JP2006190101A - Natural language analysis device, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically interpret the correspondence in meaning among a plurality of clauses or a plurality of sentences included in an input text. <P>SOLUTION: A natural language analysis device 1 for interpreting the meaning of an input text composed of a plurality of clauses or a plurality of sentences includes a problem-frame setting clause specifying means for specifying a problem-frame setting clause that sets a problem frame to present a scene for positioning and interpreting the meaning of each clause or each sentence according to a phenomenon or a predicate represented by each clause or each sentence, from among the plurality of clauses or the plurality of sentences; and a clause meaning/expression associating means for implementing a predetermined process for associating either a clause other than the problem-frame setting clause specified by the problem-frame setting clause specifying means or a sentence with the problem-frame setting clause in terms of meaning. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a natural language analysis device, a natural language analysis method, and a natural language analysis program for interpreting the meaning of an input sentence composed of a plurality of sections and a plurality of sentences.

  Conventionally, various theories and methods have been proposed in order to construct a natural language analysis system capable of realizing a dialog interface and the like. In such a natural language analysis system, when a sentence to be analyzed (hereinafter referred to as an input sentence) is input to the natural language analysis system, the meaning of the input sentence is automatically interpreted, and the semantic content of the input sentence. The purpose is to solve various problems, such as determining the truth of the message and returning the result, or if the input sentence is a question sentence, returning a response sentence containing the answer to the question. . However, a natural language analysis system that can realize a high-functional dialog interface has not been constructed yet. This is because the ability to correctly interpret the meaning of words and phrases included in the input sentence is insufficient.

  Here, the following main methods have been proposed in the past for identifying the meaning of the input sentence. The first method is a keyword-based method. In this method, a large number of keywords that are considered important for problem solving are stored in advance on the natural language analysis system side. Then, when an input sentence is input to the natural language analysis system, the keywords included in the input sentence and the keywords stored in the natural language analysis system are compared with each other. By recognizing what keywords are included in the input sentence, the meaning of the input sentence is estimated.

  The second method is a method using a template. In this method, a template that represents a skeleton of a semantic expression or a syntactic structure that seems to be important for problem solving is stored in advance on the natural language analysis system side. Then, when an input sentence is input to the natural language analysis system, it is determined whether or not all or a part of the input sentence matches a template stored in the natural language analysis system. This is a method for estimating the meaning of an input sentence based on a determination result (see Non-Patent Document 1, for example).

  The third method is a method using the similarity between the parse tree and the semantic expression. In this method, a scale for evaluating the similarity between a parse tree and a semantic expression, and various kinds of information that are considered necessary for problem solving are stored in advance on the natural language analysis system side. When an input sentence is input to the natural language analysis system, the syntax analysis having the highest similarity to the parse tree or semantic expression of the input sentence from the information stored in the natural language analysis system. Information having a tree or semantic expression is searched, and an answer is made using this information (see, for example, Non-Patent Document 2).

CONCEPTUAL INFORMATION PROCESSING, Sunk, R.C. C. , North-Holland, 1975 “Evaluation method of syntactic similarity of text”, Tetsuro Takahashi, Kentaro Inui, Yuji Matsumoto, IPSJ Natural Language Processing Study Group, NL-150-24, 2002

  However, all of the above-described conventional first to third methods have difficulty in correctly identifying the meaning of the input sentence, and the problem solving accuracy is low. For example, in the method based on the keyword, only the words that match the keyword are focused on among the words included in the input sentence, and the meanings of the other words are ignored. The possibility of correctly estimating the meaning of the input sentence was extremely low. In addition, the method using a template can identify a wider range of meanings and sentence structures than the method based on a keyword. It is necessary to prepare a different template for each structure. For this reason, it is necessary to prepare an enormous number of templates in a problem area in which various sentences including synonymous expressions may be input, which is denied in Non-Patent Document 1, and is difficult to implement. Therefore, the reality was poor. In the method using the similarity between the parse tree and the semantic expression, an essentially non-linear relationship such as the similarity (closeness) between the structured data is measured by a linear quantity. For example, expressions of various meanings included in a sentence may include very various synonymous expressions, and it is difficult to evaluate their meanings with linear quantities. Since synonymous expressions represent the same meaning despite having different parse trees and semantic representation structures, different parse trees and semantic expressions are necessary to accurately analyze such expressions. Even if the sentence has the structure, the similarity must be evaluated as 1 or a value close thereto. However, since a method for accurately performing such evaluation has not yet been proposed, it is still difficult to correctly estimate the meaning of the input sentence.

  In order to solve such a problem, the inventor of the present application can stably compare and verify the meaning of the input sentence by automatically interpreting the semantic correspondence of words and phrases included in the input sentence. A natural language analysis device, a natural language analysis method, and a program that can be performed have been proposed previously (Japanese Patent Application No. 2004-351874, hereinafter referred to as “prior patent application”). According to this natural language analysis apparatus or the like, the semantic correspondence between words and phrases included in an input sentence can be mechanically read. However, this natural language analysis device, etc., is based on the premise that the input sentence is a single sentence. When an input sentence composed of a plurality of sections or a plurality of sentences is targeted, the plurality of sections or a plurality of sections In order to read the mutual semantic correspondence between the sentences, it is necessary to further interpret the mutual semantic correspondence between a plurality of clauses and a plurality of sentences included in the input sentence.

  The present invention has been made in view of such problems, and by automatically interpreting the semantic correspondence between a plurality of clauses and a plurality of sentences included in the input sentence, the meaning of the input sentence is improved. An object of the present invention is to provide a natural language analysis apparatus, a natural language analysis method, and a natural language analysis program that enable stable comparison and collation.

  In order to solve the above-described problems and achieve the object, the natural language analysis apparatus according to claim 1 is a natural language analysis apparatus for interpreting the meaning of an input sentence composed of a plurality of clauses or a plurality of sentences. A problem of presenting a scene for positioning and interpreting the meaning of each section or each sentence based on the phenomenon or predicate indicated by each section or each sentence from among the plurality of sections or a plurality of sentences. A problem frame setting section specifying means for specifying a problem frame setting section for setting a frame (for example, a target section meaning interpretation processing unit 53a described later) and a problem frame setting specified by the problem frame setting section specifying means A clause meaning expression associating means (for example, a clause meaning expression processing unit 53d described later) that performs a predetermined process for semantically associating a clause or sentence other than the problem frame setting clause with the clause; It is characterized by providing.

  The natural language analysis apparatus according to claim 2 is the natural language analysis apparatus according to claim 1, wherein the problem frame setting clause specifying unit is configured to set the plurality of clauses or the plurality of sentences to each of the clauses or the respective sentences. Based on the phenomenon or predicate indicated by the sentence, it is classified into one of a plurality of predetermined types, and the clause or sentence corresponding to the predetermined type is specified as the question frame setting clause.

  Further, the natural language analysis device according to claim 3 is the natural language analysis device according to claim 1 or 2, wherein the clause semantic expression association unit is configured to select one of the plurality of clauses or the plurality of sentences. Based on the phenomenon or predicate indicated by each section or each sentence, the problem setting section including the problem target attribute having the value of the information queried by the input sentence as a value is specified, and the problem target attribute and the problem setting section are It is semantically associated with the problem frame setting section.

  Further, the natural language analyzing apparatus according to claim 4 is the natural language analyzing apparatus according to claim 3, wherein the clause meaning expression associating means converts the question sentence or the sentence or sentence that can be converted into the question sentence into the question sentence. It is specified as a problem setting section.

  A natural language analysis apparatus according to claim 5 is a natural language analysis apparatus according to any one of claims 1 to 4, wherein a predetermined format is used to specify the meanings of the plurality of clauses or the plurality of sentences. A semantic expression storage means for storing a plurality of semantic expressions composed of (for example, a storage unit 40 described later) and whether or not the attributes included in the semantic expressions stored in the semantic expression storage means can be associated with each other. Criteria for value propagation possibility (for example, various rules such as predicate category tense & time correspondence judgment rule, predicate category pair correspondence judgment rule, non-synthetic concept correspondence judgment rule, phenomenon semantic expression integration rule described later) ) For storing correspondence determination criteria storage means (for example, a storage unit 40 to be described later), and the clause semantic expression association means is stored in the semantic expression storage means before And semantic representation, wherein based on the corresponding permission criteria storage unit stored in the said criterion, and judging the correspondence whether or values propagation potential of the attributes to each other.

  Further, the natural language analysis apparatus according to claim 6 is the natural language analysis apparatus according to claim 5, wherein the correspondence determination criterion storage unit includes, for each attribute category (for example, “definite”, “existence”). ”,“ State retention ”,“ achievement ”,“ action ”,“ action ”,“ grant ”,“ receipt ”, etc.) are stored in a plurality of different judgment criteria, and the clause meaning expression correspondence The means obtains the judgment criteria according to the category of the attribute from the correspondence judgment criterion storage means, and judges the correspondence between the attributes or the value propagation possibility based on the judgment criteria. To do.

  Further, the natural language analysis apparatus according to claim 7 is the natural language analysis apparatus according to claim 5 or 6, wherein the clause semantic expression association means is stored in the semantic expression storage means. Among the attributes included in the attribute, an attribute having a predetermined identifier (for example, “?”, Which is a value of an attribute described later) indicating a request for an answer to the attribute is specified as a problem target attribute. Features.

  Further, the natural language analysis apparatus according to claim 8 is the natural language analysis apparatus according to any one of claims 5 to 7, wherein knowledge information including a value that becomes a response to the input sentence (for example, implementation) In the form of knowledge information storage means (for example, storage section 40 described later) is stored, and the clause meaning expression association means is stored in the meaning expression storage means. Among the attributes included in the semantic expression stored in the above, a value corresponding to the problem target attribute is acquired from the knowledge information stored in the knowledge information storage means.

  The natural language analysis apparatus according to claim 9 is the natural language analysis apparatus according to any one of claims 5 to 8, wherein the clause meaning expression association unit determines whether the attributes can be associated with each other. Based on the result, among the attributes included in the semantic expression stored in the semantic expression storage means, the attributes referred to by the plurality of clauses or the plurality of sentences (for example, included in a reference subset described later) Attributes) and attributes that are not mentioned by the plurality of sections or sentences (for example, attributes included in an unreferenced subset described later), and the problem attribute is selected from among the attributes that are not mentioned The attribute corresponding to is selected.

  The natural language analysis method according to claim 10 is a natural language analysis method for interpreting the meaning of an input sentence composed of a plurality of sections or a plurality of sentences, wherein the plurality of sections or the plurality of sentences are included. A problem that specifies a problem frame setting section that sets a problem frame that presents a scene for positioning and interpreting the meaning of each section or each sentence based on the phenomenon or predicate indicated by each section or each sentence A frame meaning that performs predetermined processing for semantically associating a frame or clause other than the problem frame setting section with the frame setting section specifying step and the problem frame setting section specified in the problem frame setting section specifying step An expression association step.

  A natural language analysis program according to claim 11 is a natural language analysis program for causing a computer to execute a natural language analysis method for interpreting the meaning of an input sentence composed of a plurality of clauses or a plurality of sentences. A problem frame presenting a scene for positioning and interpreting the meaning of each section or each sentence based on the phenomenon or predicate indicated by each section or each sentence from among the plurality of sections or a plurality of sentences. The problem frame setting section specifying step for specifying the problem frame setting section for setting the problem frame setting section and the problem frame setting section specified in the problem frame setting section specifying step mean a section or sentence other than the problem frame setting section. The computer is caused to execute a natural language analysis method including a clause meaning expression associating step for performing a predetermined process for associating with each other.

  The natural language analysis program according to claim 12 is the natural language analysis program according to claim 11, wherein the plurality of clauses or the plurality of sentences are changed to the respective clauses or the respective steps in the problem frame setting clause specifying step. Based on the phenomenon or predicate indicated by the sentence, it is classified into one of a plurality of predetermined types, and the clause or sentence corresponding to the predetermined type is specified as the question frame setting clause.

  The natural language analysis program according to claim 13 is the natural language analysis program according to claim 11 or 12, wherein, in the clause semantic expression association step, the plurality of clauses or the plurality of sentences are selected from the plurality of clauses or the plurality of sentences. Based on the phenomenon or predicate indicated by each section or each sentence, the problem setting section including the problem target attribute having the value of the information queried by the input sentence as a value is specified, and the problem target attribute and the problem setting section are It is semantically associated with the problem frame setting section.

  Further, the natural language analysis program according to claim 14 is the natural language analysis program according to claim 13, wherein the clause or sentence that can be converted into a question sentence or a question sentence in the clause meaning expression association step, It is specified as a problem setting section.

  A natural language analysis program according to claim 15 is a natural language analysis program according to any one of claims 11 to 14, wherein a predetermined format is used to specify the meaning of the plurality of clauses or the plurality of sentences. Are included in a semantic expression storing step for storing a plurality of semantic expressions configured in a predetermined semantic expression storage means (for example, corresponding to a storage unit 40 described later) and the semantic expression stored in the semantic expression storage step. A correspondence determination criterion storage step for storing a determination criterion for whether or not attributes can be supported or a value transmission possibility in a predetermined correspondence determination criterion storage means (for example, a storage unit 40 described later), and the clause meaning In the expression associating step, the semantic expression stored in the semantic expression storing step and stored in the correspondence determination criterion storing step The on the basis of said criterion, and judging the correspondence whether or values propagation potential of the attributes to each other.

  Further, in the natural language analysis program according to claim 16, in the natural language analysis program according to claim 15, a plurality of determination criteria different for each category of the attribute are stored in the correspondence determination criterion storage step. In the clause meaning expression associating step, the determination criterion corresponding to the attribute category is acquired from the correspondence determination criterion storage unit, and based on this determination criterion, the correspondence between the attributes or the possibility of value propagation It is characterized by determining.

  The natural language analysis program according to claim 17 is the natural language analysis program according to claim 15 or 16, wherein the semantic representation stored in the semantic representation storage step is stored in the semantic representation storage step. Among the included attributes, an attribute having a predetermined identifier indicating a request for an answer to the attribute as a value is specified as the problem target attribute.

  Further, the natural language analysis program according to claim 18 is the natural language analysis program according to any one of claims 15 to 17, wherein knowledge information including a value to be an answer to the input sentence is stored as predetermined knowledge. An attribute included in the semantic expression stored in the semantic expression storage step in the clause semantic expression associating step, including a knowledge information storage step stored in an information storage means (for example, corresponding to a storage unit 40 described later) The value corresponding to the problem target attribute is acquired from the knowledge information stored in the knowledge information storage step.

  Furthermore, the natural language analysis program according to claim 19 is the natural language analysis program according to any one of claims 15 to 18, wherein in the clause meaning expression associating step, determination as to whether the attributes can be associated with each other is performed. Based on the results, among the attributes included in the semantic expression stored in the semantic expression storage step, the attributes referred to by the plurality of sections or the plurality of sentences and the plurality of sections or the plurality of sentences are referred to. It is determined that the attribute is not set, and the attribute corresponding to the problem target attribute is selected from the attributes not mentioned.

  According to the present invention, a semantic correspondence between a plurality of clauses and a plurality of sentences included in an input sentence by semantically associating other clauses or sentences with a question frame setting clause for setting a question frame. The relationship can be automatically recorded as information. Therefore, a dialogue interface or the like can be realized by correctly estimating the meaning of the input sentence.

  Further, according to the present invention, a scene for positioning and interpreting the meaning of each clause or each sentence by specifying a plurality of clauses or a plurality of sentences as problem frame setting clauses based on the phenomenon or predicate type. Can be set automatically.

  Further, according to the present invention, the problem setting section including the problem target attribute is identified, and the problem target setting section and the problem setting section are semantically associated with the problem frame setting section to thereby base the problem setting section. As a result, it is possible to interpret the meaning of the problem target attribute and the problem setting section.

  Further, according to the present invention, by specifying a question sentence or a clause or sentence that can be converted into a question sentence as a question setting clause, a scene for positioning and interpreting the meaning of each clause or each sentence is automatically obtained. Can be set.

  Further, according to the present invention, it is possible to determine the correspondence between attributes or the value propagation possibility based on the semantic expression and the determination criteria for the correspondence between the attributes included in the semantic expression or the value propagation possibility. Thus, it is possible to reliably compare attributes and the like based on the semantic correspondence between words and phrases specified by semantic expressions.

  Further, according to the present invention, the optimum attribute comparison or the like is performed for each attribute category by determining whether or not the attribute can be handled or the value propagation possibility based on a criterion according to the attribute category. be able to.

  Further, according to the present invention, by specifying an attribute having a predetermined identifier indicating a request for an answer to the attribute as a questionable attribute, the content being queried in the input sentence can be automatically specified. Can do.

  In addition, according to the present invention, by acquiring the value corresponding to the problem target attribute from the knowledge information including the value that becomes the answer to the input sentence, the information asked in the input sentence is automatically acquired. Answer automatically.

  Further, according to the present invention, an attribute that is referred to by a plurality of sections or a plurality of sentences and an attribute that is not referred to by a plurality of sections or a plurality of sentences are determined, and among the attributes that are not mentioned, By selecting the attribute corresponding to the problem target attribute, the information that has already been mentioned in the input sentence is distinguished from the information that has not been mentioned yet, and the information that has not been mentioned yet is gradually limited, and the input sentence It is possible to automatically specify the contents being questioned in.

  Embodiments according to the present invention will be described below in detail with reference to the accompanying drawings. First, [I] the basic concept of the present invention will be described, then [II] an embodiment of the present invention will be described, and [III] Finally, a modification to the embodiment of the present invention will be described.

[I] Basic concept of the present invention First, the basic concept of the present invention will be described. An object of the present invention is generally to read a semantic correspondence between a plurality of clauses and a plurality of sentences included in an input sentence. In general, when the concepts represented by words or phrases included in the input sentence are in a “dependency relationship” (roughly a “modification relationship”) with each other, this dependency causes the attributes and their There is a “semantic correspondence” between values (and thus a “semantic correspondence” between concepts based on that). An object of the present application is to propose a device, a program, or the like that reads a semantic correspondence generated by such a dependency. Here, as input sentences, there are a “single sentence” composed of a single sentence and a “compound sentence” composed of a plurality of sentences having mutual dependency relations. Or multiple sections). “Reading semantic correspondence” means that the meanings of the concepts that make up phrases and clauses in the input sentence are combined by dependencies and affect each other. This means that it is mechanically recognized and recorded as information.

  In order to achieve such an object, in the natural language analysis according to the present invention, mainly (A) input sentence preprocessing, (B) semantic expression generation processing, (C) input sentence semantic interpretation processing, and (D) Response sentence generation processing is performed sequentially. Of these, in (A) input sentence preprocessing, (A-1) morpheme analysis processing is performed on the input sentence to divide the input sentence into words, and (A-2) the divided input By performing a parsing process on the sentence, a parse tree corresponding to the input sentence is generated. In (B) semantic expression generation processing, “semantic expressions” of words, phrases, and clauses included in the generated parse tree are generated. Here, the “semantic expression” is a structure for expressing the meaning of words, phrases, and clauses included in the input sentence in a predetermined format, and a specific structure thereof will be described later. It is composed of multiple expressions (hereinafter referred to as attribute expressions) in the form of “attribute values” composed of “attributes” and “values”. Thereafter, in the (C) input sentence semantic interpretation process, the generated semantic expressions are semantically associated with each other. If the input sentence is a question sentence, an answer to the question sentence is acquired. Finally, in the (D) response sentence generation process, a response sentence is generated using the answer acquired in the input sentence meaning interpretation process, and is output to the user.

  Among these processes, (A) input sentence pre-processing can be performed by a known method, and (B) semantic expression generation processing is a method already proposed by the inventors of the present application (for example, JP 2003-114886 publication etc.). In addition, the dependency interpretation processing of phrases and clauses in the (C) input sentence semantic interpretation processing can be performed using the contents described in the prior patent application already proposed by the present inventor. That is, the main feature of the present invention is that processing other than that, that is, input sentence semantic interpretation processing, in particular, excludes phrase and clause dependency interpretation processing.

(Basic concept-semantic expression)
As described above, in the present application, various processes are performed on the semantic expression generated by the semantic expression generation process. Therefore, understanding the structure of this semantic expression is important for understanding the present invention. For this reason, below, the background which came to propose this semantic expression and the outline | summary of this semantic expression structure are demonstrated.

  Even before the inventor proposes the semantic expression generation process, the natural language analysis technology maintains a semantic classification system for words, defines all possible sets of words, and By assigning semantic features, the meaning of the word is positioned in the semantic classification system, and various semantic relationships between words (synonyms, upper-lower relationships, whole-part relationships, whole-attribute relationships, etc.) are identified. I was able to do it.

  Conventionally, however, it has been impossible to prepare such a system for phrases and clauses, and to determine the meaning and identify the semantic relationship. This is because, at present, no method has been found to formally evaluate what meaning a new phrase or clause is formed when multiple words are combined in a dependency relationship. In other words, it is impossible to analyze the function of the dependency relationship connecting a plurality of words. Therefore, in the conventional natural language analysis system, when expressing the meaning of a phrase or a clause, it has been necessary to describe the semantic expression using the surface layer dependency without being analyzed.

  In this case, a plurality of sentences representing substantially the same meaning will be expressed in different semantic expression forms linked to the dependency structure of the surface layer, and both express the same meaning substantially. It becomes impossible to judge mechanically. For this reason, conventionally, information required by the system is extracted based on the meaning of the input sentence, whether the meaning of the sentence input in response to the system inquiry includes information required by the system, etc. It was not possible to make a stable determination. For this reason, for example, in a dialogue system, stable information cannot be exchanged between the system and the user, and the dialogue cannot be continued.

  In order to solve such a conventional problem, the present inventor apparently eliminates the expression of the dependency as much as possible, and for the part that cannot be excluded, only the common dependency that does not vary depending on the meaning to be expressed remains. As a result, a semantic expression that eliminates the need for analyzing the dependency relationship has been proposed (for example, Japanese Patent Application Laid-Open No. 2003-114886 described above).

  An example of the basic structure of this semantic expression is shown in FIG. In FIG. 1, a semantic expression with a concept name “predicate concept 1” and “markers”, a semantic expression with a concept name “noun concept J”, and a concept name “noun concept K” are attached. A total of four semantic expressions are shown, including the semantic expression and the semantic expression with the concept name “adverb concept M”. Each of these semantic expressions is expressed in a frame format, and a “concept name” unique to each semantic expression is given above the frame. This concept name is configured by combining “concept name” and “instance number”. In addition, “markers” are given to the concept names as necessary. Here, the markers represent the meanings of “final particle concept”, “auxiliary verb concept”, “temporal concept”, “phase concept”, etc., and in the surface sentence, the final particle, auxiliary verb, tense, and phase information are When expressed, a marker corresponding to this is added to the concept name at the head (head) of the semantic expression. These markers are shown in FIG.

  Each semantic expression includes one or more “attribute expressions”. Here, the attribute expression is a unit in semantic expression that is configured as an “attribute value” by combining “attributes” and “values” corresponding to each other. Of these, "attributes" represent the basic properties that make up a concept. The attributes that make up a predicate concept are called "predicate concept attributes", and the attributes that make up a noun concept are called "noun concept attributes". Thus, it is expressed with a conceptual name of semantic expression as necessary. This “attribute” includes a “phenomenon attribute” that represents a basic property that constitutes a phenomenon concept (or a partial phenomenon concept that is a partial phenomenon concept) and a basic property that constitutes an entity concept. “Entity attributes” that represent the basic properties that the entity concept “includes”, such as “color”, “shape”, etc. (as described later, properties related to general phenomena other than “include” are represented by phenomenon attributes) The “phenomenon attribute” is broadly classified as “phenomenon attribute i” (where i is an integer of 1 or more, the same applies hereinafter), “substance attribute” is “noun concept J $ entity attribute i” (where j is 1 or more) It is expressed as an integer.

  The “value” has a format of “[knowledge value] modifier value”. Among these, “knowledge value” is a value given from the beginning as the meaning of the concept. “Modified value” is a value specified as a result of continuous or continuous modification in a sentence (however, if it is not necessary to distinguish these knowledge values and modified values from each other, it is simply called “value”. To do). For example, the knowledge value can take as a value a pointer for specifying another semantic expression having a modification relationship (in terms of knowledge). Hereinafter, the semantic expression indicated by the pointer value is referred to as “point destination semantic expression” as necessary. In this way, when the knowledge value is a pointer value, if the same pointer value exists in the point-destination semantic expression indicated by the pointer value, a loop is formed between these semantic expressions. Become. Here, it is assumed that such a loop is not configured. Specifically, when a loop is configured (the same pointer value exists above the relationship), The point-to-point semantic expression is not expressed. When the semantic expression is illustrated, the link state based on such point values is indicated by arrows. For example, in FIG. 1, regarding the attribute expression “phenomenon attribute 1“ [knowledge value] modification value ”” of the predicate meaning expression “predicate concept I”, this modification value is a pointer value, and this point destination semantic expression is a noun concept meaning. The expression “noun concept J” is expressed by connecting the concept name of the noun concept semantic expression “noun concept J” with an arrow from the modification value of the attribute expression “phenomenon attribute 1“ [knowledge value] modification value ””. is doing.

  The semantic expression configured in this way has an arbitrary data format, for example, a list structure having a concept name as a head, an attribute, a value other than a pointer value, and a pointer value as a branch. And stored in the storage unit 40.

(Basic concept-semantic expression and dependency interpretation)
Next, with respect to a plurality of semantic expressions generated in such a structure, the background that led to the interpretation of the semantic correspondence between the plurality of semantic expressions, details of the specific configuration of each semantic expression, and An overview of dependency interpretation will be described.

  The present applicant has found that the meaning of each “phrase” and “section” included in the input sentence is expressed and specified in a predetermined format by using the semantic expression configured as described above. However, such semantic expressions are meaningful in order to individually identify the meanings of “phrase” and “word”, but the semantic correspondence between these “phrases” and “words” can be changed. It was not enough to identify. Therefore, in order to further specify the semantic correspondence between the semantic expressions, a method for interpreting the correspondence between the semantic expressions was proposed in the prior patent application.

  A predicate concept semantic expression such as “predicate concept I” shown in FIG. 1 represents a (partial) phenomenon concept. The “predicate concept attribute” constituting the predicate concept semantic expression “predicate concept I” includes “phenomenon attribute” and “attribute limit phrase”. Among them, the phenomenon attribute is the “main case”, “target case”, “partner case”, “time case”, “location case”, “quantity case”, “mode” and other continuous components of the predicate concept. Corresponding attribute. This “phenomenon attribute” is expressed by synthesizing a “predicate concept name” such as “existence” or “owned” and a “common component name” such as “subject” or “location”. It is expressed as “existing entity”, “existing location (location)”, “owning entity (owner)”, “number of ownership”, “accommodation fee”, “purchase date / time”, “production location (production area)”, and the like. At this time, it is assumed that the names of consecutive components (specifically, case names) such as “main case” and “target case” can be identified from the phenomenon attribute names. This “phenomenon attribute” is, as “value”, a pointer value to a noun conceptual meaning expression, “3 units” or “2004”, a non-pointer value such as “NIL” or “?” Indicating that no value exists. Take.

  In addition, the noun concept semantic expressions such as “noun concept J” and “noun concept K” shown in FIG. 1 are “(partial) phenomenon concept (sa variable noun etc.)”, “(partial) entity concept”, “attribute concept”. Or “value concept”. Among these, the value concept is simply expressed by a symbol (corresponding to “non-pointer value”). The meaning of the (partial) phenomenon concept indicated by the noun concept is expressed by a semantic expression of the same type as the predicate concept. Here, among the various attributes constituting the noun concept semantic expression, the attributes constituting the noun concept excluding the (partial) phenomenon concept and the value concept are referred to as “noun concept attributes”. This noun concept attribute takes one of the following forms: “noun concept $ phenomenon attribute”, “noun concept $ entity attribute”, “#”, “noun concept $ attribute limit phrase”, and “noun concept $ link modifier”. . Here, “$” represents the meaning of “no” of “inclusive”. For example, the expression “noun concept $ substance attribute” indicates a relationship in which an entity attribute is included in the noun concept.

