Knowledge based information retrieval system

What is claimed is:

1. A knowledge based document retrieval system, comprising:

user input means for inputting, in response to at least one of a user's key-typing and mouse operations, a series of words;

display means for displaying responses from said system and retrieved documents;

user input analysis means for analyzing said series of words inputted by said user through said user input means, and converting said series of words into an internal query condition based on information related to various concepts and permitting said user to edit said internal query condition;

a knowledge base for storing knowledge including said concepts and relations among said concepts;

wherein said stored knowledge is represented by concept nodes and relation links forming a network of concepts, wherein each of said nodes represents a concept, and wherein each of said links represents a relation among said concepts;

information search means for identifying concept nodes that match said internal query condition semantically; and

information retrieval means for retrieving at least one relevant document associated with said identified concept nodes;

wherein said user input analysis means includes:

lexicon storing means having contents which are an edited version of said concept network,

lexical analysis means for identifying concept nodes from said series of words inputted by said user input means by consulting the contents of said lexicon storing means,

syntactic analysis means for identifying a nominal compound in said series of words based on said concept nodes identified by said lexical analysis means, and

nominal compound interpretation means for mapping said identified nominal compound relative to said concepts and relations of said knowledge base, inferring meaning of said identified nominal compound based on said concepts and relations of said knowledge base and generating said internal query condition based on said meaning of said identified nominal compound.

2. A knowledge based file retrieval system according to claim 1, wherein said nominal compound interpretation means generates all possible interpretations, that are represented by interrelationships among said series of nouns, such that interrelationships are relations existing in said concept network, and evaluates likelihood of interpretations depending on the number of instance relations existing in said concept network, wherein instance relations represent concrete facts.

3. A knowledge based file retrieval system according to claim 1

wherein said information storage means stores files representing documents;

wherein said concept nodes includes nodes that represent said documents stored in said information storage means; and

wherein said information retrieval means retrieves documents that are associated with said identified concept nodes.

4. A knowledge-based document retrieval system, comprising:

user input means for inputting, in response to at least one of a user's key-typing and mouse operations, a series of words;

user interaction means for creating a query expression and an internal query condition from a dialogue, including said inputted series of words, between said user and said system, wherein said query expression is a displayed version of said internal query condition for said user to permit editing by said user, said query expression and said internal query condition being created based on information related to various concepts;

a knowledge base for storing knowledge including said concepts and relations among said concepts, wherein said stored knowledge is represented by concept nodes and relation links forming a concept network, wherein each of said nodes represents a concept, and wherein each of said links represents a relation among said concepts;

information search means for identifying concept nodes that match said internal query condition semantically;

information retrieval means for retrieving at least one relevant document associated with said identified concept nodes; and

display means for displaying responses from said system and retrieved documents;

wherein said user interaction means includes:

query editing means for adding to said internal query condition a new condition phrase having concepts and relations defined in said concept network, deleting one of existing condition phrases, and changing one of concepts in said condition phrases to a different concept in response to said series of words inputted by said user,

query expression display means for displaying on said display means said query expression corresponding to said internal query condition being created and edited, and

concept tree display means for displaying on said display means part of said concept network in a hierarchical tree, wherein said tree includes one of the concepts appearing in said query expression.

5. A knowledge based document retrieval system according to claim 4, wherein said user interaction means further comprises:

means for searching concepts connected by a generic relationship added to a current concept in the query expression among the concepts connected with said current concept through subsumption relationships, displaying on said display means the concept belonging to the lowest rank in the subsumption relations among these concepts and moving said current concept to the concept being displayed.

6. A knowledge based document retrieval system according to claim 4, wherein said user interaction means further comprises:

means for imposing conditions one after another to a concept in the query expression, a current concept moving freely among the concepts in the query expression.

7. A knowledge based document retrieval system according to claim 4, wherein said user interaction means further comprises:

means for adding a root to a concept, which is to be queried in the query expression and modifying the concept which is to be queried.

8. A knowledge based document retrieval system according to claim 4, wherein said user interaction means further comprises:

means for making the query of the concepts, to which the condition is imposed, in the query expression, possible one after another.

