JPH0736899A - Natural language processor - Google Patents

Abstract

PURPOSE:To automatically extract, register and utilize an operation procedure executed by a user to slove a problem. CONSTITUTION:This natural language processor has an analytical resource information storing part for storing analytical resource information, a sentence analyzing part 5 for analyzing natural language information stored in a document information storing part 2, an analyzed result storing part 6 for storing analyzed result information, an operation history storing part 8 for storing operation history, an operation script information storing part 10 for extracting and storing operation script information to be a model of a series of operation procedure by using the operation history information and analyzed result information, an adaptive script retrieving part 11 for retrieving operation script information adaptive to the analyzed result information of a document to be processed which is stored in the storing part 6, and an operation script adapting part 13 for adding information to the analytical resource information stored in the analytical resource information storing part or changing the information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language analysis device for analyzing a natural language sentence, a document processing device used for various language processing devices such as a program specification description support device in natural language, and a document revision support device. The present invention relates to a natural language processor.

[0002]

2. Description of the Related Art In recent years, computer-based systems have come to be used in many situations due to advances in information processing technology. Therefore, it has become common for users other than computer experts to use computers. However, in order to operate a computer-based system, it is necessary to learn specialized commands such as operating commands and programming languages, which makes it difficult to use a computer-based system. Responsible. As a method for solving this problem, there has been a strong demand for development of a natural language analysis technique and a device for causing a computer to process words normally used by humans.

Conventionally, this natural language analysis has been carried out using domain rules, which is knowledge about the analysis rules of the natural language grammar and the fields described by the natural language sentence to be analyzed.

However, since there are infinite kinds of expressions that appear in the natural language to be analyzed, it is very difficult to prepare an analysis rule that can correctly analyze all of them. In addition, even if the content of the processing target text is limited, a huge amount of domain knowledge is required to understand the text in that field, so it is extremely difficult to prepare sufficient domain knowledge in advance. Considered difficult. Further, in order to understand a natural language sentence, not only the domain knowledge as described above, but also the knowledge called common sense such as knowledge about time, space (place) or physical change is necessary. And it is still impossible to properly handle this common sense knowledge with the current technology of natural language processing and artificial intelligence processing.

Due to the reasons such as the incompleteness of the analysis rule, the lack of domain knowledge, and the difficulty of the common sense processing as described above, the natural language analysis device cannot expect 100% of the analysis success and the analysis error. There was a problem that problems such as the occurrence of ambiguity (a situation where the analysis result cannot be uniquely determined) and the like cannot be avoided.

Further, in order to avoid the above-mentioned problems, a method of using a restricted language in which a natural language sentence to be processed is restricted is used. However, the above-mentioned problem cannot be solved unless the restriction imposed on the document to be processed is set to a certain degree. On the other hand,
If the limit is set strongly, there is a problem that the flexibility and efficiency of natural language cannot be obtained, and the goal of helping a user who uses a computer system by performing natural language processing cannot be sufficiently achieved.

For the above-mentioned reasons, when using the natural language analysis device, the occurrence of parsing errors and ambiguity is unavoidable, and there are deficiencies in parsing rules, lack of domain knowledge, etc. Additional modification is required. However, this correction operation is complicated and increases the burden on the user. The natural language sentence to be processed may include a sentence that cannot be parsed due to a parsing error or ambiguity due to the same reason. There is a problem in that each analysis error cannot be resolved unless the same operation is performed, thus increasing the burden on the user in using the computer system and reducing the utilization efficiency.

In recent years, various document processing devices such as a natural language program specification description support device and a document revision support device analyze an input document, and correct the document and verify the contents based on the analysis result. Devices that do things such as these are attracting attention.

Conventionally, in such a document processing apparatus, when handling a document, each sentence in the document is treated as a completely independent sentence.

However, the document includes a group of sentences having various dependency relations, such as a sentence having a similar structure and a sentence using the same word.

For example, when a word A included in a sentence is replaced with another word B in a document revision device, a sentence containing another word A is searched for in the document, the word A is replaced with the word B, and reanalysis is performed. Need to do.

When performing the above-mentioned work, in the conventional method, it is necessary for the user to search for sentences, replace words, and analyze each sentence.

[0013]

As described above, in the conventional natural language analysis apparatus, an analysis error or ambiguity occurs due to reasons such as incomplete analysis rules, lack of domain knowledge, or difficulty in common sense processing. The problem of cannot be avoided. In addition, the method of using a restricted language does not provide the flexibility and efficiency of natural language, and cannot achieve the goal of helping users who use computer systems by performing natural language processing. The use of restricted languages cannot solve the above problems. Therefore, operations such as additional correction of analysis rules, dictionary items, or domain knowledge are required. Even if there is a parsing error due to the same reason in the natural language sentence to be processed, the operation must be performed for each, increasing the burden on the use of the computer system and increasing the utilization efficiency. There was a problem that caused a drop.

Further, since the conventional document processing apparatus handles each sentence in the document as an independent sentence, the sentence having a dependency relationship is also treated independently. Therefore, even when the same operation is performed on a sentence having a dependency relationship, the user needs to perform the operation for each sentence separately.
For this reason, it is difficult to maintain the consistency of the document contents, and there is a problem that the amount of work of the user increases.

The first invention is made in consideration of the former problem, and an operation procedure for solving a problem such as an analysis error made by a user is automatically extracted and registered, and then occurred. It is an object of the present invention to provide a natural language processing device that can solve the same problem without repeating the same procedure.

The second invention is made in consideration of the latter problem, and the system stores the dependency relation between each sentence in the document so that the operations on the sentence having the dependency relation are collectively performed. It is to provide a natural language processing device capable of performing.

[0017]

A natural language processing apparatus according to a first aspect of the present invention is a text input section for inputting a natural language text and a document information storage for storing the natural language text input by the text input section. Section, an analysis resource information storage section that stores analysis resource information such as analysis rules, dictionary items, and domain knowledge for analyzing natural language sentences stored in the document information storage section, and creates and edits the analysis resource information. An analysis resource information creation / editing unit, a sentence analysis unit that analyzes the natural language sentence stored in the document information storage unit using the analysis resource information stored in the analysis resource information storage unit, and this sentence An analysis result storage unit that stores the analysis result information obtained by the analysis unit, an analysis result display unit that presents the analysis result information to the user, and a history of operations performed by the analysis resource information creation / editing unit. The operation history storage unit to be stored, the operation history information stored in the operation history storage unit and the analysis result information stored in the analysis result storage unit, and operation script information which is a model of a series of operation procedures. An operation history analysis unit that extracts the operation history information, an operation script information storage unit that stores the operation script information obtained by the operation history analysis unit, and the analysis result. An applied script search unit for searching applicable operation script information from the operation script information stored in the operation script information storage unit with respect to the analysis result information of the processing target document stored in the storage unit. And an operation script adaptation unit for adding or changing the analysis resource information stored in the analysis resource information storage unit according to the operation script information.

A natural language processing apparatus according to a second aspect of the present invention is the natural language processing apparatus according to the first aspect of the present invention, inquiring of a user whether or not to apply and execute the operation script information to the document information obtained by the application script search section, and to display the result. The operation script adaptation unit has a receiving confirmation dialogue unit, and the operation script adaptation unit applies the operation script information to the document information according to the response of the user obtained by the confirmation dialogue unit, and stores the analysis resource information according to the operation script information. This is to add or change the analysis resource information stored in the department.

A natural language processing apparatus according to a third aspect of the present invention is a document editing section for inputting or editing a document, and a language dictionary storage for storing knowledge such as a dictionary for analyzing an input document, grammar, domain knowledge, and interpretation knowledge. Section, a sentence analysis unit that analyzes each sentence input to the document editing unit based on the knowledge stored in the language dictionary storage unit, and a knowledge of each sentence analyzed by the sentence analysis unit for analysis. A sentence that shares the same required knowledge consists of a sentence relation processing unit that has a dependency relation.

[0020]

[Operation] The natural language processing procedure of the first and second inventions will be described.

The sentence input unit inputs the natural language sentence to be processed into the document information storage unit.

Each sentence included in the document information stored in the document information storage unit is analyzed by the sentence analysis unit using the analysis rule and the domain knowledge prepared and stored in advance in the analysis resource information storage unit. The analysis result information indicating the analysis success result or the analysis failure status, which is the result, is stored in the analysis result storage unit and presented to the user through the analysis result display unit.

Next, when a problem such as an analysis error occurs, the user uses the analysis resource information creation / editing unit to analyze the rules, dictionary items or areas based on the information presented in the analysis result display unit. Add information to knowledge or change / correct information. Here, at the same time, the operation performed by the user is stored in the operation history storage unit.

In this way, the re-parsing of the problematic sentence is performed. Here, if the analysis fails again, the user repeats the above-described correction operation. And
If the analysis is successful, the operation history analysis unit analyzes the operation history and tries to extract an operation script that is a template of an operation procedure for solving the problem.

When the operation script is successfully extracted here, the script is stored in the operation script information storage section.