(Basic concept-Input sentence semantic interpretation processing)
Next, a basic concept of input sentence semantic interpretation processing performed with the semantic expression configured as described above as a processing target will be described. First, consider the two sentences “There are three personal computers at home” and “I want to build a LAN”, and the former is called the first sentence and the latter is called the second sentence. For example, when the first sentence and the second sentence are continuously spoken by a certain person, the meaning of the second sentence is “LAN construction” with respect to “three personal computers” expressed in the first sentence. It can be interpreted as meaning that we want to do. In other words, it can be said that the first sentence limits the “building target”, which is one of the preconditions for performing “LAN construction” expressed in the second sentence. That is, a “scene” for performing “LAN construction” is set by the second sentence, and the first sentence is positioned in this “scene”. The same applies to the relationship between nodes. For example, two clauses included in the sentence “There are three computers at home, so I want to build a LAN.” “There are three computers at home,” and The same relationship as the first sentence and the second sentence holds for “I want to build a LAN”. When this is generalized, in order to interpret multiple sections and multiple sentences, a “scene” (hereinafter referred to as “problem frame”) for embodying the meaning of these sections and sentences is defined. You can specify it.

  Here, when the input sentence is a single sentence and the question is completed with the single sentence, the “problem frame” is defined in the single sentence, and the single sentence is positioned in the “problem frame”. On the other hand, when the input sentence is a compound sentence, some of the sections and sentences that make up this input sentence are set to “problem frame” (hereinafter, this is the case). A section or sentence that sets a “problem frame” is referred to as a “problem frame setting section.” That is, a “problem frame setting section” is a section or sentence that has a meaning of directly requesting an answer from the natural language analyzer. This is a question or a clause or sentence that can be converted into a question sentence, etc. Sections and sentences other than the “question frame setting clause” are positioned in the “question frame”. In the two-sentence example, the second sentence “I want to build a LAN” becomes a “problem frame setting section”, and this “problem frame setting section” sets a “problem frame” called “LAN construction”. The first sentence “There are three computers at home.” Is positioned in “Problem”. For example, in the case where the input sentence is a question sentence related to “method”, an “action sequence” that is a specific content of “target”, “precondition” for achieving, and “method” There may be many “question frame setting clauses” for setting problem conditions.

  From these, in order to interpret an input sentence consisting of multiple sections and multiple sentences, first specify the “problem frame setting section”, and based on this “problem frame setting section”, select the “problem frame”. Set. A section or sentence other than the “problem frame setting section” may be positioned with respect to the “problem frame”. This makes it possible to interpret the input sentence. After that, the result of this interpretation is compared with information (hereinafter referred to as “knowledge” as necessary) acquired in advance and stored in a later-described storage unit 40 in a semantic expression format. You can get an answer. For example, if there is an input sentence “I want to build a LAN, what kind of cable is necessary?”, The section “What kind of cable is necessary?” Is added to the “Problem frame” called “LAN construction”. After positioning, by acquiring from the knowledge a “value” indicating the answer to the “cable type” attribute that is the subject of question in this section, for example, “LAN cable is required.” can do. Thus, the questionable attribute is referred to as “problem target attribute”, and the question type section including this “problem target attribute” is referred to as “problem setting section”. Even if the value of the “problem attribute” does not immediately exist in the knowledge, the value of the “problem attribute” can be acquired by propagating from the value of another attribute in the knowledge. The propagation of this value can be performed automatically using the contents of the above-mentioned prior patent application.

[II] Embodiment of the Present Invention Next, an embodiment of the present invention for automatically interpreting an input sentence as described above will be described. However, the present invention is not limited by this embodiment. In the following, the configuration of the natural language analysis apparatus will be described first, and then the processing contents of the natural language analysis method and the natural language analysis program executed in the natural language analysis apparatus will be described.

(Overall configuration of the device)
First, the configuration of the natural language analysis apparatus according to the embodiment will be described. FIG. 3 is a block diagram illustrating a functional concept of the configuration of the natural language analysis apparatus 1 according to the present embodiment. As illustrated in FIG. 3, the natural language analysis apparatus 1 schematically includes an input unit 10, an output unit 20, an input / output control IF (Interface) 30, a storage unit 40, and a control unit 50. These parts are connected via a bus so that data communication is possible. The natural language analyzing apparatus 1 is, for example, a personal computer, a workstation, a home game machine, an Internet attribute TV, a PDA (Personal Digital Assistant), or a mobile phone or a PHS (PHS (Personal Digital Assistant)). It can be configured as a mobile communication terminal such as a Personal Handy-Phone System) or a car navigation terminal.

  Among these, the input unit 10 is an input means for the user to input arbitrary information. For example, a pointing device and keyboard such as a mouse, optical scanning and OCR (Optical Character Reader), or voice An input microphone and a known voice analysis means can be used. The output unit 20 is an output unit for outputting arbitrary information to the user, and can be configured as, for example, a monitor, a printer, or a speaker. The input / output control IF 30 is a communication control unit that performs input / output control between the input unit 10, the output unit 20, and the control unit 50. The storage unit 40 is a storage unit that stores information such as various programs and data necessary for natural language analysis. For example, the storage unit 40 is a volatile storage medium such as a RAM (Rondom Access Memory) or a hard disk. A non-volatile storage medium can be used. In particular, the storage unit 40 stores a natural language analysis program, which is loaded into the control unit 50 and executed, thereby executing a natural language analysis method. The storage unit 40 stores various determination conditions and the like necessary for interpreting the meaning of the input sentence and generating a semantic expression. That is, the storage unit 40 corresponds to the semantic expression storage unit and the correspondence determination criterion storage unit in the claims.

  The control unit 50 is a control unit for controlling each unit of the natural language analysis apparatus 1 and includes, for example, a CPU (Central Processing Unit) and a program that is interpreted and executed on the CPU. The control unit 50 includes an input sentence preprocessing unit 51, a semantic expression generation processing unit 52, an input sentence meaning interpretation processing unit 53, and a response sentence generation unit 54 in terms of functional concept.

  Among these, the input sentence preprocessing unit 51 is a means for performing preprocessing prior to the semantic expression generation process for the target sentence, and includes a morpheme analysis process for dividing a word, a syntax analysis process for extracting a parse tree from a morpheme analysis result, Then, a phrase marker conversion process for converting a predetermined phrase included in the parse tree into a marker is performed.

  In addition, the semantic expression generation processing unit 52 decomposes the parse tree of each section to be analyzed into a set of branches composed of auxiliary verbs or predicates and the continuous modification components related thereto, and each decomposed branch In addition, a semantic expression configured by combining attributes and attribute values is generated.

  The input sentence semantic interpretation processing unit 53 is a means for semantic interpretation of a plurality of clauses and a plurality of sentences included in the input sentence, and specifically executes input sentence semantic interpretation processing. Here, as shown in FIG. 4, the input sentence semantic interpretation processing unit 53 functionally conceptually performs a focused clause semantic interpretation processing unit 53 a that performs a focused clause semantic interpretation process, a problem setting clause, and a problem target attribute identification process. A problem setting clause and problem target attribute identification processing unit 53b to perform, a problem solving processing unit 53c to perform problem solving processing, a clause semantic expression association processing unit 53d to perform clause semantic expression association processing, and a value semantic representation to perform value semantic expression processing A processing unit 53e, an all-frame correspondence determination processing unit 53f that performs all-frame correspondence determination processing, a predicate concept correspondence determination processing unit 53g that performs predicate concept correspondence determination processing, a noun concept correspondence determination processing unit 53h that performs noun concept correspondence determination processing, A non-synthetic concept correspondence determination processing unit 53i that performs a composite concept correspondence determination process, a premise correspondence determination processing unit 53j that performs a premise correspondence determination process, and a method correspondence determination process that performs a method correspondence determination process Unit 53k, attribute value pair correspondence determination processing unit 53l that performs attribute value pair correspondence determination processing, numerical concept pair correspondence determination processing unit 53m that performs numerical concept pair correspondence determination processing, and symbol concept pair that performs symbol concept pair correspondence determination processing The correspondence determination processing unit 53n is provided. Each part of these control parts 50 respond | corresponds to the problem frame setting clause specific | specification means and clause meaning expression matching means in a claim independently or jointly.

  The response sentence generation unit 54 is a response sentence generation processing unit that generates a response sentence when necessary based on the interpretation result by the input sentence meaning interpretation processing unit 53 and displays an image or a voice. Specific processing contents of each unit of the control unit 50 will be described later. In the present application, processes other than those specifically described are assumed to be performed by the control unit 50.

(Processing content)
Next, the contents of the natural language analysis process as the present method performed using the natural language analysis apparatus 1 configured as described above will be described in detail.

(Processing content-overall)
First, the flow of the entire natural language analysis process will be described. An overall flow chart of this natural language analysis processing is shown in FIG. First, when the user activates the natural language analysis apparatus 1 in a predetermined method via the input unit 10, the input sentence preprocessing unit 51 monitors whether or not an input sentence has been input (step S-1), and the user inputs When an input sentence is input via the unit 10 (step S-1, Yes), the input sentence is captured (step S-2), and input sentence preprocessing is executed (step S-3). In this input sentence preprocessing, the input sentence preprocessing unit 51 first divides the input sentence into words to generate a parse tree. However, since a known method can be used to generate the parse tree, details thereof are omitted here. Note that a dependency structure tree is assumed here as the parse tree. However, even when the phrase structure tree is derived by the parse analysis, the semantic analysis process can be performed by a procedure substantially similar to the following procedure. Next, the input sentence preprocessing unit 51 stores the generated parse tree in the storage unit 40 in an arbitrary data format, and ends the processing.

  Next, the semantic expression generation processing unit 52 executes a semantic expression generation process for the parse tree stored in the storage unit 40 (step S-4). In this semantic expression generation process, the semantic expression generation processing unit 52 generates a semantic expression. Specifically, (1) First, each clause is decomposed into a set of “branches” of “predicate + each consecutive modification component”, and semantic expressions of each branch are individually generated. The reason why the semantic analysis is performed in units of “sections” is that the basic unit of the proposition level constituting the sentence is “sections”. (2) Thereafter, in order to facilitate mutual reference between the semantic expressions, a pointer value for referring to other semantic contents is added to the attribute of the semantic expression. Generation of such a semantic expression can be performed, for example, by a method similar to that disclosed in the above-described Japanese Patent Application Laid-Open No. 2003-114886, and thus further description thereof is omitted here. Then, the semantic expression generation processing unit 52 stores the generated semantic expression in an arbitrary data format in the storage unit 40, and ends the process.

  Thereafter, the input sentence meaning interpretation processing unit 53 executes the input sentence meaning interpretation process for the semantic expression stored in the storage unit 40 (step S-5). Although details of this process will be described later, in this process, the input sentence meaning interpretation processing unit 53 interprets the meaning of a clause or sentence included in the input sentence, acquires an answer to the input sentence based on the result, and acquires The stored answer is stored in the storage unit 40. In this input sentence semantic interpretation process, among the multiple semantic expressions generated by the semantic expression generation process, the semantic expressions corresponding to each clause and each sentence are processed one by one from the beginning of the input sentence, and finally This is repeated a plurality of times until the processing for the clauses and sentences is completed (step S-6).

  Next, the response sentence generation unit 54 generates a response sentence including the answer based on the answer acquired by the input sentence meaning interpretation processing unit 53 and stored in the storage unit 40 (step S-7). ), And outputs the response sentence in a predetermined format via the output unit 20.

  Thereafter, the same steps S-1 to S-7 are repeated until an instruction to end the analysis is received from the user, and when the user gives an instruction to end the analysis via the input unit 10 (step S-8, Yes), All natural language analysis processing ends.

(Input sentence semantic interpretation processing)
Next, the input sentence semantic interpretation process will be described in detail. This processing is mainly executed by the input sentence semantic interpretation processing unit 53 and each of the units 53a to 53n in FIG. 4, and is started with the semantic representation of the input sentence as an argument, and knowledge corresponding to “problem attribute” This is a process of returning the attribute value of the side as an answer to the input sentence. This process forms a loop for sequentially interpreting the clauses included in the same sentence in the input sentence. FIG. 6 shows a flowchart of the input sentence semantic interpretation process.

  As a premise of this process, all semantic expressions as arguments are shared as global variables by all input sentence semantic interpretation processes. In addition, it is assumed that a “target clause semantic expression”, which will be described later, updated in each process inside the input sentence semantic interpretation process is stored in the storage unit 40 and is taken over by the next input sentence semantic interpretation process. The storage unit 40 stores a “problem solving table” shown in FIG. This “problem solving table” is a table for setting “problem frame”, “problem target attribute”, and “problem setting section” in the input sentence semantic interpretation process. Each sentence is shared as a global variable. In the knowledge stored in advance in the storage unit 40, attribute values (numerical values or the like) that should have the same value are linked by a link, and a change at one place is propagated to all other places as it is. Further, in the input sentence semantic interpretation process in the present embodiment, a sentence consisting only of noun phrases is not assumed as an input sentence, and the predicate concept always comes to the beginning (head) of the semantic expression. However, the “problem frame” may include a noun concept in the head. Further, in the input sentence semantic interpretation process, the knowledge side has complete knowledge (the answer to the input sentence is described in the semantic expression stored in the storage unit 40), and the input sentence does not say that it is irrelevant. It is not assumed that the input sentence is left (an irrelevant clause or sentence is included in the input sentence to obtain an answer).

  In this input sentence semantic interpretation process, first, a “restart from the beginning” flag set in the storage unit 40 is initialized (step SA-1). Then, in order to interpret sequentially from the first clause in the same sentence among the semantic expressions passed as an argument, pay attention to the clause at the head of the sentence, and proceed to the following processing with this clause as the “target clause” (step SA). -2). Next, it is determined whether or not the “restart from the beginning” flag is set in the storage unit 40 (step SA-3). If it is set (step SA-3, Yes), this flag is reset. (Step SA-4). Further, it is determined whether or not the correspondence table is stored in the storage unit 40 (step SA-5). If the table is stored (step SA-5, Yes), the correspondence table is cleared (step SA-5). Step SA-6). Here, the correspondence relationship table is a table for describing the correspondence relationship between each attribute and each value.

  Next, in order to interpret the meaning of the focused clause, the focused clause semantic interpretation process is started with the semantic representation of the focused clause (hereinafter referred to as focused clause semantic expression) as an argument (step SA-7). Then, it is determined whether or not the “restart from the beginning” flag has been set in the storage unit 40 as a result of the target clause semantic interpretation process (step SA-8). (Step SA-8, Yes), the process returns to step SA-2 to reset this flag (step SA-4). On the other hand, when the “redo from the beginning” flag is not set (step SA-8, No), the next clause (subordinate clause subordinate to the target clause in the input sentence is displayed at the end of the target clause. ) Exists (step SA-9). If the next section exists (step SA-9, Yes), this is set as a new section of interest (step SA-10), and step SA Return to -5. On the other hand, when the next section does not exist (step SA-9, No), the process is terminated.

(Input sentence semantic interpretation process--interpretation semantic interpretation process)
Next, the focused clause meaning interpretation process will be described. This node-of-interest semantic interpretation processing is mainly executed by the node-of-interest semantic interpretation processing unit 53a shown in FIG. 4, and is activated using the node-of-interest semantic expression as an argument. The flowchart of this process is shown in FIGS.

  In this process, as shown in FIG. 8A, the target section is “Tell me about”, “I don't know”, “I want to know”, “Yes / No question”, or “WH question”. It is determined whether any of the above is true (step SB-1). In this determination, for example, the concept name of the target clause meaning expression in the case of “tell me” is stored in the storage unit 40 in advance, and this concept name and the concept name of the target clause meaning expression are stored. This is done by comparing. And when it does not correspond to “tell me” (step SB-1, No), the problem setting is not completed with the focused section alone, and the preceding sentence, the preceding section, the succeeding sentence, or the succeeding section It is determined that it is necessary, and as shown in FIG. 8C, it is determined whether or not “problem frame” is set in the “problem solving table” (step SB-18). If “problem frame” is set (step SB-18, Yes), the process proceeds to step SB-14 described later, and if “problem frame” is not set (step SB-18). , No), the target section is written in the “problem solving table” (step SB-19), and the process proceeds to step SB-24 described later. On the other hand, in step SB-1 of FIG. 8-1, when it corresponds to “Tell me” (step SB-1, Yes), there is a possibility that the problem setting may be completed by the section of interest alone. The problem setting section and problem target attribute identification process are activated (step SB-2). From this problem setting section and problem target attribute identification process, a problem solving table in which “problem setting section” and “problem target attribute” are written is returned.

  After the problem setting section and the problem target attribute identification process are finished, it is determined whether or not the predicate of the section of interest is “conclusive” (step SB-3). This is because the “problem frame” determination method differs depending on whether or not it is “conclusive”. If it is not “conclusive” (step SB-3, No), the process proceeds to step SB-8 described later. On the other hand, if it is “conclusive” (step SB-3, Yes), it is determined whether or not the attribute “contained subject” of the “problem attribute” concept on the left side of this “conclusive” predicate has a pointer value. Determine (step SB-4). This is for determining whether or not the information specifying the “problem frame” is described in the section of interest. If the “inclusive subject” attribute has a pointer value, the pointer value is linked. This is because there is a possibility that the semantic expression (pointed semantic expression) is a “problem frame”.

  If the pointer value does not exist (step SB-4, No), the attribute “proposition” or the attribute “case-limited” in the “assertion” meaning expression (for example, “in the case of“ ˜ ”in a clause or sentence) ) Has a pointer value (step SB-5), and if the attribute “Proposal” or the attribute “case limited” does not have a pointer value (step SB-5, No), the node of interest Alone, it is determined that no problem is set, and the process proceeds to step SB-18 described above. On the other hand, when the attribute “proposition” or the attribute “limited to cases” has a pointer value (step SB-5, Yes), the point destination semantic representation is set as the point destination semantic representation W (step SB-6). If the pointer value is present in the previous step SB-4 (step SB-4, Yes), the point destination semantic representation is set to the point destination semantic representation W without performing step SB-5 (step SB). -6).

  Next, it is determined whether or not the point-to-point semantic expression W includes an attribute that is the same as the “problem attribute” or has an upper / lower relationship (step SB-7). If such an attribute is not included (step SB) -7, No), it is determined that the problem setting is not completed with the focused section alone, and the process proceeds to step SB-18 described above. On the other hand, when such an attribute is included (step SB-7, Yes), the process proceeds to step SB-13 described later.

  Here, when the process proceeds from step SB-3 to step SB-8, as shown in FIG. 8B, the “problem frame” includes an attribute that has a semantic correspondence with the “problem target attribute” (part). It is determined whether or not the entity concept semantic expression can be acquired (step SB-8). The acquisition of the (partial) entity concept semantic expression is performed based on the contents of the prior patent application at an arbitrary timing before the time point. For example, if the target section is “How fast does the LAN transmit data?”, Since the transmission speed is unknown, “?” Is added to the value of the “transmission speed” attribute in the semantic expression. Since it is stored, “transmission speed” is “problem attribute”. Therefore, a pointer to the (partial) entity conceptual meaning expression X including an attribute (for example, “X $ transmission speed”) that has a semantic correspondence with the “transmission speed” attribute indicates which phenomenon attribute in the “transmission” semantic expression. Is stored in the value of. In this example, “X” and “transmission” have a “subject” relationship and have a dependency relationship. Therefore, only when X is the value of the “transmission subject” attribute of “transmission”, “problem attribute” "Transmission rate" has a semantic correspondence. Accordingly, a (partial) entity conceptual meaning expression including an attribute that has a semantic correspondence with the “problem attribute” is obtained by acquiring a pointer value of the “transmission subject” attribute.

  If such a (partial) entity concept semantic expression can be acquired (step SB-8, Yes), the acquired semantic expression is set as a point destination semantic expression W, and the process proceeds to step SB-13 described later. On the other hand, if the (partial) entity concept semantic expression cannot be acquired (step SB-8, No), the attribute “proposition” or the attribute “case-limited” in the verb semantic expression of the target clause has a pointer value. If the attribute “Proposal” or the attribute “Limited to Case” does not have a pointer value (No in Step SB-10), the problem setting is not completed by the section of interest alone. And the process proceeds to step SB-18 described above. On the other hand, when the attribute “proposition” or the attribute “case-limited” has a pointer value (step SB-10, Yes), the point destination semantic representation is set as the point destination semantic representation W (step SB-11). It is determined whether the pre-semantic expression W includes an attribute that is the same as the “problem attribute” or has an upper / lower relationship (step SB-12). If such an attribute is not included (step SB-12, No), it is determined that the attribute “proposed” or the attribute “case limited” cannot be a “problem frame”, and the above-described step SB Move to -18.

  On the other hand, when such attributes are included (step SB-12, Yes), “problem frame”, “problem setting section”, and “problem target attribute” are set in a single target section. Therefore, the point destination semantic expression W is written as a “problem frame” in the problem solving table of the storage unit 40 (step SB-13). Note that there may be a case where a sentence has already been input before the target section, and some “problem frame” is set there, but here, it is assumed that it is overwritten unconditionally. Next, it is determined whether or not this “problem frame” is the focused clause meaning expression itself (step SB-14). If it is the focused clause meaning expression itself (step SB-14, Yes), the focused clause is set to “problem”. Since it is not necessary to associate with the “frame”, the process proceeds to step SB-24 described later without performing the clause meaning expression associating process (step SB-16). On the other hand, if it is not the focused clause meaning expression itself (step SB-14, Yes), the current focused clause is a “node not to be matched” stored in advance in the storage unit 40 such as “does not work”. It is determined whether or not there is (step SB-15), and if it is a “non-targeted clause” (step SB-15, Yes), the clause meaning expression association processing (step SB-16) is also performed. Without performing, the process proceeds to Step SB-24 described later.

  On the other hand, if it is not “untargeted clause” (No in step SB-15), the clause semantic expression association processing is started with the focused clause semantic representation and the “problem frame” semantic representation as arguments (step SB-16). As a result of this process, either “correspondence possible” or “impossibility of correspondence” and “correspondence table” are returned. Here, as a case where “correspondence is possible”, for example, the following examples can be given.

(1) “Problem frame” = “LAN”, focus section = “What is the LAN construction method?”, “What is the LAN transmission speed?”, Or “How fast is the LAN transmitting data?” "?" In this case, the semantic expressions of these question sentences are associated with the “LAN” semantic expression.
(2) In the case of “problem frame” = “I want to build a LAN”, and the section of interest = “How to build a LAN”. In this case, the target section is associated with the entire phenomenon meaning expression at the top of the “problem frame”.
(3) When “problem frame” = “I want to build a LAN”, and the focus section = “what is the method”. In this case, the section of interest “method?” First corresponds to the attribute “construction method NIL” in the top of the “problem frame”, and is further interpreted by the prior patent application from “construction target LAN” & “construction method NIL”. Is associated with “LAN $ construction method table” in “LAN”.
(4) When “problem frame” = “I want to build a LAN”, and the focus section = “how much does it cost?” In this case, the target clause meaning expression [question “necessary cost?”] Is associated with “LAN $ construction cost value˜” in the “LAN” meaning expression.
(5) In the case of “problem frame” = “I want to build a LAN”, and the focus section = “What are the prerequisites?” In this case, the target section is associated with the sub-attribute “premise value” semantic expression in the “method” table included below the “problem frame” semantic expression.
(6) In the case of “problem frame” = “I want to build a LAN”, and the focus section = “There are three PCs at home”. In this case, the target clause is the value of the sub-attribute “premise” or “method”, the meaning expression of “premise i” or “method j”, or the value AP or the value BP that is the value of the “set component” attribute. Or, it is associated with a phenomenon meaning expression which is a value of the “collective component” attribute.

  Then, when the fact that “correspondence is possible” is returned from the clause meaning expression association process (step SB-17, Yes), the process proceeds to step SB-24 described later. On the other hand, when “impossible to associate” is returned (step SB-17, No), conversely, it is determined whether or not the “problem frame” section can be associated with the current target section. In order to do this, as shown in FIG. 8-4, the “problem frame” semantic representation and the target clause semantic representation are used as arguments, and the clause semantic representation association process is started again (step SB-20). For example, in the first process, when “problem frame” = “3 PCs are in the house” and the focus section = “I want to build a LAN”, the focus section is set to “problem frame”. Cannot be associated with a clause in For this reason, an attempt is made to associate with each other by exchanging as “problem frame” = “I want to build a LAN”, and the focus section = “There are three PCs at home”.

  As a result, if the clause meaning expression association process returns that the association is possible (step SB-21, Yes), the “problem frame” in the “problem solving table” In addition to clearing the designation, the current section is set as the “problem frame”, the “restart from the top” flag is set (step SB-22), and the process is terminated. On the other hand, when the message “impossible to associate” is returned (step SB-21, No), neither the “problem frame” section nor the current section can be “problem frame”, and the corresponding section It is determined that the succeeding section or the succeeding sentence becomes the “problem frame”, and the “problem frame” column of the “problem solving table” is set to NIL (step SB-23). Then, it is determined whether or not the section of interest is a “problem setting section” (step SB-24). If it is not the “problem setting section” (step SB-24, No), the processing is terminated. Alternatively, if it is a “problem setting section” (step SB-24, Yes), a problem solving process is started (step SB-25).

(Input sentence semantic interpretation process-problem setting section and problem target attribute identification process)
Next, the problem setting section and problem target attribute identification processing will be described. This problem setting section and problem target attribute identification processing is mainly executed by the problem setting section and problem target attribute identification processing unit 53b shown in FIG. Identify the “setting clause” and “problem attribute”. An overall flowchart of this processing is shown in FIG.

  In this process, it is determined whether the predicate of the target clause semantic expression is “Tell me”, “I don't know”, or “I want to know” (step SC-1). SC-1, No), and the section of interest is the “problem setting section” (step SC-2). This is because, in this case, the focus clause is “YES / NO question” or “WH question” and has already become a question sentence, so that the focus clause can be directly used as the “problem setting clause”. It is. On the other hand, if either of them is satisfied (step SC-1, Yes), it is found that the target section is not yet a question sentence, and therefore, this target section can be converted into a “definite” question sentence meaning expression. (Step SC-3).

  If conversion is not possible (step SC-3, No), the general meaning predicate is converted into a question meaning expression, which is used as a "problem setting section" (step SC-4). Alternatively, if conversion is possible (step SC-3, Yes), the section of interest is converted into a “conclusive” question meaning expression, which is set as a “problem setting section” (step SC-5). Then, the “problem setting section” is extracted from the “problem setting section” semantic expression, and the “problem setting section” and the “problem target attribute” are written in the problem solving table of the storage unit 40 (step SC-6). Here, the “problem attribute” is an attribute having “?” As a value on the semantic representation of a question sentence (or a clause that can be converted into a question sentence, a sentence that is converted from a sentence into a question sentence, or a clause). Therefore, it can be extracted by searching for “?”. This is the end of the process.

(Input sentence semantic interpretation processing-problem solving processing)
Next, problem solving processing will be described. This problem solving process is mainly executed by the problem solving processing unit 53c shown in FIG. 4 and is started with the correspondence table as an argument, and the value of the attribute on the knowledge side corresponding to the “problem attribute” is set. return. An overall flowchart of this process is shown in FIG.

  In this process, the value of the attribute on the knowledge side corresponding to the “problem attribute” is acquired using the correspondence table acquired by associating the “problem setting clause” with the semantic interpretation process of the focused clause (step SD). -1). Specifically, this value is obtained by searching for a value associated with “?” In the correspondence table. Then, this value is returned (step SD-2), and the process ends.

(Input sentence semantic interpretation process-clause semantic expression correspondence process)
Next, the clause meaning expression association processing will be described. This clause meaning expression association processing is mainly executed by the clause meaning expression correspondence processing unit 53d shown in FIG. 4 and is activated with the noticed clause meaning expression X and the “problem frame” meaning expression Y as arguments. The fact that the clause meaning expression X is “correspondable” or “cannot be associated” with the “problem frame” meaning expression Y and the correspondence table are returned. An overall flow chart of this process is shown in FIG.

  In this process, first, a “location” attribute pair match flag is initialized (step SE-1). If the top semantic expression of the target clause semantic expression X is semantic expression A and the top semantic expression of the “problem frame” semantic expression Y is semantic expression B, then the semantic expression below semantic expression A and the semantic expression B below An all-frame correspondence determination process is activated for the semantic expression (step SE-2). As a result of this processing, it is returned whether or not both semantic expressions are “correspondable”. If it is “correspondable” (step SE-3, Yes), the target clause semantic expression X is The “problem frame” meaning expression Y is “can be associated” and a correspondence table ([A: B] <lower correspondence table>) is returned (step SE-15), and the processing is performed. finish. Here, <lower correspondence table> indicates a correspondence table returned from another process called in the process (the same applies hereinafter).