9. A document retrieval system using a conceptual network, comprising:

a knowledge base for storing a plurality of words representing concepts and a plurality of predicates representing relations between the plurality of words;

input means for inputting words;

display means for displaying words;

processing means, responsive to inputting of a word representing a concept which serves as a query key from said input means, for retrieving a word representing a concept in association with the inputted word from said knowledge base, and for displaying on said display means at least one word having a relation with said inputted word and at least one predicate representing said relation with said inputted word;

editing means for selecting, in response to said user, a set of a desired word and a predicate from those displayed on said display means and inputting, by said user, a new word subsumed by the selected word to produce a query condition in which said inputted word and the newly inputted word are associated with each other by said selected predicate; and

means for retrieving at least one relevant document related to said query condition.

10. A document retrieving method in an intellectual retrieval system, comprising the steps, performed by said intellectual retrieval system, of:

storing a plurality of words representing concepts and relations between the plurality of words as a knowledge base;

displaying on a screen a query condition defined by selected words and predicates representing relations between the selected words, and displaying a plurality of other words and relations between the words, said plurality of other words having a relation with a certain word in said query condition and being stored in said knowledge base;

permitting a user of said intellectual retrieval system to select one word and one relation from the displayed plurality of other words and relations;

rewriting and displaying said query condition in accordance with the selected word and the relation; and

retrieving at least one relevant document related to said query condition.

11. A method according to claim 10, wherein a certain word in said query condition is designated on the screen, and a plurality of other words having a relation with said designated word and stored in said knowledge base, relations between the plurality of other words are displayed on said screen in response to the designation.

12. A knowledge-based document retrieval system according to claim 4, further comprising:

means for rewriting the display of query expressions by substituting a current concept for a certain concept in said dialogue window representing the network by the current concept.

13. A knowledge-based document retrieval system according to claim 4, wherein said query editing means controls said query expression display means and said concept tree display means, such that, in response to a user selection of a concept in a query expression by means of a mouse operation, said concept tree display means identifies a subset of said concepts corresponding to said selected concept, and displays said subset of concepts in a hierarchical tree.

14. A knowledge based document retrieval system according to claim 13, wherein said query editing means control said query expression display means and said concept tree display means, such that, in response to user selection of a concept in said concept tree display, said query editing means substitutes a corresponding concept in a query expression by said selected concept, and said query expression display means displays the updated version of said query expression.

15. A knowledge based document retrieval system according to claim 4, wherein said query editing means controls said query expression display means to display possible relations that can be attached to a concept in response to a user's request to add new condition phrase to a concept in an existing query expression being edited.

16. A knowledge-based document retrieval system, comprising:

user input means for inputting, in response to at least one of a user's key-typing and mouse operations, a series of words;

a knowledge base for storing knowledge including concepts and relations among said concepts, wherein said stored knowledge is represented by concept nodes and relation links forming a concept network, wherein each of said nodes represents a concept, and wherein each of said links represent a relation among said concepts;

information storage means for storing said knowledge base and documents, wherein each document has a corresponding concept node defined in said knowledge base;

information search means for identifying concept nodes that match an internal query condition represented in terms of concepts and relations defined in said knowledge base;

information retrieval means for retrieving documents from said information storage means, wherein said documents are associated with said identified concept nodes;

display means for displaying responses from said system and retrieved documents; and

user interaction means for creating said internal query condition from a dialogue, including said inputted series of words, between said user and said system, wherein said internal query condition is displayed in a query expression to permit said user to edit;

wherein said user interaction means includes:

query editing means for adding to said internal query condition a new condition phrase having concepts and relations defined in said knowledge base, deleting one of existing condition phrases, and changing one of concepts in said condition phrases to a different concept in said knowledge base in response to said series of words inputted by said user,

query expression display means for displaying on said display means said query expression which is a display version of said internal query condition being created and edited, and

concept tree display means for displaying on said display means a subset of concepts defined in said knowledge base in a hierarchical tree, wherein said tree includes one of the concepts appearing in said query internal condition being edited.