When a user requests a search for an adaptable script in the process of processing an analysis error, the applicable script search unit determines a script applicable to the corresponding analysis error as the operation script information storage unit. It is searched from.

When a script that can be judged to be effective for solving the problem is found, the confirmation inquiry unit asks the user whether or not the script is applicable.

Here, when the response of applicability is obtained from the user, the operation script adaptation unit performs a process based on the corresponding script, changes and corrects the contents of the analysis resource information storage unit, and applies it. Reparsing into sentences is performed.

As described above, according to the present invention, it is possible to automatically extract and register a procedure of an operation performed by a user for solving a problem such as a parsing error in natural language analysis. Becomes

Further, by utilizing the registered operation procedure, it is possible to solve the same problem that occurs thereafter by one operation.

As a result, it becomes possible to reduce the burden on the user in using the computer system utilizing the natural language analysis and to avoid the deterioration of the utilization efficiency.

Further, in natural language analysis, it is not necessary to use a restricted language having a strong restriction that does not cause a parsing error, and natural language analysis that does not impair the flexibility and efficiency of natural language. It is possible to configure the device.

In the natural language processing apparatus of the third invention, first, a sentence is input to the document editing section. Further, the language dictionary storage unit stores knowledge of a dictionary for analyzing an input document, grammar, domain knowledge, and interpretation knowledge. The sentence analysis unit analyzes each sentence input to the document editing unit based on the knowledge stored in the language dictionary storage unit. Then, the sentence relation processing unit determines that, from the knowledge of each sentence analyzed by the sentence analysis unit, sentences sharing the same knowledge necessary for analysis have a dependency relationship. As a result, it becomes possible to collectively perform the operations on the sentences having the dependency relationship.

[0034]

Embodiments of the first and second inventions will be described using a specification document analysis system which is an example of a natural language analysis device.

First, the outline of the specification document analysis system will be described.

The specification document analysis system is a system for describing the specifications of the operation of the device in Japanese and analyzing and understanding the contents thereof. The grammar rules and dictionary for analyzing Japanese, the interpretation rules of sentences appearing in the specification, It is a system that performs analysis using knowledge about devices described in specifications and converts the contents described in specification documents into unambiguous formal semantic descriptions. This specification document analysis system is a part of a "program generator" (Japanese Patent Application No. 3-206005) for generating a computer control program from a natural language sentence.

FIG. 1 shows a block diagram of the specification document analysis system of this embodiment.

In FIG. 1, the user describes the specification document through the document editing section 1. Subsequently, the syntactic analysis unit 5a performs syntactic analysis using the grammar and dictionary information.
Further, the semantic content analysis unit 5b describes the specification sentence by using an interpretation rule for interpreting the specification sentence and domain knowledge (knowledge about parts such as devices appearing in the field described by the specification sentence). Converts the meaning content into an unambiguous meaning interpretation result.

The domain knowledge referred to here is, for example, [“L
"ED" is "device"] or [device "LE
"D" is capable of "lighting" operation].

The interpretation rule here is, for example,
"Noun phrase representing a certain device P" + "particle: wo" + "device P"
When a verb phrase representing a possible action A of "constitutes a sentence, the meaning content is to" perform action A on device P ".

The grammar and dictionary used in these processes,
The information such as the interpretation rule and domain knowledge is prepared prior to the analysis of the specification document, or is created and changed by the user through the knowledge / rule editing unit 4 in parallel with the creation of the specification document.

Information on the analysis result or analysis error in each of the above-mentioned steps is presented to the user through the result display section 7.

When an analysis error occurs in a certain sentence S, the user rewrites the sentence S, or
Through the knowledge / rule editing unit 4, grammar, dictionary, interpretation rules,
Alternatively, the sentence can be interpreted by changing the domain knowledge.

Next, the state of processing when the present invention is applied to the above-mentioned specification sentence analysis system will be described using a concrete example.

First, the sentence to be processed is input to the interactive natural language analysis apparatus. Here, it is assumed that the input sentence includes the following two sentences.

Sentence a: "Run the hydraulic pump." Sentence b: "Fix the control damper." Then the entire sentence, including both sentences, is analyzed using the rules and knowledge described above, eg sentence a. 2), the syntactic analysis result shown in FIG. 2 is obtained by the analysis using the grammar rules.

Subsequently, it is judged whether this sentence can be interpreted as a sentence of the specification document by an inspection using the interpretation rule and the domain knowledge. It is possible to prepare a plurality of syntax check rules, but here, for the sake of simplicity, it is assumed that only the following syntax check rule I exists.

Interpretation Rule I: Pattern part: "<device>" is changed to "<action>". ""==> Interpretation Result Part: [Operation: Device] The interpretation rule is composed of a pattern part representing a condition for applying the interpretation rule and an interpretation result part representing an interpretation obtained as a result. In the pattern part, the portion enclosed by the symbols “<” and “>” is rewritable, and in the above interpretation rule I, the portion of the symbol “<device>” is the domain knowledge. It is rewritten as the natural language expression L of the part P that is described as "equipment" in the
The part of "motion>" can be rewritten as the natural language expression M of the motion A described as the possible motion of the part P in the domain knowledge. The part surrounded by the symbol """indicates that the expression appears at the corresponding position in the sentence to be inspected.

Regarding the sentence a as the example this time, if the following knowledge k and knowledge l are included in the domain knowledge, the condition of the above syntax check rule I is satisfied and the sentence a is correctly analyzed. It will be. (Note that the combination of the symbol [#] and the number is an identification symbol of knowledge inside the system.) Knowledge k: "Hydraulic pump # 1" represented by the natural language expression "hydraulic pump" is "equipment" Is.

Knowledge l: Equipment: "Hydraulic pump # 1" can perform the operation: "operation # 1" represented by the expression "operate".

However, this time, it is assumed that these two pieces of knowledge do not exist in the domain knowledge, and as a result, the syntax check fails, so that it is presented to the user through the result display unit 7.

Then, based on this information, the user
The processing of additional correction for analysis rules, dictionary items, or domain knowledge is performed, but here, the user uses the knowledge / rule editing section 4 having the function of the analysis resource information creating / editing section, and Suppose you have added to your knowledge. The correction operation performed here is simultaneously stored in the operation history storage unit 8.

Following this correction, re-analysis is performed on the sentence a, and correct analysis is performed this time using the added knowledge, and the following analysis result Ra is obtained.

Analysis result Ra: [Operation # 1: hydraulic pump # 1] Next, the operation history analysis unit 9 analyzes the operation history in order to obtain an operation script representing a procedure for solving an error in analysis. An operation script representing the following information is extracted.

Operation script S1 condition: Under the interpretation rule I, C1. Failure of domain knowledge check of natural language expression "hydraulic | pump / noun phrase (noun, foreign noun)" and <device>.

C2. Failure of domain knowledge check of natural language expression "driving | suru / verb phrase (sa verb, verb)" and <action>.

==> Operation: A1. Add the following knowledge to the domain knowledge [Natural language expression [hydraulic | pump / pump / noun phrase (noun, foreign noun)] represents the equipment “hydraulic pump”] A2. The following knowledge is added to the domain knowledge [Operation "operation of device" hydraulic pump "is possible"] As described above, the operation script S1 for error correction for the adaptation failure of the interpretation rule I is obtained, and the operation script information storage unit 10 " It will be used for subsequent analysis error resolution. For example, in sentence b, the same parsing error as in sentence a occurs. Therefore, according to an instruction from the user, the application script search unit 11 searches for an operation script that can be used to eliminate an analysis error in the sentence b, and finds the above-mentioned operation script S1.

This is shown to the user by the confirmation dialogue unit 12, and an inquiry is made to the user as to whether or not the operation script S1 can be applied to the sentence b. In this example, the user responds to the application permission, the operation script adaptation unit 13 performs a series of operations according to the operation script S1, and the following knowledge m and knowledge n are automatically registered in the domain knowledge. To be done.

Knowledge m: A "control damper" represented by the natural language expression "control damper" is a "device".

Knowledge n: device: "control damper" can perform operation: "fix".

By additionally registering the knowledge m and the knowledge n, the interpretation rule I can be applied to the sentence b, and the interpretation result Rb for the reanalysis result sentence b is obtained.
Is obtained.

Analysis Result Rb: [Fixed: Control Damper] Next, an embodiment in which the above embodiment is further embodied will be described with reference to the block diagram of FIG.

In FIG. 3, reference numeral 1 is a document editing unit composed of a keyboard, a video display screen, etc., and the user inputs a natural language document to be processed into the interactive natural language analyzing apparatus, or Edit it.
The data of the natural language document input from this document editor is sent to the document information storage unit 2. The document information storage unit 2 stores the surface layer data of the input processing target natural language document and the like.

FIG. 4 shows an example of the contents of the document information storage section. The entries corresponding to each sentence of the sentence data to be processed include sentence position information A, sentence surface information B, analysis state information C, etc. Are categorized and stored. Note that D is storage address information.