  On the other hand, if it is not “correspondable” (No at Step SE-3), pay attention to the first attribute of the phenomenon attribute group of the semantic expression B, and set this attribute as the attribute T (Step SE-4). Then, it is determined whether or not this attribute T has a pointer value (step SE-5. Here, when a plurality of values exist in one attribute, these values belong to the same category. , If one value is a pointer value, the other value is also a pointer value). If it does not have a pointer value (step SE-5, No), it is determined that the target clause semantic expression X is “impossible to associate” with the phenomenon semantic expression in the “problem frame” semantic expression Y, and the meaning It is determined whether or not the next attribute of the attribute group of expression B exists (step SE-6).

  If such an attribute exists (step SE-6, Yes), paying attention to this attribute, this attribute is changed to attribute T (step SE-7), and the process returns to step SE-5. On the other hand, when such an attribute does not exist (step SE-6, No), the attribute to be associated and its value no longer exist. The fact that “correspondence is impossible” and (correspondence relationship table ([semantic expression A: NIL] [NIL: semantic expression B])) are returned (step SE-8), and the process ends.

  Further, when it is determined in step SE-5 that the attribute T has a pointer value (step SE-5, Yes), the first value of the attribute T is set as a target value (step SE-9. When the “component” attribute has a plurality of values, each value is selected in order from the top. Hereinafter, the same applies when selecting the value of attribute T). Then, the value meaning expression processing is started using the meaning expression below the meaning expression A and the target value as arguments (step SE-10). As a result, when it is returned that the target value is “correspondable” (step SE-11, Yes), the process is ended through the above-described step SE-15.

  On the other hand, when it cannot be associated with any attribute (step SE-11, No), it is determined whether or not the next value in the attribute T exists (step SE-12), and the next value is determined. Is present again as the value of interest (step SE-14), the process returns to step SE-10. On the other hand, when the next value does not exist (step SE-12, No), it is determined whether or not the next attribute of the attribute T exists (step SE-13), and when the next attribute exists. (Step SE-13, Yes), this next attribute is changed to the attribute T (Step SE-7), and the process returns to Step SE-5. On the other hand, when the next attribute does not exist (step SE-13, No), the attribute to be associated and its value no longer exist, and thus the process is ended through the above-described step SE-8.

(Input sentence semantic interpretation processing-Value semantic expression processing)
Next, value meaning expression processing will be described. This value semantic expression processing is mainly executed by the value semantic expression processing unit 53e shown in FIG. 4, and is activated with the semantic expression below the semantic expression A and the target value as arguments. ”Or“ impossible to associate ”and the correspondence table is returned. The overall flow chart of this process is shown in FIGS. 12-1 to 12-7.

  In this process, as shown in FIG. 12A, first, the top semantic expression of the point-destination semantic expression of the target value of the attribute T is changed to the semantic expression B (step SF-1). In the case of “i” or “method i” (step SF-2, Yes), as shown in FIG. 12-5, for semantic expressions below semantic expression A and semantic expressions below semantic expression B, Then, the all-frame correspondence determination process is started (step SF-34). And as a result of this processing, when it is returned that “correspondable” (step SF-35, Yes), the meaning expression A and the meaning expression B are “corresponding”, The correspondence table returned from the all-frame correspondence determination process is returned (step SF-36), and the process ends. On the other hand, when the message “impossible to associate” is returned (step SF-35, No), the meaning expression A and the meaning expression B are “impossible to associate” and the correspondence table ([ A: NIL] [NIL: B]) is returned (step SF-37), and the process ends.

  Also, in step SF-2 of FIG. 12A, if the semantic expression B is not “Premise i” or “Method i” (No in step SF-2), is the semantic expression B “table”? If it is “table” (step SF-3, Yes), the process proceeds to step SF-16 described later. On the other hand, if it is not a “table” (step SF-3, No), the all-frame correspondence determination process is activated for the semantic expression below the semantic expression A and the semantic expression below the semantic expression B (step SF-). 4). Then, as a result of this processing, when it is returned that these two semantic expressions are “correspondable” (step SF-5, Yes), as shown in FIG. The fact that the expression B is “correspondable” and the correspondence table ([semantic expression A: semantic expression B] <lower correspondence table>) are returned (step SF-39), and the processing is performed. finish.

  On the other hand, in step SF-5 in FIG. 12A, when it is returned that both semantic expressions are “cannot be associated” (step SF-5, No), the top of the attribute group of semantic expression B is returned. Focusing on the attribute, this attribute is set as the focused attribute T (step SF-6). Then, it is determined whether or not the target attribute T has a pointer value (step SF-7). If the target attribute T does not have a pointer value (step SF-7, No), the attribute group of the semantic expression B is next. It is determined whether or not the attribute exists (step SF-8). And when such an attribute exists (step SF-8, Yes), paying attention to this attribute, this attribute is changed to the attribute T (step SF-9), and the process returns to step SF-7. On the other hand, if the next attribute does not exist (step SF-8, No), as shown in FIG. 12-7, there is no longer an attribute to be associated and its value, meaning expression A and meaning expression B. Are returned to the effect that "cannot be associated" and the correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) is returned (step SF-40), and the process ends.

  If the attribute T of interest has a pointer value in Step SF-7 of FIG. 21-1 (Yes in Step SF-7), the first value of the attribute of interest T is set as the value of interest (Step SF-10). ), A value meaning expression process is started for the value of interest. If “correspondence is possible” is returned from this process (step SF-12, Yes), the process ends through the above-described step SF-39 of FIG. 12-6. On the other hand, in the case of returning in step SF-12 of FIG. 12-1 that “correspondence is impossible” (step SF-12, No), whether or not the next value of the attribute T of interest exists? (Step SF-13), and if the next value exists (step SF-13, Yes), this is again set as the value of interest (step SF-14), and the process returns to step SF-11.

  On the other hand, if the next value does not exist (step SF-13, No), as shown in FIG. 12B, it is determined whether or not the next attribute of the attribute T of interest exists (step SF-15). ). If the next attribute exists, this is changed to the attribute of interest T (step SF-15, Yes), and the process proceeds to step SF-9 in FIG. On the other hand, in FIG. 12-2, when the next attribute does not exist (step SF-15, No), the process is ended through the above-described step SF-40 of FIG. 12-7.

  Next, the case where it is determined in step SF-3 in FIG. 12A that the semantic expression B is “table” (step SF-3, Yes) will be described. When the semantic expression B is a table, the attributes are generally focused in the order of [sub-attribute 1... Sub-attribute n, main attribute]. The main attribute is an attribute having the same attribute name as that of the table type (for example, a method in the case of a method table), and the sub attribute is other attributes. In the table, since the values of each sub-attribute and main attribute are grouped for each row, it is assumed that the sub-attributes i, j,..., Main attribute values are not selected across rows. Further, the values are traced from the first row in the order of sub-attribute 1,..., Sub-attribute n, main attribute, and if the association fails in the middle (if it is determined that the association is impossible), then the sub-attribute of that line is exceeded. The value is traced from the sub-attribute 1 at the head of the next line without tracing the attribute and main attribute values. Then, when a partner to which the semantic expression A is associated is found somewhere, it is assumed that the process returns to a higher level process at that time. In addition, if it is determined that the previous sentence or clause corresponds to the value semantic representation of a certain sub-attribute or main attribute of a certain line, the subsequent sentence or clause is sub-attribute or sub-attribute of a line other than that line. It cannot have a corresponding relationship with the value attribute expression of the main attribute. Therefore, when it is determined that a certain sentence or clause has a correspondence in a certain row in the table, the row is stored in the storage unit 40 and is referred to when the correspondence is determined. Specifically, as shown in FIG. 12C, it is determined whether or not a correspondence table representing the result of associating the previous sentence and clause with somewhere in the table is added to the table semantic expression ( Step SF-16). If the correspondence relationship table is added (step SF-16, Yes), the process proceeds to step SF-26 described later, and if it is not added (step SF-16, No), the semantic expression Focusing on the first row of B (table), this is designated as row U (step SF-17). Further, paying attention to the head sub-attribute of the attribute group in the row U, this is set as the sub-attribute T (also step SF-17). In FIG. 13, the semantic expression B, which is a table, is illustrated as “method table 1”. Here, the top sub-attribute and the top row are defined as shown.

  Then, it is determined whether or not the sub-attribute T has a pointer value (step SF-18). If the sub-attribute T does not have a pointer value (step SF-18, No), the next sub-attribute of the attribute group in the row U is determined. -It is determined whether or not a main attribute exists (step SF-22). If the next sub-attribute / main attribute is present (step SF-22, Yes), the sub-attribute / main attribute is changed to the attribute T (step SF-23), and the process returns to step SF-18. On the other hand, if the next sub-attribute / main attribute does not exist (step SF-22, No), it is determined whether or not the next line of the line U exists (step SF-24). (Step SF-24, No), the process ends through the above-described step SF-40 in FIG. 12-7. On the other hand, in FIG. 12C, when there is a line next to the line U (step SF-24, Yes), the line is changed to the line U (step SF-25), and the process returns to step SF-17.

  In step SF-18, when the attribute T has a pointer value (step SF-18, Yes), the value of the attribute T is set as a target value (step SF-19), and the target value is used as an argument. The value meaning expression process is started (step SF-20). When it is returned that “cannot be matched” from this process (step SF-21, No), the process proceeds to the above-described step SF-22 and is returned that “can be matched”. (Step SF-21, Yes), as shown in FIG. 12-6, after the returned correspondence table is added to the table semantic expression B (step SF-38), the process goes through the above-described step SF-39. Then, the process ends.

  On the other hand, if it is determined in step SF-16 in FIG. 12-3 that the correspondence table is not added (step SF-16, Yes), it is added as shown in FIG. 12-4. A line in which the corresponding semantic expression is present (below) is read from the correspondence table, and this line is changed to line U (step SF-26). The method of reading the row U is arbitrary. For example, if a correspondence table ([“Existence 1”: “Premise 1”] <...>) Is added, the row U = the first row. There is a possibility that even if there is an association in “Premise 1” before the previous sentence and the previous paragraph, the subsequent sentence and the clause may be associated with “Premise 1” again unless the complete correspondence is made. This process is continued.

  Then, paying attention to the top sub-attribute of the attribute group in this row U, this sub-attribute is set as the attribute U (step SF-27). Next, it is determined whether or not this attribute T has a pointer value (step SF-28). If the attribute T has a pointer value, the value of this attribute T is set as a target value (step SF-31). The value meaning expression processing is started as an argument (step SF-32). By this value semantic expression processing, the possibility of association in the target value or lower semantic expression is searched.

  As a result of the value semantic expression processing, when it is returned that “correspondence is possible” (step SF-33, Yes), the processing is ended through the above-described step SF-38 of FIG. 12-6. To do. On the other hand, in FIG. 12-4, in the case where “impossible to associate” is returned in step SF-33 (No in step SF-33), or in the case where there is no pointer value in step SF-28. (Step SF-28, No), it is determined whether or not the next sub-attribute / main attribute of the attribute group in the row U exists (step SF-29). And when it exists (step SF-29, Yes), this is again set as the attribute T (step SF-30), and it returns to step SF-28. On the other hand, when the next sub-attribute / main attribute does not exist (step SF-29, No), the process ends through the above-described step SF-40 in FIG. 12-7.

(Input sentence semantic interpretation processing-all frame correspondence determination processing)
Next, the all-frame correspondence determination process will be described. This all-frame correspondence determination process is mainly executed by the all-frame correspondence determination processing unit 53f shown in FIG. 4, and the meaning expression below the meaning expression A and below the meaning expression B and the “location” attribute. It is activated with the pair match flag as an argument, and it is “correspondable” or “impossible to correlate”, and the correspondence table is returned. An overall flowchart of this process is shown in FIG.

  In this process, first, the concept type of semantic expression A and semantic expression B is determined (step SG-1). If both semantic expression A and semantic expression B are “predicate concepts” (step SG-1). , Yes), the predicate concept correspondence determination process is started (step SG-2). When both the semantic expression A and the semantic expression B are “noun concepts” (step SG-6, Yes), a noun concept correspondence determination process is started (step SG-7). When the semantic expression A is a predicate concept and the semantic expression B is “premise i” (Yes in step SG-8), “premise i” is a collective phenomenon concept, but it is necessary to perform another processing. Therefore, the premise correspondence determination process is activated (step SG-9).

  When semantic expression A is a predicate concept and semantic expression B is “method i” (step SG-10, Yes), “method i” is a collective phenomenon concept, but needs to be processed separately. Therefore, the method correspondence determination process is started (step SG-11). Then, since it is returned from each of these processes that “correspondable” or “impossible to correlate”, if it is “correspondable” (step SG-3, Yes), the semantic expression A means The fact that “correspondence is possible” to the expression B and the correspondence table returned from the method correspondence determination process are returned (step SG-4), and the process ends. On the other hand, if it is “impossible to associate” (No in step SG-3), the semantic expression A is “impossible to associate” with the semantic expression B and a correspondence table ([semantic expression A: NIL]). [NIL: semantic expression B]) is returned and the process is terminated.

(Input sentence semantic interpretation process-predicate concept correspondence determination process)
Next, predicate concept correspondence determination processing will be described. This predicate concept correspondence determination process is mainly executed by the predicate concept correspondence determination processing unit 53g shown in FIG. 4, and is activated with semantic expressions below semantic expression A and semantic expressions below semantic expression B as arguments. “Matchable” or “impossible to be matched” and a correspondence table are returned. The overall flow chart of this process is shown in FIGS.

  In this process, as shown in FIG. 15A, first, it is determined whether or not the conceptual categories of the semantic expression A and the semantic expression B match or have a higher-lower relationship (step SH-1). If there is no relationship (step SH-1, No), the semantic expression A is “impossible to associate” with the semantic expression B, and a correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]. ]) Is returned (step SH-13), and the process ends. On the other hand, if they are in agreement or in a superior / lower relationship (step SH-1, Yes), an “attribute pair correspondence table” is created for all phenomenon attributes of semantic expression A and semantic expression B (step SH-2). . Here, when one attribute is defined in one of the semantic expressions A and B and the corresponding attribute is not defined in the other, a pair is not created.

  Thereafter, the predicate attribute pair split & correspondence list in the storage unit 40 is initialized (step SH-3). Here, the predicate attribute pair split & correspondence list refers to the possibility of correspondence and the possibility of value propagation regarding a pair of attributes corresponding to each other among the attributes of the semantic expression A and the attributes of the semantic expression B. This list is described, and is configured in the format of FIG. 16, for example. Here, as a premise, as “predicate category”, “confirmation”, “existence”, “state retention”, “achievement”, “action”, “action”, “grant”, or “reception” are provided. In addition, when the predicate concept to be compared is conceptually in a high-level and low-level relationship, the category on the high-level concept side is used. In the “response determination result”, the returned response relationship determination table time & tense response determination result is stored, and “attribute name” is “subject”, “target”, “partner”, “means”, Or, specify the phenomenon attribute of the predicate concept such as “speed”. However, for the “time” attribute or the “location” attribute, the result of the correspondence determination is stored. In addition, as the “propagation possibility”, one of “no propagation”, “A → B propagation (can be propagated from A to B)”, and “B → A propagation (can be propagated from B to A)” is stored. The

  Next, as shown in FIG. 15A, the predicate category-specific tense & time correspondence determination rule stored in advance in the storage unit 40 is called, and the predicate category-specific tense & time correspondence determination rule is added to the predicate concept. The correspondence between the value of the “tense” marker and the value of the “time” attribute and the possibility of value propagation are determined (step SH-4). The format of this predicate category-specific tense & time correspondence determination rule is shown in FIG. Here, as a premise, as “predicate category”, “confirmation”, “existence”, “state retention”, “achievement”, “action”, “action”, “grant”, or “reception” are provided. However, as this “predicate category”, another category may be used according to problem solving. In addition, when the predicate concept to be compared is conceptually in a high-level and low-level relationship, the category on the high-level concept side is used. The respective concept names are used for “predicate A” and “predicate B”. In addition, the “temporal marker value” specifies “present”, “past”, or “future”. The predicate without the tense marker is the current tense (no tense). The “time” attribute value specifies “previous”, “now”, “next year”, “NIL”, or the like. “AB time relationship” designates a relationship between “time” attribute values of predicate A and predicate B or “NIL”. Examples of the relationship between the “time” attribute values include a context and an inclusion relationship (April 1, April, Spring, 2004, etc.), and a plurality of these may be specified. In addition, the predicate category-specific tense & time correspondence determination rule can be adjusted according to the dialogue area.

  Among the predicate category-specific tense & time correspondence determination rules set in this way, as shown in FIG. 15A, the predicate category-specific tense & Select a time correspondence determination rule, and use the selected predicate category-specific tense & time correspondence determination rule, and the correspondence and value of the value of the “tense” marker and the value of the “time” attribute added to the predicate concept The possibility of propagation is determined (step SH-4). For example, if the prescription category-specific tense & time determination rule is applied between “(There are three computers at home)” and “(There are multiple computers at any location).” 18 and it is determined that the association is impossible and there is no propagation. In addition, when the predicate category tense & time correspondence determination rule is applied between “(PC and HUB are connected by cable)” and “(PC and HUB are connected by cable)”, as shown in FIG. It is determined that the association is possible and A → B propagation.

  In this process, as a result of the determination as described above, as shown in FIG. 15A, when “correspondence is impossible” (No in step SH-5), the process is performed through step SH-13. If “correspondence is possible” (step SH-5, Yes), the fact that “correspondence is possible” and the propagation possibility are stored in the predicate attribute pair split & correspondence list (step SH-6).

Thereafter, it is determined whether or not the correspondence relationship of the “location” attribute value matches the value propagation possibility (step SH-7). Here, the case of matching is the case of the following i) to vi), and the case of mismatching is the case other than the following i) to vi) i) Both values are fixed and the same Case (no propagation. For example, “home”: “home”).
ii) When both values are “arbitrary places” (no propagation).
iii) If both values are NIL (no propagation).
iv) When the value on the semantic expression A side is fixed and the value on the semantic expression B side is “arbitrary place” (semantic expression A → semantic expression B propagation. For example, “home”: “arbitrary place”) .
v) The value on the semantic expression A side is fixed or “arbitrary place” and the value on the semantic expression B side is NIL (semantic expression A → semantic expression B propagation).
vi) When the value on the semantic expression A side is NIL and the value on the semantic expression B side is fixed or “arbitrary place” (semantic expression B → semantic expression A propagation).

  If the result of this determination is that they do not match (step SH-7, No), the process is terminated through step SH-13 described above. Further, when they match (step SH-7, Yes), the matching type is specified such as the determination result (for example, “a) -iv” when both values are fixed and the same). Information in a predetermined format) and the possibility of propagation (for example, “if both values are fixed and are identical”, “no propagation”) are stored in the predicate attribute pair split & correspondence list of the storage unit 40 (Step SH-8). If it is determined as iv) and v), the value is propagated by substituting the value on the semantic representation A side into the value on the semantic representation B side.

Then, the correspondence determination result of the “location” attribute value pair is set as a “location” attribute pair match flag (step SH-9), and this is used as an argument to “location” in the “attribute pair correspondence table” created above. Or, by starting the repeated attribute value pair correspondence determination process for each phenomenon attribute pair other than the “time” attribute, such as “subject” attribute, “target” attribute, “means” attribute, etc., the correspondence relationship and propagation of the value pair The possibility is determined (step SH-10). That is, first, an attribute value pair correspondence determination process is started for the “subject” attribute pair, a “subject” attribute pair matching flag indicating the result is stored in the storage unit 40, and then the “subject” attribute pair. Then, the attribute value pair correspondence determination process is activated, and the “target” attribute pair matching flag indicating the result is stored in the storage unit 40. Thereafter, similarly, the attribute value pair correspondence determination process is activated for all attribute pairs other than the “location” or “time” attribute, and an attribute pair match flag indicating the result is stored in the storage unit 40 (in the drawing). The attribute value pair correspondence determination process is shown only once). From this process, one of the following results is returned for each attribute. That is, i) When the pointer value pair is “correspondable”, if A and B are “correspondable”, it is “correspondable” and the correspondence table of 1) below When the pointer value pair is “can be associated”, B is divided, A and B mention subsets are “can be associated”, and B unmentioned subsets are “cannot be associated” anywhere The following 2) correspondence table is returned, and when the non-pointer value is “correspondable”, “correspondence is possible” and the following 3) correspondence table is returned respectively. The
1) Correspondence table ([A: B] <lower correspondence table T>)
2) Correspondence table ([A: B mention subset] <lower correspondence table T>
[NIL: B unmentioned subset]
3) Correspondence relationship table ([A: NIL] [NIL: B])

For example, a specific example in the case where the pointer value pair is “can be associated” will be described. After the correspondence determination result of the “place” attribute corresponding pair is set as the “location” attribute pair match flag, “existence 1” is set. The attribute “subject” value = “PC1 & cable 1” and the “existence 2” attribute “subject” value = “PC2 & cable 2 & HUB1” are passed as arguments to the attribute value pair correspondence determination process, and the attribute value pair correspondence Start the judgment process. Here, when “PC1” is part of “PC2” and “cable 1” is part of “cable 2” and “correspondable” respectively, the attribute value pair correspondence determination processing In the “non-composite concept correspondence determination process” further called from the above, the following correspondence table is returned to the “noun concept correspondence determination process” for performing the “PC1 & cable”: “PC2 & cable 2 & HUB1” correspondence determination.
Correspondence table ([PC1: PC2 mention subset] <NIL>
[NIL: PC2 unmentioned subset]
Correspondence relationship table ([cable 1: cable 2 mention subset] <NIL>
[NIL: Cable 2 unmentioned subset])
Correspondence table ([PC1: NIL]
[NIL: HUB1])
Correspondence table ([Cable 1: NIL]
[NIL: HUB1])

Then, the correspondence table is divided into a corresponding part and an uncorresponding part, and the following correspondence table is obtained.
Correspondence table ([PC1: PC2 mention subset] <NIL>
[Cable 1: Cable 2 mention subset] <NIL>) (TI)
Correspondence table ([NIL: PC2 unmentioned subset]
[NIL: Cable 2 unmentioned subset]
[NIL: HUB1]) (TII)

Among these, the following correspondence table is returned with the correspondence table (TI) as the “lower correspondence table”.
Correspondence Relationship Table ([PC1 & Cable 1: PC2 & Cable 2 & HUB1 Reference Subset] <Lower Correspondence Relationship Table TI>
[NIL: PC2 & Cable 2 & HUB1 unmentioned subset])

  Similarly, a correspondence determination result of an attribute value other than “location” or “time” of “existence 1” and an attribute value other than “location” or “time” of “existence 2” is returned. Since the correspondence table is filled in for each attribute based on the predicate category-by-predicate correspondence determination rule described later, it is not necessary to integrate the table for each attribute here.

  As a result of such processing, when it is returned that all return values are not “correspondable” (step SH-11, No), the processing is ended through the above-described step SH-13. On the other hand, when it is returned that all return values are “correspondable” (step SH-11, Yes), the correspondence determination result for each attribute value pair and the propagating possibility are expressed as a predicate attribute pair. It stores in the division & correspondence list (step SH-12). Then, as shown in FIG. 15-2, using the predicate category-specific pairing determination rule stored in advance in the storage unit 40, the correspondence between predicate concepts is determined from the contents of this list, and a correspondence table is created. (Step SH-14).

The format of this predicate category-specific pairing determination rule is shown in FIG. In this predicate category-specific pairing determination rule, the “attribute name” specifies, as a premise, a phenomenon attribute of the predicate concept such as “subject”, “target”, “partner”, “means”, “speed”, and the like. In addition, the “correspondence determination result” designates the correspondence table 1) to 3) below or 4) that is assumed to be returned from the attribute value pair correspondence determination process for each attribute value pair. Alternatively, when the attribute is “location”, one of “match type” obtained as the “location” correspondence determination result is designated. Then, the following combinations 1) to 4) designated for each attribute are conditions for predicate concept division & association determination.
1) Correspondence table ([A: B] <lower correspondence table>)
2) Correspondence table ([A: NIL] [NIL: B])
3) Correspondence table ([A: B mention subset] <lower correspondence table> [NIL: B unmentioned subset])
4) “Matchable” / “Matchable”

Further, the “predicate correspondence” is a correspondence table representing a correspondence determination result between the predicate concept U and the predicate concept V when the two predicate concepts to be matched are the predicate concept U and the predicate concept V. Is specified. This “correspondence table” is determined as follows based on the correspondence determination result of each attribute (here, <lower correspondence table i> is a correspondence table returned in each attribute value pair). is there).
When U and V are “matchable”:
(1) Correspondence table ([U: V] <lower correspondence table 1>... <Lower correspondence table n>)
When U and V are “matchable”:
(2) Correspondence relationship table ([U: NIL] [NIL: V])
When V is divided and U and V mention subsets are “correspondable”, but V unreferenced subsets are “unmatchable” everywhere:
(3) Correspondence table ([U: V mention subset] <lower correspondence table 1>... <Lower correspondence table n> [NIL: V unreferenced subset])

  For example, in the “existence” phenomenon, when the correspondence table specified as the correspondence determination result of the “subject” attribute is the correspondence table of (1) or (2) above, the “existence” meaning expression on the knowledge side is The original semantic expression is not divided. On the other hand, in the case of the correspondence table in (3) above, the “existence” meaning expression on the knowledge side is divided into a V mention subset and a V unreference subset. Further, the value of each attribute in the V mention subset and the V unreference subset is set for each individual rule based on the correspondence tables 1 to n.

  For example, in the case of the “existence” phenomenon, the “time” attribute is “matchable” and the “location” attribute correspondence is “match”. When the value of the “subject” attribute is divided, the “existence” concept is also divided into two “existence” concepts each having the value of the “subject” attribute. As a result, it becomes as shown in FIG. Here, the value of the “subject” attribute in the V mention subset is substituted with the B mention subset in the correspondence table ([A: B mention subset] [NIL: B unreferenced subset]). Further, the B unreferenced subset in the correspondence table ([A: B reference subset] [NIL: B unreferenced subset]) is substituted for the value of the “subject” attribute in the V unreferenced subset. Here, the values of the other attributes including the “time” or “location” attribute have the same value in both the V mention subset and the V unreferenced subset. For example, when “place” is “match” and B is “arbitrary place” and replaced with the value of A, the value replaced with both of the divided predicates is taken. For attribute values that are determined to be propagateable, the value is propagated here.

  Using such predicate category-specific pairing determination rules, as shown in FIG. 15-2, a correspondence table is created (step SH-14), and predicate B is divided and attribute values are propagated as necessary. After performing (step SH-15), the semantic expression A is “can be associated” with the semantic expression B, and the correspondence table created so far is returned (step SH-16), and the processing is performed. Exit.

(Noun concept correspondence determination process)
Next, the noun concept correspondence determination process will be described. This noun concept correspondence determination process is mainly executed by the noun concept correspondence determination processing unit 53h shown in FIG. 4. The semantic expression below the semantic expression A, the semantic expression below the semantic expression B, and “location”. It is activated with the attribute pair match flag as an argument, and it is “correspondable” or “impossible to associate” and a correspondence table is returned. Here, since the noun concept has a composite concept, it is first decomposed to determine the correspondence of the combination of each element, and a process of collecting the entire result is performed. However, at the time of determining the correspondence of the noun concept, the correspondence relationship of the “place” attribute in the upper “predicate concept pair” has already been determined and propagated as the “place” attribute pair match flag. It is assumed that the set component of the composite concept is not a composite concept. The overall flow chart of this process is shown in FIGS. 22-1 to 22-4.

  In this process, as shown in FIG. 22A, it is determined whether or not each concept is a composite concept (whether or not it has a “composite” flag). Specifically, it is determined whether both the semantic expression A or the semantic expression B is a parallel phrase composition concept (step SI-1). Although the semantic expression A is not a parallel phrase composition concept, the semantic expression B is a parallel phrase composition concept. It is determined (step SI-2), and it is determined whether the semantic expression A is a parallel phrase composition concept but the semantic expression B is not a parallel phrase composition concept (step SI-3).

