JP2008145769A - Interaction scenario creation system, its method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently create an interaction scenario which is required for a natural language interaction system for carrying out operation for a device and a computer program based on user's request. <P>SOLUTION: The interaction scenario is formed from a document in which functions and operation methods of the device and the computer program are described in natural language. First, in a document item extraction section 106, only object items are extracted from the document according to a predetermined rule. Then, in an extraction item analysis section 108, a content of the extracted item is analyzed and required information is extracted, according to an expression rule which is specific to an object document. A command and a parameter corresponding to the extracted information are obtained from a developer. In an interactive scenario creation section 110, the interactive scenario is created, based on the information extracted from the document, the information obtained from the developer, and the rule which is provided beforehand. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a dialogue scenario in which information related to interaction with a user is described in a natural language dialogue system in which a user interacts with the user through a natural language dialogue to operate a device on behalf of the user or provide a service to the user. This is related to a technique for easily creating a natural language dialogue system developer.

  A number of technologies related to a natural language dialogue system have been proposed that interact with a user in natural language and provide information and services required by the user.

  In a natural language dialogue system, information related to the interaction between the user and the system is stored as a database called a dialogue scenario, and a response to a question from the user is generated and a service is provided according to the dialogue scenario. However, in order to provide a variety of services and realize a flexible dialogue between the system and the user, it is necessary to assume a number of situations. Therefore, there is a need for a means to efficiently build a good-quality dialogue scenario. It becomes.

  In Patent Document 1, text data corresponding to a table portion is extracted from hypertext data, and the extracted table data is inserted into a scenario template to generate a voice conversation scenario.

  Further, in Patent Document 2, it is intended for a dialogue service with a pseudo personality, and is information for accepting and responding to a user's question based on the content of the speech or dialogue of the subject person's mail, copyrighted work, etc. Generate a basic document for interaction.

  On the other hand, Patent Document 3 aims to improve work efficiency and reduce work load when constructing a speech interface manually by preparing tools that can easily combine functions such as speech recognition and speech synthesis. .

JP 2004-334369 A JP 2005-196356 A JP-A-06-161704

  In the prior art, information on basic interaction between the user and the natural language dialogue system is given as a template, or based on the premise that information that is highly compatible with dialogue scenarios such as user speech and dialogue recording can be acquired in advance. Yes.

  On the other hand, when a request is received from a user and it is assumed that the natural language dialogue system operates a device such as an information appliance or a computer program on behalf of the user, the exchange between the user and the natural language dialogue system is an operation target. Since it takes various forms depending on the type of device or computer program, it is difficult to create a template in advance assuming all situations. In addition, there are generally few exchanges in the situation where a user asks another person to operate a device or a computer program, and since it is rarely recorded, the content of the user's remarks and dialogues It is very difficult to build a dialogue scenario based on it.

  In addition, the prior art also shows a technique for improving work efficiency when manually generating a conversation scenario. However, recently developed devices and computer programs have a wide variety of functions, so individual functions Assuming the detailed situation of interaction with the user every time and describing it as a dialogue scenario, even if an efficient tool is used, it requires a lot of labor for manual work and creates a dialogue scenario It is considered that there is a high possibility that a function to forget will occur.

  The present invention solves the above problems and efficiently constructs a dialogue scenario capable of realizing a natural language dialogue system capable of flexibly responding to user requests for various services provided by the device and computer program. A technology for constructing dialogue scenarios from documents such as operation manuals and instruction manuals provided to users by program developers and providers.

  In the present invention, in order to achieve the above-described problem and construct a dialogue scenario, first, only items related to the dialogue scenario are extracted from a document provided for the user according to a predetermined rule. The items to be extracted are the items corresponding to the user's request and the items corresponding to the detailed exchange required to execute the user's request, and the items that are listed in the heading or item that indicate the contents of the operation The item representing the operation procedure corresponds to this.

  Next, the contents of the extracted items are analyzed according to the specific expression used in documents provided for users, such as operation manuals and instruction manuals, and the units corresponding to user requests, their relationships, Extract information about items, operation targets, and operation contents for which information needs to be acquired. For the analysis of the expression form, a rule regarding the expression form is prepared in advance.

  When a natural language dialogue system operates a device or a computer program, it is necessary for the natural language dialogue system to transmit a command that can be interpreted by the device or the computer program and parameters associated therewith. The commands and parameters correspond to the contents of the operation extracted from the document, the operation target, and information input from the user, but are not usually described in the document provided to the user. For this reason, an interface is provided for the developer to supplementally input commands and parameters corresponding to the operation content and operation target extracted from the document and information input from the user. In addition, when the natural language dialogue system allows the user to input information, there are cases where selection is made from a plurality of options because the items supported by the device and the computer program are limited. Information about options in this case may not be described in the document provided for the user. For this reason, it is possible for the developer to input the above interface including such information.

  A dialog scenario is generated based on the items extracted from the document, information supplemented by the developer, and rules for generating a dialog scenario prepared in advance. In dialog scenario generation, information related to basic interaction in the dialog is generated by correcting and supplementing information extracted from the document based on rules, and further, response sentences and speech recognition grammar are generated. As a result, a dialogue scenario is generated.

  A description indicating the contents of the operation performed by the user described in a document provided for the user such as an operation manual or an instruction manual, for example, “turn on TV” in the case of a TV operation, It can be thought of as representing a request. In addition, these documents describe all the basic operations that a user performs to use a device or receive a service. For this reason, by generating a conversation scenario based on the contents of a document for users, a conversation scenario that can realize a natural language conversation system that can flexibly respond to user requests for various services provided by devices and computer programs. Efficient construction is possible.

  Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS.

  FIG. 1 is a conceptual block diagram showing a configuration of a dialogue scenario generation system and a natural language dialogue system using the dialogue scenario generation system according to the first embodiment. In FIG. 1, reference numeral 101 denotes a dialogue scenario generation system. An operation target related document 102 is used by the dialog scenario generation system 101 to generate a dialog scenario. The operation target related document 102 includes documents provided to users by developers and providers of devices and computer programs such as operation manuals and instruction manuals for devices such as information appliances and computer programs. Reference numeral 103 denotes a dialogue scenario generated by the dialogue scenario generation system 101 according to the present embodiment. 104 is a natural language dialogue system for performing a dialogue with a user in a natural language based on the contents of the dialogue scenario 103, performing an operation on an operation target based on a user request, and providing a service based on the operation target to the user. There are various input / output functions such as a voice recognition function, a voice synthesis function, and a text display function, and a dialog control function for realizing a dialog with a user using them. A natural language dialogue system that has an input / output unit that performs voice recognition and voice output, interacts with the user in natural language, and acts as an operation target device or computer program based on the user's request For example, a technique already known, for example, a technique disclosed in Japanese Patent Application Laid-Open No. 2001-188668 “Conversation processing apparatus and method, information processing apparatus, conversation processing system, and recording medium” can be used.

  Further, the operation target 105 in FIG. 1 is a device such as an information home appliance or a computer program operated by the natural language dialogue system 104, and assumes a television, a video recorder, a word processor, or the like. The natural language dialogue system 104 and the operation target 105 are connected using means generally used for operating the operation target, such as a wired or wireless network or an infrared remote controller.

  In the dialogue scenario generation system 101 of FIG. 1, the document item extraction unit 106 analyzes the structure of the operation target related document 102 based on the document structure analysis rule 107 and extracts items necessary for generating the dialogue scenario 103. The document structure analysis rule 107 describes information relating to the document format, keywords representing necessary items, and the like. The extracted item analysis unit 108 analyzes the contents of the items extracted by the document item extraction unit 106 based on the expression analysis rules 109, and extracts information necessary for the dialogue between the user and the natural language dialogue system 104. In the expression analysis rule 109, keywords and rules for classifying the extracted items and extracting information such as operation contents and operation targets are described. The dialogue scenario generation unit 110 generates the dialogue scenario 103 based on the result of the analysis performed by the extraction item analysis unit 108 and the dialogue scenario generation rule 111. In the dialogue scenario generation rule 111, a rule necessary for converting information obtained by analysis in the extraction item analysis unit 108 into a dialogue scenario and a rule for capturing information are described. In addition, the information input / output unit 112 is used to supplement the information lacked by the developer of the natural language dialog system when generating the dialog scenario.