Here, the sentence position information A represents the position of the corresponding sentence in the specification document, and is information used for identifying each sentence. The sentence surface layer information B stores the natural language surface layer of the sentence to be analyzed. Further, the analysis state information C stores information such as a processing state of a corresponding sentence such as “analysis success”, “analysis failure” or “unanalyzed”.

Reference numeral 3 in FIG. 3 is a knowledge / rule storage unit for storing area knowledge and interpretation rules for analyzing a sentence to be processed.
The expression correspondence rule storage unit 3b and the interpretation rule storage unit 3c are included.

By the way, in the specification document analysis system according to the present embodiment, the domain knowledge is obtained by using a framework called "semantic network" which is a directed graph composed of nodes and arcs which are effective arrows representing the relation between the nodes. It describes.

FIG. 5 shows an example of domain knowledge represented by a semantic network. The circles represent nodes and the arrows represent arcs. For example, an arc (arrow) labeled "equipment" is extended from the "parts" node to the "compressor" node, and this represents the knowledge that "compressor" is "equipment".

From the "Compressor" node,
A "working" arc is stretched to the "run", "start", and "stop" nodes, which allows the equipment "compressor" to run, start or stop. It expresses something. Furthermore, since the "state" arc is extended from the "compressor" node to the "running" node and the "stopping" node, the device "compressor" is said to be "stopping" or "operating". It represents the knowledge that it can be in a state.

Knowledge of the above "compressor"
It is expressed in the semantic network shown in FIG.

Further, in FIG. 5, "F chamber temperature sensor"
It can be seen that similar knowledge is described for and “LED”. Note that the dashed-dotted rectangles in FIG. 5 indicate the parts related to the “compressor”, the “F room temperature sensor”, and the “LED”, respectively, but these are added for convenience of description. Yes, it is not included in the knowledge representation by the semantic network.

FIG. 6 shows an example of the contents of the area knowledge storage unit 3a. The entry corresponding to the knowledge about the parts such as the apparatus described in the specification document to be processed is the arc start point information A and the arc. It is classified and stored as type information B, arc end point information C, and the like. Note that D is storage address information. In each entry of the area knowledge storage unit 3a, the column of the arc start point information A and the arc end point information C stores the label of the node of the start point and the end point of the corresponding arc. It should be noted that the node label here is provided with an identification symbol (a set of symbol “#” and a number) for identifying arcs having the same name. Further, in the field of arc type information B of each entry of the area knowledge storage unit 3a, a label indicating the type of the corresponding arc is stored.

The example of the contents of the area knowledge storage unit shown in FIG. 6 expresses the contents of the semantic network shown in FIG. 5, and the storage addresses Q21 to Q25 express the knowledge about the "compressor". Similarly, storage address Q3
1 to Q42 represent knowledge about "F room temperature sensor", and storage addresses Q51 to Q54 represent knowledge about "LED".

FIG. 7 shows an example of the contents of the expression correspondence rule storage section 3b. In each entry of the expression correspondence rule storage unit 3b, knowledge representing a correspondence relationship between a node in the domain knowledge and the natural language expression and its syntactic analysis result is classified and stored in the node label information A, the correspondence expression information B, and the like. I am trying to do it. Note that C is storage address information. In the example of the expression correspondence rule storage unit 3b in FIG. 7, for example, the entry of the storage address R11 is the node "compressor # 1" in the domain knowledge is expressed by the noun "compressor" in the natural language document. In the entry of R13, the operation "start" of the device "compressor" represented by the node "start # 1" is expressed as "start" in the natural language document. There is.

FIG. 8 shows an example of the contents of the interpretation rule storage unit 3c. Each of these entries represents an interpretation rule used for interpreting the specification sentence, and the contents of the interpretation rule are called interpretation rule number information A, application pattern information B, area knowledge condition information C, and interpretation result information D. The data is classified and stored as follows. Note that E is the storage address information.

A number for identifying each interpretation rule is entered in the field of the interpretation rule number information A of each entry of the interpretation rule storage unit 3c. This information is used as information for selecting a script for an appropriate operation in the resolution process in which a failure occurs in the analysis performed later.

Then, in the column of application pattern information B, syntactic conditions for applying the interpretation rule corresponding to the entry are described. For example, in the entry of the storage address T11 in FIG. 8, the contents of the adaptive pattern information B are as follows.

<Action sentence: X> → <noun phrase: N> ”is changed to“ <
Verb phrase: V> ”.” This means that the target of this rule is noun phrase (N),
It indicates that it is necessary to be an action sentence (X) consisting of "wo", verb phrase (V), and ".". Note that the symbols N, V, and X here are variable symbols for representing corresponding parts in a sentence (and an analysis result thereof) to which this rule is applied.

Further, in the field of area knowledge condition information C, the conditions to be satisfied in view of the area knowledge are described. The contents of the entry T11 in FIG. 8 are as follows.

N = expression (device (P)) & V = expression (A = motion (P)) This is a conjunctive of two conditions. The former represents the condition that the expression of the part corresponding to the variable symbol N matches the expression obtained by referring to the expression correspondence rule storage unit 3b of a certain device P included in the domain knowledge. Similarly, the latter expresses the condition that the expression of the part corresponding to the variable symbol V is the expression corresponding to the possible operation of the device P.

When the application pattern information B and the area knowledge condition information C described above are satisfied, the corresponding interpretation rule can be applied.

Then, in the column of the interpretation result information D of each entry of the interpretation rule storage unit 3c, information about the interpretation result obtained by applying the corresponding interpretation rule is described.

In the entry T11 in FIG. 8, the contents of the interpretation result information D are as follows.

Interpretation (X) = [A: P] This expresses that the overall interpretation of the analysis target corresponding to the variable symbol X is “to operate the device P”.

In the entry T21 in FIG. 8, the value of the interpretation result information D is as follows.

Interpretation (X) = ([S: P] ==> Interpretation (R)) This means that the overall interpretation of the analysis target corresponding to the variable symbol X is “if the device P is in the state S, < The content of the interpretation of the action sentence R> ”is expressed. It should be noted here that one interpretation rule can be filled with information obtained as a result of another interpretation rule. That is,
In this example, the interpretation result of the action sentence R is obtained by another interpretation rule.

Interpretation processing of the specification sentence is performed according to the interpretation rules described above.

Now, returning to FIG. 3, the description will be returned to the respective parts constituting the interactive natural language analyzing apparatus according to the present embodiment.

In FIG. 3, reference numeral 4 is a knowledge / rule editing unit, which allows the user to create and modify the knowledge and rules stored in the knowledge / rule storage unit 3. It should be noted that all the operations performed here are stored in the operation history storage unit 8.

Reference numeral 5 is a sentence analysis unit, which performs a syntactic syntactic analysis process on the natural language specification sentence stored in the document information storage unit 2 according to a user's instruction, and a knowledge / rule storage unit. By using the domain knowledge and the interpretation rules stored in No. 3, the description contents are analyzed, and the obtained information about the interpretation result or the interpretation error situation is obtained.
It is stored in the interpretation result storage unit 6. In addition,
The syntactic parsing performed here is the same as that performed by a normal natural language processor, and will not be described below.

Reference numeral 6 is an interpretation result storage unit, which stores the result of the sentence interpretation processing performed by the sentence analysis unit 5.

FIG. 9 shows an example of the contents of the interpretation result storage unit 6. In each entry of the interpretation result storage unit 6, information on the analysis performed on each sentence information stored in the document information storage unit 2 is syntactic analysis result information B, interpretation rule number information, and application rule number information C. , Unresolved condition information D, interpretation result information E, etc., and are stored. Note that F is storage address information.

The position information of the corresponding sentence is stored in the column of the sentence position information A of each entry of the interpretation result storage unit 6, so that it can be associated with the document information storage unit 2. I have to. For example, since the sentence position information A of the entry of the storage address V11 in FIG. 9 is S11, this entry has the same value as the sentence position information A among the entries of the document information storage unit 2 shown in FIG. It is understood that the information is the interpretation information for the sentence of the entry of the storage address P11.

In the column of the syntactic analysis result information B of each entry of the interpretation result storage unit 6, the result of the syntactic analysis for the corresponding sentence performed by the sentence analysis unit 5 is stored in a list expression. For example, the contents described in the entry of V11 in FIG. 9 represent the syntactic analysis result shown in FIG. Note that, for an entry for which the corresponding sentence has not been analyzed, such as the entry of V41,
The symbol "-" is stored.

In the column of the adaptive rule number information C of each entry of the interpretation result storage unit 6, the rule number of the interpretation rule which is tried to be applied to the sentence corresponding to this entry is stored. , V11 entry, rule number U
11 indicates that an attempt was made to apply the 11 interpretation rules.
In this case, too, if there is no interpretation rule that is tried to be applied, such as the entry of V41, the symbol "-" is used.
Will be remembered.