  Here, when “No” is selected in all steps of step SI-1, step SI-2, and step SI-3, that is, both semantic expression A and semantic expression B are parallel phrase composition concepts. In this case, if there is a component that modifies the whole semantic expression A, it is distributed to each value of the set component, and if there is a component that modifies the entire semantic expression B, it is assigned to the set component. Distribute to each value (step SI-4). Next, the value of the “collective component” attribute of the semantic expression A is set as a value Ai, and the value of the “collective component” attribute of the semantic expression B is set as a value Bj (step SI-5). Here, it is determined whether or not the value Ai and the value Bj exist (step SI-6). When the value Ai and the value Bj exist (step SI-6, Yes), the value Ai and the value Bj For all combinations, a non-synthetic concept correspondence determination process described later is started to determine the correspondence between the two (step SI-7).

  Further, from the combination of the results of the non-synthetic concept correspondence determination process, it is determined whether the semantic expression A and the semantic expression B are complete match, partial match, or complete non-correspondence (step SI-8). If the semantic expression A and the semantic expression B completely match and it is not necessary to divide the semantic expression B (step SI-8, complete matching), the semantic expression A and the semantic expression B are “corresponding”. The fact and the correspondence table returned from the non-synthetic concept correspondence determination process are returned (step SI-9), and the process ends.

Specifically, the value of the “set component” attribute of the semantic expression A and the semantic expression B is the same number (A1 to An, B1 to Bn), and each [Ai: Bi] pair (i = 1 to n) is In the case of “can be associated” without division, that is, the following table 1) is returned from the non-synthetic concept correspondence determination processing for each [Ai: Bi] pair, and the result of integrating these pairs for all pairs When the following 2) is obtained, A and B are “corresponding to each other” and the correspondence table 3) below is returned, and the process is terminated.
1) Correspondence relationship table ([Ai: Bi] <lower correspondence relationship table Ti>) (i = 1 to n)
2) Correspondence table ([A1: B1] <lower correspondence table T1>
...
[An: Bn] <lower correspondence table Tn>) (T)
3) Correspondence table ([A: B] <lower correspondence table T>)

  Further, in the determination process of step SI-8, since the semantic expression A matches a part of the semantic expression B, it is necessary to divide and associate the semantic expression B according to the result (step SI-8, part). Match), the synthetic concept B semantic expression is divided, and the semantic expression A and the semantic expression B are “correspondable” and a correspondence table is returned (step SI-10).

Specifically, the semantic representation B is divided into a portion associated with the semantic representation A and a portion that is not (this time, since the division on the A side is not considered). That is, the result of the non-synthetic concept correspondence determination processing for each pair of Aj (j = 1 to w) and Bk (k = 1 to z) includes the following correspondence relationship tables 1) to 4). If the result obtained by integrating the part and the uncorresponding part is the following correspondence table 5) to 6), the synthetic concept B semantic expression is expressed as “Bf,... Br, Bl reference subset,. B mention subset having “referenced subset” as the value of “set constituent component” attribute, and “Bl unreferenced subset,..., Bo unreferenced subset, Bp,. Into B unmentioned subsets. Then, the semantic expression A and the semantic expression B are “can be associated” and the following correspondence table 7) is returned, and the process is terminated.
1) Correspondence table ([Ah: Bf] <lower correspondence table Th>) (a table corresponding to a pair that can be “associated” without division. However, h is the same for Ah and Th. There is no meaning.)
2) Correspondence table ([Aj: Bl mention subset] <lower correspondence table Tj> [NIL: Bi unsubscribed subset]) (for a pair that becomes “correspondable” by dividing B side table.)
3) Correspondence relationship table ([Ad: NIL]) (table for the A-side component that cannot be associated with any of the B-side components)
4) Correspondence relationship table ([NIL: Bp]) (a table for the B side component that cannot be associated with any of the A side components)
5) Correspondence relationship table ([Ah: Bf] <lower correspondence table Th>
...
[Ai: Br] <lower correspondence table Ti>
[Aj: Bl mention subset] <lower correspondence table Tj>
...
[Ak: Bo mention subset] <lower correspondence table Tk>) (TI)
6) Correspondence table ([NIL: Bl unmentioned subset]
...
[NIL: Bo unmentioned subset]
[Ad: NIL]
...
[Am: NIL]
[NIL: Bp]
...
[NIL: Bq]) (TII)
7) Correspondence table ([A: B mention subset] <lower correspondence table TI>
[NIL: B unmentioned subset]

  On the other hand, in step SI-1 of FIG. 22-1, when it is determined that both semantic expression A and semantic expression B are not parallel phrase composition concepts (step SI-1, Yes), as shown in FIG. 22-4, In order to determine the correspondence between the two, a non-synthetic concept correspondence determination process is activated for semantic expression A and semantic expression B (step SI-27). Next, it is determined whether or not the two can be “correspondable” (step SI-28). If “correspondable” (step SI-28, Yes), the semantic expression A is “correspondable to the semantic expression B”. ”And the correspondence table returned from the non-synthetic concept correspondence determination process is returned (step SI-29), and the process ends. On the other hand, if it is not “correspondable” (step SI-28, No), the meaning expression A is “impossible to associate” with the meaning expression B, and a correspondence table ([A: NIL] [NIL: B]). ) Is returned (step SI-30), and the process ends.

  Further, in step SI-2 of FIG. 22-1, if it is determined that semantic expression A is not a parallel phrase composition concept but semantic expression B is a parallel phrase composition concept (step SI-2, Yes), FIG. As shown in -3, if there is a component that modifies the entire semantic expression B, it is distributed to each value of the collective component (step SI-20). Next, the non-synthetic concept correspondence determination process is activated for all pairs of the “collective component” attribute Bj of the semantic expression B and the semantic expression A to determine the correspondence (step SI-21). Based on the entire result, the semantic expression B is divided as necessary to create a correspondence table.

When a certain [A: Bq] pair is “no division between both” and “can be associated”, a correspondence table as in 1) below is returned, and each other [A: : Bk (k ≠ q)] pair is “impossible to associate”, a table as in 2) below is returned, and the result of integrating this into a corresponding part and an uncorresponding part is the following 3 ) To 4) (step SI-22, Yes), the combined concept B semantic expression is represented by a B reference subset having the Bq reference subset as a value of the “set component” attribute, and “B1,. “Bz (z ≠ q)” is divided into B unmentioned subsets having the value of the “set component” attribute. Then, the semantic expression A is “can be associated with” the semantic expression B, and the following correspondence table 5) is returned (step SI-23), and the process ends.
1) Correspondence table ([A: Bq] <lower correspondence table Tq>)
2) Correspondence relationship table ([A: NIL] [NIL: Bk]) (k ≠ q)
3) Correspondence table ([A: Bq] <lower correspondence table Tq>) (TI)
4) Correspondence table ([NIL: B1]... [NIL: Bz (1, z ≠ q)]) (TII)
5) Correspondence table ([A: B mention subset] <lower correspondence table TI>
[NIL: B unmentioned subset]

Alternatively, when a certain [A: Bq] pair is “partitionable by Bq” and “can be associated”, a table such as 1) below is returned, and each other [A: Bk ( k ≠ q)] If the pair is “unmatchable”, the following table 2) is returned. When the result of integrating the corresponding part and the non-corresponding part is a table like the following 3) to 4) (step SI-25, Yes), the combined concept B meaning The B reference subset with the Bq mention subset as the value of the “set component” attribute, and “B1,..., Bz (z ≠ q), Bq unreferenced subset” with the “set component” attribute The value is divided into B unmentioned subsets. Then, the semantic expression A is returned as “separable” to the semantic expression B and the following 4) table (step SI-24), and the process ends.
1) Correspondence table ([A: Bq mention subset] <lower correspondence table Tq> [NIL: Bq unmentioned subset])
2) Correspondence relationship table ([A: NIL] [NIL: Bk]) (k ≠ q)
3) Correspondence table ([A: Bq mention subset] <lower correspondence table Tq>) (TI)
4) Correspondence relationship table ([NIL: Bq unreferenced subset]
[NIL: B1] ... [NIL: Bz (z ≠ q)]) (TII)
5) Correspondence table ([A: B mention subset] <lower correspondence table TI>
[NIL: B unmentioned subset]

Alternatively, if not all of the set components of the semantic expression B are associated with A, that is, each [A: Bk] pair is “cannot be associated”, the following table 1) is returned. When the result of integrating the table becomes the table 2) below (duplicate [A: NIL] is reduced to one) (step SI-25, No), semantic expression A corresponds to semantic expression B. “Not attachable” and the following 3) correspondence table are returned (step SI-26), and the process is terminated.
1) Correspondence table ([A: NIL] [NIL: Bk]) (k = 1 to z)
2) Correspondence relationship table ([A: NIL] [NIL: B])
3) Correspondence relationship table ([A: NIL] [NIL: B])

  Furthermore, in step SI-3 of FIG. 22-1, when it is determined that the semantic expression A is a parallel phrase composition concept but the semantic expression B is not a parallel phrase composition concept, as shown in FIG. If there is a component that modifies the entire expression A, it is distributed to each value of the collective component (step SI-11). Next, a non-synthetic concept correspondence determination process is activated for all pairs of the semantic expression B and the value Ai of the “collective component” attribute of the semantic expression A to determine the correspondence between them (step SI− 12). Based on the entire result, the semantic expression B is divided as necessary to create a correspondence table.

Then, when a certain [Ai: B] pair is “correspondable” without division between the two, the following table 1) is returned, and each other [Aj (j ≠ i): B] When the pair is “impossible to associate” (step SI-13, Yes), the following table 2) is returned, and the result of integrating these is the table 3) below Since the A side is not divided (in this application, on the input sentence side, the context, the knowledge and the uncorresponding part do not matter), the following 4) table is returned (step SI-15) ), The process ends.
1) Correspondence relationship table ([Ai: B] <lower correspondence relationship table Ti>),
2) Correspondence relationship table ([Aj: NIL] [NIL: B]) (j ≠ i)
3) Correspondence relationship table ([A1: NIL]... [Aw (1, w ≠ i): NIL]
[Ai: B] <lower correspondence table Ti>) (T)
4) Correspondence table ([A: B] <lower correspondence table T>)

Alternatively, when one [Ai: B] pair is “corresponding” in the division of B (step SI-16, Yes), the following table 1) is returned, and each other [ When the Aj (j ≠ i): B] pair is “unmatchable”, the following table 2) is returned. When the result of integrating the corresponding parts into the corresponding parts and the non-corresponding parts is as shown in the following 3) to 4), the following 5) table is returned (step SI-18). ), The process ends.
1) Correspondence table ([Ai: B mention subset] <lower correspondence table Ti>
[NIL: B unmentioned subset]],
2) Correspondence relationship table ([Aj: NIL] [NIL: B]) (j ≠ i)
3) Correspondence table ([Ai: B mention subset] <lower correspondence table Ti>) (TI)
4) Correspondence relationship table ([A1: NIL]... [Aw (w ≠ i): NIL]
[NIL: B unmentioned subset]) (TII)
5) Correspondence table ([A: B mention subset] <lower correspondence table TI>
[NIL: B unmentioned subset]

Alternatively, when all set components of the non-synthetic concept A are not associated with the non-synthetic concept B (step SI-16, No), that is, each [Aj: B] pair is “unmatchable”, When a table such as 1) is returned and the result of integrating the table is as shown in 2) below (duplicate [NIL: B] is reduced to one), FIG. As shown in FIG. 4, the non-composite concept A returns “unmatchable” to the non-composite concept B and the following 3) table (step SI-30), and the process ends.
1) Correspondence relationship table ([Aj: NIL] [NIL: B]) (j = 1 to w)
2) Correspondence table ([A1: NIL] ... [Aw: NIL] [NIL: B])
= Correspondence table ([A: NIL] [NIL: B])
3) Correspondence relationship table ([A: NIL] [NIL: B])

(Non-composite concept correspondence determination process)
Next, the non-synthetic concept correspondence determination process will be described. This non-synthetic concept correspondence determination process is mainly executed by the non-synthetic concept correspondence determination processing unit 53i shown in FIG. 4, and includes a semantic expression below the non-synthetic concept A, a semantic expression below the non-synthetic concept B, and , “Location” attribute pair match flag is started as an argument, and the correspondence table is returned. The overall flow chart of this process is shown in FIGS. 23-1 to 23-3.

  In this process, as shown in FIG. 23-1, the conceptual categories of the semantic expression A and the semantic expression B are compared (step SJ-1). Then, when the semantic expression A and the semantic expression B do not match or do not have an upper-lower relationship (step SJ-1, No), the semantic expression A is “unmatchable” with the semantic expression B as shown in FIG. And the correspondence table ([A: NIL] “NIL: B”) are returned (step SJ-9), and the process ends.

  On the other hand, in FIG. 23A, when the semantic expression A and the semantic expression B match or are in a higher-order subordinate relationship (step SJ-1, Yes), the attribute pair for all the attributes of the semantic expression A and the semantic expression B is shown. A correspondence table is created (step SJ-2). Here, when one attribute is defined in one of the semantic expressions A and B and the corresponding attribute is not defined in the other, a pair is not created.

  Next, similarly to step SH-10 in FIG. 15A, the correspondence relationship between value pairs and the propagation possibility are determined by starting the repeated attribute value pair correspondence determination process for each phenomenon attribute pair (step SJ). -3.In the figure, the attribute value pair correspondence determination process is shown only once). As a result of the attribute value correspondence determination process, for each pair Ai: Bi, it is possible to associate, one of the following correspondence tables (when the target pair has a pointer value), and the possibility of attribute value propagation ( No propagation, A → B propagation, B → A propagation) is returned. Here, the “lower correspondence table” is a correspondence table returned as a result of the correspondence determination regarding the lower-level semantic expression pairs that contributed to the determination of the correspondence relationship of each pair. Ai and Bi are synthetic concepts or non-synthetic concepts.

When Ai and Bi are “correspondable”, the correspondence table returned here becomes 1) or 2) below (Bi is divided, and Ai and Bi mention subsets are “corresponding”. If “Attachable” and “Ai” and “Bi” are “Unable to associate”, the following table 3) is obtained. Furthermore, the general form of the correspondence table in which this is divided into a corresponding part and an uncorresponding part and integrated is as shown in 4) or 5) below.
1) Correspondence table ([Ai: Bi] <lower correspondence table Ti>)
2) Correspondence table ([Ai: Bi mention subset] <lower correspondence table Ti>
[NIL: Bi not mentioned subset])
3) Correspondence table ([Ai: NIL] [NIL: Bi])
4) Correspondence table ([Ac: Bc] <lower correspondence table Tc>
...
[Ad: Bd] <lower correspondence table Td>
[Ae: Be mentioning subset] <lower correspondence table Th>
...
[Af: Bf mention subset] <lower correspondence table Tk>) (TI)
5) Correspondence relationship table ([Al: NIL]
...
[Ao: NIL]
[NIL: Bl]
...
[NIL: Bo]
[NIL: Be unsubscribed subset]
...
[NIL: Bf unmentioned subset]) (TII)

  When there is an attribute value pair that is not “correspondable” in the return value from such attribute value pair correspondence determination processing (when “correspondence impossible” or the correspondence table TII is included) (step SJ-4, No), The process is terminated through step SJ-9 in FIG.

  On the other hand, in FIG. 23-1, when all the return values from the attribute value pair correspondence determination processing are “corresponding” (“cannot be correlated” or does not include the correspondence table TII) (step SJ-4, Yes) After storing this return result in the non-synthetic concept attribute pair split & correspondence list, the relationship between the values of the “place” attribute in the non-synthetic concept A and the non-synthetic concept B (the value of the “location” attribute pair match flag) Is used to determine the ideal correspondence between the non-synthetic concept A and the non-synthetic concept B (step SJ-5). As a result, when the correspondence relationship between the values of the “location” attribute is “match” (“home”: “any location”, etc.) (step SJ-5, Yes), the attribute pair split & correspondence list includes the storage unit. The correspondence relationship as a whole concept is determined by applying the non-synthetic concept correspondence determination rule stored in 40, and a correspondence table is created (step SJ-7).

The format of this non-synthetic concept correspondence determination rule is shown in FIG. Here, as a premise, “attribute name” designates a phenomenon attribute possessed by a noun concept. The “correspondence determination result” designates one of the following correspondence tables 1) to 4) that is assumed to be returned from the attribute value pair correspondence determination process for each attribute value pair. Thus, the combinations of the correspondence relationship tables 1) to 4) specified for each attribute are the conditions for the non-synthesis concept division & association determination.
1) Correspondence table ([A: B] <lower correspondence table>)
2) Correspondence relationship table ([A: NIL] [NIL: B])
3) Correspondence table ([A: B mention subset] <lower correspondence table>
[NIL: B unmentioned subset]
4) “Matchable” / “Matchable”

Here, the “non-synthetic concept correspondence” is the correspondence between the non-synthetic concept U and the non-synthetic concept V when the two non-synthetic concepts to be matched are defined as a non-synthetic concept U and a non-synthetic concept V. Specifies the correspondence table that represents the determination result. This “correspondence relationship table” is one of the following 1) to 3) based on the correspondence determination result of each attribute.
1) Correspondence table ([U: V] <lower correspondence table 1>... <Lower correspondence table n>) (when U and V are “corresponding”)
2) Correspondence relationship table ([U: NIL] [NIL: V]) (when U and V are “cannot be associated”)
3) Correspondence table ([U: V mention subset] <lower correspondence table 1>... <Lower correspondence table n>
[NIL: V unreferenced subset]) (when V is divided, U and V mention subsets are “matchable”, and V unreferenced subsets are “unmatchable” everywhere)

  In the above, <lower correspondence table i> is a correspondence table returned in each attribute value pair. Further, the value of each attribute in the V mention subset and the V unreference subset is set for each individual rule based on the correspondence tables 1 to n. The values of the other attributes, including the “time” and “location” attributes, have the same value for both the V mention subset and the V unreferenced subset. For example, when “place” is “match” and B is “arbitrary place” and replaced with the value of A, the value replaced with both of the divided predicates is taken. For attribute values that are determined to be propagateable, the value is propagated here.

  By creating a correspondence table based on the determination result based on the non-synthetic concept correspondence determination rule, the non-synthetic concept B is divided as necessary, and attribute values are propagated as necessary. Thereafter, as shown in FIG. 23-2, the non-synthetic concept A returns “not compatible” to the non-synthetic concept B and the correspondence table (step SI-8), and the process ends. To do. On the other hand, in step SJ-5 of FIG. 23-1, when the correspondence relationship of the value of the “location” attribute is “mismatch” (“home”: “company”, etc.) (step SJ-5, No), The process ends through step SJ-10 in FIG.

(Prerequisite correspondence determination processing)
Next, the premise correspondence determination process will be described. This premise correspondence determination processing is mainly executed by the premise correspondence determination processing unit 53j shown in FIG. 4, and is activated with semantic expressions below semantic expression A and semantic expressions below semantic expression B as arguments. The fact that “correspondence is possible” or “impossibility of correspondence” and the correspondence table are returned. The overall flow chart of this process is shown in FIGS.

  In this premise correspondence determination processing, as shown in FIG. 25A, it is determined whether or not the value of the “collective component” attribute of the “premise i” semantic expression B exists (step SK-1). Here, the concept category of “Premise i” corresponds to “phenomenon” and is positioned at the top. Accordingly, since the semantic expression A is not directly associated with the semantic expression B, the correspondence between the semantic expression A and the semantic expression B is not determined, and the value of the “collective component” attribute of the semantic expression B is directly focused on. . And when such a value does not exist (step SK-1, No), as shown to FIG. 25-2, the meaning expression A cannot be matched with the meaning expression B, and a correspondence table ( [A: NIL] [NIL: premise i]) is returned (step SK-15), and the process is terminated. On the other hand, in FIG. 25A, when such a value exists (step SK-1, Yes), it is determined whether or not the “reference subset AP” or “unreferenced subset BP” is included in this value. (Step SK-3).

  Then, when “reference subset AP” or “unreferenced subset BP” is not included (step SK-3, No), that is, when “premise i” has never been mentioned so far, The number of values of the “collective component” attribute is n, j = 1 to n, the jth value of the “collective component” attribute is Bj, and the semantic representation below the semantic representation A and the semantic representation below each Bj. On the other hand, predicate concept correspondence determination processing is started (step SK-4).

As a result of this processing, if the semantic expression A is “uncorresponding to Bj” with respect to the semantic expressions A and Bj, “the semantic expression A is“ uncorresponding to Bj ”, When the following table 1), meaning expression A is “corresponding to Bj”, “A can be“ corresponding to Bj ”, and the following tables 2) to 5) Either is returned.
1) Correspondence table ([A: NIL] [NIL: Bj])
2) Correspondence relationship table ([A: Bj] <lower correspondence relationship table T>) (j = 1 to n) (complete correspondence)
3) Correspondence relationship table (["existence 1": "existence 2"] <correspondence relationship table (["PC1 & cable 1": "PC2 & cable 2"]>)
4) Correspondence table ([A: Bj mention subset] <lower correspondence table T>
[NIL: Bj unmentioned subset]) (j = 1 to n)
5) Correspondence relationship table ([“Existence 1”: “Existence 2 mention subset”]]
<Correspondence table (["PC1 & cable 1": "PC2 & cable 2 & HUB1 mention subset"]>)
[NIL: “Existence 2 unmentioned subset”])

  When the semantic expressions A and Bf completely correspond, the remaining “Bg,..., Bn” remain unsupported. When a part of the semantic expression A completely corresponds to “Bf,..., Bg” and further corresponds to “Bh mention subset,..., Bk mention subset,” “Bh unreferenced subset,. The reference subset, B1,..., Bm ”remains uncorresponding (At this time, a part of the semantic expression A may remain uncorresponding to any Bj. However, in this application, it is assumed that sufficient knowledge is given. (Because it is assumed, I don't think that part of the semantic expression A is associated but not part). If the semantic expression A does not correspond to any Bj, all Bj remain unsupported.

For example, if A = “existence 1” completely corresponds to B1 = “existence 2” and does not correspond to B2 = “existence 3” (j = 1, 2), 1) or 2) below, A = When the entire “existence 1” corresponds to a part of B1 = “existence 2” and does not correspond to B2 = “existence 3” (j = 1, 2), the following 3) or 4): = “Existence 1” does not correspond to any of B1 = “Existence 2” and B2 = “Existence 3” (j = 1, 2), either of 5) or 6) below is returned .
1) Correspondence relationship table (["existence 1": "existence 2"] <correspondence relationship table (["PC1 & cable 1": "PC2 & cable 2"]>)
2) Correspondence relationship table ([“Existence 1”: NIL]
[NIL: “Existence 3”])
3) Correspondence table (["existence 1": "existence 2 mention subset"]
<Correspondence table (["PC1 & cable 1": "PC2 & cable 2 & HUB1 mention subset"]>)
([NIL: “existence 2 unmentioned subset”])
4) Correspondence table (["Existence 1": NIL]
[NIL: “Existence 3”])
5) Correspondence table ([“Existence 1”: NIL]
[NIL: “Existence 2”
6) Correspondence relationship table (["Existence 1": NIL]
[NIL: “Existence 3”])

Thereafter, the results for j = 1 to n are divided into corresponding parts and uncorresponding parts and integrated (step SK-5). Thereafter, it is determined whether or not all return results are “impossible to associate” (step SK-6). Here, when A completely corresponds to Bf and the remaining Bg,..., Bm does not correspond, 1) and 2) below, and a part of A is completely Bf,. Corresponding to a Bh mention subset,..., Bk mention subset (Bh unreferenced subset,..., Bk unreferenced subset, Bl,..., Bm remain uncorresponding). In the following 3) and 4), when A does not correspond to any Bj (all Bj remains unsupported), the following table 5) is returned.
1) Correspondence table ([A: Bf] <lower correspondence table Tf>) (TI)
2) Correspondence table ([NIL: Bg] ... [NIL: Bm]) (TII)
3) Correspondence table ([A: Bf] <lower correspondence table Tf>
...
[A: Bg] <lower correspondence table Tg>
[A: Bh mention subset] <lower correspondence table Th>
...
[A: Bk mention subset] <lower correspondence table Tk>) (TI)
4) Correspondence relationship table ([NIL: Bh unmentioned subset] ... [NIL: Bk unmentioned subset]
[NIL: Bl] ... [NIL: Bm]) (TII)
5) Correspondence table ([A: NIL] [NIL: Bf]... [NIL: Bm]) (TII)

For example, when A = “existence 1” completely corresponds to B1 = “existence 2” and does not correspond to B2 = “existence 3” (j = 1, 2), the following 1) and 2) = “Existence 1” as a whole corresponds to a part of B1 = “Existence 2” and does not correspond to B2 = “Existence 3” (j = 1, 2), the following 3) and 4): When A = “existence 1” does not correspond to any of B1 = “existence 2” and B2 = “existence 3” (j = 1, 2), the following table 5) is returned.
1) Correspondence relationship table (["existence 1": "existence 2"] <correspondence relationship table (["PC1 & cable 1": "PC2 & cable 2"]>) (TI)
2) Correspondence table ([NIL: “existence 3”]) (TII)
3) Correspondence table (["existence 1": "existence 2 mention subset"]
<Correspondence table (["PC1 & cable 1": "PC2 & cable 2 & HUB1 reference subset"]>) (TI)
4) Correspondence table ([NIL: “existence 2 unmentioned subset”]]
[NIL: “Existence 3”]) (TII)
5) Correspondence table ([“Existence 1”: NIL]
[NIL: “Existence 2”]
[NIL: “Existence 3”]) (TII)

  As a result of such a determination, if the return value does not include “correspondable” or the correspondence table TI (all return values are “unmatchable” or only the correspondence table TII) (step SK− 6, No), the process is terminated through step SK-15. On the other hand, if the return value includes “can be matched” or TI (step SK-6, Yes), a “reference subset AP” is generated (step SK-7). Here, the attributes are “collective component” and “number of collective components”. In addition, the element of the correspondence table TI, that is, [A: Bj] or [A: Bj mention subset] in the integrated correspondence table is added to the value of the “set component” attribute of the reference subset AP. And a pointer to a Bj reference subset. In addition, the number corresponding to this is stored in the value of the “number of collective components” attribute. Then, the “set component” attribute of “premise i” is cleared, and a pointer to the “reference subset AP” is newly stored.

  Next, “unreferenced subset BP” is generated (step SK-8), where the attributes are “set component” and “number of set components”. In the integrated correspondence table, elements of the correspondence table TII, that is, Bj unmentioned subsets and Bk having the form [NIL: Bj unmentioned subset] and [NIL: Bk] are represented by A And whether it belongs to the same or higher-order concept category category. Also, “HUB needs to exist” and “Protocol needs to exist” are integrated. Here, a phenomenon semantic expression integration rule described later is used to collect all the Bj unreferenced subsets and Bk whose concept category is the same as A or belonging to the upper and lower categories and integrate them into one phenomenon semantic expression P. Further, in the returned [NIL: Bj], the value of the “set component component” attribute of “unknown subset BP” is the same as A or the concept category of Bj that does not belong to the same or higher / lower category, and , The pointer to P is stored. Then, the number corresponding to this is stored in the value of the “set component number” attribute, and a pointer to “unreferenced subset BP” is stored in the “set component” attribute of “premise i”.

Here, the phenomenon semantic expression integration rule will be described. First, in the premise, it is determined that phenomena having the same phenomenon category, the same tense and phase, and the same subject can be integrated. However, “existence” is determined to be able to integrate “existence” having different subjects if the “location” is the same or has an inclusion relationship. Further, all Bj and Bk having the same concept category as A and belonging to the upper and lower categories are collected and integrated into one phenomenon meaning expression P.
1) By referring to a dictionary stored in advance in the storage unit 40, a phenomenon concept semantic expression P corresponding to the category is generated. Adjust the instance number so that it does not overlap (same as other cases). If markers are assigned to each predicate concept before integration, it is assigned to P.
2) A list of pairs (sets) is created for each corresponding phenomenon attribute (the “set component” attribute does not exist in the phenomenon concept) of each Bj unmentioned subset, Bk.
3) All attribute pairs (groups) are selected in order from the top of the list of pairs (groups), and respective values in Bj and Bk are acquired.
4) A composite value T in which the acquired values are bundled is generated. When each value is a pointer value, a composite value T of the pointer value is generated, and further, a composite concept semantic expression Tp is generated as the point destination semantic expression. The composite value T has a name obtained by combining the values acquired in 3) with “&” (other naming methods may be used as long as they can be identified). The combined concept semantic expression Tp is a semantic expression having the concept name of the same name and having “set component” as an attribute. If each value is the same (such as “house 1” and “house 1”), the generated value is the same as the original value. If the value is a pointer value, the point-to-point semantic representation of the original value is copied and used as the combined concept semantic representation Tp. When one is a fixed value and the other value is “NIL” or “arbitrary place”, the values are integrated into the fixed value. If the value is a pointer value, the point-to-point semantic representation of the original value is copied and used as the combined concept semantic representation Tp. 5) When the composite value T is a pointer value combined by “&”, each value acquired in the above 3) is stored in the value of the “set constituent component” attribute of the point destination semantic expression Tp. 6) The composite value T is stored in the value of the corresponding phenomenon attribute of P.