Description Submit all comments and votes

BACKGROUND OF THE INVENTION

This invention relates to knowledge based information retrieval system and in particular to a human interface of an intellectual query system permitting the end user to query efficiently information stored in a network structure in an electronic file.

This interface can be divided into the natural language interface and the visual interface. The natural language interface is suitable for a global search, by which the search is effected by deduction from the natural language and the visual interface is suitable for a search from a domain, which can be seen by eyes, i.e. for a local search.

Heretofore, as a human interface using the natural language, there is known a natural language interface for the database. Therefor there are references, G. G. Hendrix et al., "Developing a Natural Language Interface to Complex Data", ACM Trans. Database Systems, Vol. 3, 1978, pp. 105-147, etc. In these systems a data model for the database (method, by which the relation between different data items, which are to be memorized, is expressed) and a grammar and a dictionary for interpreting the natural language are set independently. That is, when the natural language interface is added to an existing database, it is necessary to construct newly a grammar and a dictionary. Or it has a problematical point that it is necessary to modify the grammar and the dictionary for the natural language interface, when the object database is changed.

Further, heretofore, a database, to which the natural language interface is given, is a relational database and the formal language for the search used therefor, i.e. a quasi-standardized SQL (Structural Query Language), is weak in the capability for describing high order knowledge. Since a query expression by the natural language is translated usually into a formal language as such an intermediate language, it has a problematical point that the function of the whole system is restricted by the expressing capacity of this formal language. In particular, although the relational database is useful for uniform data, it cannot be said that it is satisfactorily suitable for a heterogeneous database, by which various kinds of matters are dealt with, or an object-oriented database. For example, it is not suitable for describing a matter, based on ambiguous memory of a user, and querying information concerning it, based on that description.

Further, in these systems means other than the natural language is used for inputting data (new knowledge and information) and the data input is carried out by a specialist. Consequently there is a problem that it is difficult for an end user to input and register directly the data.

Furthermore, as small scale and large capacity memory devices such as optical disk storage units have been realized, document filing devices directed to offices, for which the end user operates directly the processing of supervision and search of the database, capable of storing and querying a large amount of information, which has been effected heretofore by a specialist, have been realized.

As a method for facilitating memory and search of information in such a filing device, e.g. JP-A-61-220027 can be referred to. This literature discloses an information querying method enabling the end user to query easily desired documents etc. from ambiguous and fragmental information and at the same time to facilitate their registration. However, by this method, it is very difficult to form query conditions, under which information required by the user can be appropriately taken out, when the query conditions for effecting the query from the knowledge base are formed.

SUMMARY OF THE INVENTION

A first object of this invention is to solve the problematical points as described above and to enable the end user to query desired information from a description by a natural language even on the basis of fragmental memory. Furthermore it is to enable the user itself to register new information and knowledge similarly by using the natural language.

A second object of this invention is to provide a system, by which the user finds a concept, which he seeks, without any feeling that he is querying, by facilitating modification of concept in query expressions, enabling him to modify the object to be queried, to query one after another even in the course of formation of query expressions and to query locally the query expressions.

In order to achieve the above first object, this invention is characterized in that a common knowledge expression base is given to the knowledge base and the natural language interface so that the query and the registration of knowledge and information can be effected by using the natural language.

Concretely speaking, this invention gives a knowledge representation method (corresponding to a data model in the data base) called "concept relation model" expressing a system of matters and the fact with "concept" and "relation" as a method for constructing the knowledge base, and further provides a method, by which knowledge of language can be memorized also in the knowledge base. Here a "concept" means a "data item" in a computer representing matters, events or abstract concept and a "relation" means a "data item" defined between different concepts. The concept can be represented by a node (apex) and the relation can be represented by a link (side). Knowledge represented by a concept relation model constitutes therefore a network of concepts. Here this is called a conceptual network.

That is, the knowledge base according to this invention is characterized in that the knowledge, which is originally desired to be stored, is stored in the conceptual network as one body together with the knowledge for expressing it by using a language and that the natural language interface uses the same knowledge in common. Consequently, in principle it is not necessary to construct newly a dictionary, etc. for the natural language interface.