  FIG. 2 is a configuration diagram when the dialogue scenario generation system 101 according to the present embodiment is realized on a commonly used computer. Reference numeral 201 in FIG. 2 denotes an information processing unit for executing various programs necessary for generating a dialogue scenario, and includes a normal CPU (Central Processing Unit). Reference numeral 202 denotes an input unit for inputting information required by the developer of the natural language dialogue system in the information input / output unit 112 in FIG. 1, and a well-known device such as a keyboard or a mouse can be used. In the information input / output unit 112 in FIG. 1, the display unit 203 serves as an output unit for prompting the developer of the natural language dialogue system to input information necessary for generating a dialogue scenario, and confirming the input information. A well-known device such as a monitor can be used.

  In FIG. 2, reference numeral 204 denotes a storage unit for storing various programs necessary for generating a dialogue scenario and information on the progress of processing. 205 is a document item extraction program for performing processing corresponding to the document item extraction unit 106 in FIG. 1, 206 is an extraction item analysis program for performing processing corresponding to the extraction item analysis unit 108, and 207 is This is a dialogue scenario generation program for performing processing corresponding to the dialogue scenario generation unit 110 and the information input / output unit 112, and is stored in the storage unit 204 and executed by the information processing unit 201 or the like. Further, the operation target related document 208, the document structure analysis rule 209, the expression analysis rule 210, the dialog scenario generation rule 211, and the dialog scenario 212 correspond to 102, 107, 109, 111, and 103 in FIG. Stored in the storage unit of the computer. Therefore, the storage unit 204 that stores various programs and the storage unit / storage unit that stores and stores them may be referred to as an information storage unit.

  FIG. 3 shows an example of the operation target related document assumed in this embodiment. FIG. 3 exemplifies a part of a TV operation manual. As operations for “watch TV (301)”, “turn on TV (303)”, “adjust volume” (310), “sound” Switching operation (315) "is possible, and the specific operation procedure is described for each item (304, 306, 307, 309, 311, 316, 321). Further, supplemental information as described in 302, 305, 308, 312, 313, 314, 317, 318, 319, 320, and 322 is described for operations and procedures. FIG. 3 is an example of an operation manual described only by characters, but a more general operation manual describes an operation procedure with figures and tables. However, the basic configuration is almost the same as the example shown in FIG.

  FIG. 4 shows the configuration of the document structure analysis rule 107 of the dialogue scenario generation system 101. In FIG. 4, format information 401 represents a format related to items necessary for generating a dialogue scenario in the operation target related document 102. That is, the document item extraction unit 106 extracts items corresponding to the format described in the format information 401 from the operation target related document 102.

  FIG. 5 shows information described in the format information 401. In FIG. 5, a hierarchy number 501 is a number that indicates which hierarchy item in the hierarchical structure of the document the extracted item is. For example, in FIG. 3, “2. Watch TV (301)” is the highest hierarchy, and the hierarchy number 501 is “1”. Since “(1) Press the button on the remote control (304)” is a layer below “2. Watch TV (301)” and “2.1 Turn on TV (303)”, the layer number is “3”. The type 502 in FIG. 5 represents the type of the corresponding item. “Heading” indicates an item indicating an operation classification such as “2. Watch TV (301)” or “2.1 Turn on TV (303)”, and “Procedure” indicates “(1) Remote control button An item representing a specific operation procedure such as “Press (304)” is shown. Other types of items can be added according to the structure of the operation target related document 102. The start format 503 and the end format 504 are conditions for determining whether or not an item in the operation target related document 102 is to be extracted. The description of the start format 503 is included in the start part of the item, and the end part is included in the end part. When the description of the end format 504 is included, the items in the operation target related document 102 are extracted.

  In the case of the line indicated by 505 in FIG. 5, if the notation “/ NO /.” Is included in the start part of the item, it becomes an extraction target. Here, the notation “/ NO /” represents an arbitrary number. In addition, “−” indicates that the format is not defined. As a result, “2. Watch TV (301)” is an extraction target under the condition 505. Similarly, in the condition 506, “2.1 Turn on the television (303)” or the like is selected, and in the condition 507, “(1) Press the button on the remote controller (304)” or the like becomes the extraction target.

  Further, an item exclusion keyword 402 in FIG. 4 describes a keyword included in an item to be excluded exceptionally among the items extracted by the format information 401. FIG. 6 shows an example of keywords described in the item exclusion keyword 402. In the case of the example shown in FIG. 6, when keywords such as “Introduction (601)” and “Appendix (602)” are included in the item, even if the condition described in the format information 401 is satisfied, the extraction target And not.

  FIG. 7 shows an example of the configuration of the expression analysis rule 109. The expression analysis rule 109 includes an operation content extraction rule 701, a condition extraction rule 702, an operation type determination rule 703, an operation object extraction rule 704, and an option detection rule 705.

FIG. 8 shows an example of the format of the operation content extraction rule 701. The operation content extraction rule 701 is a rule for detecting an expression (operation content) representing an operation such as “push” or “input” from items extracted from the operation target related document. . In the column indicated by reference numeral 801 in FIG. 8, a pattern of expression representing the operation content appearing in the item extracted from the operation target related document is described. In the column 802, an operation content extraction method is described. For example, in the pattern shown in 803, “*” indicates that an arbitrary expression is described, and the description “(verb)” indicates that a word that is a verb is described in that portion. As a result, in the pattern 803, when the sentence ends with a verb, it is determined that the operation content is included. Further, in the description of the operation content 803, it means that a verb detected by the pattern 803 is detected as the operation content. Therefore, the item extracted from the operation target related document is
“2. Watch TV”
In this case, the verb “see” is detected as the operation content. In the pattern 804 in FIG. 8, the verb “see” is detected as the operation content when the sentence ends with the expression “(verb) + function”, for example, “view function”. Also, in 805 and 806, sub-nouns such as “recording” and “playback” are targeted. In the case of 805, “reserve and record” and in the case of 806 “recording function”. , It is determined that the operation content is included. Further, in 805 and 806, “record” is detected as the operation content in accordance with the description in the right column. In the operation content extraction rule shown in FIG. 8, specific expressions and parts of speech are included in the pattern.

FIG. 9 shows an example of the format of the condition extraction rule 702. The condition extraction rule is a rule for detecting a part representing a condition for executing an operation from items extracted from the operation target related document. In addition, since it is often described in a similar format, it also includes rules for extracting the part corresponding to the content (user request) requested by the user as a request from the item representing the operation procedure. . In the rule shown in FIG. 9, it is assumed that a portion between the start keyword 901 and the end keyword 902 is detected as a condition. When the keyword is “−”, it indicates that the corresponding keyword does not exist. In 903, information indicating whether the part corresponds to a part indicating a condition or a part corresponding to a user request is described. In FIG. 9, “condition” corresponds to the former, and “user request” corresponds to the latter. In each keyword, a combination of a part of speech and a specific word can be described as a pattern as in the operation content extraction rule shown in FIG. For example, the item extracted from the operation target related document is “Switch voice [in case of BS]”
In this case, the “in case of BS” part is detected as a part representing a condition according to the rule shown at 904. Also, if the item extracted from the operation target related document is “If it is terrestrial, press the [Main Audio / Sub Audio / Main Sub Audio] button”
In this case, the part “in the case of terrestrial wave” conforms to the rule “in the case of (noun) +” shown in 905, and the part “terrestrial wave” is detected as a part representing a condition. Here, the part extracted as the part representing the condition is a part other than the specific keyword in the start keyword and the end keyword among the parts that match the rule. That is, in the above example, “terrestrial waves” that match “(part of speech)” in the rule are extracted. Further, the extracted part may be detected including the start keyword and the end keyword.

  FIG. 10 shows an example of the format of the operation type determination rule 703. The operation type determination rule 703 is a rule for determining the type of operation content detected from items extracted from the operation target related document, and classifies the operation content as either “automatic execution” or “option”. . Here, “automatic execution” represents an operation that can be executed unconditionally when requested by the user, such as an operation of pressing a specific button. “Option” represents an operation that can be executed only after the user selects one of a plurality of options or after the user performs some input such as a character string or a symbol string. In FIG. 10, keywords indicating operation contents are described in a column 1001, and types for the respective keywords are described in 1002. The rules 1003, 1004, and 1005 describe the operation contents classified as “automatic execution”, and the rules 1006, 1007, and 1008 describe the operation contents classified as “options”.

FIG. 11 shows an example of the format of the operation target extraction rule 704. The operation target extraction rule 704 is a rule for detecting a part representing an operation target such as a button or an icon from items extracted from the operation target related document. The rule shown in FIG. 11 is a rule for detecting a place between the start keyword 1101 and the end keyword 1102 as an operation target, like the condition extraction rule shown in FIG. For example, if the item extracted from the operation target related document is
“Press BS button for BS or CS”
In this case, 1104 is applied from the rules 1103, 1104, 1105, and 1106 shown in FIG. 11, and “voice” or “[voice] button” is detected as an operation target.