  Using such a phenomenon semantic expression integration rule, all Bj unreferenced subsets, Bk are collected and integrated into one phenomenon semantic expression P, and a pointer to “unreferenced subset BP” is stored. As shown in FIG. 25-2, A returns to the effect that “can be matched” to B and the correspondence table ([A: “premise i”] <lower correspondence table TI>) ( Step SK-14), the process ends.

  On the other hand, in step SK-3 of FIG. 25-1, when “referenced subset AP” or “unreferenced subset BP” is included (step SK-3, Yes), that is, in the input sentence so far When a part of the value meaning expression of the “set component” attribute of “premise i” has already been mentioned, as shown in FIG. 25-2, the “set component” attribute of “unreferenced subset BP” It is determined whether or not the value of NIL is NIL (step SK-9). If it is not NIL (No in step SK-9), the number of “aggregate component” attribute values is n, j = 1 to n, and the jth value of the “aggregate component” attribute is Bj. The predicate concept correspondence determination process is activated for the following semantic expressions and semantic expressions below Bj (step SK-10).

As a result, when A and Bj are “unmatchable”, “A and Bj are“ unmatchable ”” and the table 1) below, A and Bj are completely “associated”. If “Available”, “A and Bj are“ corresponding ”” and the following table 2), Bj is divided, and A and Bj mention subsets are “correspondable” When the Bj unreferenced subset is “cannot be associated”, the following 3) table is returned.
1) Correspondence table ([A: NIL] [NIL: B])
2) Correspondence table ([A: Bj] <lower correspondence table>) (j = 1 to n)
3) Correspondence table ([A: Bj mention subset] <lower correspondence table T>
[NIL: Bj unmentioned subset]) (j = 1 to n)

Next, the results for j = 1 to n are divided and integrated into corresponding portions and uncorresponding portions (step SK-11). Thereafter, it is determined whether or not all return results are “impossible to associate” (step SK-12). Here, in the case where A completely corresponds to Bf and the remaining “Bg,..., Bm” does not correspond, 1) and 2) below, and a part of A is “Bf,. Bg ”corresponds to“ Bh mention subset,..., Bk mention subset ”(Bh unreferenced subset,..., Bk unreferenced subset, Bl,. In the following 3) and 4), when A does not correspond to any Bj (all Bj remains unsupported), the following 5) is obtained.
1) Correspondence table ([A: Bf] <lower correspondence table Tf>) (TI)
2) Correspondence table ([NIL: Bg] ... [NIL: Bm]) (TII)
3) Correspondence table ([A: Bf] <lower correspondence table Tf>
...
[A: Bg] <lower correspondence table Tg>
[A: Bh mention subset] <lower correspondence table Th>
...
[A: Bk mention subset] <lower correspondence table Tk>) (TI)
4) Correspondence relationship table ([NIL: Bh unmentioned subset] ... [NIL: Bk unmentioned subset]
[NIL: Bl] ... [NIL: Bm]) (TII)
5) Correspondence table ([A: NIL] [NIL: Bf]... [NIL: Bm]) (TII)

  As a result of such a determination, if the return value does not include “correspondable” or the correspondence table TI (all return values are “unmatchable” or only the correspondence table TII) (step SK− 12, No), the process ends through the above step SK-15. On the other hand, when the return value includes “correspondable” or the correspondence table TI (step SK-12, Yes), [A: Bj] and [A: Bj mention subsets in the correspondence table ] Is added to the value of “set component” of “reference subset AP” (step SK-13). In accordance with this, the value of the “number of collective components” attribute is modified. Further, “set component” of “unreferenced subset BP” is cleared. Further, it is determined whether or not the Bj unreferenced subset Bk having the form [NIL: Bj unreferenced subset] and [NIL: Bk] belongs to the same or higher / lower category as A. Then, using the phenomenon semantic expression integration rule, all the Bj unreferenced subsets Bk that belong conceptually the same as A or belong to the higher and lower categories are collected and integrated into one phenomenon semantic expression P. In the returned [NIL: Bj], the value of the “collection component” attribute of “unknown subset BP”, and each Bj in which A and the conceptual category do not belong to the same or higher / lower category, and the above Add a pointer to P. In accordance with this, the value of the “number of collective components” attribute is modified.

  Thereafter, the fact that the semantic expression A is “correspondable” to the semantic expression B and a correspondence table ([A: “premise i”] <lower correspondence table TI>) are returned (step SK-14). ), The process ends.

(Method correspondence judgment processing)
Next, the method correspondence determination process will be described. This method correspondence determination process is mainly executed by the method correspondence determination processing unit 53k shown in FIG. 4, and is activated with semantic expressions below the semantic expression A and semantic expressions below the semantic expression B as arguments. “Matchable” or “impossible to be matched” and a correspondence table are returned. The entire flowchart of this process is shown in FIGS. 26-1 to 26-4. In this process, in the “method”, it is assumed that the same category is not bundled in the case of a phenomenon at different time points. In addition, it is assumed that phenomena having the same time point and the same subject are integrated in advance and knowledge is given to the storage unit 40. In the “action sequence set component”, pointers to the actions constituting the “method” are stored in the order of execution.

  In this process, as shown in FIG. 26A, the conceptual categories of the semantic expression A and the semantic expression B are compared. When the semantic expression A conceptually belongs to a lower category than B (step SL-1, Yes), the semantic expression A does not directly correspond to the semantic expression B. Focusing on the value of the “component” attribute, the process proceeds to step SL-5 described later. On the other hand, if the semantic representation A is conceptually equivalent to the semantic representation B or belongs to a higher category than the semantic representation B (step SL-1, No), the semantic representation A may directly correspond to the semantic representation B. First, in order to determine whether or not it is possible, a predicate concept correspondence determination process is activated for semantic expressions below semantic expression A and semantic expressions below semantic expression B (step SL-2).

  As a result, when it is returned that “correspondence is possible” (step SL-3, Yes), the lower correspondence table is set to NIL (step SL-4), meaning as shown in FIG. 26-3. The expression A is “can be associated” with the semantic expression B and returns a correspondence table (A: [“method i”] <lower correspondence table TI>) (step SL-23), and the processing Exit. On the other hand, in FIG. 26A, when “impossible to associate” is returned (step SL-3, No), the semantic expression A does not directly correspond to “method i”. Focusing on the value of the “method i” semantic expression B “action sequence set component” attribute, it is determined whether or not this value exists (step SL-5). If the value does not exist (step SL-5, No), as shown in FIG. 26-4, the semantic expression A is “impossible to associate” with the semantic expression B, and a correspondence table ([ A: NIL] [NIL: “method i”]) is returned (step SL-24), and the process ends.

  On the other hand, in FIG. 26A, when a value exists (step SL-5, Yes), it is further determined whether or not “referenced subset AP” or “unreferenced subset BP” is included (step S1). SL-6). If “reference subset AP” or “unreferenced subset BP” is included (step SL-6, Yes), that is, the “set component” of “method i” has already been entered in the input sentence so far. When a part of the value meaning expression of the attribute is mentioned, the process proceeds to Step SL-12 described later. On the other hand, when “reference subset AP” or “unreferenced subset BP” is not included (step SL-6, No), that is, when “method i” is mentioned for the first time, “action sequence set component” The number of attribute values is n, the semantic expression of the head of the semantic expression pointed to by each pointer is Bj (j = 1 to n), and the predicates for semantic expressions below A and below Bj The concept correspondence determination process is started (step SL-7). At this time, all Bj are processed in order from the top, and if one is associated, the process is terminated, and the remaining Bj is set to [NIL: Bj].

As a result, for each A and Bj, if A and Bj are “cannot be associated”, the following table 1), and if A and Bj are “can be associated”, “ A and Bj are “correspondable” and the following 2) table is returned. Here, in the “method”, it is not assumed that the semantic expression B is divided and the semantic expression A and the B mention subset are “correspondable”. Further, it is assumed that the “method” element does not correspond to the semantic expression A at all or only one of the “method” elements corresponds. Therefore, the correspondence table for partial correspondence ([A: Bj mention subset] [NIL: Bj unreferenced subset]) (j = 1 to n) is never returned.
1) Correspondence relationship table ([A: NIL] [NIL: Bj]) (j = 1 to n)
2) Correspondence table ([A: Bk] <lower correspondence table T>) (1 ≦ k ≦ n)

Next, the results for j = 1 to k (≦ n) are divided into a corresponding part and an uncorresponding part and integrated (step SL-8). Thereafter, it is determined whether or not all return results are “impossible to associate” (step SL-10). Here, when A does not correspond to any Bj, the following 1) table, when A completely corresponds to Bk, and before B1,... 2) and 3).
1) Correspondence relationship table ([A: NIL] [NIL: B1]... [NIL: Bn])
= Correspondence table ([A: NIL] [NIL: B]) (TII)
2) Correspondence table ([A: Bk]] <lower correspondence table T>...) (TI)
3) Correspondence relationship table ([NIL: B1]
...
[NIL: Bf]) (1, f ≠ k) (TII)

  As a result of such determination, if the return value does not include “correspondable” or the correspondence table TI (all the return values are “unmatchable” or only TII) (step SL-10, No) ) After the step SL-24, the process is terminated. On the other hand, if the return value includes “correspondable” or the correspondence relationship table TI (Yes in step SL-10), a “reference execution action subset AP” is generated (assuming that A is associated with Bk). (Step SL-11). Also, the set attributes “action sequence set component” and “number of action sequence set components” are given. In addition, pointers to all action sequence expressions from B1 to Bk are stored in the value of the “action sequence set component” attribute of the reference execution action subset AP while storing the original order. The number corresponding to this is stored in the value of the “number of constituent elements of the action sequence set” attribute. As the phenomenon attributes, the attributes that are commonly included in Bj from B1 to Bk are abstracted to the “execution action” level, that is, “execution subject”, “execution time”, “execution location”, “execution method”, and the like. Give. The attribute value common to B1 to Bk is set as the value of the corresponding phenomenon attribute of the mentioned subset AP. For non-common items, the value is NIL. However, the value of the “execution time” attribute stores the value of the “execution time” attribute of Bk. The “action sequence set component” of “method i” is cleared, and a pointer to “reference execution action sequence subset AP” is newly stored.

  Further, “unmentioned execution action subset BP” is generated (also in step SL-11). Also, the set attributes “action sequence set component” and “number of action sequence set components” are given. A pointer to Bk + 1,..., Bn is stored in the value of the “action sequence set component” attribute of the unmentioned subset BP while storing the original order. The number corresponding to this is stored in the value of the “number of constituent elements of the action sequence set” attribute. Among the attributes abstracted in the “execution action” level as attributes commonly included in Bk + 1,..., Bn, the attributes “execution subject”, “execution point”, “execution location”, “execution” appearing in each action in common Give way "and others. The attribute value common to Bk + 1,..., Bn is set as the value of the corresponding phenomenon attribute of the unmentioned subset BP. For non-common items, the value is NIL. The “execution time” refers to the “execution time tx” (non-pointer value) of the mentioned subset AP, and stores the expression “t> tx” of the time t based on this. This represents “t after time tx”. Further, a pointer to “unmentioned execution action subset BP” is added to “action sequence set component” of “method i”. Thereafter, as shown in FIG. 26-3, the processing is terminated through the above-described step SL-23.

  On the other hand, when it is determined in step SL-6 shown in FIG. 26A that “reference subset AP” or “unreferenced subset BP” is not included (step SL-6, No), FIG. As shown in FIG. 5, predicate concept correspondence determination processing is activated for the semantic expressions below the semantic expression A and the semantic expressions below the reference execution action subset AP (step SL-12). That is, the mention execution action subset AP and the unreferenced execution action subset BP are phenomenon concepts belonging to the concept category of the “execution action” level, and depending on the input sentence, the mention execution action subset AP and the unreferenced execution action part. Since it may be associated with the set BP, first, the correspondence between the semantic expression A and the reference execution action subset AP or the reference execution action subset BP is determined. As a result, when it is returned that “correspondence is possible” (step SL-13, Yes), the returned correspondence table is set as a lower correspondence table (step SL-21), and the above-described FIG. The process is terminated through step SL-23 in step 3.

  On the other hand, in FIG. 26-2, when the message “impossible to associate” is returned (step SL-13, No), this time, the semantic expression below the semantic expression A and the unexecuted execution action subset BP. The predicate concept correspondence determination process is activated for the semantic expression (step SL-14). As a result, when it is returned that “correspondence is possible” (step SL-15, Yes), the returned correspondence table is set as a lower correspondence table (step SL-22), and the above-described FIG. The process is terminated through step SL-23 in step 3.

  In addition, in step SL-15 of FIG. 26-2, when “impossible to associate” is returned (step SL-15, No), “action sequence set configuration of“ unreferenced execution action subset BP ” Focusing on the value of the “component” attribute, it is determined whether or not the value of the “action sequence set component” attribute of the “unmentioned execution behavior subset BP” is NIL (step SL-16). And when it is NIL (step SL-16, Yes), the said process is complete | finished through step SL-24. On the other hand, if it is not NIL (No in step SL-16), the number of values of the “action sequence set component” attribute is n, and the head of the semantic expression pointed to by each pointer is Bj (j = 1 to n) Then, the predicate concept correspondence determination process is activated for semantic expressions below A and below Bj (step SL-17). Here, the following is executed in order from j = 1 for all Bj. However, it ends when one correspondence is attached. [NIL: Bj] is set for the remaining Bj.

As a result of this processing, if A and Bj are “unmatchable” for each A and Bj, “A and Bj are“ unmatchable ”” and the following table 1) When A and Bj are “correspondable”, the table “2) below is returned that“ A and Bj are “uncorrespondable” ”. It is not assumed that B is divided and A and B mention subsets are “correspondable”. Assume that no “method” element corresponds to A or only one of the “method” elements. Therefore, the partial correspondence table ([A: Bj mention subset] [NIL: Bj unreferenced subset] <lower correspondence table>) (j = 1 to n) is never returned.
1) Correspondence relationship table ([A: NIL] [NIL: Bj]) (j = 1 to n)
2) Correspondence table ([A: Bj] <lower correspondence table T>) (j = 1 to n)

Next, the results for j = 1 to k (≦ n) are divided into corresponding parts and uncorresponding parts and integrated (step SL-18). Thereafter, it is determined whether or not all the return results are “unmatchable” (step SL-19). Here, when A does not correspond to any Bj, the following 1) table, when A completely corresponds to Bk, and when the remaining B1,. It becomes the table of 2) and 3).
1) Correspondence relationship table ([A: NIL] [NIL: B1]... [NIL: Bn])
= Correspondence table ([A: NIL] [NIL: B]) (TII)
2) Correspondence table ([A: Bk]] <lower correspondence table T>) (TI)
3) Correspondence relationship table ([NIL: Bd]
...
[NIL: Bf]) (TII)

  As a result of such a determination, if the return value does not include “correspondable” or the correspondence table TI (all return values are “impossible to associate” or only the correspondence table TII) (step SL− 19, No), the process is terminated through the step SL-24. On the other hand, if the return value includes “correspondence possible” or the correspondence relationship table TI (Yes in step SL-19), modify “reference execution action subset AP” (assuming that A corresponds to Bk). (Step SL-20). In addition, pointers to all action sequence expressions from B1 to Bk are added to the end of the “action sequence set component” attribute of the “reference execution action subset AP” while storing the original order. Correspondingly, the value of the “number of components of the action sequence set” attribute has been modified. As the phenomenon attributes, attributes that are commonly included in the value of the “action sequence set component” of the mentioned subset AP are abstracted to the “execution action” level, that is, “execution subject”, “execution point”, “execution location” Give "Execution method" and others. Here, an attribute value common to the value of the “action sequence set component” of the reference subset AP is set as the value of the corresponding phenomenon attribute of the reference subset AP. For those not common, the value is NIL.

  Next, “unmentioned execution behavior subset BP” is modified (also in step SL-20). The values of “action sequence set component” and “number of action sequence set components” of “unknown execution behavior subset BP” are cleared. A pointer to Bk + 1,..., Bn is stored in the value of the “action sequence set component” attribute of the unmentioned subset BP while storing the original order. Correspondingly, the value of the “number of components of the action sequence set” attribute has been modified. Among the attributes obtained by abstracting the phenomenon attributes commonly included in Bk + 1,..., Bn as “phenomenon actions” as the phenomenon attributes, the attributes “execution subject”, “execution point”, “execution place” that appear in common in each action Give "execution method" and others. Here, the attribute value common to Bk + 1,..., Bn is set as the value of the corresponding phenomenon attribute of the unmentioned subset BP. For those not common, the value is NIL. Further, the “execution time” attribute value tk of the value phenomenon located at the end of the value string of the “action sequence set component” attribute of the mentioned subset AP is stored in the “execution time” value.

  Thereafter, the process is terminated through the above-described step SL-23 in FIG.

(Attribute value pair correspondence determination processing)
Next, attribute value pair determination processing will be described. This attribute value pair correspondence determination process is mainly executed by the attribute value pair correspondence determination processing unit 53l shown in FIG. 4 and includes attribute value pairs (value on the semantic expression A side: value on the semantic expression B side. Is received with the attribute name included) and the “location” attribute pair match flag as arguments, indicating that “can be matched” or “cannot be matched”, the correspondence table, and the value propagation Return the possibility. FIG. 27 shows an overall flowchart of this process.

  Here, the attribute value includes a numerical value, a symbol, and a semantic expression. In this process, when the value is a numerical value or a symbol, the correspondence is determined by directly comparing them. If the value is a pointer to a semantic expression, an all-frame correspondence determination process is activated for the two semantic expressions to determine the correspondence. Since all the component modification components corresponding to attribute adjectives are transferred to the attribute values of the entity concept and the partial entity concept, even if there is a pointer to the “attribute noun”, it is not regarded as a branch to be determined. Since all the component modification components that limit the partial entity concept are transferred into the attribute expression of the partial entity concept, even if there is a pointer to the partial entity concept, it is not regarded as a branch to be determined. Therefore, the branch to be determined is a phenomenon concept (predicate, noun) or a pointer toward the entity concept.

  In this process, first, whether the value on the semantic expression A side and the value on the semantic expression B side are pointers (step SM-1), and are numerical values (step SM-6). , Whether or not they are symbols (step SM-10), whether one is NIL or between NILs (step SM-13), the value on the A side is “?”, And the value on the B side is It is determined whether it is not NIL (step SM-15).

  And when it is pointers to a semantic expression (step SM-1, Yes), the top semantic expression of each point destination semantic expression of the value of the semantic expression A side and the value of the semantic expression B side, Again, the semantic representation A and the semantic representation B are set (step SM-2), and the all-frame correspondence determination process is activated for the semantic representation A and the semantic representation B with the “location” attribute pair match flag as an argument ( Step SM-3). As a result, when it is returned that both are “correspondable” (step SM-4, Yes), the semantic expression A is “correspondable” to the semantic expression B, and all-frame correspondence determination processing The correspondence table returned from step S3 and that the value propagation possibility is “non-propagation” (step SM-5), and the process ends.

  Further, in step SM-6, when the values are numerical values (step SM-6, Yes), a numerical concept pair correspondence determination process is activated for the attribute value pair (step SM-7). As a result, when both indicate that they can be matched (step SM-8, Yes), the fact that the attribute value pair is “matchable” and the value propagation possibility are returned. (Step SM-9), the process is terminated.

  In step SM-10, when the symbols are symbols (step SM-10, Yes), “N” representing the number is regarded as a numerical value, and the symbol concept pair correspondence determination process is activated for the attribute value pair. (Step SM-11). If both are returned as “correspondable” (step SM-12, Yes), the process is terminated through step SM-9.

  In step SM-13, when one is NIL or between NILs (step SM-13, Yes), the fact that the attribute value pair is “can be associated” and the value propagation possibility are returned. (Step SM-14), and the process ends.

  In step SM-15, if the value on the semantic expression A side is “?” And the value on the semantic expression B side is not NIL (Yes in step S-15), the semantic expression A is associated with the semantic expression B. "Possible", the correspondence table (["?": B]) and the value propagation possibility are returned (step SM-16), and the process is terminated.

  In step SM-15, if the value on the semantic expression A side is not “?” Or the value on the semantic expression B side is not NIL (No in step S-15), the attribute value pair is “cannot be associated”. ”And a correspondence table ([A: NIL] [NIL: B]) and propagation possibility = no propagation are returned (step SM-17), and the process is terminated.

(Numeric concept versus correspondence determination process)
Next, the numerical concept pair correspondence determination process will be described. This numerical concept pair correspondence determination processing is mainly executed by the numerical concept pair correspondence determination processing unit 53m shown in FIG. 4, and includes attribute value pairs (values on the semantic expression A side: values on the semantic expression B side. Here Is received with the attribute name included) as an argument, and the fact that it is “matchable” or “cannot be matched” and the possibility of propagation are returned. FIG. 28 shows an overall flowchart of this process. In this process, even if the attribute value pair is the same numerical value, a special determination is required in the case of the value of the “collective component” attribute, so this determination is performed.

  In this process, first, the numerical values of the attribute value pairs are compared. If at least one of the value on the semantic expression A side and the value on the semantic expression B side is N (indefinite plural) (step SN-1, Yes), the process proceeds to step SN-7 described later. When at least one of the value on the semantic expression A side and the value on the semantic expression B side is not N (indefinite plural) (step SN-1, No), the value on the semantic expression A side and the value on the semantic expression B side Whether or not the two numerical values are equal (step SN-2), whether the value on the semantic representation A side> the value on the semantic representation B side (step SN-4), the value on the semantic representation A side <the value on the semantic representation B side Whether the value on the semantic representation A side is N and the semantic representation B side value is fixed (step SN-7), or the semantic representation A side value is fixed And whether the value on the semantic representation B side is N (step SN-9), whether the value on the semantic representation A side and the value on the semantic representation B side are both N (step SN-11), and Determine whether the attribute name of semantic expression A and the attribute name of semantic expression B are set components (step SN-13) That.

  When the two values of the value on the semantic representation A side and the value on the semantic representation B side are equal (step SN-2, Yes), the value on the semantic representation A side and the value on the semantic representation B side are both N. If there is a case (step SN-11, Yes), or if the attribute name of the semantic expression A and the attribute name of the semantic expression B are set components (step SN-13, Yes), "Attachable" and no propagation are returned (steps SN-3, SN-12, SN-14), and the process is terminated.

  Alternatively, if the value on the semantic expression A side> the value on the semantic expression B side (step SN-4, Yes), or if none of the above applies (step SN-13, No), "Not attachable" and no propagation are returned (steps SN-5 and SN-15), and the process is terminated.

  Alternatively, when the value on the semantic expression A side is N and the value on the semantic expression B side is fixed (step SN-7, Yes), it indicates that “correspondence is possible” and B → A propagation. Is returned (step SN-8), and the process is terminated.

  Alternatively, if the value on the semantic expression A side is fixed and the value on the semantic expression B side is N (step SN-9, Yes), it indicates that “correspondence is possible” and A → B propagation. Are returned (step SN-10), and the process is terminated.

(Symbol concept pair correspondence judgment processing)
Next, the symbol concept pair correspondence determination process will be described. This symbol concept pair correspondence determination process is mainly executed by the symbol concept pair correspondence determination processing unit 53n shown in FIG. 4, and includes attribute value pairs (value on the semantic expression A side: value on the semantic expression B side. Here Is received with the attribute name included) as an argument, and the fact that it is “matchable” or “cannot be matched” and the possibility of value propagation are returned. An overall flowchart of this process is shown in FIG.

  In this process, first, the values of attribute value pairs are compared. When the value of this attribute value pair is a correspondence determination between “t> tx” and “tj” (step SO-1, Yes), the fact that “correspondence is impossible” and no propagation are returned. (Step SO-2), and the process ends.

  Alternatively, when it is not the correspondence determination between “t> tx” and “tj” (step SO-1, No), when it is the correspondence determination between “t> tx” and “t> tk” (step SO−). 3, Yes), it is determined whether or not the value of the attribute value pair is the same type (including “>”) time expression (step SO-4), and if it is the same type (step SO-4, Yes) ), “Matchable” and non-propagation are returned (step SO-5), and the process ends. On the other hand, if it is not the same type (step SO-4, No), the fact that “correspondence is impossible” and no propagation are returned (step SO-6), and the process is terminated.

  Further, when it is not the correspondence determination between “t> tx” and “t> tk” (step SO-3, No), when it is the correspondence determination between “?” And NIL (step SO-7, Yes), If it is a correspondence determination between “?” And a specific value (step SO-9, Yes), or if the meanings of the two symbols are equivalent (step SO-10, Yes), “correspondence is possible”. The fact and no propagation are returned (step SO-8), and the process is terminated.

  If none of the above applies (step SO-10, No), it is determined whether A> B (step SO-11). If A> B (step SO-11, Yes), it returns “impossible to associate” and B → A propagation (step S-13), and the process is terminated. . On the other hand, if A> B is not satisfied (step SO-11, No), “correspondence is possible” and A → B propagation are returned (step SO-12), and the process is terminated.

  Next, specific examples of natural language analysis processing applied to input sentences will be described. Here, the input sentence shown in FIG. 30 “There are three PCs and three cables at home. So, trying to construct a LAN, I connected the PC to the HUB with a cable. What should I do?” An example will be described. In the following, as shown in the figure, in the input sentence, “There are three PCs and three cables in the house” is sentence 1 and section 1, and “HUB tries to construct a LAN and connects the PC with a cable. "Connected to the HUB" in sentence 2, "Let's try to build a LAN" in section 2, "Connect the PC to the HUB with a cable" in section 3, and "What should I do after that?" In sentences 3 and 4 Called.

  When the user inputs an input sentence via the input unit 10 (FIG. 5, step S-1, Yes), the input sentence preprocessing unit 51 captures this input sentence (step S-2) and performs input sentence preprocessing. Execute (Step S-3). In this input sentence preprocessing, the input sentence preprocessing unit 51 first divides the input sentence into words to generate a parse tree. Next, the input sentence preprocessing unit 51 stores the generated parse tree in the storage unit 40 in an arbitrary data format, and ends the processing.

  Next, the semantic expression generation processing unit 52 executes a semantic expression generation process for the parse tree stored in the storage unit 40 (step S-4). As a result, a plurality of semantic expressions corresponding to the sentences 1 to 3 are generated, the semantic expressions are stored in the storage unit 40, and the process ends.

  After that, the input sentence meaning interpretation processing unit 53 has the first sentence 1 (= section 1) “there are three PCs and three cables in the house among the semantic expressions of the clause 1 stored in the storage unit 40. The input sentence semantic interpretation process is activated with the semantic expression of "" as an argument (step S-5). The semantic expression of sentence 1 is shown in FIG.

  First, in the input sentence semantic interpretation process, “NIL” is set in order to initialize the redo flag from the beginning (FIG. 6, step SA-1), and the clause at the head of the sentence is the focused clause (step SA- 2). Here, sentence 1 “there are 3 PCs and 3 cables in the house” is a simple sentence, so “the section on the most head side” is the section 1 itself. Since the “Redo from top” flag is not set (step SA-3, No), and the correspondence table does not exist (step SA-5, No), sentence 1 “3 PCs at home” And the three cables are started ", the target clause semantic interpretation process is activated (step SA-7).

  In the target section semantic interpretation process, the target section “There are three PCs and three cables at home.” Is “Tell me about”, “I don't know”, “I want to know”, “Yes / Since it does not correspond to “NO question” or “WH question” (FIG. 8-1, step SB-1, No), it is determined that the problem setting cannot be completed with the focused section alone. It is determined whether or not “problem frame” is set in the “problem solving table” (FIG. 8-3, step SB-18). Here, since the “problem frame” has not yet been set, for the time being, the notice section “There are three PCs and three cables in the house” is written as the “problem frame” in the “problem solving table” (steps). SB-19). The problem solving table in this state is shown in FIG. Next, because the section of interest “There are 3 PCs and 3 cables at home” is not a “problem setting section” (not a question sentence) (FIG. 8-4, step SB-24, No). The process ends.