Further this invention provides a natural language understanding method, by which the meaning of a query expression expressed by a natural language is interpreted by effecting deduction from the matters, etc. stored in the knowledge base. In particular, it gives a method for interpreting the meaning of nominal compounds consisting of a plurality of series of nouns, which we use frequently. In order to interpret the meaning of the nominal compounds, it is necessary that the system deduces relations between different nouns and this invention gives a method, by which only significant relations are deducted from the concepts and the relations stored in the knowledge base.

Furthermore, the knowledge representation method according to this invention restricts the part depending on the language so as to facilitate the application to a plurality of languages. In addition, it makes the coexistence of expressions by different languages possible. Consequently this invention provides a method, by which it is possible to query and register information in e.g. both English and Japanese in a same knowledge base.

In order to achieve the above second object, according to this invention, display of a superconcept of a current concept in a query expression together with the query expression and a network display from the current concept and changeable superconcepts to subconcepts satisfying conditions added to the current concept are effected by utilizing multiple-window functions. Further the object can be achieved by the fact that displacement of the current concept due to shifting between different concepts in the query expression, modification of the current concept within concepts satisfying the added conditions, addition of restrictions to the current concept, addition of a root and query of concepts satisfying the added conditions can be always executed.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a functional block diagram representing a natural language expression interpreting program by the method according to this invention;

FIG. 2 is a scheme of a concept network for explaining a knowledge expressing method according to the concept relation model;

FIG. 3 is a scheme indicating names of concept, FIG. 3 and the followings being schemes illustrating the knowledge memory according to the same model;

FIG. 4 is a scheme indicating subsumption relations;

FIG. 5 is a scheme indicating the generic relationship definition;

FIG. 6 is a scheme indicating the relations;

FIG. 7 is a scheme for explaining the principle of the method for interpreting the meaning of nominal compounds;

FIG. 8 is a scheme for explaining a method for analyzing the structure of sentences;

FIGS. 9, 10 and 11 show examples of the analysis of the structure of sentences;

FIG. 12 is a table indicating prepositions;

FIG. 13 is a table indicating relational descriptor;

FIG. 14 is a block diagram illustrating the construction of the hardware for a system, which is an embodiment of this invention;

FIG. 15 is a scheme expressing concepts and relation knowledge stored in the data base;

FIGS. 16 and 17 show images on a screen when concept matching is effected;

FIG. 18 shows an example of the table added, when conditions are added;

FIG. 19 shows an example of the table added, when a root is added;

FIG. 20 is a scheme illustrating the construction of a system according to this invention;

FIGS. 21 to 32 show images on the screen appearing in the process of a query of information; and

FIGS. 33 to 36 are flow charts for the processing according to this invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

Hereinbelow a concrete embodiment of this invention will be explained. At first an embodiment concerning the natural language interface will be described.

At first the fundamental principle of this embodiment will be explained. The knowledge expressing method by using the concept relation model serving as the base of this invention will be explained. FIG. 2 indicates a part of a conceptual network. In the figure ellipses represent concepts (nodes) and arrows represent relations (links). A node 201 "UNIVERSAL" is a root node representing all within the knowledge base. For each of the nodes it is possible to define more than one series of letters as the name of the relevant concept. For example a synonym or a corresponding word in a foreign language can be added thereto.

On the other hand, in the links connecting different nodes there are subsumption relations (IS-A link 202) defined between two nodes, between which a property is inherited, "generic relationships" (link 203) defined generally between different concepts, and "instance relations" (link 204) as concrete examples of the same generic relationships. The subsumption relation represents the class of matters. Consequently the conceptual network consisting of the concepts and the subsumption relations constitutes a conceptual tree representing a taxonomic hierarchy.