FIG. 12 shows an example of the format of the option detection rule 705. The option detection rule 705 is a rule for detecting all the options from items extracted from the operation target related document by regarding them as options when there are a plurality of operation targets. The rule in FIG. 12 describes a pattern of expression when there are a plurality of operation targets in items extracted from the operation target related document. In FIG. 12, the description ($ 1), ($ 2), or ($ n) indicates that some operation target is described in that part, and the numbers are assigned to the included operation targets. It is a serial number. In addition, a portion corresponding to ($ 1), ($ 2), or the like is detected as an operation target to be an option. For example, if the item extracted from the operation target related document is
“Press any of the [Ground] button, [BS] button, [CS] button”
, 1201 among the rules 1201, 1202, and 1203 shown in FIG. 12 is applied, and the “[ground] button”, “[BS] button”, and “[CS] button” are selected and detected. It will be.

  FIG. 13 shows the format of the dialogue scenario generation rule 111. In FIG. 13, 1301 describes a character string or a symbol representing the type of rule described in the dialogue scenario generation rule. The types of rules assumed in FIG. 13 are three types: “response sentence”, “dialogue step”, and “grammar generation”. The “response sentence” represents a rule required when generating a sentence that becomes a message indicating a question or information presentation to the user from the natural language dialogue system. The “dialogue step” is a rule necessary for correcting / adding an interactive step included in an after-mentioned dialogue scenario. The “grammar” is a rule that is necessary when generating a speech recognition grammar that is necessary for recognizing an utterance from a user to a natural language dialogue system by a voice recognition function in the natural language dialogue system. Details of each rule are described in 1302 of FIG. In 1303, 1304, and 1305, a rule whose type is “response sentence” is described. In 1306, 1307 and 1308, a rule whose type is “dialogue step” is described. In 1309, 1310, 1311 and 1312, a rule whose type is “grammar” is described. In FIG. 13, the rules are described in Japanese for ease of explanation. However, the description is based on various information described later, and information obtained as a result of well-known morphological analysis techniques and grammatical analysis techniques. Also, since only descriptions corresponding to simple character string conversion are included, it is easy to describe them as rules in a format that can be understood by a computer, or as a program that can be executed by a computer. Can do. How the rules described in FIG. 13 are used will be described later.

FIG. 14 shows the format of the dialogue scenario 103. The dialogue scenario shown in FIG. 14 is created for each user request such as “Turn on TV” or “Adjust volume”. Among them, it is necessary to execute an operation corresponding to each user request. Stores information about user interaction. In FIG. 14, a dialogue name 1401 is a unique identifier assigned to each dialogue scenario, and an arbitrary character string, symbol string, number, or the like can be used. The dialogue name 1401 is used to manage the dialogue scenario 103 in the natural language dialogue system 104 in FIG. The execution determination formula 1402 is a conditional expression for determining whether or not to execute the selected dialog when a dialog scenario is selected by a user request. Since functions and services provided by devices and computer programs may be executable only in specific situations, it is desirable to determine whether or not they can be executed before the start of the dialogue in order to improve usability. The execution determination formula 1402 is information for realizing this, and can describe a condition regarding the value of a variable used in a natural language dialogue system or an operation target. As a format to be described, a format according to a variable used in a natural language dialogue system or a target device and a method of referring to the variable may be used. For example, if a variable named “STATUS” is used in a natural language dialogue system and its value is “true”, the dialogue can be executed.
$ STATUS == “true”
Such a description can be used. Here, “$ STATUS” indicates that the value of the variable “STATUS” is referred to, and the symbol “==” indicates that the condition is satisfied when the values are equal.

  The speech recognition grammar name 1403 in FIG. 14 stores a character string representing a name of a file storing a speech recognition grammar used by the speech recognition function constituting the natural language dialogue system 104 or a name indicating a storage location. The The speech recognition grammar stores information about the content of the user's utterance to the natural language dialogue system when the user makes a request to the natural language dialogue system. A speech recognition grammar is also generated based on information extracted from a document. The number of dialog steps 1404 in FIG. 14 is a numerical value indicating the number of dialog steps constituting the dialog scenario. Here, the dialog step is defined as a basic unit of interaction between the user and the natural language dialog system, and specific contents of the dialog step are described in 1405 to 1406.

  FIG. 15 shows a dialog step format. The dialogue step consists of a response sentence for question and confirmation, information presentation, etc. to the user from the natural language dialogue system, information input from the user to the natural language dialogue system (the result of the speech recognition function recognizing the user's utterance). The basic element is an input variable name indicating the name of the variable in which the content of the user's utterance is stored) and a command for operating a device or a computer program based on information from the user. Moreover, since there are executable dialog steps and non-executable dialog steps based on the situation, an execution determination formula for determining whether or not execution is possible is also included. In FIG. 15, 1501 describes an execution determination formula. As the format of the execution determination formula, the same format as that described in the format of the execution determination formula 1402 in FIG. 14 can be used. Reference numeral 1502 denotes a dialog step type indicating which one of a response sentence, an input variable name, and a command described in the dialog step is used in the dialog step. As the dialog step types, four types of “operation”, “slot”, “confirmation”, and “guidance” are assumed. When the interaction step type is "Operation", a response statement and a command are used. In the case of “slot”, the response text, input variable name, and command are used. In the case of “confirmation”, a response sentence and a command are used. In the case of “guidance”, only the response text is used.

In FIG. 15, a response statement 1503 describes a character string representing the response statement. It is also possible to write a response statement that refers to the value of a variable used in a natural language dialog system or operation target. For example,
"Are you sure you want to use $ CHANNEL channel?"
Can be used. Here, “CHANNEL” is a variable for storing the channel number of the TV, and represents that the content of the variable CHANNEL is referred to by the notation “$ CHANNEL”, and the value of CHANNEL is “1”. The response sentence presented to the user by example is
"Are you sure you want one channel?"
It becomes. Further, an input variable name 1504 in FIG. 15 is a character string representing the name of a variable in which information input by the user when information is input from the user to the natural language dialogue system. In this case, it is necessary to substitute the information from the user obtained from the input means such as the voice recognition function into a variable. However, in the voice recognition function generally used in a personal computer or the like, the voice input by the user is recognized. Since the information contained in the result can be easily output in the form of the corresponding variable name and its value, the function can be used. Further, in 1505 in FIG. 15, a character string representing a command for operating the operation target is described. As the command format, a format that can be directly interpreted by the operation target is used. Similarly to the format described in the response sentence, a command including a reference to a value of a variable used in a natural language dialogue system or an operation target can be described.

Processing performed in the document item extraction unit 106 will be described with reference to FIGS. 16 and 17. The document item extraction unit 106 analyzes the operation target related document 102 based on the document structure analysis rule 107 and extracts items necessary for generating a dialogue scenario. In the description here, the example shown in FIG. 3 is used as the operation target related document 102, and the examples shown in FIGS. 5 and 6 are used as the document structure analysis rule 107. FIG. 16 is a diagram showing the flow of processing performed by the document item extraction unit 106. In step 1601 in FIG. 16, one line is read from the operation target related document 102. In the case of the operation target related document shown in FIG.
“2. Watching TV (301)”
Is read. In step 1602, it is determined whether or not the line has been successfully read in step 1601. If the line has been successfully read, that is, if a readable line remains in the operation target related document, the process proceeds to step 1603. If reading of a line fails in step 1601, it is assumed that processing has been executed for all lines in the operation target related document, and the process ends. In step 1603, the layer number and type of the line read in step 1601 are determined based on the format information in the sentence structure analysis rule. For example, in the case of “2. Watch TV (301)” in FIG. 3, if the format information shown in FIG. 5 is applied, the hierarchy number is determined to be “1” and the type is “heading” according to the rule shown in 505. It will be. In addition, a line for which the corresponding hierarchy number and type do not exist according to the format information is determined to be excluded. In step 1604, if it is determined in step 1603 that it is not applicable, and if the corresponding hierarchy number and type exist but an item exclusion keyword is included, the processing for that line is interrupted, and step 1601 is entered. Return to the next line. Otherwise, go to step 1605. For example, in the case of “2. Watch TV (301)”, since the corresponding hierarchy number and type exist and the item exclusion keyword shown in FIG. 6 is not included, the process proceeds to step 1605. In step 1605, when a line includes a punctuation mark “.” Or “.”, A process of dividing the line into a plurality of lines is performed. In the case of equipment such as information appliances and operation manuals for computer programs, there are particularly cases where a plurality of sentences are included in one line in the description of the operation procedure. For example, in the example shown in FIG. 3, the operation procedures shown in 316 and 322 are described as separate procedures.
“For terrestrial waves, press the [Main Audio / Sub Audio / Main Sub Audio] button. For BS or CS, press the [Audio] button.”
In some cases, such a description is made. In step 1605, a line as shown in the above example is divided into a plurality of sentences. This processing can be easily performed by searching for the phrase “.” Or “.” In the line and dividing the line at the point “.” Or “.”. The hierarchy number and type corresponding to each sentence after division are the same as the hierarchy number and type corresponding to the original line. Further, in step 1606, a storage area is temporarily secured in the storage unit 204 or the like of FIG. 2 as information to be used by the extraction item analysis unit 108 for the row after the processing in step 1605 and the hierarchy number and type for the row. And store. Alternatively, it may be stored as a file.