  Returning to the input sentence semantic interpretation process, since the “Redo from top” flag has not been set yet (FIG. 6, step SA-8, No), the section of interest “There are three PCs and three cables at home. It is determined whether or not a clause exists at the end of the sentence (step SA-9). Here, sentence 1 “There are 3 PCs and 3 cables in the house” is a simple sentence and there is only one section. Therefore, it is determined that there is no section at the end of the sentence, and the processing Exit. With the processing so far, the processing of sentence 1 (section 1) ends, and section 1 is temporarily set as a “problem frame”.

  Next, returning to the overall processing, the input sentence interpretation process is started with the semantic expression of the next sentence or section 2 (= section 2 and section 3) as an argument (FIG. 5, steps S-5, S-). 6). The semantic expression of Section 2 is shown in FIG. 33, and the semantic expression of Section 3 is shown in FIG. In this input sentence interpretation process, first, the redo flag from the beginning is initialized with “NIL” (FIG. 6, SA-1), and clause 2 “Let's try to build a LAN” which is the clause at the head of the sentence is the focus clause. (SA-2). Furthermore, since the “restart from the beginning” flag is not set (step SA-3, No), and the correspondence table does not exist (step SA-5, No), the focus section “Let's build a LAN” The target clause semantic interpretation process is activated for “Yes” (step SA-6).

  In the target clause semantic interpretation process, the target clause “Let's build a LAN” is “Tell me about”, “I do n’t know”, “I want to know”, “Yes / No question”, or “WH Since it does not correspond to any of “question” (FIG. 8-1, step SB-1, No), it is determined whether or not “problem frame” is set by referring to the “problem solution table” in the storage unit 40. (FIG. 8-3, Step SB-18). Here, since the clause 1 semantic expression is set in the “problem frame” of FIG. 32 (step SB-18, Yes), it is determined whether or not this “problem frame” is the focused clause semantic expression itself. Is determined (FIG. 8-2, step SB-14). Here, the clause 1 semantic expression that is the “problem frame” is different from the node 2 that is the target node (step SB-14, No), so whether or not the current target node is a node that is not the association target. Determine. Here, since it is not a section not subject to association (step SB-15, No), in order to determine whether the target section is “correspondable” on the “problem frame”, this target section is used as an argument. The clause meaning expression association process is started (step SB-16).

  In the clause semantic expression association processing, the “location” attribute pair matching flag is initialized, the semantic expression “Top” of the semantic expression X is set as the semantic expression A, and the semantic expression “Top” of the semantic expression Y is set as “existence 1”. The semantic expression is B (FIG. 11, step SE-1). Then, an all-frame correspondence determination process is started for the semantic expression below the semantic expression A “construction action 1” and the semantic expression below the semantic expression B “existence 1” (step SE-2).

  In the all-frame correspondence determination process, since the concept types of the semantic expression A “constructing action 1” and the semantic expression B “existence 1” are both “predicate concepts” (FIG. 14, step SG-1, Yes), the predicate The concept correspondence determination process is started (step SG-2).

  In the predicate concept correspondence determination process, the category of the semantic expression A “construction action 1” is “action” and the category of the semantic expression B “existence 1” is “existence”, and both are not in agreement or in a higher-order relationship ( In FIG. 15-1, step SH-1, No), the meaning expression A “constructing action 1” cannot be associated with the meaning expression B “existence 1”, and a correspondence table ([meaning expression A: NIL] [ NIL: semantic expression B]) is returned (step SH-13), and the process ends.

  Returning to the all-frame correspondence determination process, the fact that “correspondence is impossible” is returned from the predicate concept correspondence determination process (FIG. 14, step SG-3, No), and therefore semantic expression A corresponds to semantic expression B. "Not attachable" and a correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) are returned (step SG-5), and the process ends.

  Returning to the clause semantic expression association process, the fact that “correspondence is impossible” is returned from the all frame correspondence determination process (FIG. 11, step SE-3, No), so that the top of the attribute group of the semantic expression B The attribute “existence 1 subject” is set as the attribute T (step SE-4), and it is determined whether or not the attribute T “existence 1 subject” has a pointer value (step SE-5). Here, since the attribute T “existence 1 subject” has the pointer value “PC1 & cable 1” (step SE-5, Yes), the first value “PC1 & cable 1” of this attribute T “existence 1 subject” is the value of interest. (Step SE-9), the value meaning expression processing is activated for the target value “PC1 & cable 1” (Step SE-10).

  In the value semantic expression processing, the point destination semantic expression “PC1 & cable 1” of the target value “PC1 & cable 1” is changed to the semantic expression B (FIG. 12-1, step SF-1). Since this semantic expression B “PC1 & cable 1” is not “premise i”, “method i”, or a table (SF-2, SF-3, No), the meaning below the semantic expression A “constructing action 1” The all-frame correspondence determination process is activated for the expression and the semantic expression below the semantic expression B “PC1 & cable 1” (step SF-4).

  In the all-frame correspondence determination process, since the conceptual type of the semantic expression A “construction action 1” and the semantic expression B “PC1 & cable 1” is “others” (FIG. 14, step SG-10, No), the semantic expression A returns “not possible to associate” to semantic expression B, and the correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) is returned, and the process ends (step SG-5). ).

  Returning to the value semantic expression processing, the fact that semantic expression A and semantic expression B are “cannot be associated” is returned from the all-frame correspondence determination process (FIG. 12, Step SF-5, No), meaning expression The first attribute “set component” of the B attribute group is set as attribute T (step SF-6), and it is determined whether or not this attribute T “set component” has a pointer value (step SF-7). Here, since the attribute T “set component” has the pointer value “PC1” (step SF-7, Yes), this “PC1” is set as the target value (step SF-10), and the target value “PC1” is set. On the other hand, another value meaning expression process is started (step SF-11).

  In the value semantic expression processing, the point destination semantic expression “PC1” of the attribute T “collective component” is changed to the semantic expression B “PC1” (step SF-1), and the semantic expression B “PC1” is “premise i”. , “Method i” or because it is not a table (step SF-2, SF-3, No), meaning expression below semantic expression A “constructing action 1” and meaning expression below semantic expression B “PC1” On the other hand, the all-frame correspondence determination process is activated (step SF-4).

  In the all-frame correspondence determination process, since the conceptual type of the semantic expression A “constructing action 1” and the semantic expression B “PC1” is “others” (FIG. 14, step SG-10, No), the semantic expression A is a meaning. The fact that “cannot be associated” and the correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) is returned to the expression B (step SG-5), and the process ends.

  Returning to the value semantic expression process, the fact that “correspondence is impossible” is returned from the all-frame correspondence determination process (FIG. 12, Step SF-5, No), so the top of the attribute group of the semantic expression B “PC1” The attribute “set component” is focused on, and this attribute “set component” is set as attribute T (step SF-6). Thereafter, since this attribute T “set component” does not have a pointer value (step SF-7, No), it is determined whether or not the next attribute of the attribute group of the semantic expression B exists (step SF- 8). Here, since there is the next attribute “number of aggregate components” (step SF-8, Yes), this attribute “number of aggregate components” is again set as attribute T (step SF-9). Since this attribute T “the number of set components” does not have a pointer value (step SF-7, No), it is further determined whether or not the next attribute of the attribute group of the semantic expression B exists (step SF). -8). Here, since the next attribute does not exist (step SF-8, No), the semantic expression X is “impossible to associate” with the semantic expression below the semantic expression B, and a correspondence table ([semantic expression] A: NIL] [NIL: semantic expression B]) is returned (FIG. 12-7, step SF-40), and the process ends.

  Further, returning to the value semantic expression process, the fact that “correspondence is impossible” is returned from the previous value semantic expression process (FIG. 12-1, step SF-12, No), attribute T “set component” Next, the value “cable 1” is set as a target value again (steps SF-13 and SF-14), and the value meaning expression processing is started (step SF-11). The fact that it is “impossible to associate” is returned, and this process ends (No at step SF-12).

  Further, returning to the value meaning expression processing, the next value of the attribute T “set component” does not exist (step SF-13, No), but the attribute “number of set components” next to the attribute T “set component” is set. Since it exists (FIG. 12-2, step SF-15, Yes), this attribute “the number of set components” is changed to the attribute T (FIG. 12-1, step SF-9). Since this attribute T “the number of set components” does not have a pointer value (No in step SF-7), it is determined whether or not the next attribute in the attribute group of the semantic expression B exists (step SF−). 8). Here, since the next attribute does not exist (step SF-8, No), the semantic expression X is “impossible to associate” with the semantic expression below the semantic expression B, and a correspondence table ([semantic expression A : NIL] [NIL: semantic expression B]) is returned (FIG. 12-7, step SF-40), and the process ends. By the processing so far, it is interpreted that the semantic expression below the semantic expression X = “construction action 1” cannot be associated with the semantic expression below “PC & cable 1”.

  Returning to the clause meaning expression associating process, the fact that “cannot be associated” is returned from the previous process (FIG. 11, step SE-11, No), so the next value of the attribute T “existence 1 subject” Is selected (step SE-12), but since this next value does not exist (step SE-12, No), the attribute “existence 1 place” next to the attribute T “existence 1 subject” is changed to the attribute T. (Steps SE-13 and SE-7). Since the value “house 1” of the attribute T “existence 1 place” has a pointer value (step SE-5, Yes), the first value “house 1” of the attribute T is set as a target value (step SE−). 9) The value meaning expression process is activated for the target value “house 1” (step SE-10).

  Since the value semantic expression process returns “impossible to associate” as in the previous value semantic expression process (step SE-11, No), the process returns to the clause semantic expression association process, and the attribute T An attempt is made to select the next value of “Existence 1 location” (step SE-12). Here, since the next value does not exist (step SE-12, No) and the attribute next to the attribute T “existence 1 place” does not exist (step SE-13, No), the semantic expression X is a semantic expression. The fact that it is “impossible to associate” with the phenomenon meaning expression in Y and the correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) are returned (step SE-8), and the process is performed. finish. By the processing so far, it is interpreted that the semantic expression below the semantic expression X = “construction action 1” did not correspond to the “problem frame” = section 1 semantic expression.

  Returning to the section-interpretation processing of the target section, the fact that it is “cannot be associated” is returned from the previous section-semantic expression association processing (FIG. 8-2, step SB-16). In order to determine whether or not the current node can be associated with the current node of interest (step SB-17, No), the new “problem frame” is the node 2 semantic expression, and the new node of interest semantic expression is the node 1 semantic expression. Then, the clause meaning expression associating process is started (FIG. 8-4, step SB-20).

  In the clause semantic expression association process, as a result of this process, after initializing the “location” attribute pair match flag, the top semantic expression “existence 1” of the semantic expression X is set as the semantic expression A, and the top of the semantic expression Y is set. The semantic expression “constructing action 1” is defined as semantic expression B (FIG. 11, step SE-1). Then, an all-frame correspondence determination process is activated for the semantic expression below the semantic expression A “existence 1” and the semantic expression below the semantic expression “constructing action 1” semantic expression B (step SE-2).

  In this all-frame correspondence determination process, the concept types of the semantic expression A “existence 1” and the semantic expression B “constructing action 1” are both “predicate concepts” (FIG. 14, step SG-1, Yes). ), Predicate concept correspondence determination processing is started (step SG-2).

  In the predicate concept correspondence determination process, the concept categories of the semantic expression A “existence 1” and the semantic expression B “constructing action 1” do not match or do not have an upper-lower relationship (FIG. 15-1, step SH-1). The semantic expression A cannot be associated with the semantic expression B, and a correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) is returned (step SH-13), and the process is performed. finish.

  Returning to the all-frame correspondence determination process, the fact that “correspondence is impossible” is returned from the previous predicate concept correspondence determination process (FIG. 14, step SG-3, No), and semantic expression A is changed to semantic expression B. The fact that “correspondence is impossible” and the correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) are returned (step SG-5), and the process is terminated.

  Returning to the clause semantic expression association processing, the fact that “correspondence is impossible” is returned from the previous all frame correspondence determination processing (FIG. 11, step SE-3, No), meaning expression B “constructing action 1 Attention is focused on the first attribute “construction act 1 subject” of the attribute group, and this attribute “construction act 1 subject” is attribute T (step SE-4). Since this attribute T “construction action 1 subject” does not have a pointer value (step SE-5, No), the next attribute “construction action 1 target” of the attribute group of the semantic expression B “construction action 1” is selected. Again, attribute T is set (steps SE-6 and SE-7). Here, since this attribute T “construction action 1 target” has a pointer value “LAN1” (step SE-5, Yes), the leading value “LAN1” of this attribute T “construction action 1 target” is the focus value. (Step SE-9), the value meaning expression processing is activated for the target value “LAN1” (Step SE-10).

  In the value semantic expression processing, the semantic expression B at the top of the point destination semantic expression of the target value “LAN1” = “LAN1” is changed to the semantic expression B (FIG. 12-1, step SF-1). Since this semantic expression B is not “premise i”, “method i”, or a table (step SF-2, SF-3, No), the semantic expression and semantic expression B below semantic expression A “existence 1” The all-frame correspondence determination process is activated for the semantic expression below “LAN1” (step SF-4).

  In the all-frame correspondence determination process, since the concept type of the semantic expression A “existence 1” and the semantic expression B “LAN1” is “other” (FIG. 14, step SG-10, No), the semantic expression A “existence 1”. "Returns to the semantic expression B" LAN1 "that" cannot be associated "and the correspondence table ([semantic expression A: NIL] [NIL: semantic expression B]) (step SG-5) The process ends.

  Returning to the value semantic expression processing, since the fact that “correspondence is impossible” is returned from the all-frame correspondence determination processing (FIG. 12-1, step SF-5, No), the attribute group of the semantic expression B “LAN1” The attribute “set constituent component” at the head of the attribute is attribute T (step SF-6). Since attribute T “set component” does not have a pointer value (step SF-7, No), it is determined whether or not the next attribute of the attribute group of the semantic expression B exists (step SF-8). ). Thereafter, through the same processing, the attribute “LAN1 $ construction method” next to the attribute group of the semantic expression B is changed to the attribute T (step SF-9). Here, since the attribute T “LAN1 $ construction method” has a pointer value (step SF-7, Yes), the first value “method table 1” of this attribute T “LAN1 $ construction method” is used as a target value ( In step SF-10), a value meaning expression process is further activated for the target value “method table 1” (step SF-11).

  In the value semantic expression processing, the semantic expression “method table 1” at the top of the point destination semantic expression of the target value “method table 1” is changed to the semantic expression B (step SF-1). Since this semantic expression B “method table 1” is a table (SF-3, Yes), there is a correspondence table that represents the result of mapping the previous sentence or clause somewhere in the table to the table semantic expression. It is determined whether or not it has been added (FIG. 12-3, step SF-16). Here, since such a correspondence table is not added (step SF-16, No), paying attention to the first line (first line) of the semantic expression B “method table 1”, this line is Let us focus on the sub-attribute “premise” at the beginning of the attribute group in this row U, and let this sub-attribute be attribute T (step SF-17). Since the attribute T “premise” has a pointer value (step SF-18, Yes), the value “premise 1” of the attribute T “premise” is set as a target value (step SF-19). Further, a value meaning expression process is activated for “1” (step SF-20).

  In the value semantic expression processing, the top semantic expression “Premise 1” of the point-point semantic expression of the target value “Premise 1” is changed to the semantic expression B (FIG. 12-1, step SF-1). Since this semantic expression B “premise 1” is “premise i” (step SF-2, Yes), the semantic expression below semantic expression A “existence 1” and semantic expression below semantic expression B “premise 1”. In response to this, the all-frame correspondence determination process is started (FIG. 12-5, step SF-34).

  In the all-frame correspondence determination process, the semantic expression A “existence 1” is “predicate concept”, and the semantic expression B “premise 1” is “premise i” (FIG. 14, step SG-8, Yes). The determination process is started (step SG-9).

  In the premise correspondence determination process, the value “premise 1” of the “set component” attribute of the “premise i” semantic expression B exists (FIG. 25-1, step SK-1, Yes), but “reference subset AP”. Or “unknown subset BP” is not included (so far, “premise i” has never been mentioned) (step SK-3, No), the number of values of the “set component” attribute = n, j = 1 to n, j-th value of “set component” attribute = semantic expression Bj, and predicate concept determination corresponding processing is activated for semantic expression A and each semantic expression Bj (step SK−) 4).

  In the predicate concept correspondence determination process, since the concept categories of semantic expression A and semantic expression B are both “exist” (FIG. 15-1, step SH-1, Yes), semantic expression A and semantic expression B An attribute pair correspondence table (hereinafter referred to as attribute pair correspondence table (1) is created (step SH-2) for all the phenomenon attributes of FIG. 35. This attribute pair correspondence table (1) is illustrated in FIG. 35. Next, the predicate attribute The pair split & correspondence list is initialized (step SH-3. This predicate attribute pair split & correspondence list is shown in FIG. 36), and a predicate category tense & time correspondence determination rule is used (in the rule group having the same category). The rule that matches the distribution of values is selected and applied (this predicate category tense & time correspondence determination rule is shown in FIG. 37), the “tense” marker value added to the predicate concept, and the “time” The correspondence between the attribute values and the propagation possibility are determined (step SH-5), where the tense markers “existence 1” and “existence 2” are both “NIL”, that is, “non-temporal”. Are “correspondable” and “non-propagation” (step SH-5, Yes), and store “presence of association” and “non-propagation” in the predicate attribute pair split & correspondence list ( Step SH-6 This predicate attribute pair split & correspondence list is shown in Fig. 38.) Then, the correspondence relationship between the values of the "location" attribute is the semantic expression A) -iv), and the propagation possibility of the value is "A → B propagation ”(step SH-7, Yes), and this determination result (for example, semantic expression A) -iv)) and“ A → B propagation ”are predicate attribute pair split & correspondence list (Step SH-8. This predicate attribute pair split & correspondence list The door shown in FIG. 39). After that, the semantic expression A) -iv), which is the correspondence determination result of the “location” attribute value pair, is set as the “location” attribute pair match flag (step SH-9), and this is used as an argument in FIG. In the attribute pair correspondence table (1), the attribute value pair correspondence determination process is started for each phenomenon attribute pair such as “subject”, “target”, “means” other than “location”, “time”, and the correspondence between the value pairs The relationship and propagation possibility are determined (step SH-10).

  In the attribute value pair correspondence determination process, the value “PC1 & cable 1” on the semantic expression A side and the value “PC2 & cable 2 & HUB1” on the semantic expression B side are pointers to the semantic expression (FIG. 27, step SM-1, Yes), the semantic expression A side value “PC1 & cable 1”, and the semantic expression B side value “PC2 & cable 2 & HUB1”, the top semantic expressions “PC1 & cable 1” and “ “PC2 & cable 2 & HUB1” are changed to semantic expression A and semantic expression B (step SM-2). Then, for all the semantic expressions A “PC1 & cable 1” and the semantic expressions B “PC2 & cable 2 & HUB1”, the “location” attribute pair match flag is used as an argument to start the all-frame correspondence determination process (step SM-3). .

  In the all-frame correspondence determination process, since the conceptual types of the semantic expression A “PC1 & cable 1” and the semantic expression B “PC2 & cable 2 & HUB1” are both “noun concepts” (FIG. 14, step SG-6, Yes), The concept correspondence determination process is started (step SG-7).

  In the noun concept correspondence determination process, both the semantic expression A “PC1 & cable 1” and the semantic expression B “PC2 & cable 2 & HUB1” are parallel phrase composition concepts (FIG. 22-1, steps SI-1 to SI-3, No. ) Because there is a value in the semantic expression Ai that is the value of the “collective component” attribute of the semantic representation A, and there is a value in the semantic expression Bj that is the value of the “collective component” attribute of the semantic representation B (step SI-5, Yes), for all combinations of the semantic expression Ai and the semantic expression Bj, the non-synthetic concept correspondence determination process is activated to determine the correspondence between them (step SI-7).

  In the non-synthetic concept correspondence determination process, the concept categories of the semantic expression A “PC1 & cable 1” and the semantic expression B “PC2 & cable 2 & HUB1” coincide with each other in that they are parallel phrase composition concepts (FIG. 23-1, step SJ). −1, Yes), an attribute pair correspondence table (hereinafter, attribute pair correspondence table (2)) is created for all attributes of semantic expression A “PC1 & cable 1” and semantic expression B “PC2 & cable 2 & HUB1” (step SJ). -2.). This attribute pair correspondence table (2) is shown in FIG. Then, an attribute value pair correspondence determination process is activated for the first attribute value pair (set component attribute), and the correspondence between the value pair and the propagation possibility are determined (step SJ-3).

  In the attribute value pair correspondence determination processing, the value of the set component attribute of the semantic expression A “PC1 & cable 1” side and the semantic expression B side “PC2 & cable 2 & HUB1” are both NILs (FIG. 27, In step SM-13, Yes), the fact that the attribute value pair is “can be associated” and “no propagation” is returned (step SM-14), and the processing is terminated.

  Returning to the non-synthetic concept correspondence determination process, it is returned from the attribute value pair correspondence determination process that the attribute value pair is “matchable” and “no propagation” (FIG. 23-1, step SJ-4, Yes), the attribute value correspondence determination process is activated for the next attribute value pair “set component number 3”: “set component number N” (step SJ-3).

  In the attribute value correspondence determination process, the value “3” on the semantic expression A side and the value “N” on the semantic expression B side are numerical values (FIG. 27, step SM-6, Yes). On the other hand, a numerical concept pair correspondence determination process is started (step SM-7).

  In the numerical concept pair correspondence determination process, the value on the semantic expression A side is the fixed value “3” and the value on the semantic expression B side is “N” (FIG. 28, step SN-9, Yes). "Attachable" and "A → B propagation" are returned (step SN-10), and the process is terminated.

  Returning to the attribute value correspondence determination process, the fact that “correspondence is possible” is returned from the previous numerical concept pair correspondence determination process (FIG. 27, step SM-8, Yes). And “A → B propagation” are returned (step SM-9), and the process ends.

  Returning to the non-synthetic concept correspondence determination processing, the attribute value correspondence determination processing is started for the third attribute value pair “PC1 $ manufacturer NIL”: “PC2 $ manufacturer NIL”. Hereinafter, similarly, the non-composite concept correspondence determination process to the attribute value correspondence determination process are repeated (step SJ-3), and each time an indication that “association is possible” is returned for all attribute pairs.

Thereafter, in the non-synthetic concept correspondence determination process, all return values for each attribute pair are “corresponding” (when “corresponding impossible” or the correspondence table TII is not included). The association result and propagating possibility are stored in the non-synthetic concept attribute pair split & correspondence list (step SJ-5). Further, the relationship between the values of the “place” attribute in the semantic expression A “PC1” and the semantic expression B “PC2” (the value of the “location” attribute pair match flag, here the semantic expression A) -iv)) Expression A “PC1” and semantic expression B “PC2”
] Is determined (step SJ-6). In this example, since the semantic expression A) -iv) (“home”: “arbitrary place”), this corresponds to the case where the correspondence of the value of the “place” attribute is “match”. Therefore, the non-synthetic concept determination rule is applied to the attribute pair split & correspondence list, the correspondence as the whole concept is judged, and a correspondence table is created (step SJ-7). In this rule, a value is propagated for an attribute pair that is determined to be propagated. For example, for the set component number attribute, the propagation possibility is “A → B propagation”, and therefore the value “3” on the A side is propagated to the value “N” on the B side. As a result, “N = 3” is propagated to the full semantic expression. FIG. 41 shows the semantic expression B “PC2” of PC2 in a state where the number of set components N is rewritten to 3 by this. Then, the semantic expression A is “can be associated” with the semantic expression B, and the correspondence table (“PC1”: “PC2”] <NIL>) is returned, and the process ends.

Returning to the noun concept correspondence determination process, the non-synthetic concept correspondence determination process is similarly activated for semantic expressions A1 to A2: semantic expressions B2 to B3, and the following correspondence table is obtained.
1) Correspondence table (["PC1": "PC2"] <NIL>)
2) Correspondence table ([“PC1”: NIL],
3) [NIL: “Cable 2”])
4) Correspondence table ([“PC1”: NIL]
5) [NIL: “HUB1”]))
6) Correspondence table (["Cable 1": NIL]
7) [NIL: “PC2”])
8) Correspondence table (["Cable 1": "Cable 2"] <NIL>)
9) Correspondence table ([“Cable 1”: NIL]
10) [NIL: “HUB1”]))

Here, the above 2) and 4) are deleted because it is understood that “PC1” corresponds to “PC2” in the above 1). The above 7) is deleted because “PC2” corresponds to “PC1” in 1) above, and “cable 1” corresponds to “cable 2” in 6) and 9) above in 8). The above 3) is deleted. Since it is understood that “cable 2” corresponds to “cable 2” in 8) above, the deletion and the above 5) and 10) are left without being deleted. However, since these 5) and 10) are the same expression, they are combined into one. This is designated as 5). As a result, the following correspondence table remains.
1) Correspondence table (["PC1": "PC2"] <NIL>)
8) Correspondence table (["Cable 1": "Cable 2"] <NIL>)
5) Correspondence relationship table ([NIL: “HUB1”])

Next, the semantic representation B is divided into a portion associated with the semantic representation A and a portion that is not (separated on the semantic representation A side). Here, the above 1) and 8) are correspondence tables ([semantic expression Ah: semantic expression Bf] <lower correspondence table attribute Th>) (a table for a pair that can be “correspondable” without division). 5) corresponds to the table for the semantic expression B side component that cannot be associated with any of the components on the semantic expression A side, and the other tables were not returned. Separate and integrate the corresponding parts. As a result, the following correspondence table is obtained.
Correspondence table (["PC1": "PC2"] <NIL>
["Cable 1": "Cable 2"<NIL>) (T1I)
Correspondence table ([NIL: “HUB1”]) (T1II)

  “PC2 & Cable 2 & HUB1” is divided into “PC2 & Cable 2 & HUB1 mention subset” and “PC2 & Cable 2 & HUB1 unreferenced subset”. Here, the value of “component component” attribute in “PC2 & cable 2 & HUB1 reference subset” = “PC2”, “cable 2”, value of “number of component components” attribute = 2, in “PC2 & cable 2 & HUB1 unreferenced subset” The value of the “component” attribute = “HUB1” and the value of the “number of components” attribute = 1. FIG. 42 shows this “PC2 & cable 2 & HUB1 reference subset” and “PC2 & cable 2 & HUB1 unreferenced subset”.

Thereafter, the semantic expression A is “can be associated with” the semantic expression B, and the following correspondence table is returned (step SI-10), and the process is terminated.
Correspondence relationship table (["PC1 & cable 1": "PC2 & cable 2 & HUB1 reference subset"]) <attribute T1I>
[NIL: “PC2 & Cable 2 & HUB1 unmentioned subset”]) (T2I)

  Returning to the all-frame correspondence determination process, the fact that “correspondence is possible” is returned from the noun concept correspondence determination process (FIG. 14, step SG-3, Yes), so the semantic expression A “PC1 & cable 1” means The expression B “PC2 & cable 2 & HUB1” is “correspondable” and the returned correspondence table is returned (step SG-4), and the process is terminated.

  Returning to the attribute value pair correspondence determination process, the fact that “correspondence is possible” is returned from the all frame correspondence determination process (FIG. 27, step SM-4, Yes), so the semantic expression A “PC1 & cable 1” is The meaning expression B “PC2 & cable 2 & HUB1” is “corresponding” and the returned correspondence table is returned (step SM-5), and the process is terminated.

  Returning to the predicate concept correspondence determination process and returning from the attribute value pair correspondence determination process that "can be matched" and "no propagation", for other attribute pairs other than "location" and "time" Then, the attribute value pair correspondence determination process is activated (FIG. 15-1, SH-10).

  Thereafter, similarly, as a result of starting the attribute value pair correspondence determination processing for the values of all attribute pairs other than “location” and “time” of “existence 1” “existence 2”, here, for all attribute pairs The value is determined to be “correspondable”, and a message “correspondable” is returned (FIG. 27, steps SM-5 and SM-9). However, since all are “NIL”: “NIL”, the correspondence table is not returned. In addition, the possibility of value propagation is all “non-propagation”.

  Returning to the predicate concept correspondence determination processing, since the correspondence determination results for all attribute value pairs are returned from the attribute value pair correspondence determination processing, it can be determined that all the return values are “correspondable” (see FIG. 15-1, step SH-11, Yes), the correspondence determination result for each attribute value pair and the value propagation possibility are stored in the predicate attribute pair division & correspondence list (step SH-12). This predicate attribute pair split & correspondence list is shown in FIG. Next, by using the predicate category pair correspondence determination rule, the correspondence relationship between the predicate concepts is determined from the contents of the predicate attribute pair split & correspondence list, and the predicate semantic representation B is divided as necessary to obtain the semantic representation A and A correspondence and correspondence table is created (step SH-14).