For example the conceptual tree within the conceptual network indicated in FIG. 2 represents the following knowledges;

(PAPER-MATERIAL (is-a THING))

(BOOK (is-a PAPER-MATERIAL))

(BOOK#0051 (is-a BOOK))

(LIVING-THING (is-a THING))

(PERSON (is-a LIVING-THING))

(NEWTON (is-a PERSON))

These are called frame expressions. These are expressed by the symbolic expression used for the LISP language. These can be expressed also in usual English and written as follows;

______________________________________ Paper material is a thing. Book is a paper material. etc. ______________________________________

In FIG. 2, an example of the generic relationship is the relation 203 defined between the concept 211 (BOOK) and the concept 212 (PERSON). This means that there can be a relation "author" or "work" between "book" and "person". This generic relationship can be read in both the directions, i.e. towards either right or left, as follows;

(BOOK (is-written-by PERSON))

(PERSON (has-written BOOK))

The instance relation is a relation as indicated by a link 204 indicated by a broken line in FIG. 2, which represents a concrete example (called also instance value for the data base) of a certain generic relationship. For example, in FIG. 2, an instance relation 204 is defined between a concept 213 and a concept 214 as a concrete example of the generic relationship 203, which is "author". With the frame expression it can be written as follows;

______________________________________ (BOOK#0051 (is-a BOOK) (is-written-by NEWTON)) (NEWTON (is-a PERSON) (has-written BOOK#0051)) ______________________________________

With the natural language they can be written;

______________________________________ BOOK#0051 is a book which is written by NEWTON. NEWTON is a person who has written BOOK#0051. ______________________________________

The knowledge described above is memorized, according to this invention, in a data structure, as stated below. At first the concept and the name thereof are memorized in a concept name table 221 (FIG. 3). The same table 221 consists of three columns 222, 223 and 224. The column 222 indicates the unique number C# of the concept and the name CNAME thereof is defined in the column 223. The language LANG of the name is prescribed in the column 224. For example, when the value of LANG is "J", the name is written in Japanese and when it is "E", the name is written in English.

Further, a plurality of concept names can be defined for one language. For this reason, for the data structure in the column 223 it is allowed to repeat the data. For example the name of the concept C#0004 is "book" (in Japanese) and "BOOK" and "printed matter" (in Japanese) can be added also thereto.

Now, the subsumption relation of the concept is represented in a subsumption relation table 231 indicated in FIG. 4. Columns 232 and 233 indicate unique numbers C# and SC# for respective concepts and represent that the superconcept of the concept C# is the concept SC#. For example, the second record in the table 231 indicates that the superconcept of the concept C#0002 ("paper material") is the concept C#0001 ("thing"). The "relation" such as a property, which is defined for each concept, is inherited from a higher rank to a lower rank through a link of the subsumption relation. In this case, it is possible to define a plurality of superconcepts for one concept. Consequently multiple property inheritance is realized.

Various kinds of relations between different concepts other than the subsumption relation can be defined in a generic relationship defining table 241 indicated in FIG. 5. Each of the generic relationships represents the kind of the relation. Basically there is no limit for the number of such kinds of relations and it is possible to define an arbitrary number of generic relationships.

The generic relationship defining table 241 defines principally "reading", when the "relation" is expressed by a natural language. A column 244 indicates "reading" LR, when the relevant relation is read from left to right, and a column 245 defines "reading" RL read from right to left contrarily thereto. In the data structure of these columns it is allowed to repeat data so that it is possible to define a plurality of readings. Further, just as for the concept name table, it is possible to specify the language for the reading. Consequently it is possible to express same data in a plurality of languages.

In the example indicated in FIG. 5 the relation "AUTHORSHIP" can be expressed by a natural language (English in this case) as follows;

PERSON who is author of BOOK

PERSON who is the author of BOOK

PERSON who wrote BOOK

PERSON who has written BOOK

or

BOOK whose author is PERSON

BOOK by PERSON

BOOK from PERSON

BOOK of PERSON

This is the same also for the Japanese expression (not described).

The existance of relations between different concepts is memorized according to a relation table 251 indicated in FIG. 6. As explained above, in the relation links, there are generic relationships and instance relations. These are distinguished by a column 256 in the table 251. When the value in a column CLASS is GR, it is a generic relationship and when the value is INST, it is an instance relation. In the example indicated in FIG. 6, the first record represents the generic relationship 203 in FIG. 2 and the second record indicates the instance relation 204 in FIG. 2. Further a column C#L defines the concept on the left side and a column C#R the concept on the right side. In this case, on which side a certain concept is located, right or left, depends on the definition and as far as there is no contradiction for the tables 241 and 251, it may be defined on either side.