  FIG. 17 shows items extracted from the operation target related information shown in FIG. 3 using the format information shown in FIG. 5 and the item exclusion keyword shown in FIG. In FIG. 17, reference numeral 1701 denotes a hierarchical number, 1702 denotes an item type, and 1703 denotes extracted item contents. Further, 1704 in FIG. 17 is 301 in FIG. 3, 1705 is 303, 1706 is 304, 1707 is 306, 1708 is 307, 1709 is 309, 1710 is 310, 1711 is 311 and 1712 is 315, 1713 corresponds to 316, and 1714 corresponds to 321.

  FIG. 18 shows the flow of processing performed by the extraction item analysis unit 108. In the extracted item analysis unit 108, in step 1801, an analysis process is performed on items (hereinafter referred to as “headings”) classified as “headings” in the items extracted by the document item extraction unit 106. After that, in step 1802, “procedure” Analysis processing is performed on items classified as (hereinafter referred to as operation procedures). In addition, when there are types other than “Heading” and “Procedure” as item classifications, analysis processing may be appropriately added for each type based on the hierarchical order relation. Through the processing performed by the extraction item analysis unit 108, information on the user request, the operation content, the operation target, the execution condition, etc. is extracted from the information in the format as shown in FIG. 17, and as shown in FIG. 19 for each user request. Stored in format. As a storage method, a method of temporarily securing a storage area in the storage unit 204 of FIG. 2 or a method of storing as a file can be used.

  In FIG. 19, a user request content 1901 is a character string representing the content of the user request. The execution condition 1902 is a character string representing a condition for executing a user request indicated by 1901. In the operation procedure number 1903, a numerical value representing the number of operation procedures for executing the user request indicated by 1901 is described. In 1904 to 1905, the content of a specific operation procedure is described.

  FIG. 20 shows a format of information described in the operation procedures 1904 to 1905. In FIG. 20, an execution condition 2001 is a character string representing a condition for executing an operation procedure. The operation procedure type 2002 describes a symbol or a character string representing the type of operation procedure. The type of operation procedure includes “automatic execution” that indicates that the operation can be executed unconditionally when requested by the user, such as pressing a specific button, and the user can select one of multiple options. Or “option” indicating that the operation can only be executed after the user has made some input such as a character string or symbol string. In the operation content 2003, a character string representing the content of the operation is described. An operation target 2004 represents an operation target such as a button or a menu. The number of options 2005 represents the number of options when the user selects an operation target from a plurality of operation targets. In the options 2005 to 2006, the contents of specific options are described.

  FIG. 21 shows the flow of analysis processing for a headline. In step 2101 in FIG. 21, it is checked whether or not a heading to be processed remains. If not, the process ends. If there is a heading to be processed, one heading is selected in step 2102 and set as a processing target. In step 2103, the condition extraction rule 702 is applied to extract a portion representing the execution condition from the target heading. In step 2104, the operation content extraction rule 701 is applied to portions other than the portion representing the execution condition in the target heading. As a result of applying the operation content extraction rule, if it matches the pattern representing the operation content, the process proceeds to step 2105 to create a storage area for information represented in the format of FIG. 19 with the target heading as a new user request. , Store information. In the storage, the execution condition extracted from the target heading is stored as 1902 in FIG. 19, and the part excluding the execution condition from the target heading is stored as the user request content 1901. Other information is 0 or blank at this stage. In step 2104, if the pattern does not match the operation content, it is regarded as a non-target heading and the process returns to step 2101. When the operation content extraction rule shown in FIG. 8 is applied to the items in the example shown in FIG. 17, the headings 1704, 1705, 1710, and 1712 conform to the rule shown in 803 in FIG. Information in the format shown in Fig. 6 is created and stored.

FIG. 22 shows the flow of analysis processing for the operation procedure. In FIG. 22, in step 2201, it is checked whether or not there is an operation procedure to be processed. If not, the process is terminated. If there remains an operation procedure to be processed, one operation procedure is selected in step 2202 and set as a processing target. In step 2203, the condition extraction rule 702 is applied to extract a portion representing the execution condition from the target operation procedure. At this time, if the target operation procedure is the first operation procedure for the heading immediately above and the execution condition is not included in the next operation procedure for the same heading, the extracted execution condition is set immediately above. Are stored as execution conditions 1902 in the format information shown in FIG. For example, from the operation procedure 1713 in FIG. 17, “terrestrial” is detected as an execution condition by the rule 905 in FIG. Furthermore, since “BS or CS” is similarly detected as an execution condition from the next operation procedure 1714, the execution condition “terrestrial wave” extracted from 1713 is used as the execution condition of the operation procedure indicated by 1713. Stored. On the other hand, when there is no 1714, the execution condition “terrestrial” is stored as the execution condition of the heading 1712 immediately above 1713. In step 2204, the execution condition extraction rule is applied again to a part other than the part representing the execution condition in the target operation procedure, a part that can become a user request is searched, and a part that can become a user request exists. In this case, an area for storing the information of the format shown in FIG. 19 is newly generated, the corresponding portion is stored as a user request, and the remaining part in the target operation procedure is newly generated. Stored as the operation procedure for. For example, in the case of 1709 in FIG. 17, the “when TV is turned off” portion is determined as a “user request” by the rule 907 in FIG. 9, so “turn off TV” is generated as a new user request, The remaining “Press the [Power] button again” portion is stored as the “turn off TV” operation procedure. In addition, the hierarchy number of the newly generated user request is set to a value indicating that it is lower than the “user request” immediately above the original operation procedure. For example, in the case of 1709 in FIG. 17, the layer number of the user request 1705 immediately above the original operation procedure is “2”, so that “3”, which is 1 larger than that, is the “turn off TV” layer. Register as a number. In step 2205, when a description of a plurality of operations is included in a portion excluding a portion that becomes a condition and a portion that becomes a user request in the target operation procedure, the operation is divided into individual operations. This processing can be easily performed by extracting a verb phrase that does not modify other phrases from the processing target and dividing the processing target at a position where the verb phrase exists. Extraction of verb phrases and analysis of modification relations with other phrases can be easily realized by using well-known morphological analysis programs and grammatical analysis programs. If the execution condition is extracted in step 2203, the same execution condition is added to each divided operation. For example, in the case of 1708 in FIG. 17, “input” and “select” are extracted as verb phrases. By dividing 1708 by each verb phrase and removing the connected particles, etc.,
"Enter the desired channel number"
"Select a channel"
Two types of operation procedures will be obtained. Expression shaping can be easily performed by using word information obtained from the results of a morphological analysis program and a grammar analysis program. Further, the operation type determination rule 704 is applied to the last operation procedure in the operation procedure obtained as a result of the division, and it is checked whether or not the keyword in the rule is included. If it is included, the process proceeds to step 2206. If it is not included, an area for storing the information of the format shown in FIG. 19 is newly generated, and the last operation procedure of the operation procedure obtained as a result of the division is stored as a new user request. Information extracted from the remaining operation procedures is stored as an operation procedure corresponding to a new user request by the processing from 2206 onward. In the case of the operation procedure extracted from 1708, “select channel” does not conform to the rules in FIG. 10, and therefore “select channel” is registered as a user request and extracted from “enter desired channel number”. Information to be stored is stored as an operation procedure for “select channel”. In step 2206, it is determined whether or not the processing from step 2207 has been performed on all the operation procedures obtained as a result of the division. If it is determined that the processing has been performed, the process returns to step 2201. If it is determined that it has not been performed, the process proceeds to step 2207, and one operation procedure to be processed is selected. In step 2208, the operation type determination rule 704 is applied to the selected operation procedure to determine whether the operation procedure type is classified as “automatic execution” or “option”. If it is not classified into any of these, the process returns to step 2206. If it is classified into any of the above, in step 2209, the keyword that matches the operation type determination rule 704 is removed from the target operation procedure. Further, by extracting the noun from the remaining part and applying the operation object extraction rule 705 and the option detection rule 706, specific operation objects such as buttons and menus are extracted from the operation procedure. For example, in 1706 in FIG. 17, the “[power] button” is detected as the operation target by the operation target extraction rule shown in FIG. In addition, in 1707 in FIG. 17, “[ground] button”, “[BS] button”, and “[CS] button” are detected as operation targets, and all these operation targets are detected by the option detection rule shown in FIG. Is considered an option. In step 2210, as the user-requested operation procedure corresponding to the heading immediately above the target operation procedure, the execution condition, operation procedure type, operation target, and options obtained by the above processing are respectively shown in FIG. Store as 2002, 2004 and 2005-2006. Further, the contents of the target operation procedure are stored as 2003 in FIG. At this time, when an option is detected in the operation procedure classified as “automatic execution” by the operation type determination rule 704, or a noun that is not detected as an operation target by the operation target extraction rule 705 is displayed in the operation type determination rule 704. If it is described immediately before the corresponding keyword, the operation procedure type 2002 is changed to “option” and stored. For example, the operation procedure is “Select one of the [Ground] button, [BS] button, [CS] button”.
In this case, it is determined as “automatic execution” by the operation type determination rule shown in FIG. 10, but the option is also detected by the option detection rule shown in FIG. Also,
"Select program name"
In this case, the operation type determination rule shown in FIG. 10 determines “automatic execution”, but the noun “program name” does not correspond to the operation target extraction rule shown in FIG. .