Here, (predicate category-specific correspondence determination rule 1 (predicate category = “exist”)) is a predicate category-specific correspondence determination rule. For attribute values determined to be propagated, value propagation between A and B After that, the value of the “subject” attribute in the V mentioning subset is the meaning in the correspondence table ([semantic expression A: semantic expression B mention subset] [NIL: semantic expression B unreferenced subset]). The expression B mention subset is substituted, and the value of the “subject” attribute in the V unreferenced subset includes the above correspondence table ([semantic expression A: semantic expression B mention subset] [NIL: semantic expression B not yet]. Substituting an unreferenced subset of semantic expression B) in the referenced subset]), and the other attribute values have the same value in both the V-referenced subset and the V-unreferenced subset. In "match" When the taste expression B is replaced with the value of the semantic expression A at “any place”, the value replaced with both of the divided predicates is taken. It is divided into “subset” and “subject to which existence 2 is not mentioned.” “PC2 & cable 2 & HUB1 mention subset” is substituted for the value of the “subject” attribute of “existence 2 mention subset”. The value of “PC2 & cable 2 & HUB1 unreferenced subset” is assigned to the value of “subject” attribute of “subset.” The value of “location” attribute is the “existence 2 mention subset” or “existence 2 unreferenced subset”. Both take “House 1.” At this point, this value is propagated in the semantic representation of knowledge, and “any place” is replaced by “House 1” for attribute values that should take the same value. , As the predicate correspondence, the following correspondence Bull is obtained. Further, shown in FIG. 44 the results so far.
Correspondence relationship table (["Existence 1": "Existence 2 mention subset"] <T2I>
[NIL: “Existence 2 unmentioned subset”] (T3I)

  Next, the semantic expression A is “can be associated with” the semantic expression B, and the created correspondence table is returned (step SH-16), and the process is terminated.

  Returning to the premise correspondence determination processing, the j-th value of the “collective component” attribute is set as the semantic expression Bj, that is, the second value of the “collective component” attribute is set as the semantic expression B2 “existence 3” here. Then, predicate concept correspondence determination processing is activated for semantic expression A and each semantic expression Bj (FIG. 25-1, step SK-4).

  In the predicate concept correspondence determination process, since the concept categories of the semantic expression A “existence 1” and the semantic expression B “existence 3” are both “exist” (FIG. 15-1, step SH-1), the semantic expression A “ An attribute pair correspondence table (hereinafter, attribute pair correspondence table (3)) is created for all phenomenon attributes of “existence 1” and semantic expression B “existence 3” (step SH-2). This attribute pair correspondence table (3) is shown in FIG. Next, the predicate attribute pair split & correspondence list is initialized to the state shown in FIG. 36, and the value of the “temporal” marker added to the predicate concept and the “time” attribute using the predicate category-specific tense & time correspondence determination rule. The correspondence between the values of and the possibility of propagation are determined (step SH-5). Here, the tense markers of “existence 1” and “existence 3” are both “NIL”, that is, “no tense”. As the “existing” category tense & time correspondence determination rule, the same rule as shown in FIG. 37 is adopted. The fact that “correspondence is possible” and “no propagation” is stored in the predicate attribute pair split & correspondence list (step SH-6), and the correspondence relationship between the values of the “location” attribute corresponds to any of the following: (Step SH-7), and the determination result (for example, semantic expression A) -iv) / "A-> B propagation" is stored in the predicate attribute pair split & correspondence list (step SH-8). . Furthermore, the “location” attribute value pair correspondence determination result (semantic expression A) -i)) is set as a “location” attribute pair match flag (step SH-9), and this is used as an argument to create the “attribute pair” In the "correspondence table", the attribute value pair correspondence determination process for each phenomenon attribute pair other than "location" and "time", such as "subject", "target", and "means", can be started repeatedly to propagate the value pair correspondence and propagation The sex is determined (step SH-11).

  In the attribute value pair correspondence determination process, the value “PC1 & cable 1” on the semantic representation A side and the value “network card 1 & protocol 1” on the semantic representation B side are pointers to the semantic representation (FIG. 27, step SM-1, Yes), the meaning expression A side value “PC1 & cable 1”, and the meaning expression B side value “network card 1 & protocol 1”, the top semantic expressions of the point-to-point semantic expressions are changed to meanings. Let expression A and semantic expression B (step SM-2). Then, for all the semantic expressions A and B, the all-frame correspondence determination process is started using the “location” attribute pair match flag as an argument (step SM-3).

  In the all-frame correspondence determination process, the concept types of the semantic expression A “PC1 & cable 1” and the semantic expression B “network card 1 & protocol 1” are both “noun concepts” (FIG. 14, step SSG-6, Yes), the noun concept correspondence determination process is started (step SG-7).

  In the noun concept correspondence determination process, the semantic expression A “PC1 & cable 1” and the semantic expression B “network card 1 & protocol 1” are both parallel phrase composition concepts (FIG. 22-1, steps SI-1 to SI-3). ), Because there is a value of the “collective component” attribute of semantic expression A and semantic expression B (step SI-5, Yes), non-composite concept correspondence determination is performed for all combinations of semantic expression Ai and semantic expression Bj. The process is started (step SI-7).

  In the non-synthetic concept correspondence determination process, the semantic expression A “PC1” and the semantic category B “network card 1” do not match the concept category (FIG. 23-1, step SJ-1, No), so the semantic expression A “PC1”. The following semantic expressions and semantic expressions B “Network card 1” and the following semantic expressions are “impossible to be associated” and a correspondence table ([“PC1”: NIL] [NIL: “network card 1”]) Is returned (FIG. 23-3, step SJ-9), and the process ends.

  Returning to the noun concept correspondence determination process, similarly, the non-synthetic concept correspondence determination process is activated for other combinations of the semantic expressions Ai and Bj to determine the correspondence between the two (FIG. 22-1, step SI- 8) Here, all “unmatchable” are returned, and for the semantic expression A1: semantic expression B2, the correspondence table ([“PC1”: NIL] [NIL: “protocol 1”]), semantic expression A2 : Semantic expression B1 for the correspondence table (["Cable 1": NIL] [NIL: "Network card 1"]), Semantic expression A2: For the semantic representation B2, the correspondence table (["Cable 1 ": NIL] [NIL:" Protocol 1 "]) are sequentially returned. In this way, after finishing the determination of the correspondence relationship for all combinations of the semantic representation Ai and the semantic representation Bj, the division processing of the semantic representation B is performed as necessary from the combination of the obtained result division results. Create a correspondence table. Here, since all combinations of the semantic expression Ai and the semantic expression Bj are “unmatchable”, the semantic expression below the semantic expression A and the semantic expression below the semantic expression B are “unmatchable”. Then, the correspondence table (["PC1 & cable 1": NIL] [NIL: "network card 1 & protocol 1"]) is returned (FIG. 22-4, step SI-30), and the process is terminated.

  Returning to the all-frame correspondence determination process, since the fact that “no association” is returned from the noun concept correspondence determination process, the meaning expression A is “not compatible” with the semantic expression B, and the correspondence table (["PC1 & cable 1": NIL] [NIL: "network card 1 & protocol 1"]) is returned (FIG. 14, step SG-3, No), and the process ends.

  Returning to the attribute value pair correspondence determination process, the fact that “correspondence is impossible” is returned from the all frame correspondence determination process (FIG. 27, step SM-4, No), and therefore the attribute value pair is “impossible to associate”. In addition, it returns “no propagation” and a correspondence table ([“PC1 & cable 1”: NIL] [NIL: “network card 1 & protocol 1”]) (step SM-17), and ends the processing. To do.

  Returning to the predicate concept correspondence determination processing, the attribute value pair correspondence determination processing is activated for other attribute pairs other than “place” and “time” (step SH-10 in FIG. 15A).

  In the attribute value pair correspondence determination process, similarly, the values of all the attribute pairs are determined to be “correspondable” (FIG. 27, step SM-4, Yes), respectively, and are respectively stored in the correspondence relationship table or “correspondable”. A message to that effect is returned (step SM-5), and the process is terminated.

  Returning to the predicate concept correspondence determination processing, it is determined from the attribute value pair correspondence determination processing whether or not the values of all the attribute pairs are “matchable”, but here the “subject” attribute pair is “unmatchable” Since the return value includes “impossible to associate” (FIG. 15-1, step SH-11, No), the fact that semantic expression A cannot be associated with semantic expression B and the correspondence table ([ “Existence 1”: NIL] [NIL: “Existence 3”]) is returned (step SH-13), and the process is terminated.

Returning to the premise correspondence determination processing, since there is no value in the “set component” attribute of “premise 1” any more, the results for j = 1 to n are converted into corresponding portions and uncorresponding portions. Dividing and integrating (FIG. 25-1, step SK-5), the following correspondence table is obtained.
Correspondence relationship table (["existence 1": "existence 2 mention subset"] <T2I>)
[NIL: “Existence 2 unmentioned subset”] (T4I)
Correspondence relationship table ([NIL: “existence 3”]) (T4II)

  Since the return value includes “can be associated” (step SK-6, Yes), the reference subset semantic expression AP is generated (step SK-7). Then, the “set component” attribute of “premise 1” is cleared, and a pointer to the “reference subset AP” is newly stored. This semantic expression “Premise 1” and the reference subset semantic expression AP are shown in FIG. Further, an unreferenced subset semantic expression BP is generated (step SK-8). Here, “existence 2 unsubscribed subset”] and “existence 3” belong to the same concept category, so the phenomenon semantic expression integration rule is used, and the concept category belongs to the same or higher / lower category as the semantic expression A. All the semantic expressions Bj not yet mentioned and the semantic expressions Bk are integrated into one phenomenon semantic expression P.

Specifically, (1) referring to a dictionary, a phenomenon concept meaning expression P corresponding to the category is generated. At this time, the instance numbers are adjusted so as not to overlap. If markers are assigned to each predicate concept before integration, it is assigned to the phenomenon concept semantic expression P. This phenomenon conceptual meaning expression P is shown in FIG.

  Also, (2) a pair (set) list is created for each corresponding phenomenon attribute (the “set component” attribute does not exist in the phenomenon concept) of each semantic expression Bj unmentioned subset and semantic expression Bk. As a result, the attribute pair correspondence list (1) of FIG. 48 is obtained.

  Further, (3) all attribute pairs (groups) are selected in order from the top of the list of pairs (groups), and the respective values in each semantic expression Bj unreferenced subset and semantic expression Bk are acquired. Here, “protocol 1”, “cable 2 & HUB 1” are acquired as the “subject” value, “house 1” is acquired as the “location” value, and “NIL” is acquired as the “time” value. This attribute pair is shown in FIG.

  Further, a composite value T in which the acquired values are bundled is generated. When each value is a pointer value, a composite value T of the pointer value is generated, and further, a composite concept semantic expression Tp is generated as the point destination semantic expression. The composite value T has a name obtained by combining the values acquired in (3) with “&” (other naming methods may be used as long as they can be identified). The composite concept semantic expression Tp is a semantic expression having the concept name of the same name and having “component” as an attribute. If each value is the same (such as “house 1” and “house 1”), the generated value is the same as the original value. If the value is a pointer value, the point-to-point semantic representation of the original value is copied and used as the combined concept semantic representation Tp. As a result, the composite value T = “house 1 & house 1” = “house 1” and the composite value T = “NIL & NIL” = “NIL” are obtained. Further, when one is a fixed value and the other value is “NIL” or “arbitrary place”, the values are integrated into the fixed value. If the value is a pointer value, the point-to-point semantic representation of the original value is copied and used as the combined concept semantic representation Tp. As a result, the composite value T = “NIL & house 1” = “house 1” and the composite value T = “house 1 & house 1” = “house 1” are obtained. This synthetic concept semantic expression Tp is shown in FIG.

  Further, (5) when the attribute T is a pointer value combined with “&”, each value acquired in (3) above is stored in the value of the “set constituent component” attribute of the point destination semantic expression attribute Tp. To do. Here, the value of the “aggregate component” attribute = “protocol 1”, “cable 2 & HUB 1”. As a result, this synthetic concept semantic expression Tp is shown in FIG.

  Further, (6) the attribute T is stored in the value of the corresponding phenomenon attribute of the phenomenon concept semantic expression P. Furthermore, a pointer to “unknown subset semantic expression BP” is stored in the “set component” attribute of “premise i”. As a result, the phenomenon concept semantic expression P is as shown in FIG. In addition, a pointer to “unreferenced subset BP” is stored in the “set component” attribute of “premise i”. This semantic expression “premise i” and the like are shown in FIG. Then, the semantic expression A is “can be associated” with the semantic expression B, and a correspondence table ([“existence 1”: “premise 1”] <attribute T4I>) (attribute T5I) is returned (see FIG. 25-2, Step SK-14), and the process ends.

  Returning to the all-frame correspondence determination process, the fact that the association is possible is returned from the premise correspondence determination process (step SG-3, Yes) in FIG. "Responsible" and the returned correspondence table (["Existence 1": "Premise 1"] <Attribute T4I>) (Attribute T5I) is returned (Step SG-4), and the process ends. .

  Returning to the value semantic expression process, the fact that “correspondence is possible” is returned from the all-frame correspondence determination process (FIG. 12-5, step SF-35, Yes). "Responsible", and the returned correspondence table (["Existence 1": "Premise 1"] <attribute T4I>) (attribute T5I) is returned (step SF-36) finish. Also,

  Further, returning to the value semantic expression process, the fact that “correspondence is possible” has been returned from the previous value semantic expression process (FIG. 12-3, step SF-21, Yes), the returned correspondence table is It is added to the semantic expression B (table) (FIG. 12-6, step SF-38). In addition, the semantic expression A is “can be associated with” the semantic expression B and returns a correspondence table ([“existence 1”: “method table 1”] <attribute T5I>) (attribute T6I) (step T6I). SF-39), the process is terminated.

  Further, returning to the value semantic representation processing, since the fact that “correspondence is possible” is returned from the previous value semantic representation processing (FIG. 12-1, step SF-12, Yes), the semantic representation A is the semantic representation B. And a correspondence table (["existence 1": "method table 1"] <attribute T5I>) (attribute T6I) (attribute T7I) is returned (FIG. 12-6). Step SF-39), the process is terminated.

  Returning to the clause semantic expression association processing, the fact that “correspondable” is returned from the value semantic expression processing (FIG. 11, step SE-11, Yes), the semantic expression X corresponds to the semantic expression Y. "Attachable" and a correspondence table (["Existence 1": "Construction action 1"] <attribute T7I>) are returned (step SE-15), and the process ends.

  In the “problem solving table”, since it is returned to the section-of-interest semantic interpretation process and the fact that “association is possible” is returned from the section semantic expression association process (FIG. 8-4, step SB-21, Yes) The designation of the “problem frame” up to that point is cleared, the current section is stored, the “restart from the beginning” flag is set (step SB-22), and the process is terminated.

  Then, returning to the input sentence semantic interpretation process, since the “restart from the beginning” flag is set (FIG. 6, step SA-8, Yes), the first sentence clause 1 is set as the target clause (step SA-2). Then, the “restart from the beginning” flag is reset (step SA-4), the correspondence table is cleared (step SA-6), and then the target clause semantic interpretation process is activated for the target clause semantic expression.

  In the focused clause semantic interpretation process, the focused clause does not correspond to “tell me”, “I don't know”, “I want to know”, “YES / NO question”, or “WH question” (see FIG. 8-1, Step SB-1, No), referring to the “problem solving table” to determine whether “problem frame” is set (FIG. 8-3, step SB-18). Here, the clause 2 semantic expression is set as the “problem frame” as shown in FIG. 54 (step SB-18, Yes), but the “problem frame” is not the focused clause semantic expression itself (FIG. 8-2, In step SB-14, No), since the current clause is not a non-association clause (step SB-15, No), the clause meaning expression association processing is started (step SB-16).

  As described above, the clause meaning expression associating process returns “correspondable” and a correspondence table ([“existence 1”: “premise 1”] <...>) (T9I). Therefore, returning to the focused section semantic interpretation process, the focused section is not the “problem setting section” (FIG. 8-2, step SB-17, FIG. 8-4, SB-24), so that the problem solving process is not started. Then, the process ends.

  Then, the process returns to the input sentence semantic interpretation processing, and the processing is continued with the node 2 at the end of the sentence of the target section as a new target section (FIG. 6, SA-9, SA-10).

  Hereinafter, other inter-node relationships are associated in the same manner, but the detailed processing description and drawing numbers are omitted, and only the outline is described. With the processing so far, the processing of Section 2 “Trying to build a LAN” is completed, and Section 1 “There are 3 PCs and 2 cables at home.” It was interpreted to be mapped above. Thereafter, in order to associate the clause 1 with the clause 2, the clause interpretation processing is executed with the clause 2 as the “problem frame” and the clause 1 as the focus clause. Further, after this processing is completed, in order to associate clause 2 with clause 2, clause 2 is “problem frame” and clause 1 is the focused clause. In the final section-interpreting semantic interpretation process, the section ends because the section of interest is not the “problem setting section”.

  Next, the process returns to the input sentence semantic interpretation process, and the process of associating clause 3 with clause 2 is performed. With the section 2 as the “problem frame” and the section 3 as the target section, the target section semantic interpretation processing is executed. From this target semantic interpretation process, the clause semantic expression matching process is started, the clause 3 semantic representation is the target clause semantic representation X, the clause 2 semantic representation is “problem frame” semantic representation Y, and the semantic representation “connecting action 2”. The all-frame correspondence determination process is started with the expression A and the meaning expression “constructing action 1” as the meaning expression B.

  And since the semantic expression A “connection action 2” and the semantic expression B “construction action 1” are both predicate concepts, the predicate concept correspondence determination process is activated, The fact that “cannot be associated” is returned, and the process returns to the clause meaning expression associating process through the all-frame corresponding determining process. Here, among the attributes of the semantic expression B “construction action 1”, the attribute “construction action 1 subject” having a pointer value first is used as a target value, and value semantic expression processing is started. In this process, the top semantic expression “LAN1” of the point-point semantic expression of the target value “LAN1” of the attribute T “construction action 1 target” is changed to the semantic expression B, and the semantic expression below the semantic expression A “connection act 2”. And the all-frame correspondence determination process for the semantic expressions below the semantic expression B “LAN1”, but the conceptual types of the semantic expression A “connection act 2” and the semantic expression B “LAN1” are other. Therefore, the fact that it is “cannot be associated” is returned, and the process returns to the value meaning expression processing.

  Hereinafter, the same processing is sequentially performed with each attribute of the semantic expression B “LAN1” as the attribute T. When the attribute “LAN1 $ construction method” is set as the attribute T, this attribute T “LAN1 $ construction method” is a pointer value. Therefore, the value meaning expression process is started with this value “method table 1” as a target value. Here, the top semantic expression “method table 1” of the target value “method table 1” is referred to as semantic expression B. This semantic expression B “method table 1” is a table, and this table includes a correspondence table. Is added, and the first line where the semantic expression of the corresponding destination exists below the table is defined as line U. The leading sub-attribute “premise” of the attribute group of the row U is set as an attribute T. Since this attribute T “premise” has a pointer value, the value meaning expression process is started with this value as a target value.

  In the value semantic expression process, the top semantic expression “Premise 1” of the point-point semantic expression of the target value “Premise 1” is changed to the semantic expression B, and this semantic expression B is “Premise i”. All-frame correspondence determination processing is activated for “connection act 2” and semantic expression B “premise 1”. In this process, since the concept type of the semantic expression A “connection act 2” is “predicate” and the concept type of the semantic expression B “premise 1” is “premise i”, the premise correspondence determination process is started. The value of the “set component” attribute of the “premise 1” semantic expression B includes “reference subset semantic expression AP” and “unreferenced subset semantic expression BP”, and “unreferenced subset meaning” Since the value “existence 4” of the “collection component” attribute of the expression BP is other than that (the value exists), the number “1” of the value of the “collection component” attribute is n, j = 1 to n Then, the j-th value of the “set component” attribute is set as the semantic expression Bj, and the predicate concept correspondence determination process is started for the semantic expression below the semantic expression A and the semantic expression below the semantic expression Bj. In this process, the semantic expression A “connection act 2” and the semantic expression B “existence 4” are processed, but it returns “impossible to associate” and returns to the premise correspondence determination process. Also in this process, since all the results of j = 1 to n cannot be wiped, the fact is sequentially returned to the all-frame correspondence determination process, the value meaning expression process, and the value meaning expression process.

  In the value semantic expression processing, the next sub-attribute “method 1” in the row U of the semantic expression B “method table 1” is changed to the attribute T, and the value “method 1” of the attribute T is set as the target value. Start up. In this process, the top semantic expression “method 1” of the point-point semantic expression of the target value “method 1” is changed to the semantic expression B, and the semantic expression B “method 1” is “method i”. For the expression A “connection act 2” and the semantic expression B “method 1”, the all-frame correspondence determination process is activated. In this process, since the concept type of the semantic expression A “connection act 2” is “predicate” and the concept type of the semantic expression B “method 1” is “method i”, the method correspondence determination process is started. In this process, the semantic expression B “method 1” is a phenomenon concept belonging to the concept category of the “execution action” level, and the semantic expression A “connection action 2” is conceptually subordinate to the semantic expression B “method 1”. Belonging to. Since the value of the attribute “action sequence set component” of the semantic expression B “method 1” is “others (when it has a value other than the semantic representation AP and the semantic representation BP)”, the “action sequence set component” The number of attribute values “3” is n, the semantic representation of the head of the semantic representation pointed to by each pointer is the semantic representation Bj (j = 1 to n), and for all semantic representations Bj, j = In order from 1, the predicate concept correspondence determination process is executed for the semantic expressions below the semantic expression A and the semantic expressions below the semantic expression Bj.

  As a result of the predicate concept correspondence determination process, the meaning expression A = “connection action 2” and the meaning expression B1 = “attachment action 1” are “not compatible” and a correspondence table ([“connection action 2 ”: NIL] ([NIL:“ attachment action 1 ”]), meaning expression A =“ connection action 2 ”, meaning expression B2 =“ connection action 1 ”, and“ The attribute pair correspondence table (4) is obtained 55. At this time, the value of the place attribute of “connection act 1” (knowledge side) is “house 1” due to the propagation of the value.

  Next, the predicate attribute pair split & correspondence list is initialized to FIG. 56, and a predicate category tense & time correspondence determination rule is used (value distribution matches among the rule groups having the same category). And the correspondence between the value of the “tense” marker and the value of the “time” attribute added to the predicate concept and the possibility of propagation are determined. As a result, the fact that “can be associated” and “no propagation” is stored in the predicate attribute pair split & correspondence list. Since the correspondence relationship between the values of the attribute “location” corresponds to the semantic expression A) -v) and the propagation possibility corresponds to “B → A propagation”, the determination result is stored in the predicate attribute pair split & correspondence list. To do. Then, the correspondence determination result of the “location” attribute value pair is set as a “location” attribute pair match flag, and this is used as an argument in the “attribute pair correspondence table” created above, except for “location” and “time”. By repeatedly starting the attribute pair correspondence determination process for each phenomenon attribute pair such as “subject”, “target”, and “means”, the correspondence between the value pair and the propagation possibility are determined. As a result, the attribute “subject” is “correspondable” and “non-propagation”, and the attribute “target” is “correspondable” and “non-propagation”. The table ([“PC5”: “PC2”] <NIL>) (attribute T10I) and the attribute “destination” are “matchable” and “non-propagation” and the correspondence table ([“HUB5 ":" HUB1 "] <NIL>) (attribute T11I), the attribute" means "is" matchable "and" non-propagating ", and a correspondence table ([" cable 5 ":" cable 2 ”] <NIL>) (attribute T12I) The fact that“ can be matched ”and“ no propagation ”is returned for the attribute“ purpose ”. As a result, since all return values are “correspondable”, the correspondence determination result and propagation possibility for each attribute value pair are stored in the predicate attribute pair division & correspondence list. FIG. 58 shows the predicate attribute pair split & correspondence list. Next, the correspondence relationship between predicate concepts is determined from the contents of this list using the predicate category-specific pairing determination rule of FIG. 59, and the predicate semantic representation B is divided and associated with the semantic representation A as necessary. Create a correspondence table. At this time, an attribute value that can be propagated is propagated (for example, connection act 2 place = house 1). The semantic expression A is “can be associated with” the semantic expression B, and the created correspondence table ([“connection action 2”: “connection action 1”] <attribute T10I> <attribute T11I> <attribute T12I>) (attribute T13I) is returned, and the process ends.

  Next, returning to the method correspondence determination process, since “connection action 2” has already been associated with one action sequence element, there is no need to search for a correspondence destination any more, so j = 1 to k (≦ n). The results of are divided into corresponding parts and non-corresponding parts and integrated. Here, “reference execution action subset AP” is generated, “action sequence component” of “method 1” is cleared, and a pointer to “reference execution action sequence subset AP” is newly stored. Here, since the return value includes “can be matched” or the attribute TI, the method is divided, and an unmentioned execution action subset semantic expression BP is generated. The result is shown in FIG. Then, the fact that “correspondence is possible” and the correspondence table ([“connection act 2”: “method 1”] <attribute T14I>) (attribute T15I) are returned, and the process ends. After that, the correspondence table ([“connection act 2”: “method 1”] <attribute T14I>) (attribute T15I) is set to “all frame correspondence determination process-value semantic expression process-value”. Return sequentially to the semantic expression process. In this process, the returned correspondence table is added to the semantic expression B (table) to indicate that “correspondence is possible” and that the correspondence table ([“connection act 2”: “method table 1”]] <attribute T15I>) (attribute T16I) is returned, and the process ends. Further, from the value meaning expression process to the clause meaning expression associating process, “correspondence is possible” and a correspondence table ([“connection act 2”: “LAN1”] <attribute T16I>) (attribute T17I) is returned. From the clause meaning expression matching process to the focused clause meaning interpretation process, the fact that it is “matchable” and the correspondence table ([“connection action 2”: “construction action 1”] <attribute T17I>) (attribute T18I) is returned. In this process, because the target section is not the “problem setting section”, the process ends. In the input sentence semantic interpretation process, the “Redo from beginning” flag is not set, and there is a section at the end of the target section. The processing is terminated because it does not. By the processing so far, clause 3 is associated with clause 2, and the method in the clause 2 semantic expression is divided.

  Thereafter, the input sentence interpretation process is activated for the semantic expression of sentence 3 “What should I do after that?”. Here, “Bayo” is an auxiliary verb equivalent phrase, and its meaning is represented by a “recommended” marker. Also, “how” (adverb) represents the meaning of “how” and asks “modify” that it modifies, that is, “content” (element of partial phenomenon) of “executive action”. In the concept dictionary of “how”, as the meaning of “how”, a semantic expression and “executive action sequence set component?” Are stored. In the “semantic expression generation process” that precedes this process, the semantic expression of the adverb “how” is positioned as the corresponding attribute expression of the “executive action” semantic expression that means “do”.

  As a result, “?” Is assigned to the value of the “execution action sequence set component” attribute of the “execution action” semantic expression. “Subsequently” means “at a time point after the time when it occurred”. In addition, “it” is the subject of “occurs”, and thus represents the meaning of “the phenomenon (group) mentioned (immediately before)”. Here, for convenience, the meaning of “after” is expressed as a semantic expression, “time attribute T> attribute T semantic expression X”. “Attribute T semantic expression X” is the point in time when the phenomenon (group) mentioned (immediately before) occurs. In the preceding “semantic expression generation process”, the semantic expression of “after” is positioned as the corresponding attribute expression of the “executive action” semantic expression that means “do”. As a result, “attribute T> attribute T semantic expression X” is assigned to the value of the attribute “execution action time point” of the “execution action” semantic expression.