In the column of unresolved condition information D of each entry in the interpretation result storage unit 6, among the domain knowledge conditions specified by the interpretation rule that is attempted to be applied to the sentence corresponding to this entry, the condition is satisfied. Not trying to remember the conditions. For example, in the column of the unresolved condition information D of the entry V11 in FIG. 9, “N = noun phrase (noun (hydraulic) foreign noun (pump)), V = verb phrase (sa verb (start) verb (do)) ) ”Is described. This corresponds to the interpretation rule stored in the entry of T11 of the content of the interpretation rule storage unit 3c shown in FIG. 8, and in order to apply this interpretation rule U11, “noun phrase (noun (hydraulic The expression "hydraulic pump" having the syntactic structure of ") foreign noun (pump)" needs to correspond to the variable symbol N, and therefore the expression "hydraulic pump" is obtained from the domain knowledge condition of the interpretation rule U11.
Is an expression of a certain “equipment” (here, “hydraulic pump”), and similarly, the expression “start” which has a syntactic structure of “verb phrase (sa verb (start) verb (do))”
Expresses that two conditions of this device “hydraulic pump”, that is, one possible operation (here, “starting”) have not been achieved. As described above, the interpretation rule storage unit 3
The column of unresolved condition information C of each entry of c stores the condition that must be achieved in order to apply the corresponding interpretation rule. Again, here
For example, if there is no unfulfilled condition like the entries of V21 and V22, or the entry to which the interpretation rule is not yet applied, such as the entry of V41, the symbol "-" is stored. I have to.

The interpretation result information E of each entry in the interpretation result storage unit 6 stores the interpretation result for the sentence corresponding to this entry. For example, in the entry of V21, "[lighting: LED] ”is stored. Again, the symbol "-" is stored if no interpretation result is obtained.

In FIG. 3, reference numeral 7 is a result display unit, and the result of the interpretation performed by the sentence analysis unit 5 stored in the interpretation result storage unit 6 is presented to the user.

Reference numeral 8 is an operation history storage unit, which stores the status of the analysis processing on the processing target sentence and the operation performed by the user through the knowledge / rule editing unit 4 in order to solve the analysis error. ing.

FIG. 10 shows an example of the contents of the operation history storage unit 8. In each entry of the operation history storage unit 8, the history information of the operation performed by the user and the result thereof is stored in the column of the operation history information A in the order of the time of occurrence. Note that B is storage address information.

In FIG. 10, for example, the storage address W
The entry 11 indicates that the sentence whose sentence position information is S21 is analyzed, and the entry W12 indicates that the analysis is successful. Also, W2
The entry 5 indicates that knowledge has been registered in the area knowledge storage unit 3a using the knowledge / rule editing unit 4. Here, the symbol “New” indicates that the node having the label enclosed by the following parentheses is newly generated. The entry of W26 stores that expression registration has been performed in the expression correspondence rule storage unit 3b.

In FIG. 3, reference numeral 9 represents an operation history analysis section, and here, referring to the contents of the interpretation result storage section 6 and the operation history storage section 8, an operation for eliminating an analysis error is performed. An operation script representing a procedure is extracted, and the result is stored in the script storage unit 10. The script extraction process performed by the operation history analysis unit 9 has a central role in this embodiment, and will be described later together with the flow of the process.

In FIG. 3, reference numeral 10 is an operation script information storage unit, which stores the script obtained by the operation history analysis unit 9.

FIG. 11 shows the operation script information storage unit 10.
The information of the operation procedure that may be used to solve the parsing failure using a certain interpretation rule is the target interpretation rule number information A, the script application condition information B, and the operation script. It is classified and stored as information C and the like. Note that D is storage address information.

In the column of the target interpretation rule number information A of each entry of the operation script information storage unit 10, there is an interpretation rule number of an interpretation rule which may solve the parsing error by the script described in this entry. Remembered For example, it can be seen that the entry of the storage address X11 in FIG. 11 should be used for the failure of the interpretation using the entry of the storage address T21 in the example of the content of the interpretation rule storage unit 3C.

In the column of the script application condition information B of each entry of the operation script information storage unit 10, the entry of the interpretation rule storage unit corresponding to the interpretation rule specified by the target interpretation rule number information A is applied. Among the conditions described in the pattern information B or the domain knowledge condition information C, the conditions that have not been achieved are described. This indicates that the operation script may be available when the application of the corresponding interpretation rule is blocked by the failure to meet the condition shown in the script application condition information B.

That is, for example, in the entry of X11 of FIG. 11, the content of the script application condition information B is “N2 = expression (S)”, so that the entry of T21 shown in FIG. , That part of the domain knowledge condition information C has not been achieved. And the contents use this script when the condition that "the second noun phrase N2 must be an expression of a state S (which the device P can take)" is not achieved. It is possible to do.

Further, for example, in the entry X21 of FIG. 11, the content of the script application condition information B is “S = state (P)”, so that the area of the entry T21 of FIG. 8 corresponding to the interpretation rule U21 is stored. When a part of the knowledge condition information C has not been achieved and the condition "a certain state S (expressed by the second noun phrase N2) is a state that the device P can take" has not been achieved, Indicates that this script is applicable.

Also, a plurality of conditions can be described in the column of the applicable condition information B like the entry of X31.

Further, in the column of the applicable condition information B, a variable symbol represented by the symbol "L" and a number set can be used. This means that you can put an arbitrary surface string in the part where the variable symbol appears,
For example, in the entry X31, the symbols L1 to L4 are the variable symbols.

In each entry of the operation script information storage unit 10, the operation script information C column stores the procedure of the operation to be performed when the application of the script is decided.

For example, in the entry of X11 in FIG.
In the field of the operation script information C, "Noun phrase N2 is state S
Is a language expression of the expression correspondence rule storage unit 3
The operation procedure of registering to b is stored, and X
In the entry of 21, the operation procedure of storing the knowledge “S is a state that the device P can take” in the area knowledge storage unit 3a is stored. This operation script information C
In the column, a series of a series of operations can be stored. For example, in the entry of X31 in FIG. 11, an operation procedure including two knowledge registrations and two expression registration operations is stored. The notation “New (C)” that appears here means that a new node is generated in the domain knowledge and its identification label is substituted for the variable symbol C.

In FIG. 3, reference numeral 11 is an applied script search unit. The applied script search unit 11 refers to the contents of the interpretation result storage unit 6 according to a user's instruction or the like and stores a script that may be applicable to a sentence in which an analysis error occurs, in the operation script information storage unit. Part 1
Information such as the presence or absence of the corresponding script is sent to the confirmation dialogue unit 12 by searching from 0.

In FIG. 3, reference numeral 12 is a confirmation dialogue unit. Here, the information regarding the search result and the applicability of the script for solving the analysis error in the processing target sentence is transmitted and received between the user and this device. Here, an inquiry is made to the user regarding the application of the script provided from the application script search unit 11 to the confirmation dialogue unit 12, and the information about the result of this inquiry is sent to the operation script adaptation unit 13. There is.

Reference numeral 13 is an operation script adaptation section.
Here, based on the information from the confirmation dialogue unit 12, a script whose application is permitted by the user is fetched from the operation script information storage unit 10, and the knowledge / rule storage unit 3 is operated. .

The processes performed by the application script search unit 11, the confirmation dialogue unit 12, and the operation script adaptation unit 13 will be described in the subsequent process flow.

Next, the processing in this embodiment will be described.

The entire processing is performed according to the following processing procedure A. The processing procedure A includes a result display procedure B, a script extraction procedure C, a script processing procedure D, etc., which will be described later. In addition, FIG.
Is a flow chart showing the flow of the processing procedure A.

[Processing Procedure A] (a1) The document information to be processed is input to the document information storage unit 2 through the document editing unit 1.

(A2) The processes of (a3) to (a9) are repeated.

(A3) If there is no sentence S to be analyzed in the document information storage section 2, proceed to (a10).

(A4) The sentence analysis unit 5 analyzes the sentence S using the knowledge and rules for sentence analysis stored in the knowledge / rule storage unit 3, and stores the document information corresponding to the sentence S. “Analysis success” or “analysis failure” is stored in the analysis state information C of the entry Ps of the part 2 according to the analysis result.

(A5) The interpretation result information Rs obtained from the sentence analysis unit 5 is stored in the interpretation result storage unit 6.

(A6) According to the result display procedure B, the information Is concerning the analysis result is presented to the user through the result information display section 7 in accordance with the interpretation result information R.

(A7) Analysis state information C of entry Ps
When the content of “is analysis successful”, the operation history analysis unit 9
Thus, the processing according to the script extraction processing procedure C is performed, and the processing for the sentence S ends.

(A8) Analysis state information C of entry Ps
If the content is “analysis failure”, the user uses the knowledge / rule editing unit 4 to add the content to the content of the knowledge / rule storage unit 3 based on the information I on the analysis result. Perform operations such as changing. In this process, when the user gives an instruction, the processing according to the script processing procedure D is performed.

(A9) Proceed to (a3).

(A10) The processing procedure A is completed.

Incidentally, (a2) to (a1) of this processing procedure A
All the history information of the operations performed in 0) is stored in the operation history storage unit 8.

The result of the sentence interpretation process performed by the sentence analysis unit 5 is performed by the result display unit 7 according to the following result display procedure B.