  FIG. 23 to FIG. 25 show an example of results obtained by processing the extraction item analysis unit 108 on 1705 to 1709 in FIG. FIG. 23 shows a user request stored as a result of processing the heading 1705, and the user request content “Turn on TV” is described in 2301. Since 1705 does not include a portion that becomes an execution condition, “-” indicating a blank is described in the execution condition 2302. In FIG. 17, there are four operation procedures 1706 to 1709 for the heading 1705. 1708 is stored as a new user request in step 2205 in FIG. 22, and 1709 is stored in step 2204 in FIG. Therefore, since only the operation procedure corresponding to 1706 and 1707 is registered as the operation procedure for “turn on TV”, the operation procedure number 2303 is “2”. An operation procedure corresponding to 1706 is described in 2304, and an operation procedure corresponding to 1707 is described in 2305. FIG. 24 shows information regarding the operation procedure corresponding to 1706. Since 1706 in FIG. 17 does not include a condition part, “-” indicating a blank is described in the execution condition 2401. Further, when “automatic execution” is determined in step 2208, “automatic execution” is described in the operation procedure type 2402. In the operation procedure contents 2403, the contents of the line 1706 are described as they are. The operation target 2404 stores the “[power] button” extracted in step 2209. In addition, since there is no portion corresponding to the option in 1706, the number of options 2405 is “0”, and information regarding the option is not described. FIG. 25 shows information regarding the operation procedure corresponding to 1707. Since 1707 in FIG. 17 also does not include a condition part, “-” indicating a blank is described in the execution condition 2501. In step 2208, “automatic execution” is determined, but since an option is detected in step 2209, the operation procedure type 2502 is described as “option”. In the operation procedure content 2503, the content of the line 1707 is described as it is. Since only options are extracted in step 2209, “-” indicating a blank is described in the operation target 2504. In Step 2209, since “[Ground] button”, “[BS] button”, and “[CS] button” are detected as options, the number of options 2505 is “3”, and the options 2506, 2507, and 2508 are respectively selected. Described in

  FIG. 26 shows a flowchart of processing performed in the dialogue scenario generation unit 110. In the dialogue scenario generation unit 110, as a result of the processing of the extraction item analysis unit 108, the information stored in the format shown in FIGS. Based on the information input by the developer, an interactive scenario having the format shown in FIGS. 14 and 15 is generated. The dialogue scenario is generated for each user request, that is, in units of information stored in the format shown in FIG. In step 2601 in FIG. 26, it is checked whether or not a dialogue scenario has been generated for all user requests. If processing is performed for all user requests, the process ends. Otherwise, in step 2602, one user request to be processed is selected. In step 2603, an area for storing a new dialogue scenario is secured, and a dialogue name 1401 is generated from the user request content 1901 and registered. As the dialog name, a character string representing the user request content can be used as it is. Alternatively, any character string, symbol string, number, or the like corresponding to the user request content may be used. In step 2604, if the execution condition 1902 is not blank, the information input / output unit 112 presents the contents of the execution condition to the developer of the natural language dialogue system, and the natural language dialogue system interprets the execution condition corresponding to the execution condition. A possible determination formula is acquired and registered as an execution determination formula 1402. Alternatively, if the content of the execution condition is known, a correspondence table between the execution condition and the execution judgment expression is prepared, and by referring to it, the execution condition is automatically converted into the execution judgment expression, and the interactive scenario. You may make it register inside. In step 2605, it is checked whether or not the processing shown in step 2606 and step 2607 has been performed for all the operation procedures in the target user request, and if it is determined that the processing has been performed, the processing proceeds to step 2608. If not, the process proceeds to step 2606, where the information input / output unit 112 presents information on the operation procedure to the developer of the natural language dialogue system, and the execution judgment expression and input variables that the natural language dialogue system can interpret from the developer. Acquires supplementary information about names, commands, and options. As for the input variable name, a unique name in the dialogue scenario can be automatically generated and registered. Based on the information acquired from the developer in step 2606 and the dialog scenario generation rule 111, the operation procedure is converted into a dialog step in step 2607 and registered in the dialog scenario, and then the process returns to step 2605. Alternatively, if information about operation procedures, that is, execution conditions, operation targets, and options are known, prepare a correspondence table with them, an execution judgment expression, an input variable name, and a command. Information related to the operation procedure may be converted into an execution judgment formula, input variable name, and command and registered in the dialogue scenario.

  FIG. 27 shows an example of a user interface for acquiring information from the developer of the natural language dialogue system in Step 2604 and Step 2606. In step 2604 and step 2606, the execution determination formula in the user request and the information related to the operation procedure are input separately, but the user interface shown in FIG. It has become. Further, in FIG. 27, the hatched fields indicate information that cannot be corrected by the developer of the natural language dialogue system. In FIG. 27, a list 2701 is a list of user requests generated as a result of processing in the extraction item analysis unit 108, and user request contents 1901 in each user request are displayed in the list. When a user request in the list 2701 is selected with a mouse or the like, detailed information regarding the selected user request is displayed on the right side of the list. Reference numeral 2702 denotes an execution condition for the user request, which displays an execution condition extracted from the corresponding item in the operation target related document. Reference numeral 2703 denotes a field for the natural language dialogue system developer to input an execution determination expression that can be interpreted by the natural language dialogue system for the execution conditions shown in 2702. Reference numeral 2704 denotes a list of operation procedures in the selected user request. In the list, operation contents 2003 in the operation procedure are displayed. When an operation content in the list 2704 is selected with a mouse or the like, detailed information regarding the selected operation procedure is displayed below the list. 2705 is an execution condition for the operation procedure, and the execution condition extracted from the corresponding item in the operation target related document is displayed. Reference numeral 2706 denotes a field for the developer of the natural language dialogue system to input an execution judgment expression that can be interpreted by the natural language dialogue system for the execution condition shown in 2705. 2707 indicates the operation procedure type 2002 and 2708 indicates the contents of the operation target 2004. The contents of the operation target displayed in 2708 can be corrected by the developer of the natural language dialogue system. Reference numeral 2709 denotes a field for a natural language dialogue system developer to input a command to be executed by a device or a computer program that is an operation target for the operation procedure. 2710: When the type of operation procedure is “option”, the input variable name, which is the name of the variable in which the content uttered by the user is stored in the result of the speech recognition function recognizing the user's utterance, is This is a field for input by the dialog system developer. As the input variable name, an arbitrary character string that is unique in the dialogue scenario may be automatically registered. 2711 is a list of options extracted from the corresponding items in the operation target related document. The content of each option can be modified by the developer of the natural language dialogue system. Since 2710 and 2711 are not necessary when the operation procedure type is “automatic execution”, they can be displayed and changed in the case of “automatic execution”. Further, in the user interface of FIG. 27, information registered for each user request and information registered for each operation procedure can be performed on one screen, but can be performed on separate screens. Also good. In the user interface of FIG. 27, unnecessary user requests and operation procedures may be deleted, or new user requests and operation procedures may be added.