  First, in the input sentence semantic interpretation process, the most important clause 4 of sentence 3 is the focused clause, the clause 4 semantic representation is the focused clause semantic representation, and the clause 2 semantic representation is the “problem frame” semantic representation. Start up. Then, since the target clause semantic expression is “WH question”, the problem setting clause and the problem target attribute identification process are activated. In this process, section 4 as the target section is defined as “problem setting section”, “problem attribute” is extracted from this “problem setting section” semantic expression, and both “problem setting section” are written to the problem solving table. Return the problem solving table and end the process. This problem solving table is shown in FIG. In the target clause semantic interpretation process, because the predicate of the target clause is not "Conclusive", the (partial) entity concept semantic expression including the attribute that has a semantic correspondence with "problem versus target attribute" is acquired as the "problem frame" To do. The acquisition of the (partial) entity concept semantic expression is automatically performed based on the contents of the prior patent application (here, the fact that the (partial) entity concept semantic expression cannot be acquired is returned). Thereafter, in order to check whether or not the node of interest is “can be associated” on the “problem frame”, the clause meaning expression association processing is started.

  In the clause semantic expression association process, the top semantic expression “execution action 2” of the semantic expression X is the semantic expression A, and the top semantic expression “constructing action 1” of the semantic expression Y is the semantic expression B. Start up. In this process, since the concept types of the semantic expression A “execution action 2” and the semantic expression B “construction action 1” are both “predicate concepts”, the predicate concept correspondence determination process is started. In this process, since the concept category “execution action” of the semantic expression A “execution action 2” is a superordinate concept of the concept category “construction action” of the semantic expression A “execution action 2”, the semantic expression A “execution action 2”. ”And an attribute pair correspondence table (hereinafter, attribute pair correspondence table (5)) for all phenomenon attributes of the semantic expression B“ construction action 1 ”. This attribute pair correspondence table (5) is shown in FIG. 62 (“construction action 1” (knowledge side has place = “house 1” by value propagation)). Next, the predicate attribute pair split & correspondence list is initialized to FIG. 63, and the value of the “temporal” marker and the “time” added to the predicate concept using the predicate category tense & time correspondence determination rule of FIG. Judge the correspondence between attribute values and propagating possibilities. Here, since “correspondence is impossible”, the fact that the association is impossible and the correspondence relationship table ([execution action 2: NIL] [NIL: construction action 1]) are returned, and the process is terminated. Returning to the all-frame correspondence determination process, the fact that the association is impossible and the correspondence table ([execution action 2: NIL] [NIL: construction action 1]) are returned, and the process ends.

  Returning to the clause semantic expression association processing, the first attribute “construction action 1 subject” of the phenomenon attribute group of the semantic expression B “construction action 1” is attribute T, and this attribute T “construction action 1 subject” has a pointer value. Since there is no attribute, the next attribute “construction act 1 subject” is set as attribute T. Since this attribute T “construction act 1 subject” has a pointer value, the value meaning expression process is started with the leading value “LAN1” of this attribute T “construction act 1 subject” as a focus value. In this process, the top semantic expression “LAN1” of the point destination semantic expression of the target value “LAN1” is changed to the semantic expression B, and this semantic expression B “LAN1” is “premise i”, “method i”, or table. Therefore, the all-frame correspondence determination process is activated for the semantic expression below the semantic expression A “execution action 2” and the semantic expression below the semantic expression B “LAN1”.

  In the all-frame correspondence determination process, since the conceptual type of the semantic expression A “execution action 2” and the semantic expression B “LAN1” is “other”, the fact that “correspondence is impossible” and the correspondence table ([ Execution act 2: NIL] ([NIL: LAN1]) is returned to end the process, returning to the value semantic expression process, and changing the attribute “set component” at the head of the attribute group of the semantic expression B “LAN1”. Although the attribute T is an attribute T, the attribute T “collective component” does not have a pointer value, and sequentially determines whether the attribute T has the pointer value with the next attribute as the attribute T. Then, the attribute T “LAN1 $ Since the “construction method” has a pointer value, the value meaning expression process is started with the first value “method table 1” of the attribute T “LAN1 $ construction method” as the target value. 1 ”point ahead The semantic expression “method table 1” at the top of the taste expression is changed to the semantic expression B, the semantic expression B “method table 1” is a table, and the previous sentence and clause are added to the table semantic expression “method table 1”. Since a correspondence table representing a result corresponding to somewhere in the table is added, the row (first row) where the semantic representation of the correspondence destination exists (below) is designated as row U. Since the first sub-attribute “premise” of the attribute group in this row U is attribute T, and this attribute T “premise” has a pointer value “premise 1”, the value “premise 1” is regarded as a value of interest and the value meaning Start expression processing.

  In the value semantic expression processing, the top semantic expression “Premise 1” of the point-point semantic expression of the target value “Premise 1” is changed to the semantic expression B, and this semantic expression B “Premise 1” is “Premise i”. The all-frame correspondence determination process is started for the semantic expression below the semantic expression A “execution action 2” and the semantic expression below the semantic expression B “premise 1”. Then, since the concept type of the semantic expression A “execution action 2” is “predicate concept” and the concept type of the semantic expression B “premise 1” is “premise i”, the premise correspondence determination process is started. In this process, attention is paid to the value of the attribute “set component” attribute of the semantic expression B “Premise 1”, and this value includes “reference subset semantic expression AP” and “unreferenced subset semantic expression BP”. Therefore, attention is paid to the value of the “set component” attribute of “unreferenced subset semantic expression BP”. Since this value is “other than that”, the number “1” of the “collective component” attribute value is n, j = 1 to n, and the j-th value of the “collective component” attribute is the semantic expression Bj. The predicate concept correspondence determination process is activated for semantic expressions below semantic expression A and semantic expressions below semantic expression Bj.

  In the predicate concept correspondence determination process, since the concept categories of the semantic expression A “execution action 2” and the semantic expression B “premise 1” do not match or do not have a higher-lower relationship, the fact that the association cannot be made and the correspondence table ([execution Action 2: NIL] [NIL: Existence 4]) is returned and the process is terminated. Returning to the premise correspondence determination process, the results for j = 1 to n are divided into corresponding portions and uncorresponding portions and integrated. Here, since all the return results are “unmatchable”, the fact that it is “unmatchable” and the correspondence table ([execution action 2: NIL] [NIL: premise 1]) are returned, The process ends. Similarly, in the all-frame correspondence determination process and the next value meaning process, the fact that “correspondence is impossible” and the correspondence table ([execution action 2: NIL] [NIL: premise 1]) are returned. Further, in the next value semantic processing, the next sub-attribute “method” of the row U (first row) is changed to the attribute T again. Since the attribute T has a pointer value, the value semantic expression process is started with the value “method 1” of the attribute T as a target value.

  In the value semantic expression processing, the top semantic expression “method 1” of the point destination semantic expression of the target value “method 1” is changed to the semantic expression B, and this semantic expression B “method 1” is “method i”. The all-frame correspondence determination process is activated for the semantic expression below the semantic expression A “execution action 2” and the semantic expression below the semantic expression B “method 1”. In this process, since the concept type of the semantic expression A “execution action 2” is “predicate concept” and the concept type of the semantic expression B “method 1” is “method i”, the method correspondence determination process is started. In this process, since the semantic expression A “execution action 2” conceptually belongs to a higher category than the semantic expression B “method 1”, the semantic expression A and the semantic expression B “method 1” below the semantic expression A “execution action 2”. The predicate concept correspondence determination process is activated for the following semantic expressions.

  In the predicate concept correspondence determination processing, the semantic expression A “execution action 2” and the semantic category B “method 1” are both “execution actions”. An attribute pair correspondence table (hereinafter, attribute pair correspondence table (6)) is created for all phenomenon attributes of “method 1”. This attribute pair correspondence table (6) is shown in FIG. 65 (the location of “method 1” (knowledge side) is “house 1” by value propagation). Next, the predicate attribute pair split & correspondence list is initialized to FIG. 66, and the value of the “temporal” marker and the “time” added to the predicate concept using the predicate category tense & time correspondence determination rule of FIG. Determine correspondence between attribute values and propagating value. In this case, since it is determined that “cannot be associated”, the fact that the association is impossible and the correspondence table ([execution action 2: NIL] ([NIL: method 1])) are returned and the process is performed. finish.

  Returning to the method correspondence determination processing, here, since the semantic expression A “execution action 2” does not directly correspond to the “method i”, next, the value of the “action sequence set component” attribute of the semantic expression B is set. Pay attention. Since this value includes “reference subset semantic expression AP” and “unreferenced subset semantic expression BP”, the semantic expression below semantic expression A and the semantic expression below semantic expression AP are: Predicate concept correspondence determination processing is activated (semantic expression AP and semantic expression BP are phenomenon concepts belonging to the concept category of “executive action” level. Depending on the input sentence, it may be associated with semantic expression AP and semantic expression BP. Since it is possible, first, the correspondence between the semantic expression A, the semantic expression AP, and the semantic expression BP is determined). In this process, since the value of the “time point” attribute is “unmatchable”, “unmatchable” is returned. Then, predicate concept correspondence determination processing is activated for semantic expressions below semantic expression A and semantic expressions below semantic expression BP.

  In the predicate concept correspondence determination process, both the semantic expression A and the semantic expression BP belong to the “execution act” category, and therefore, an attribute pair correspondence table (hereinafter referred to as attribute pair) is assigned to all phenomenon attributes of the semantic expression A and the semantic expression B. A correspondence table (7)) is created. This attribute pair correspondence table (7) is shown in FIG. (The place of “executive action 3” (knowledge side) is “house 1” by the propagation of the value). Thereafter, the predicate attribute pair split & correspondence list is initialized to the above-described FIG. 67, and the value of the “tense” marker added to the predicate concept using the predicate category-specific tense & time correspondence determination rule of FIG. 68 described above. And the correspondence between the value of the attribute “time” and the propagation possibility. As a result, since it is determined that “can be associated” and “no propagation”, this fact is stored in the predicate attribute pair split & correspondence list. Further, since the correspondence relationship between the values of the attribute “location” is the semantic expression A) -v) and the propagation possibility is determined as “A → B propagation”, the determination result is stored in the predicate attribute pair split & correspondence list. . Next, the correspondence determination result of the “location” attribute value pair is set as a “location” attribute pair match flag, and this is used as an argument in the “attribute pair correspondence table” created above, except for “location” and “time”. The attribute value pair correspondence determination process is repeatedly activated for each phenomenon attribute pair such as “subject”, “target”, and “means”, and the correspondence between the value pairs and the propagation possibility are determined.

  As a result, it is returned to the attribute “subject”, the attribute “target”, the attribute “destination”, and the attribute “means” that “can be matched” and “no propagation”. For the attribute “action sequence set component”, the fact that it can be “corresponding” and the correspondence table ([“?”: “Installation action 1”] <NIL>) (attribute T19I) are returned. Furthermore, since all the return values are “correspondable” here, the correspondence determination result for each attribute value pair is stored in the predicate attribute pair division & correspondence list. FIG. 69 shows the predicate attribute pair split & correspondence list. Then, using the predicate category-specific pairing determination rule of FIG. 70, the correspondence between the predicate concepts is determined from the contents of this list, and the predicate semantic expression B is divided as necessary and associated with the semantic expression A. Create a correspondence table. At this time, the attribute value determined to be propagated is propagated. For example, regarding the “place” attribute, “arbitrary place” of the semantic expression B is replaced with the value of the semantic expression A. Thereafter, the meaning expression A is “can be associated with” the meaning expression B, and the created correspondence table ([“execution action 2”: “semantic expression BP”] <attribute T19I>) (attribute T20I)) Is returned to finish the process.

  Returning to the method correspondence determination process, since the fact that it is “correspondable” was returned in the previous predicate concept correspondence determination processing, the returned correspondence table is set as a lower correspondence table, and “correspondence is possible” The fact and the correspondence table (["execution action 2": "method 1"] <attribute T20I>) (attribute T21I)) are returned, and the process is terminated. Similarly, in the all-frame correspondence determination process to the value meaning expression process, “correspondence is possible” and a correspondence table ([“execution action 2”: “method 1”] <attribute T20I>) (attribute T21I)) Is returned to finish the process. Further, in the next value semantic expression processing, since the association was successful in the semantic expressions below the value semantic expression, the returned correspondence table is added to the semantic expression B. Then, the fact that “correspondence is possible” and the correspondence table ([“execution action 2”: “method table 1”] <attribute T21I>) (attribute T22I)) are returned, and the process is terminated.

  Returning to the value semantic expression processing, since the association was successful in the semantic expressions below the value semantic expression, the fact that “correspondence is possible” and the correspondence table ([“execution action 2”: “LAN1”] < Attribute T22I>) (attribute T23I)) is returned, and the process ends. Further, returning to the clause semantic expression association processing, since the association has succeeded in the semantic expressions below the value semantic expression, the fact that the semantic expression X is “can be associated” with the semantic expression Y and the correspondence table ( [“Execution Act 2”: “Construction Act 1”] <Attribute T23I>) (Attribute T24I)) is returned, and the process ends.

  Returning to the target clause semantic interpretation process, the fact that “correspondence is possible” is returned from the clause semantic expression correlation process, and it is checked whether or not the target node is the “problem setting section”. As a result, since the target section is the “problem setting section”, the problem solving process is started using the association result. In this problem solving process, using the correspondence table obtained as a result of the association of “problem setting section”, obtain the value of the attribute on the knowledge side corresponding to “problem attribute”, and return this value To end the process. Here, since the value “installation action 1” associated with “?” Is described in the returned correspondence table (attribute T24I), this value “installation action 1” is acquired as an answer. After that, returning to the input sentence semantic interpretation process, the “Redo from beginning” flag is not set, and there is no section at the end of the sentence of the target section. It passes to the production | generation part 54, and a process is complete | finished. This completes the association of sentence 3 (= clause 4).

  Then, the response sentence generation unit 54 generates a response sentence with this value as an answer. This response sentence may be, for example, the simplest sentence that responds only to an answer such as “installation action 1”, or suitable for an answer such as “need to perform installation action 1”. It may be a more complete sentence constructed by adding a predicate or the like.

  Thereafter, the same steps (S-1 to S-8 in FIG. 5) are repeated until an instruction to end the analysis is received from the user, and when the user gives an instruction to end the analysis via the input unit 10, all natural The language analysis process ends.

[III] Modifications to the Embodiments Each embodiment of the present invention has been described above, but the specific configuration and method of the present invention are within the scope of the technical idea of each invention described in the claims. Modifications and improvements can be made arbitrarily. Hereinafter, such a modification will be described.

(About problems to be solved and effects of the invention)
First, the problems to be solved by the invention and the effects of the invention are not limited to the above-described contents, and the present invention solves the problems not described above or has the effects not described above. There are also cases where only some of the described problems are solved or only some of the described effects are achieved. For example, even when the dependency relations of the semantic expressions of all input sentences cannot be completely interpreted, some of the problems of the present application are solved as long as the scope and accuracy of interpretation are improved compared to the conventional case. ing.

(About control)
Further, all or any part of the control described as being automatically performed in each of the embodiments may be performed manually, and conversely, all or any of the control described as being performed manually is performed. The part may be automated based on a known technique or the idea described above. Each functional block in the control unit 50 shown in each embodiment can actually be configured as a CPU and a computer program read and executed by the CPU, or a hard wire. -It can be configured by logic. Alternatively, this program can be stored in any computer-readable storage medium, such as CR-ROM, DVD, HD-DVD, or blu-disk. Further, each of the electrical components described above is functionally conceptual and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each unit is not limited to the one shown in the figure, and all or a part thereof is functionally or physically distributed / integrated in arbitrary units according to various loads and usage conditions. Can be configured. In addition, the processing procedure or the control procedure shown in the document or the drawing can be arbitrarily changed unless otherwise specified.

  As described above, the natural language analysis device, the natural language analysis method, and the natural language analysis program according to the present invention are useful for automatically interpreting the semantic correspondence between words and phrases included in an input sentence. In particular, it is suitable for realizing a dialog interface by interpreting the semantic correspondence between a plurality of semantic expressions.

It is a figure which shows an example of the basic structure of a semantic expression. It is a figure which shows markers. It is a block diagram which illustrates functionally the structure of the natural language analyzer which concerns on embodiment. It is a block diagram which illustrates the structure of an input sentence meaning interpretation process part functionally. This is a flowchart of the entire natural language analysis process. This is a flow chart of input sentence semantic interpretation processing. It is a figure which shows an example of a problem solution table. This is a flow chart of semantic interpretation processing of a target clause. This is a flow chart of semantic interpretation processing of a target clause. This is a flow chart of semantic interpretation processing of a target clause. This is a flow chart of semantic interpretation processing of a target clause. It is a flowchart of a problem setting clause and problem target attribute identification processing. This is a flowchart of problem solving processing. It is a flowchart of a clause meaning expression matching process. This is a flowchart of value semantic expression processing. This is a flowchart of value semantic expression processing. This is a flowchart of value semantic expression processing. This is a flowchart of value semantic expression processing. This is a flowchart of value semantic expression processing. This is a flowchart of value semantic expression processing. This is a flowchart of value semantic expression processing. It is a figure which illustrates semantic expression B "method table 1". This is a flowchart of the all-frame correspondence determination process. It is a flowchart of predicate concept correspondence determination processing. It is a flowchart of predicate concept correspondence determination processing. It is a figure which illustrates the format of a predicate attribute pair division & correspondence list. It is a figure which illustrates the format of the predicate category classified tense & time correspondence determination rule. It is a figure which shows the example of application of the tense & time correspondence determination rule according to predicate category. It is a figure which shows the example of application of the tense & time correspondence determination rule according to predicate category. It is a figure which illustrates the format of the predicate category classified correspondence determination rule. It is a figure which shows the example of an application result of the pair corresponding | compatible determination rule according to predicate category. It is a flowchart of a noun concept correspondence determination process. It is a flowchart of a noun concept correspondence determination process. It is a flowchart of a noun concept correspondence determination process. It is a flowchart of a noun concept correspondence determination process. It is a flowchart of a non-synthetic concept correspondence determination process. It is a flowchart of a non-synthetic concept correspondence determination process. It is a flowchart of a non-synthetic concept correspondence determination process. It is a figure which illustrates the format of the non-synthetic concept correspondence determination rule. It is a flowchart of a premise correspondence determination process. It is a flowchart of a premise correspondence determination process. It is a flowchart of a method corresponding | compatible determination process. It is a flowchart of a method corresponding | compatible determination process. It is a flowchart of a method corresponding | compatible determination process. It is a flowchart of a method corresponding | compatible determination process. It is a flowchart of an attribute value pair correspondence determination process. It is a flowchart of a numerical concept pair correspondence determination process. It is a flowchart of a symbol concept pair correspondence determination process. It is a figure which shows the input sentence in an Example. It is a figure which shows the semantic expression of the sentence 1 (section 1). It is a figure which illustrates a problem solution table. It is a figure which shows the semantic expression of clause 2. It is a figure which shows the semantic expression of clause 3. It is a figure which illustrates an attribute pair corresponding | compatible table (1). It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates the prescription category classified tense & time correspondence determination rule. It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates an attribute pair corresponding | compatible table (2). It is a figure which illustrates the semantic expression B. It is a figure which illustrates a reference subset, an unreferenced subset, etc. It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates a reference subset, an unreferenced subset, etc. It is a figure which illustrates an attribute pair corresponding | compatible table (3). It is a figure which illustrates semantic expression "Premise 1", mention subset semantic expression AP, etc. It is a figure which illustrates phenomenon concept semantic expression P. It is a figure which illustrates an attribute pair corresponding list (1). It is a figure which illustrates an attribute pair. It is a figure which illustrates synthetic concept semantic expression Tp. It is a figure which illustrates synthetic concept semantic expression Tp. It is a figure which illustrates phenomenon concept semantic expression P. It is a figure which illustrates semantic expression "premise i" etc. It is a figure which illustrates a problem solution table. It is a figure which illustrates an attribute pair corresponding | compatible table (4). It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates the prescription category classified tense & time correspondence determination rule. It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates the predicate category classified correspondence determination rule. It is a figure which illustrates mention execution action sequence subset AP and unreferenced execution action subset semantic expression BP. It is a figure which illustrates a problem solution table. It is a figure which illustrates an attribute pair corresponding | compatible table (5). It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates the prescription category classified tense & time correspondence determination rule. It is a figure which illustrates an attribute pair corresponding | compatible table (6). It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates the prescription category classified tense & time correspondence determination rule. It is a figure which illustrates an attribute pair corresponding | compatible table (7). It is a figure which illustrates a predicate attribute pair division & correspondence list. It is a figure which illustrates the predicate category classified correspondence determination rule.

Explanation of symbols

1 Natural Language Analyzer 10 Input Unit 20 Output Unit 30 Input / Output Control IF
40 Storage Unit 50 Control Unit 51 Input Sentence Preprocessing Unit 52 Semantic Expression Generation Processing Unit 53 Input Sentence Interpretation Processing Unit 53a Target Clause Semantic Interpretation Processing Unit 53b Problem Setting Clause and Problem Object Attribute Identification Processing Unit 53c Problem Solution Processing Unit 53d Section Semantic expression association processing unit 53e Value semantic expression processing unit 53f All-frame correspondence determination processing unit 53g Predicate concept correspondence determination processing unit 53h Noun concept correspondence determination processing unit 53i Non-synthetic concept correspondence determination processing unit 53j Premise correspondence determination processing unit 53k Method correspondence Determination processing unit 53l Attribute value pair correspondence determination processing unit 53m Numerical concept pair correspondence determination processing unit 53n Symbol concept pair correspondence determination processing unit 54 Response sentence generation unit

Claims (19)

A natural language analyzer for interpreting the meaning of an input sentence composed of a plurality of clauses or a plurality of sentences,
From among the plurality of sections or sentences, a problem frame is set that presents a scene for positioning and interpreting the meaning of each section or each sentence based on the phenomenon or predicate indicated by each section or each sentence. A problem frame setting section identifying means for identifying a problem frame setting section;
Clause meaning expression associating means for performing predetermined processing for semantically associating a clause or sentence other than the question frame setting clause with the question frame setting clause specified by the question frame setting clause specifying means;
A natural language analyzing apparatus comprising: The problem frame setting section specifying means includes:
The plurality of clauses or sentences are classified into one of a plurality of predetermined types based on a phenomenon or predicate indicated by each of the clauses or sentences, and the clause or sentence corresponding to a predetermined type is classified as Identify it as a problem frame setting section,
The natural language analysis apparatus according to claim 1. The clause meaning expression association means includes:
A question setting clause including a question target attribute having information as questioned by the input sentence as a value based on a phenomenon or a predicate indicated by each clause or each sentence from the plurality of clauses or a plurality of sentences. Identify and semantically associate these questionable attributes and question setting clauses with the question frame setting clauses;
The natural language analysis apparatus according to claim 1, wherein: The clause meaning expression association means includes:
Specifying a question sentence or a clause or sentence that can be converted into a question sentence as the question setting clause,
The natural language analysis apparatus according to claim 3. Semantic expression storage means for storing a plurality of semantic expressions configured in a predetermined format for specifying the meaning of the plurality of clauses or the plurality of sentences;
Correspondence propriety determination criterion storage means for storing a determination criterion of whether or not the attributes included in the semantic expression stored in the semantic expression storage means are compatible or value propagation possibility,
The clause semantic expression associating unit is configured to determine whether or not the attribute is compatible based on the semantic expression stored in the semantic expression storage unit and the determination criterion stored in the correspondence determination criterion storage unit. Determining propagation possibilities,
The natural language analysis apparatus according to claim 1, wherein: A plurality of determination criteria different for each category of the attribute are stored in the correspondence determination criterion storage means,
The clause meaning expression associating unit obtains the determination criterion according to the attribute category from the correspondence determination criterion storage unit, and based on the determination criterion, determines whether the attributes can be associated with each other or whether value propagation is possible. Judging,
The natural language analysis apparatus according to claim 5. The clause semantic expression associating means includes an attribute having a value of a predetermined identifier indicating a request for an answer to the attribute among attributes included in the semantic expression stored in the semantic expression storage means. Identifying it as an attribute,
The natural language analyzer according to claim 5 or 6. Comprising knowledge information storage means for storing knowledge information including a value to be an answer to the input sentence,
The clause semantic expression associating means is a knowledge stored in the knowledge information storing means for a value corresponding to the problem target attribute among the attributes included in the semantic expression stored in the semantic expression storing means. To get from the information,
The natural language analysis apparatus according to claim 5, wherein The clause semantic expression associating means, based on the determination result of whether or not the attributes can be matched, out of the attributes included in the semantic expression stored in the semantic expression storage means, the plurality of clauses or the plurality of sentences. Determining attributes that are not mentioned by the plurality of sections or sentences and selecting an attribute corresponding to the problem attribute from among the attributes that are not mentioned,
The natural language analysis apparatus according to any one of claims 5 to 8, wherein A natural language analysis method for interpreting the meaning of an input sentence composed of a plurality of clauses or a plurality of sentences,
From among the plurality of sections or sentences, a problem frame is set that presents a scene for positioning and interpreting the meaning of each section or each sentence based on the phenomenon or predicate indicated by each section or each sentence. A problem frame setting section identifying step for identifying a problem frame setting section;
A clause meaning expression associating step for performing predetermined processing for semantically associating a clause or sentence other than the question frame setting clause with the question frame setting clause specified in the question frame setting clause specifying step;
A natural language analysis method comprising: A natural language analysis program for causing a computer to execute a natural language analysis method for interpreting the meaning of an input sentence composed of a plurality of clauses or a plurality of sentences,
From among the plurality of sections or sentences, a problem frame is set that presents a scene for positioning and interpreting the meaning of each section or each sentence based on the phenomenon or predicate indicated by each section or each sentence. A problem frame setting section identifying step for identifying a problem frame setting section;
A clause meaning expression associating step for performing predetermined processing for semantically associating a clause or sentence other than the question frame setting clause with the question frame setting clause specified in the question frame setting clause specifying step;
A natural language analysis program for causing a computer to execute a natural language analysis method including In the problem frame setting section specifying step,
The plurality of clauses or sentences are classified into one of a plurality of predetermined types based on a phenomenon or predicate indicated by each of the clauses or sentences, and the clause or sentence corresponding to a predetermined type is classified as Identify it as a problem frame setting section,
The natural language analysis program according to claim 11. In the clause meaning expression matching step,
A question setting clause including a question target attribute having information as questioned by the input sentence as a value based on a phenomenon or a predicate indicated by each clause or each sentence from the plurality of clauses or a plurality of sentences. Identify and semantically associate these questionable attributes and question setting clauses with the question frame setting clauses;
The natural language analysis program according to claim 11 or 12, characterized in that: In the clause meaning expression matching step,
Specifying a question sentence or a clause or sentence that can be converted into a question sentence as the question setting clause,
The natural language analysis program according to claim 13. A semantic expression storage step of storing in a predetermined semantic expression storage means a plurality of semantic expressions configured in a predetermined format for specifying the meanings of the plurality of clauses or the plurality of sentences;
A correspondence availability judgment criterion storage step for storing the judgment criteria of the possibility of correspondence between the attributes included in the semantic representation stored in the semantic expression storage step or the possibility of value propagation in a predetermined correspondence availability judgment criterion storage means,
Based on the semantic expression stored in the semantic expression storage step and the determination criterion stored in the correspondence determination criterion storage step, the attribute availability / value of the attributes in the clause semantic expression association step Determining propagation possibilities,
The natural language analysis program according to claim 11, wherein: In the correspondence determination criteria storage step, a plurality of different determination criteria are stored for each category of the attribute,
In the clause meaning expression associating step, the determination criterion corresponding to the attribute category is acquired from the correspondence determination criterion storage unit, and based on this determination criterion, whether the attribute is compatible or whether value propagation is possible is determined. Judging,
The natural language analysis program according to claim 15. In the clause semantic expression association step, among the attributes included in the semantic expression stored in the semantic expression storage step, an attribute having a predetermined identifier indicating a response request for the attribute as a value is the problem target attribute To identify as,
The natural language analysis program according to claim 15 or 16, wherein: A knowledge information storage step of storing knowledge information including a value to be an answer to the input sentence in a predetermined knowledge information storage means,
Among the attributes included in the semantic expression stored in the semantic expression storage step, a value corresponding to the problem target attribute is extracted from the knowledge information stored in the knowledge information storage step in the clause semantic expression association step. Getting,
The natural language analysis program according to any one of claims 15 to 17, wherein: In the clause semantic expression association step, based on the determination result of the correspondence between the attributes, among the attributes included in the semantic expression stored in the semantic expression storage step, by the plurality of clauses or the plurality of sentences Determining an attribute that is mentioned and an attribute that is not mentioned by the plurality of sections or sentences, and selecting an attribute that corresponds to the problem attribute from among the attributes that are not mentioned;
The natural language analysis program according to any one of claims 15 to 18, wherein:

Images