[Result Display Procedure B] (b1) When the content of the analysis state information C of the entry Ps of the document information storage unit 2 corresponding to the sentence S to be processed is “analysis success”, the analysis success is notified. The message is presented to the user, and the interpretation result stored in the interpretation result information D of the entry Vs of the interpretation result information storage unit 6 corresponding to the sentence S is presented to the user.

(B2) Entry P in the document information storage unit 2
When the content of the analysis state information C of s is “analysis failure”, a message notifying the analysis failure is presented to the user, and the applicable rule number information B of the entry Ts of the interpretation result information storage unit 6 and the unsolved information are also presented. The contents of the condition information D are presented to the user.

(B3) The result display procedure B is ended.

An example of the information presented by the result display procedure B will be shown in the description of the processing example later.

Subsequently, the operation script extraction processing performed by the operation history analysis unit 9 will be described.

The script extraction process is performed according to the following script extraction procedure C. Note that FIG. 13 is a flow chart showing the flow of the processing procedure C.

[Script Extraction Procedure C] (c1) This processing is started by (a7) of the processing procedure A.

(C2) Operation script information storage unit 10
To create a new entry Xs.

(C3) With reference to the contents of the operation history storage unit 8, the latest entry W whose operation history information A is "analysis failure (S)" within a certain search range H from the latest storage.
Search for s.

(C4) If the search for the entry Ws fails, the process proceeds to (c12).

(C5) Entry W in the operation history storage unit 8
Of the operation history from s to the latest description, those related to the knowledge / rule editing unit 4 are copied in the order of time into the operation script information C column of the entry Xs.

(C6) If the operation script information C of the entry Xs is empty, proceed to (c12).

(C7) The entry Vs of the interpretation result storage unit 6 corresponding to the sentence S to be processed is obtained. Also, the entry V
The content of the applicable rule number information C of is copied to the target interpretation rule number information A of the entry Xs.

(C8) The content of the unresolved condition information D of the entry Vs is changed to the script application condition information B of the entry Xs.
Copy to.

(C9) If the operation script information C of the entry Xs includes a node of area knowledge having an identifier consisting of the symbol "#" and a number, it is rewritten into a variable symbol based on the label. .

(C10) Through the column of the script application condition information B and the column of the operation script information C of the entry Xs, the surface expression that appears there is rewritten into a variable symbol based on the label.

(C11) Operation script information storage unit 1
When another entry having the same content as the entry Xs does not exist in 0, the process proceeds to (c13).

(C12) The entry Xs is erased.

(C13) The script extraction procedure C is ended.

The search range H used here is a constant designated in advance.

Next, the processing of the applied script search section 11, the confirmation dialogue section 12, and the operation script adaptation section 13 performed based on the following script processing procedure D will be described.

[Script Processing Procedure D] (d1) In (a8) of the processing procedure A, the processing for the sentence S is script-started by a request from the user.

(D2) The applied script search unit 11 makes an entry V of the sentence interpretation result storage unit 6 corresponding to the sentence S.
The adaptive rule number C of s and the contents of the unresolved condition information D, the target interpretation rule number information A of each entry of the operation script information storage unit 10, and the script adaptive condition information B
By comparing the contents of each of the above, the entry Xs of the operation script information storage unit 10 storing the script applicable to the target sentence S is searched.

(D3) If the entry Xs cannot be retrieved, the user is informed of it and the process proceeds to (d7).

(D4) The confirmation dialogue unit 12 refers to the entry Vs and presents a series of operations Ms in which the variable symbol portion of the contents of the operation script information C of the entry Xs is changed to a concrete value, and Inquire the user about the availability.

(D5) If there is a negative response from the user in (d4), the process proceeds to (d7).

(D6) The operation script adaptation section 14 executes a series of operations Ms.

(D7) The script adaptive processing procedure D is terminated, and the processing procedure A returns to (a8).

The above is the configuration of this apparatus and its function.

The processing in the interactive natural language analyzing apparatus described above will be described in more detail using a specific example.

First, it is assumed that the contents shown in FIG. 4 are stored in the document information storage unit 2. Here, the sentences of the sentence position information S21 and S22 have been successfully analyzed, and S1
The statements 1 and S31 have failed in the analysis. Also,
The sentence of S41 has not been analyzed.

Further, it is assumed that the area knowledge storage unit 3a, the expression correspondence rule storage unit 3b, and the interpretation rule storage unit 3c store the information shown in FIGS. 6, 7, and 8, respectively.

Furthermore, it is assumed that the interpretation result storage unit 6 has stored the information shown in FIG. 9 by the analysis up to that point using these knowledge.

It is also assumed that the operation history storage unit 8 has stored information up to the storage address W24 in FIG. Further, it is assumed that the content of the operation script information storage unit 10 is empty.

With the above settings, as an example, the user performs an operation for recovery of an analysis error in the sentence "Operate hydraulic pump" (hereinafter: sentence a) whose sentence position information is S11. The processing in this apparatus will be described. Note that FIG. 14 illustrates the process of extracting the operation script from the operation on the sentence a.

1: As a result of the analysis of the sentence a, this apparatus notifies the user that there is an analysis error, and the user uses the knowledge / rule editing section 4 to express the area knowledge storage section 3a. It is assumed that the correspondence rule storage unit 3b is operated twice, for a total of four times. Then, this series of operations is stored in the entries W25 to W29 of the operation history storage unit 8 shown in FIG. 10 by the configuration of the present apparatus described above.

2: Then, the user re-analyzes the sentence a, and the knowledge added in (1) succeeds in the analysis this time, and the result display unit 12 displays, for example, FIG.
A message as shown in (a) is presented to the user.

3: Then, in step (a7) of the processing procedure A, an attempt is made to extract an operation script for the correction operation performed by the user.

4: The script extraction procedure C is activated,
At (c2), a new entry Xa is created in the operation script information storage unit 10.

5: At (c3), W22 of FIG. 10 is obtained,
(C5) Information related to the operation of the internal knowledge / rule of the operation history of W22 to W29 is copied to the column of the script information C of the entry Xa.

6: The entry V11 in FIG. 9 is searched at (c6), and the target interpretation rule information A is copied to the entry Xa.

7: At (c8), the unsolved condition information D of the entry Vs is copied to the entry Xa.

8: At (c9), the label of the domain knowledge is rewritten to the variable symbol.

9: At (c10), the variable symbol of the surface expression is rewritten.

10: After the inspection of (c11), the operation script (Xa) is extracted from the correction operation of the parsing error of the sentence a by the above procedure.

Further, regarding the sentence of "fix control damper" of the sentence position information S31 of the document information storage unit 2 shown in FIG. 2, it is assumed that the user requests the retrieval of the applicable script. The usage process will be described.

11: (a8) of processing procedure A for sentence b
Then, the script processing procedure D is activated at the request of the user.

12: By the search of (d2), the entry Xa is searched from the operation script information storage section 10.

13: At (d4), as shown in FIG. 15B, the operation procedure in which the operation script of the entry Xa is applied to the sentence b is shown, and the user is inquired whether or not the application is possible.

14: In response to this, the user gives an affirmative response, and a series of operations consisting of knowledge, knowledge registration, and expression registration are automatically performed, and the area knowledge storage unit 3a and expression correspondence rule storage are carried out. The information shown in FIGS. 16A and 16B is additionally registered in the section 3b.

15: Subsequently, the reanalysis of the sentence b is performed, and the result analysis of the above operation is successful.

According to the present apparatus thus configured, the procedure of the operation performed by the user for solving the problem such as the parsing error in the natural language analysis can be automatically extracted and registered. It will be possible.

Furthermore, by using the registered operation procedure, it is possible to solve the same problem that occurs thereafter by one operation.

This solves the conventional problem that the user had to repeatedly perform the same operation in using the computer system using the natural language analysis, and reduces the burden on the user. It is possible to avoid a decrease in utilization efficiency.

Further, since the present invention provides an effective part for solving a parsing error in natural language analysis, it is not necessary to use a restricted language having a strong restriction that does not cause a parsing error. As a result, it is possible to configure a natural language analysis device that does not impair the flexibility and efficiency of natural language.

The present invention is not limited to the examples shown in the above embodiments.

In the above-described embodiment, the user is requested to search for a script applicable to the solution of the parsing error. However, the applicable script search unit 11 is expanded so that this device It is also possible to automatically perform the processing and, when there is an applicable script, propose the application of the script to the user through the confirmation dialogue unit 12 from this apparatus.

When a plurality of adaptable scripts are searched for in the applicable script in the applicable script retrieving section 11, the operating procedure in which the operation script information is integrated is used. It is also possible to present it to the user and inquire whether it is applicable.

If only some of the adaptive conditions of the script have not been achieved, only the operation procedure related to the unachieved condition is extracted from the contents of the operation script information of the script and presented to the user. However, it is also possible to inquire whether or not the adaptation is possible.

Alternatively, even when the adaptive script search unit 11 fails in the search for an adaptable script, the script is expanded and interpreted by removing a part of the script application condition, and the adaptation is performed. It is also possible to inquire the user whether or not.