  FIG. 28 shows a flowchart of the processing performed in step 2607 of FIG. In FIG. 28, in step 2801, it is checked whether or not the operation procedure type is “option”. If it is “option”, the process proceeds to step 2802; otherwise, the process proceeds to step 2804. In step 2802, the dialogue step type 1502 corresponding to the target operation procedure is set to “slot”. In step 2803, the execution judgment formula, input variable name, and command acquired from the natural language dialogue system developer by the information input / output unit 112 are the dialogue step execution judgment formula 1501, input variable name 1504, and command 1505 shown in FIG. Register with. In step 2803, the option is corrected and added using the supplementary information of the option acquired from the developer of the natural language dialogue system, and used in the processing of step 2608 described later. In step 2804, the dialogue step type 1502 corresponding to the target operation procedure is set to “operation”. In step 2805, the information input / output unit 112 registers the execution determination formula and the command acquired from the developer of the natural language dialog system in the dialog step execution determination formula 1501 and the command 1505 shown in FIG. In step 2806, a response sentence 1503 is generated by applying the rule “response sentence” type described in the dialogue scenario generation rule 111 to the operation content 2003 of the operation procedure. For example, when the dialog scenario generation rule shown in FIG. 13 is applied to the operation procedure shown in FIG. 24, it is the first operation procedure of the user request shown in FIG. “Turn on TV” changed from “Turn on TV (2301)” is added as a response to the operation procedure shown in FIG. In the case of the operation procedure shown in FIG. 25, the rule shown at 1304 is applied. Here, “(option item)” in the rule represents that all options are listed, and “[” and “] buttons” are deleted for the options described in 2506 to 2508. Thus, if the modification is made by the developer, the response sentence will be "Which is the ground, BS, or CS?" Further, the case where the rule indicated by 1305 in FIG. 13 is applied is a case where the information regarding the operation procedure has the contents as shown in FIG. When the rule 1305 shown in FIG. 29 is applied, “Please tell me the channel number” is generated as a response sentence using the operation target 2901. Through the above processing, the user request shown in FIG. 23 is converted into the dialog scenario shown in FIG. 30, and the operation procedure shown in FIGS. 24 and 25 is converted into the dialog steps shown in FIGS. At this time, the speech recognition grammar name 3001 is not described. It is assumed that commands 3101 and 3202, and variable name 3201 are input by the developer of the natural language dialogue system. Further, the description “$ VAL0001” represents that the value of the variable VAL0001 is referred to.

  In step 2807 of FIG. 28, the content of the dialog step is corrected by applying the “dialog step” type rule described in the dialog scenario generation rule 111 to the dialog step generated from the operation procedure. Alternatively, an interactive step is added. For example, the rules shown in 1306 are applied to the dialog steps shown in FIG. 32, so that the dialog steps shown in FIG. 33 are modified, and the dialog steps shown in FIG. 34 and FIG. Added after the interaction step shown in. In FIG. 33, the command 3301 is deleted and moved to the command 3501 in the dialog step of FIG. The dialogue step shown in FIG. 34 is a dialogue step of the type “confirmation”, presents a response sentence for confirming the contents of the input variable to the user, and obtains a response from the user. The dialog step in FIG. 35 executes a command 3501 when the value of “YESNO (3402)” that is an input variable from the user is “yes”. Also, the rule indicated by 1307 in FIG. 13 is always applied, and the dialog step as shown in FIG. 36 is added as the last dialog step of each dialog scenario. A variable “RET” described in 3601 in FIG. 36 represents a common variable indicating a command execution result. Further, since only the response sentence is presented to the user, the dialogue step type 3602 is “guidance”. The response sentence is generated from the user request contents. For example, by applying a rule such as “(user request content) + cannot be done”, a response sentence “Cannot turn on TV” shown in 3603 can be generated. In addition, when a dialogue scenario is generated from 1704 in FIG. 17, since there is no operation procedure corresponding to 1704, the rule indicated by 1308 in FIG. 13 is applied. As a result, for example, a dialog step as shown in FIG. 37 is generated. The response sentence in this example is generated by enumerating only the contents of the user request one level below. Furthermore, all the contents of user requests at lower levels than the target user request may be listed.

In step 2608 in FIG. 26, a grammar for speech recognition necessary for executing the dialogue scenario generated from the target user request is generated. FIG. 38 shows a flowchart of processing performed in step 2608. In FIG. 38, in step 3801, a speech recognition grammar name (1403 in FIG. 14), which is the name of a file for storing speech recognition grammar, is generated and registered in the dialogue scenario. As the speech recognition grammar name, an arbitrary character string or symbol string can be used as long as it is not the same as the speech recognition grammar name in another dialogue scenario. For example, a name obtained by adding an arbitrary extension to the dialog name (1401 in FIG. 14) may be used. In step 3802, a speech recognition grammar for accepting an utterance from a user for starting a dialogue described in the dialogue scenario is generated. This grammar is generated by applying a rule of the type “grammar” in the dialog scenario generation rule to the user request content (1901 in FIG. 19) in the user request that is the basis of the dialog scenario. For example, the user request content for the user request generated from 1705 in FIG. 17 is “turn on TV”, but when the dialog scenario generation rule shown in FIG. 13 is applied to this, the rule shown in 1309 is applied first. as a result,
"Turn on TV"
"Turn on the TV"
Are generated. Furthermore, the rules shown in 1312 are applied,
"Turn on the TV"
"I want to turn on TV"
"Turn on the TV"
"Turn on the TV"
"I want to turn on the TV"
"Turn on the TV"
6 sentences are generated. Here, the description “verb (basic form)” in the rule shown at 1312 represents that a basic form verb exists at that location. In the case of “verb (continuous form)”, it indicates that there is a verb in the same way. These sentences are processed into words by using a well-known morphological analysis program or the like, and then converted into a speech recognition grammar. As the format of the grammar for speech recognition, a format that can be interpreted by the speech recognition function to be used may be used. The generated speech recognition grammar is stored in a file indicated by the speech recognition grammar name generated in step 3801. In step 3803, a grammar for recognizing the user's utterance for inputting an option is generated in the dialog step of the type “slot”. This grammar is generated by using the information after the developer of the natural language dialog system modifies and adds to the option information described in the operation procedure that is the basis of the dialog step. For example, if the developer modifies the options included in the operation procedure shown in FIG. 25 so as to delete the “[] and“ ”buttons”, the contents of the target options in step 3803 are “ “Ground”, “BS”, “CS”. Since the rules shown in 1311 apply to the options,
"Ground"
"It's on the ground"
"On the ground"
"BS"
"This is BS"
“To BS”
"CS"
"It is CS"
"CS"
Nine sentences are generated. Here, the description “(option)” in the rule shown in 1311 represents that the place is replaced with each content of the option. As in the case of step 3802, these sentences are processed into words using a well-known morphological analysis program or the like, and then converted into a speech recognition grammar. In the speech recognition grammar for the options, it is necessary that the speech recognition function can notify the natural language dialogue system of the result of associating the contents of the options spoken by the user with the input variable name 1504. The speech recognition function commonly used in personal computers can easily output the information contained in the recognition result of speech uttered by the user in the form of the corresponding variable name and its value. It is only necessary to generate the grammar in a format that allows the function to be used. In the above description, the speech recognition grammar generated from the user request and the speech recognition grammar generated from the operation procedure are stored in one storage area or file, but the speech recognition grammar used for each dialog step is used. When a natural language dialogue system capable of switching grammars is used, the speech recognition grammar generated from the user request and each operation procedure may be stored in a separate storage area or file. In addition, the speech recognition grammar generated from the user request and operation procedure described above assumes a type of dialog that prompts the user to input necessary options in order after starting the dialog requested by the user. However, in a natural language dialogue system, the user's utterance including information about the dialogue to be activated and the necessary choices is recognized, and the dialogue is activated and the choices are set (ie, assigned to input variables). There is also a natural language dialogue system that can be used simultaneously. Such a speech recognition grammar that can be used in a natural language dialogue system can be easily generated by describing a rule that generates a sentence including user request contents and option contents in a dialog scenario generation rule. it can. In addition, a grammar for inquiring about functions that can be requested by the user may be added to the grammar for speech recognition for starting a conversation scenario. For example, a grammar for a sentence such as “What is a function for watching TV?” Is added to the grammar for speech recognition that starts the dialogue scenario with the dialogue step shown in FIG. 37 added. be able to. The generation of this grammar can be easily realized by preparing a rule for changing the end of a sentence such as the contents of a user request in the same way as the dialog scenario generation rule described above. As a result, not only the user can request the operation of the device or the computer program, but also the user can inquire about a function that can be executed by the device or the computer program.