Now the principle of the natural language understanding method based on the knowledge expressing method described above will be explained.

At first the method for understanding nominal compounds, which is the most important in the object-oriented knowledge base will be explained. Here a nominal compound means a noun phrase consisting of a series of nouns including partially adjectives. For example, the following are examples of the nominal compounds;

______________________________________ supercomputer article (1) ElectronicsWeek article (2) Japanese personal computer company (3) Americal personal computer (4) software packages ______________________________________

In this case understanding the significance means to obtain positively the relation among these adjectives and nouns.

For example, although the nominal compounds (1) and (2) have a same structure, they have different significances. They should be interpreted as follows; (1) means; "article whose subject is supercomputer" and (2) means; "article which is part of ElectronicsWeek". That is, it is necessary to deduce that in (1) "article" and "supercomputer" are combined through a relation "subject-is" and that in (2) "article" and "ElectronicsWeek" are combined through a relation "is-part-of". Understanding the significance is to extract automatically following structures, when they are described in the frame form;

(ARTICLE (subject-is SUPERCOMPUTER)) (5)

(ARTICLE (is-part-of ElectronicsWeek)) (6)

By the method for natural language understanding according to this invention, the significance is interpreted, as follows, on the basis of the knowledge indicated in FIG. 7. At first, as knowledge making this deduction possible, relations RS#0011 as generic relationships;

(ARTICLE (subject-is UNIVERSAL)) (7a)

(UNIVERSAL (is-subject-of ARTICLE)) (7b)

and relations RS#0012

(ARTICLE (is-part-of JOURNAL)) (8a)

(JOURNAL (has-part-of ARTICLE)) (8b)

should be defined. That is, it is necessary that "anything can be a subject of an article" and "the article is a part of a journal (the article is published in a part of a journal)" are memorized as knowledge.

Further, as a subsumption,

(SUPERCOMPUTER (is-a THING)) (9)

(THING (is-a UNIVERSAL)) (10)

(ElectronicsWeek (is-a JOURNAL)) (11)

should be memorized.

By using these memories it is possible to interpret "supercomputer article". At first, following the subsumption relation towards the higher rank from SUPERCOMPUTER, it is possible to understand;

(SUPERCOMPUTER (is-a UNIVERSAL))

(UNIVERSAL (is-subject-of ARTICLE)).

As the result, by the property inheritance, it is deduced that there can be relations;

(SUPERCOMPUTER (is-subject-of ARTICLE))

or

(ARTICLE (whose Subject-is SUPERCOMPUTER)).

That is, it is deduced that "a supercomputer can be a subject of an article". In this case, since there is no other interpretation, the interpretation;

"article whose subject is supercomputer"

is adopted.

The interpretation of the significance of the nominal compound (2) is a little more complicated.

In this case, since

(ElectronicsWeek (is-a JOURNAL))

(JOURNAL (has-part-of ARTICLE))

and at the same time

(ElectronicsWeek (is-a UNIVERSAL))

(UNIVERSAL (is-subject-of ARTICLE))

as it can be clearly seen from FIG. 7, it is deduced that there can be two relations;

(ElectronicsWeek (has-part-of ARTICLE))

and

(ElectronicsWeek (is-subject-of ARTICLE)).

That is, it is understood that there can be two interpretations;

"article which is part of ElectronicsWeek"

and

"article whose subject is ElectronicsWeek"

In the case where there exist a plurality of candidates of interpretation, the method according to this invention utilizes a heuristic method, by which the likelihood of the interpretations is evaluated, depending on which interpretation has more concrete examples.

Concretely speaking, in the preceding example, the numbers of instance relations for the relation RS#0011 and the relation RS#0012, which are registered, are counted, respectively, while querying the subconcepts of the concept "ARTICLE" and the concept "ElectronicsWeek", including themselves. In the example indicated in FIG. 7, O for the former and one concrete relation for the latter are registered. That is, there is no article, whose subject is "ElectronicsWeek", but there is one article, ARTICLE #0101, which is published in "ElectronicsWeek". Consequently the relation RS#0012 (is-part-of) is selected as the more suitable interpretation. That is, it is interpreted as follows;

"article which is part of ElectronicsWeek".