Further, in the above-described embodiment, the processing relating to the additional registration of the domain knowledge and the expression correspondence rule has been mainly described. For example, the registration change of the grammar rule and the interpretation rule by the processing using the syntactic analysis result is performed. It can also be applied to proposals and proposals of rewriting candidates for the original text.

Further, in the above-mentioned embodiment, the proposed operation is presented to the user by the information embodying the contents of the operation history, but the presentation of this information is expanded to be performed in Japanese sentences. It is also possible.

For each script that this apparatus has proposed to apply to the user, the result of the response from the user to the proposal is stored, and a higher success rate (the degree to which the adaptive response is obtained is high) is stored. It is possible to make more effective proposals by preferentially selecting scripts.

In the present embodiment, the applicability of the script is determined by inquiring the user through the confirmation dialogue unit about the adaptation of each script. However, the applicability of the script is automatically determined. It is also possible to configure. For example, based on the judgment criteria such as the success rate described above, for a script that is judged to be sufficiently valid for application, the application is automatically determined without any inquiry to the user, and the processing is performed. It is also possible to further reduce the burden on the user.

Since this apparatus performs the processing based on the script, it is possible to incorporate a redone operation according to the script.

The third embodiment of the invention will be described with reference to a document processing device.

FIG. 17 is a block diagram of the entire document processing apparatus 100 which is an embodiment of the present invention.

This system includes a document editing unit 101 for inputting or editing a document, a language dictionary storage unit 102 referred to in the sentence analysis unit, a target knowledge storage unit 103, a sentence analysis unit 104 for analyzing an input document, and a It is composed of a sentence relation processing unit 105 for storing and searching the relation of each sentence and an analysis result storage unit 106 for accumulating the output result of the analysis unit.

The structure of each part will be described below.

The document editing section 101 is a section for inputting or editing a document. The appearance of the screen display is shown in FIG. 20
1 is a document display unit, 202 is an end button, 203 is an analysis button, 204 is a partial analysis button, 205 is a dependency relationship button, 206 is a language dictionary dependency relationship display unit, 207 is a target knowledge dependency relationship display unit, and 208 is a language dictionary. Error dependency relationship display unit 209 is a target knowledge error dependency relationship display unit 210
Is an auxiliary display / input section. When the user inputs documents from the keyboard, the documents are sequentially displayed on the document display unit 201.
When the analysis button 203 is clicked with the mouse, the input document is output to the sentence analysis unit 104. Further, when an arbitrary sentence is clicked with the mouse and then the dependency relation button 205 is clicked with the mouse, an inquiry is made to the sentence relation processing unit 105 using the sentence number of the sentence as a parameter, and the dependence is obtained based on the obtained information. The word or knowledge that causes the relationship is a language dictionary dependency relationship display unit 206, a target knowledge dependency relationship display unit 207, a language dictionary error dependency relationship display unit 208,
It is displayed on the target knowledge error dependency display unit 209. When you click one of them with the mouse, the sentence that has a dependency is underlined, and the word that causes the dependency is highlighted. At this time, when the partial analysis button 204 is clicked with the mouse, only the sentence with the underline displayed is output to the sentence analysis unit 104. At this time, the auxiliary display / input unit 2
When a word is input in 10 and the return key is pressed, the highlighted word in the document is replaced with the input word.
When an error occurs during the document analysis, the error message sent from the sentence analysis unit 104 is displayed as an auxiliary display.
It is displayed on the input unit 210. When the end button 202 is clicked with the mouse, the document editing ends.

The language dictionary storage unit 102 stores knowledge such as a dictionary for analyzing an input document, grammar, domain knowledge, and interpretation knowledge, that is, language information. It stores three tables for morphological analysis, syntax analysis, and semantic analysis.

The morphological analysis table stores information on a plurality of words, and the information on each word consists of the number of the word, natural language expression, part of speech, and utilization information. FIG. 19 shows an example of the contents of the morphological analysis table.

The syntactic analysis table means a rewriting rule in the phrase structure grammar, and an operation of rewriting the symbol string on the right side that has been successfully matched with the part-of-speech string in the sentence to the symbol string on the left side is performed. FIG. 20 shows an example of the contents of the syntax analysis table.
Shown in.

In the semantic analysis table, rules for generating a semantic structure based on the syntax analysis result are described.
It is referred to in the semantic analysis process of the sentence analysis unit 104. FIG. 21 shows an example of the contents of the semantic analysis table.

The target knowledge storage unit 103 stores the semantic information for analyzing the input document and is referred to in the semantic analysis process of the sentence analysis unit 104. An example of the contents stored in this is shown in FIG.

The morphological analysis table corresponds to a dictionary, the syntax analysis table corresponds to grammar, the target knowledge storage unit 103 corresponds to area knowledge, and the semantic analysis table corresponds to interpretation knowledge. Further, the target knowledge storage unit 103 is independent of the language dictionary storage unit 102, but these two storage units may be one.

The sentence analysis section 104 is a section for analyzing the input document output from the document editing section 101. This has a sentence number register that stores which sentence in the input document the sentence under analysis corresponds to.

The flow of processing in the sentence analysis unit will be described with reference to the flowchart of FIG.

At step 701, sentences are taken out one by one from the input document. At this time, the number of sentences in the entire document is stored in the sentence number register.

In the morphological analysis processing step 702,
The input sentence is divided into words by referring to the information in the morphological analysis table in the language dictionary storage unit 102, and the number, original form, part of speech, and utilization information of the word in the input sentence are obtained. When the part-of-speech of each word is any of a noun, a verb, and a number, the number of each word and the content of the sentence number register are combined and output to the sentence relation processing unit 105.

In the morphological analysis inspection processing step 703, it is examined whether or not there is an unknown word in the input sentence. The unknown word is a word that does not exist in the morphological analysis table in the language dictionary storage unit 102. If there is an unknown word,
The word and the content of the sentence number register are paired and output to the sentence relation processing unit 105, and an error message generation processing step 7
Move to 07.

In the syntax analysis processing step 704, by referring to the information of the syntax analysis table in the language dictionary storage unit 102, the syntax of the input sentence is obtained based on the word obtained in the morphological analysis processing step 702 and its part-of-speech information. To parse and generate the syntactic structure of the input sentence.

Specifically, a syntax structure having a tree structure is generated from the information of the syntax analysis table in the language dictionary storage unit 102 according to the following rules 1 to 3.

1. Of the terms on the right side, the term with * at the end is the parent node.

2. Among the terms on the right side, the term with + at the end is the arc name. When there is no term with + at the end, the term of the arc name is the arc name.

3. Of the terms on the right side, terms other than 1 and 2 are child nodes.

In the semantic analysis processing step 705,
Information on the semantic analysis table in the language dictionary storage unit 102,
Also, the syntactic structure of the input sentence is converted into a semantic structure by referring to the information in the target knowledge storage unit 103 and output to the analysis result storage unit 106. At this time, the referenced knowledge number and the contents of the sentence number register are paired and output to the sentence relation processing unit 105.

A detailed description will be given below. The set of parent node / arc / child node in the tree structure of the syntax analysis result is collated with the elements in each row of the table in FIG. 21 in order.
Necessary and sufficient conditions for matching shall comply with the following 1 and 2.

1. The set of parent node / arc / child node in the tree structure matches the set of parent node / arc / child node in FIG. However, blank items are optional.

2. The parent knowledge node / knowledge arc / child knowledge node terms in FIG. 21 must have rows that match the set of knowledge node 1, knowledge arc, and knowledge note 2 in FIG. In that case, in FIG. 22, knowledge node 1 must correspond to node 1 and knowledge node 2 must correspond to node 2. Here, if there is a term including a "?" Symbol in FIG. 22, it is assumed that "?" Matches any character string. However, blank items in FIG. 21 are optional.

The matching process according to this rule is performed,
If there is a rule that has been successfully matched in the semantic analysis rule of FIG. 21, the arc in the tree structure of the syntax analysis result is shown in FIG.
The process of replacing with the item of the new arc of the matched rule in 1 is performed. By performing the above processing, the syntactic structure of the input sentence is converted into a semantic structure.

In the semantic analysis check processing step 706, when the relation between the nodes in the input sentence does not exist in the target knowledge, the word of the parent node, the word of the child node and the contents of the sentence number register are paired. It is output to the sentence relation processing unit 105, and the process proceeds to the error message generation processing step 707.

Error message generation processing step 707
In step (1), an error message is generated based on the information regarding the processing and location where the error has occurred, and is output to the document editing unit 101. Here, the error message is generated using the table shown in FIG. The part enclosed by [] is replaced with the word that caused the error in each process.

In step 708, it is checked whether all the input documents have been parsed. If there is an input sentence that has not been analyzed yet, the process returns to step 701. If all the input sentences have been analyzed, the process ends.

The sentence relation processing unit 105 stores and retrieves the dependency relation of each sentence in the input document. This is a language dictionary dependency list, a target knowledge dependency list, a language dictionary error dependency list, for storing dependency relationships between sentences,
It has a target knowledge error dependency list.