  By performing the processing described above, it is possible to convert the operation target related documents such as operation manuals and instruction manuals of information appliances and computer programs to dialog scenarios that can be used by natural language dialog systems. It becomes.

  A second embodiment will be described with reference to FIGS.

  In the first embodiment, the user makes a request to the natural language dialogue system by voice or the like, and based on the contents, the dialogue scenario necessary for operating the device or computer program such as information home appliances on which the natural language dialogue system is the operation target In this embodiment, the user inquires about the method of using the function of the device or the computer program in the natural language dialogue system, and the natural language dialogue system sequentially instructs the user about the operation method. It is also possible to generate an interactive scenario.

In order to do this, for example, in the example described in the first embodiment, a dialogue scenario generation rule as shown in FIG. 39 is prepared. When the dialog scenario generation rule shown in FIG. 39 is applied to the user request and operation procedure shown in FIG. 23, FIG. 24 and FIG. 25 to generate a dialog scenario, first, the rule shown in 3901 is applied. The response text of the operation step corresponding to the operation procedure shown is
“I will explain how to turn on the TV”
Will be added. Next, by applying the rules shown in 3902, the following response sentences are generated from the operation contents 2403 and 2503 shown in FIGS.
"Press the [Power] button on the remote control"
"Press any of the [Ground] button, [BS] button, [CS] button"
In the rule shown in 3902, a response sentence is generated from the operation content, but a rule for generating a response sentence with a unique expression is prepared using the operation target and options extracted from the operation content, and the response sentence is generated. You can also Also, by applying the rules shown in 3903 and 3904, the dialog steps shown in FIGS. 40 and 41 are generated from the operation procedure shown in FIG.

In FIG. 40, “guidance” is registered in the dialog step type 4001 according to the rule 3903. At the same time, the command 4003 is blank. The response sentence 4002 has contents to which 3901 and 3902 are applied. FIG. 41 shows a dialog step generated as a result of applying the rule shown in 3904, and is inserted immediately after the dialog step shown in FIG. The contents of the dialogue step shown in FIG. 41 are for confirming whether or not the user who has received the instruction in the dialogue step shown in FIG. 40 has executed the instruction. “Yes”, “What is next?”, Etc. When the user inputs an utterance indicating that the previous instruction has been completed, the process proceeds to the next dialog step. Here, the variable YESNO is the name of the variable into which the recognition result of the user's utterance for the confirmation from the natural language dialogue system is substituted. Further, the grammar for speech recognition is generated only from the user request contents according to the rule shown in 3905, and is not generated from the information regarding the operation procedure. When the rule shown in 3905 is applied to the user request content of the user request shown in FIG.
"Tell me how to turn on the TV"
"How can I turn on the TV?"
Are generated, and these sentences are converted into a speech recognition grammar. Or, a question sentence from the user regarding the operation target and options, for example,
“Where is the [BS] button?”
A speech recognition grammar for a question sentence such as, and a dialogue scenario for them can also be generated by preparing necessary dialogue scenario generation rules.

  Based on the above, dialogue scenarios for realizing dialogues with users explaining the functions and operation methods of devices and computer programs should also be generated from operation target related documents such as operation manuals and instruction manuals for devices and computer programs. Is possible.

  A third embodiment will be described with reference to FIG.

  In the dialogue scenario generation system according to the first embodiment and the second embodiment, the dialogue scenario generation unit includes a processing result of the extraction item analysis unit 108, a dialogue scenario generation rule 111, and a natural language dialogue system input from the information input / output unit 112. A dialogue scenario 103 is generated based on information from the developer. On the other hand, there are cases in which information on executable commands and available parameters (information corresponding to options in the operation procedure) can be acquired from the outside among devices and computer programs to be operated. When such an operation target is used, information that can be acquired from the operation target can also be used when generating a dialogue scenario.

  FIG. 42 is a conceptual block diagram showing the configuration of a dialogue scenario generation system according to the third embodiment. 42, the components of the dialogue scenario generation system are the same as those in the first embodiment shown in FIG. The difference between FIG. 1 and FIG. 42 is that the dialogue scenario generation unit 110 is connected to the operation target 105. The dialogue scenario generation unit 110 requests the operation target 105 to send information relating to executable commands and available parameters, and uses the acquired information to generate a dialogue scenario. The information acquired from the operation target 105 is, for example, the executable information acquired from the operation target 105 when the developer of the natural language dialogue system is requested to input the execution determination formula in step 2604 of the flowchart shown in FIG. A list of commands can be presented to the developer as candidates, and the developer can select a required command from the list. In step 2606, when a developer of the natural language dialogue system is requested to input an execution determination formula, a command, and option correction / additional information, a list of executable commands acquired from the operation target 105 and available parameters are obtained. Can be presented to the developer as candidates, and the developer can select the necessary commands and parameters from the list. Further, information on commands that can be executed by devices and computer programs to be operated and parameters that can be used may be acquired from other servers or file systems connected via a network (not shown).

  As a result, it is possible to reduce the developer's work required to generate a dialogue scenario and to efficiently generate a dialogue scenario.

  As described above, the natural language dialogue system described in detail using various embodiments can be expected as an interface for improving the operability of various devices such as information appliances and computer programs. The dialogue scenario generation system according to the present invention can easily expand the range of use of the natural language dialogue system, and thus can be applied to products in various fields having a user interface.

The conceptual block diagram which shows the structure of the dialogue scenario generation system in a 1st Example, and the natural language dialogue system using the same. The block diagram at the time of implement | achieving the dialogue scenario production | generation system in a 1st Example on a computer. The figure which shows an example of the operation target related document assumed in the 1st Example. The figure which shows the structure of the document structure analysis rule in a 1st Example. The figure showing the content of the format information in the document structure analysis rule in a 1st Example. The figure showing the content of the item exclusion keyword in the document structure analysis rule in a 1st Example. The figure which shows the structure of the expression analysis rule in a 1st Example. The figure which shows the format of the operation content extraction rule contained in the expression analysis rule in a 1st Example. The figure which shows the format of the condition extraction rule contained in the expression analysis rule in a 1st Example. The figure which shows the format of the operation kind determination rule contained in the expression analysis rule in a 1st Example. The figure which shows the format of the operation target extraction rule contained in the expression analysis rule in a 1st Example. The figure which shows the format of the choice detection rule contained in the expression analysis rule in a 1st Example. The figure which shows the format of the dialogue scenario production | generation rule in a 1st Example. The figure which shows the format of the dialogue scenario in a 1st Example. The figure which shows the format of the dialogue step in the dialogue scenario in a 1st Example. The figure explaining the flow of the process performed in the document item extraction part in a 1st Example. The figure which shows an example of the process result performed in the document item extraction part in a 1st Example. The figure explaining the flow of the process performed in the extraction item analysis part in a 1st Example. The figure which shows the format which stores the information obtained as a result of the process performed in the extraction item analysis part in a 1st Example. The figure which shows the format which stores the operation procedure contained in the information obtained as a result of the process performed in the extraction item analysis part in a 1st Example. The figure which shows the flow of the analysis process with respect to the item classified as "Heading" in the extraction item analysis part. The figure which shows the flow of the analysis process with respect to the item classified as "procedure" in the extraction item analysis part in a 1st Example. The figure which shows an example of the processing result performed in the extraction item analysis part in a 1st Example. The figure which shows an example of the processing result performed in the extraction item analysis part in a 1st Example. The figure which shows an example of the processing result performed in the extraction item analysis part in a 1st Example. The figure which shows the flow of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the screen used by the information input / output part in a 1st Example. The figure which shows the flow of the process which converts an operation procedure into a dialog step in the process performed in the dialog scenario production | generation part in a 1st Example. The figure which shows an example of the information regarding the operation procedure in a 1st Example. The figure which shows an example of the middle progress of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the middle progress of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the middle progress of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the result of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the result of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the result of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows an example of the result of the process performed in the talk scenario production | generation part in a 1st Example. The figure which shows an example of the result of the process performed in the dialogue scenario production | generation part in a 1st Example. The figure which shows the flow of the process which produces | generates the grammar for speech recognition in the dialogue scenario production | generation part in a 1st Example. The figure which shows the dialogue scenario production | generation rule used in a 2nd Example. The figure which shows an example of the dialogue scenario produced | generated in the 2nd Example. The figure which shows an example of the dialogue scenario produced | generated in the 2nd Example. The conceptual block diagram which shows the structure of the dialogue scenario production | generation system in a 3rd Example.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 101 ... Dialog scenario generation system, 102 ... Operation object related document, 103 ... Dialog scenario, 104 ... Natural language dialog system, 105 ... Operation object, 106 ... Document item extraction part, 107 ... Document structure analysis rule, 108 ... Extraction item analysis 109: Expression analysis rule, 110 ... Dialog scenario generation unit, 111 ... Dialog scenario generation rule, 112 ... Information input / output unit.