As explained above, the interpretation of nominal compounds is based on a deduction processing of the relation between 2 nouns. That is, the basic processing of the interpretation of a nominal compound consisting of more than 3 words, as explained below, consists of extracting the relation between 2 words described above. This will be explained below, taking the nominal compound (3) as an example.

At first, the concept corresponding to each of the words is selected, while examining whether there are concept names consisting of a composite word among the words constituting the nominal compound or not. That is, the words are cut-off one after another from the beginning and it is examined whether they are registered or not, referring to the concept name table.

In the case of the nominal compound (3), partial series of words such as;

______________________________________ "Japanese" "Japanese personal" "Japanese personal computer" "Japanese personal computer company" "personal" "personal computer" "personal computer company" "computer company" "company" ______________________________________

are cut-off and it is examined whether each of them is a concept name or not.

At this time, the method according to this invention is characterized in that an adjective is registered as a synonym of the concept, whose name is the noun form corresponding thereto, and the adjective is dealt with as a same concept as the noun. For example, the adjective "Japanese" is registered as a synonym of the concept "JAPAN" or the concept "Japanese people" and dealt with as the same concept.

Consequently, as the result, supposing that "personal computer" is defined as a concept name "PERSONAL-COMPUTER" the nominal compound (3) is at first recognized as;

(JAPAN PERSONAL-COMPUTER COMPANY)

(JAPANESE-PEOPLE PERSONAL-COMPUTER COMPANY).

However, in the following explanation, in order to facilitate understanding, explanation will be made, omitting the latter, for which it is understood finally to be a meaningless interpretation.

That is, at this step, it is understood that the nominal compound is a combination of substantially three concepts. This can be expressed by using parentheses as follows;

(Japanese (personal computer) company) (12)

Therefore the following processing is to examine how these three concepts are related with each other. In this case it can be seen that there are the following two possibilities;

(Japanese ((personal computer) company)) (13)

((Japanese (personal computer)) company) (14)

At first, in the case of (13), it is necessary to deduct two relations, which can be connected between COMPANY and PERSONAL-COMPUTER and between COMPANY and JAPAN. In this case, by the method described previously for deducing the relation, following relations;

______________________________________ (COMPANY (15a) (produces PERSONAL-COMPUTER) (is-located-in JAPAN)) (COMPANY (15b) (has-developed PERSONAL-COMPUTER) (is-located-in JAPAN)) ______________________________________

are extracted. Here, in order to evaluate the priority (likelihood) of a plurality of interpretations, the total numbers of concrete examples of the two relations between COMPANY and PERSONAL-COMPUTER and between COMPANY and JAPAN (concrete relations defined in the subconcepts) are counted for (15a) and (15b), respectively, so as to obtain weights for these relations. In order to obtain the evaluation for all the relations, it is normalized by dividing each of the numbers of the instance relations (weight of the relation) by the number of the generic relationships. For the examples of (15a) and (15b) the number of the generic relationships is 2.

Then, the relation is extracted for the second possibility (14). In this case two relations between COMPANY and PERSONAL-COMPUTER and between PERSONAL-COMPUTER and JAPAN should be obtained. For the former, two relations;

(COMPANY (produces PERSONAL-COMPUTER)) (16a)

and (COMPANY (has-developed PERSONAL-COMPUTER)) (16b)

can be found (on the presumed knowledge base). In the same way, for the latter two relations

______________________________________ (PERSONAL-COMPUTER (17a) (is-produced-by (COMPANY (is-located-in JAPAN))) and (PERSONAL-COMPUTER (17b) (was-developed-by (COMPANY (is-located-in JAPAN))) ______________________________________

are found. At this time, since there is no relation connecting directly PERSONAL-COMPUTER and JAPAN, the concept COMPANY relating indirectly these two is found automatically.

By the method according to this invention, when no relation relating directly two concepts is found, as stated abo