When a pair of a dictionary number and a sentence number register is input from the morphological analysis processing of the sentence analysis unit 104, that information is registered in the language dictionary dependency list. At this time, if the line of the dictionary number already exists, the information of the sentence number register is added to the line. FIG. 25 shows an example of the contents of the language dictionary dependency list.

When a set of a knowledge number and a sentence number register is input from the semantic analysis processing of the sentence analysis unit 104, that information is registered in the target knowledge dependency relationship list. At this time, if the line of the knowledge number already exists, the information of the sentence number register is added to the line. FIG. 26 shows an example of the contents of the target knowledge dependency relationship list.

When a set of the unknown word and the sentence number register is input from the morphological analysis inspection process of the sentence analysis unit 104, the information is registered in the language dictionary error dependency list. At this time, if the line including the unknown word already exists, the information of the sentence number register is added to the line. FIG. 27 shows an example of the contents of the language dictionary error dependency list.

From the semantic analysis inspection processing of the sentence analysis unit 104,
When a pair of the word of the parent node, the word of the child node and the sentence number register is input, that information is registered in the target knowledge error dependency relationship list. At this time, if a line including the word of the parent node and the word of the child node already exists, the information of the sentence number register is added to the line. FIG. 28 shows an example of the contents of the target knowledge error dependency relationship list.

When an inquiry is received from the document editing unit 101 using the sentence number as a parameter, first the line containing the sentence number is extracted from the language dictionary dependency list,
From the dictionary number included in each line, the morphological analysis table in the language dictionary storage unit 102 is referenced to extract the word having that dictionary number, and the document editing unit 1 with the list of sentence numbers.
Output to 01. FIG. 29 shows an example of the output information.
Also, the line containing the sentence number is extracted from the target knowledge dependency relationship list, and from the knowledge numbers contained in each line, the target knowledge storage unit 103 is referred to and the knowledge nodes 1 and 2 having that knowledge number are identified. It is taken out and output to the document editing unit 101 together with the list of sentence numbers. An example of the output information is shown in FIG. Also, the line containing the sentence number is extracted from the language dictionary error dependency list, and the list of words and sentence numbers included in each line is output to the document editing unit 101. An example of the output information is shown in FIG. Also, the line containing the sentence number is extracted from the target knowledge error dependency relationship list, and the list of the word of the parent node, the word of the child node and the sentence number included in each line is output to the document editing unit 101. FIG. 32 shows an example of the output information.

The analysis result storage unit 106 includes the sentence analysis unit 104.
The analysis result output from is stored.

The operation of the document processing apparatus 100 according to the embodiment of the present invention will be specifically described below with reference to examples.

Here, the operation when the user inputs the document of FIG. 33 will be described.

Here, it is assumed that the language dictionary storage unit 102 stores the contents of FIG. 19 as a morphological analysis table, the contents of FIG. 20 as a syntax analysis table, and the contents of FIG. 21 as a semantic analysis table. . Further, it is assumed that the target knowledge storage unit 103 stores the contents of FIG.

The user edits the document shown in FIG. 33 with the document editing unit 101.
When the input document is input to, the input document is analyzed by the sentence analysis unit 104.

First, by the processing of step 701, "when the temperature of the sensor is 0 ° C., the compressor is turned on".
Is taken out as the first sentence. The content of the sentence number register becomes "1".

At step 702, morphological analysis processing is performed, and the morphological analysis result shown in FIG. 34 is obtained.
Further, the dictionary number and the content of the sentence number register (“1” in this case) for each word whose part of speech is noun, verb, or number are output to the sentence relation processing unit 105. The contents of the language dictionary dependency list of the sentence relation processing unit 105 after step 702 are as shown in FIG.

At step 703, a morphological analysis inspection process is performed. Here, since all the words in the input sentence are present in the morphological analysis table, it is determined that the morphological analysis is correctly performed, and the process proceeds to step 704.

In step 704, the syntax analysis process is performed, and the syntax analysis result shown in FIG. 36 is obtained.

In step 705, the semantic analysis process is performed, and the semantic analysis result shown in FIG. 37 is obtained.

At step 706, a semantic analysis inspection process is performed. Here, for each node in the input sentence, since the relation exists in the target knowledge, it is determined that the semantic analysis has been correctly performed, and the process proceeds to step 708.

The contents of the target knowledge dependency relationship list of the sentence relationship processing unit 105 after the step 706 is as shown in FIG.

Similar operations are performed for the remaining two sentences. After the analysis of the entire document is completed, the contents of the language dictionary dependency relationship list of the sentence relationship processing unit 105 are as shown in FIG. 35, and the contents of the target knowledge dependency relationship list are as shown in FIG. 38.

Next, after the analysis operation described above is completed,
The case where the user performs the operation of searching the dependency relation of the sentence will be specifically described.

In the document editing section 101, the user reads the first sentence "When the sensor temperature is 0 ° C., the compressor is turned on.
N ", And click the dependency button 20
A case where 5 is clicked with a mouse will be described.

Since the clicked sentence is the first sentence, the document editing unit 101 inquires the sentence relation processing unit 105 about the dependency relation by setting the sentence number to 1.

The sentence relation processing unit 105 first extracts lines in the language dictionary dependency list that include "1" in the sentence number portion. In this case, the rows of dictionary numbers 1, 2, 3, 8, 12 are extracted. Next, the word having the dictionary number is extracted from the dictionary number included in each line by referring to the morphological analysis table in the language dictionary storage unit 102. For example, the word "compressor" is obtained from the dictionary number 1. Then, it outputs to the document editing unit 101 together with the list of sentence numbers. The output information is as shown in FIG.

Next, in the target knowledge dependency relationship list, a line including "1" in the sentence number portion is extracted. In this case, the rows of knowledge numbers 1, 4, and 5 are extracted. From the knowledge numbers included in each row, the target knowledge storage unit 103 is referenced and the knowledge nodes 1 and 2 having that knowledge number are extracted. For example, from knowledge number 1 node 1 "compressor", node 2
"ON" is obtained. Then, it outputs to the document editing unit 101 together with the list of sentence numbers. The output information is shown in Figure 3.
It becomes like 0.

Since the language dictionary error dependency list and the target knowledge error dependency list have no contents, no information is output to the document editing unit 101.

The document editing unit 101 includes a sentence relation processing unit 105.
The screen is displayed based on the information obtained from. in this case,
The screen shown in FIG. 18 is displayed.

[0251] Here, for example, when the user clicks "Sensor: Temperature" on the target knowledge dependency relationship display section 207, from the information obtained from the statement relationship processing section 105, the dependent statements are the first sentence and the second sentence. Since it is understood that it is a sentence, an underline is displayed in the first sentence and the second sentence, "sensor" of each sentence
And “Temperature” are highlighted.

When the user clicks the partial analysis button 204, only the first and second sentences are output to the sentence analysis unit 104,
Analysis is performed.

[0253] Further, the user selects the target knowledge dependency relationship display unit 2
When "Compressor: ON" is clicked at 07, the first sentence and the third sentence are underlined, and "Compressor" and "ON" of each sentence are highlighted. At this time, when "compressor: operation" is input to the auxiliary display / input unit 210 and the return key is pressed, "ON" that is highlighted in the first and third sentences is replaced with "operation". Here, when the user clicks the partial analysis button 204, only the first and third sentences are output to the sentence analysis unit 104 and analysis is performed.

The operation when an error occurs when the input document is analyzed will be described. Here, the operation when the user inputs the document of FIG. 39 will be described.

The user edits the document shown in FIG. 39 by the document editing unit 101.
When the input document is input to, the input document is analyzed by the sentence analysis unit 104.

In step 701, "When the temperature of the sensor is 0 ° C., start the compressor." Is taken out as the first sentence. The content of the sentence number register becomes "1".

In step 702, morphological analysis processing is performed, and the morphological analysis result shown in FIG. 40 is obtained. Further, the dictionary number and the content of the sentence number register (“1” in this case) for each word whose part of speech is noun, verb, or number are output to the sentence relation processing unit 105. The contents of the language dictionary dependency list of the sentence relation processing unit 105 after step 702 are as shown in FIG.

At step 703, a morphological analysis inspection process is performed. Here, since the word "start" in the input sentence does not exist in the morphological analysis table, it is determined to be an unknown word and is output to the sentence relation processing unit 105 together with the content of the sentence number register (in this case, "1"). Then, the process proceeds to step 707. The contents of the language dictionary error dependency list of the sentence relation processing unit 105 after step 703 are as shown in FIG.

At step 707, an error message is generated. Here, since an error has occurred in the morphological analysis inspection process 703, a process of replacing the [unknown word] portion of the morphological analysis template of the error message template 8 with the word “start”. FIG. 43 shows an example of the generated error message. The generated error message is output to the document editing unit 101, and the process proceeds to step 708.

In step 708, it is checked whether all input sentences have been parsed. Since there is an input sentence that has not been analyzed yet, the process proceeds to step 701.

By the processing of step 701, "The compressor stops when the sensor temperature is -10 degrees."
Is taken out as the second sentence. The content of the sentence number register is "2".

