a device extends knowledge representation,

Interpreting the predicate-argument structure of the text,

Matching the predicate-determination structure of the text to a ternary relation of the knowledge expression language, and

Adding a first predicate extracted from the predicate-dissertation structure of the text as a predicate of the knowledge expression language based on a matching similarity;

Knowledge expression expansion method comprising a.
In claim 11,

Adding as a predicate of the knowledge expression language

Extracting a second predicate matching the first predicate of the predicate-nonserial structure of the text from the ternary relation of the knowledge expression language;

Comparing the similarity between the first predicate and the second predicate, and

If the similarity is equal to or less than a reference value, adding the first predicate to the knowledge expression language.

Knowledge expression expansion method comprising a.
In claim 11,

Expressing the text in a ternary relationship using the first predicate

Knowledge expression expansion method further comprising.
In claim 11,

Matching with the ternary relation of the knowledge expression language

And extending the predicate-non-argument structure of the text into the ternary relationship based on the similarity between the domains and the range of the ternary relationship extracted from the predicate-terminal structure of the text.