The morphological analysis process 702 to the semantic analysis process 7
The same process is performed up to 06.

At step 706, a semantic analysis inspection process is performed. Here, since the relation between "stop" and "compressor" in the input sentence does not exist in the target knowledge, each word serves as a parent node and a child node together with the contents of the sentence number register (in this case, "1") and the sentence relation. Processing unit 10
5, and the process proceeds to step 707. Step 706
The contents of the target knowledge error dependency relationship list of the sentence relationship processing unit 105 after the end are as shown in FIG.

At step 707, an error message is generated. Here, since an error has occurred in the semantic analysis inspection processing 706, the words “stop” and “parent node” and “child node” in the error analysis template of FIG.
The process of replacing with "compressor" is performed. An example of the generated error message is shown in FIG. The generated error message is output to the document editing unit 101, and the process proceeds to step 708.

Similar operations are performed for the remaining two sentences. After the analysis of the entire document is completed, the contents of the language dictionary error dependency relationship list of the sentence relationship processing unit 105 are as shown in FIG. 27, and the contents of the target knowledge error dependency relationship list are as shown in FIG.

The search for the dependency relation can be performed in the same manner as the example of the dependency relation regarding the language dictionary and the target knowledge.

In the embodiment thus constructed,
It becomes possible to analyze the sentences that have a dependency relationship in a batch, maintain the consistency of the document, and reduce the workload of the user.

In this embodiment, the document editing unit 101 is a document input device having a user interface as shown in FIG. 18, but the user interface is not limited to this example. An input device having only a keyboard may be used. Further, the document may be input from a magnetic disk, a magnetic tape, an OCR or the like.

In the present embodiment, the sentence relation processing section 105 is not provided with a display device, but a display device such as a CRT display for displaying the dependency relation may be provided. Further, a display device for displaying the analysis result may be provided at the subsequent stage of the analysis result storage unit 106.

As described above, the present invention is not limited to the above-mentioned embodiments, and can be carried out by appropriately modifying it without departing from the scope of the invention.

[0271]

The natural language processing apparatus of the first and second inventions automatically extracts and registers the procedure of the operation performed by the user in order to solve a problem such as a parsing error in the natural language analysis. can do. Further, by using the registered operation procedure, it is possible to solve the same problem that occurs thereafter by one operation.

This eliminates the need for the user to repeatedly perform the same operation when using the computer system utilizing the natural language analysis, reducing the burden on the user and avoiding the decrease in utilization efficiency. In addition, in order to provide an effective part for solving parsing errors in natural language analysis, it is not necessary to use a restricted language that has a strong restriction that does not cause parsing errors. It is possible to configure a natural language analysis device that does not impair flexibility and efficiency.

The natural language processing apparatus according to the third aspect of the invention makes it possible to collectively perform operations on sentences having a dependency relationship, thereby maintaining the consistency of documents and reducing the amount of work of the user. It will be possible.

[Brief description of drawings]

FIG. 1 is a specification document analysis system.

FIG. 2 is an example of a syntactic analysis result for sentence a.

FIG. 3 is a configuration of an embodiment of an interactive natural language analysis device according to the present invention.

FIG. 4 shows an example of the contents of the document information storage unit 2.

FIG. 5 is an example of domain knowledge represented by a semantic network.

FIG. 6 shows an example of the contents of a domain knowledge storage unit 3a.

FIG. 7 is an example of contents of an expression correspondence rule storage unit 3b.

FIG. 8 shows an example of the contents of an interpretation rule storage unit 3c.

9 is an example of contents of an interpretation result storage unit 6. FIG.

FIG. 10 shows an example of the contents of the operation history storage unit 8.

FIG. 11 shows an example of the contents of the operation script information storage unit 10.

FIG. 12 is a flow chart of processing procedure A.

FIG. 13 is a flow chart of processing procedure C.

FIG. 14 is an explanatory diagram of script extraction processing.

FIG. 15 is an example of information presented to the user.

FIG. 16 is an example of information automatically registered by applying a script.

FIG. 17 is a block diagram showing a schematic configuration of an example.

FIG. 18 is an example of a screen display of the document editing unit.

FIG. 19 shows an example of a morphological analysis table.

FIG. 20 shows an example of a syntax analysis table.

FIG. 21 is an example of a semantic analysis table.

FIG. 22 is an example of contents of a target knowledge storage unit.

FIG. 23 is a flowchart showing a processing flow of a sentence analysis unit.

FIG. 24 is a content of an error message template.

FIG. 25 is an example of the contents of a language dictionary dependency list.

FIG. 26 is an example of the contents of the target knowledge dependency relationship list.

FIG. 27 is an example of the contents of a language dictionary error dependency relationship list.

FIG. 28 is an example of the contents of a target knowledge error dependency relationship list.

FIG. 29 is an example of dependency relationship information regarding a language dictionary output from the sentence relationship storage unit.

FIG. 30 is an example of dependency relationship information regarding target knowledge output from a sentence relationship storage unit.

FIG. 31 is an example of dependency relationship information regarding a language dictionary error output from the sentence relationship storage unit.

FIG. 32 is an example of dependency relationship information regarding a target knowledge error output from a sentence relationship storage unit.

FIG. 33 shows an example of an input document.

FIG. 34 is an example of a morpheme analysis result in the sentence analysis unit.

FIG. 35 is an example of the contents of a language dictionary dependency list.

FIG. 36 is an example of a syntax analysis result in the sentence analysis unit.

FIG. 37 is an example of a semantic analysis result in the sentence analysis unit.

FIG. 38 is an example of the contents of the target knowledge dependency relationship list.

FIG. 39 is an example of an input document.

FIG. 40 is an example of a morpheme analysis result in the sentence analysis unit.

FIG. 41 is an example of the contents of a language dictionary dependency list.

FIG. 42 is an example of contents of a language dictionary error dependency relationship list.

FIG. 43 is an example of an error message.

FIG. 44 is an example of the contents of a target knowledge error dependency relationship list.

FIG. 45 is an example of an error message.

[Explanation of symbols]

 1 Document Editing Section 2 Document Information Storage Section 3 Knowledge / Rule Storage Section 4 Knowledge / Rule Editing Section 5 Sentence Analysis Section 6 Interpretation Result Storage Section 7 Result Display Section 8 Operation History Storage Section 9 Operation History Analysis Section 10 Operation Script Information Storage Section 11 Applied Script Search Unit 12 Confirmation Dialogue Unit 13 Script Adaptation Unit 101 Document Editing Unit 102 Language Dictionary Storage Unit 103 Target Knowledge Storage Unit 104 Sentence Analysis Unit 105 Sentence Relation Processing Unit 106 Analysis Result Storage Unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Katsumi Tanaka 1-1-30 Oyonanaka, Kita-ku, Osaka-shi, Osaka Umeda Sky Building Tower West Co., Ltd. Toshiba Kansai Branch Office

Claims (3)

[Claims] 1. A sentence input unit for inputting a natural language sentence, a document information storage unit for storing the natural language sentence input by the sentence input unit, and a natural language sentence stored in the document information storage unit. An analysis resource information storage unit for storing analysis resource information such as analysis rules, dictionary items, and domain knowledge for analyzing; an analysis resource information creation / editing unit for creating / editing the analysis resource information; A sentence analysis unit that analyzes the natural language sentence stored in the document information storage unit using the analysis resource information stored in the storage unit, and an analysis result storage that stores the analysis result information obtained by this sentence analysis unit Section, an analysis result display section that presents the analysis result information to the user, an operation history storage section that stores a history of operations performed by the analysis resource information creation / editing section, and an operation history record An operation history analysis unit that extracts operation script information, which is a template of a series of operation procedures, by using the operation history information stored in the analysis unit and the analysis result information stored in the analysis result storage unit; An operation script information storage unit that stores operation script information obtained by the analysis unit, and an operation script stored in the operation script information storage unit for the analysis result information of the processing target document stored in the analysis result storage unit An application script search unit for searching applicable operation script information from information, and an operation for adding or changing the analysis resource information stored in the analysis resource information storage unit according to the operation script information. A natural language processing apparatus having a script adaptation unit. 2. A confirmation dialogue unit for inquiring of a user as to whether or not to apply the operation script information to the document information obtained by the application script search unit and receiving a result thereof, the operation script adaptation unit: And applying or changing the operation script information to the document information according to the response of the user obtained by the confirmation dialogue unit, and adding or changing the analysis resource information stored in the analysis resource information storage unit according to the operation script information. The natural language processing device according to claim 1, wherein 3. A document editing unit for inputting and editing a document, a language dictionary storing unit for storing knowledge such as a dictionary for analyzing an input document, grammar, domain knowledge, and interpretation knowledge, and the document editing unit. The same knowledge necessary for analysis is shared from the sentence analysis unit that analyzes each sentence input to the sentence based on the knowledge stored in the language dictionary storage unit and the knowledge of each sentence analyzed by this sentence analysis unit. A natural language processing device characterized in that a sentence is composed of a sentence relation processing unit that has a dependency relation.