Claims (17)

A dialogue scenario generation system that generates a dialogue scenario necessary for a natural language dialogue system that interacts with a user in a natural language and performs operation of a device or a computer program to be operated based on a request from the user.
A document storage unit for storing an operation target related document in which a function or an operation method of the operation target is described in a natural language;
A document structure analysis rule storage unit for storing a document structure analysis rule that is a rule necessary for analyzing the structure of the operation target related document;
A document item extraction unit that analyzes the structure of the operation target related document based on the document structure analysis rule, and selects necessary items from the operation target related document;
An expression analysis rule storage unit for storing an expression analysis rule, which is a rule for extracting information necessary for dialogue with the user from the content of the selected item;
An extracted item analysis unit that analyzes the selected item based on the expression analysis rule and extracts information necessary for the dialog with the user;
An interaction scenario generation rule storage unit for storing an interaction scenario generation rule that is a rule for generating the interaction scenario;
An information input / output unit for prompting the developer of the natural language dialogue system to input information and acquiring information from the developer;
A dialog scenario generation unit that generates the dialog scenario based on the information extracted in the extraction item analysis unit, the dialog scenario generation rule, and information acquired from the developer;
A dialogue scenario generation system having: In the dialogue scenario generation system according to claim 1,
The document structure analysis rule is:
A heading at the beginning of a chapter or section,
Bulleted items,
Items that describe the operating procedure,
Including information related to the format in the operation target related document and information related to the hierarchical relationship between the items.
The document item extraction unit is a dialog scenario generation system that extracts items corresponding to the format and a hierarchical relationship between items in the operation target related document from the operation target related document. In the dialogue scenario generation system according to claim 1,
The expression analysis rule is:
Operation content extraction rules that extract expressions representing what operations (operation content) to perform,
Condition extraction rules for extracting expressions that are prerequisites (execution conditions) for executing operations,
An operation type determination rule for determining the type of operation,
An operation target extraction rule for extracting an expression indicating what operation is performed (operation target),
Choice extraction rules for extracting choices when multiple operation targets are choices,
Including any one or more of
The extracted item analysis unit, based on the expression analysis rule, from items extracted from the document item extraction unit,
Extraction of execution conditions,
Extraction of operations,
Extraction of operation target,
Choice extraction,
Determining the type of operation,
Dialog scenario generation system that performs any one or more of the above. In the dialogue scenario generation system according to claim 1,
The dialog scenario generation rule is:
Rules for generating a response sentence that is a message for asking a question, confirmation, explanation, etc. to the user from the natural language dialogue system;
Rules for generating, modifying and adding dialogue steps which are units of interaction in dialogue between the user and the natural language dialogue system;
Rules for generating a speech recognition grammar, which is information for recognizing speech input by the user,
Including any one or more of
Based on the result of the process in the extraction item analysis unit, the dialogue scenario generation unit, based on the dialogue scenario generation rule and the information acquired from the developer,
Response sentence generation,
Creating, modifying and adding interactive steps,
Generation of grammar for speech recognition,
Dialog scenario generation system that performs any one or more of the above. In the dialogue scenario generation system according to claim 1,
The dialogue scenario generation unit, as the dialogue scenario,
A dialogue scenario for substituting the operation to be operated based on a request from the user;
A dialogue scenario for presenting the user with a function executable by the operation target based on a request from the user;
A dialogue scenario for teaching the user how to execute the operation target based on a request from the user;
Dialog scenario generation system that generates one or more of the above. In the dialogue scenario generation system according to claim 4,
The dialogue scenario generation unit
Information obtained as a result of processing in the extraction item analysis unit is presented to the developer, and the developer inputs a command executed by the operation target input from the information input / output unit to the presented information by the developer. Register during conversation scenario,
The information obtained as a result of the processing in the extraction item analysis unit and the information indicating the correspondence relationship between the command to be executed by the operation target are stored in advance, and the information indicating the correspondence relationship and the processing result obtained in the extraction item analysis unit are obtained. Register the command in the dialogue scenario from the information
Dialog scenario generation system that performs any one or more of the above. In the dialogue scenario generation system according to claim 4,
The dialogue scenario generation unit
A device or computer program to be operated;
Servers and file systems installed on the network,
A dialog scenario generation system that acquires information on commands and parameters that can be executed by the operation target using any one or more of the above, and generates the dialog scenario using the acquired information. A dialogue scenario used in a natural language dialogue system that has an information processing unit, an information storage unit, and an information input / output unit, and performs operations of devices and other computer programs to be operated based on requests from users in natural language A dialogue scenario generation method in a system for generating
The information processing unit is
An operation target related document describing functions and operation methods of the operation target stored in the information storage unit in a natural language, a document structure analysis rule for analyzing the structure of the operation target related document, and the operation Using expression analysis rules for extracting information necessary for dialogue with the user from the contents of the item selected from the target related document, and dialogue scenario generation rules for generating the dialogue scenario,
Analyzing the structure of the operation target related document based on the document structure analysis rule, and selecting a necessary item from the operation target related document;
Analyzing the selected item based on the expression analysis rule, extracting information necessary for dialogue with the user,
Using the information input / output unit, prompting the developer of the natural language dialogue system to input information, obtaining information from the developer,
A dialog scenario generation method for generating the dialog scenario based on the extracted information, information acquired from the developer, and the dialog scenario generation rule. The dialog scenario generation method according to claim 8,
The document structure analysis rule is:
Including information on the format in the operation target related document and information on the hierarchical relationship between items,
The information processing unit
When selecting necessary items from the operation target related document based on the document structure analysis rule, an interactive scenario for selecting the items corresponding to the format and the hierarchical relationship between items in the operation target related document. Generation method. The dialog scenario generation method according to claim 8,
The expression analysis rule is:
Operation content extraction rules that extract expressions representing what operations (operation content) to perform,
Condition extraction rules for extracting expressions that are prerequisites (execution conditions) for executing operations,
An operation type determination rule for determining the type of operation,
An operation target extraction rule for extracting an expression indicating what operation is performed (operation target),
Choice extraction rules for extracting choices when multiple operation targets are choices,
Including any one or more of
The information processing unit
From the information extracted based on the expression analysis rule,
Extraction of execution conditions,
Extraction of operations,
Extraction of operation target,
Choice extraction,
Determining the type of operation,
Dialog scenario generation method for performing any one or more of the above. The dialog scenario generation method according to claim 8,
The dialog scenario generation rule is:
Rules for generating a response sentence that is a message for asking a question, confirmation, explanation, etc. to the user from the natural language dialogue system;
Rules for generating, modifying and adding dialogue steps which are units of interaction in dialogue between the user and the natural language dialogue system;
Rules for generating a speech recognition grammar, which is information for recognizing speech input by the user,
Including any one or more of
The information processing unit
Based on the dialogue scenario generation rule and the information acquired from the developer, from the information necessary for the dialogue with the extracted user,
Response sentence generation,
Creating, modifying and adding interactive steps,
Generation of grammar for speech recognition,
Dialog scenario generation method for performing any one or more of the above. The dialog scenario generation method according to claim 8,
The information processing unit, as the dialogue scenario,
A dialogue scenario for performing the processing target operation on the basis of a request from the user;
An interactive scenario for presenting to the user a function that can be executed by the processing target based on a request from the user;
A dialogue scenario for teaching the user how to execute the processing target based on a request from the user;
A dialogue scenario generation method for generating any one or more of the above. The dialogue scenario generation method according to claim 11,
The information processing unit
Information necessary for dialogue with the user is presented to the developer, and the developer inputs a command to be executed by the operation target with respect to the presented information, and the entered command is displayed in the dialogue scenario. Register with
The information necessary for the dialogue with the user and the information representing the correspondence relationship between the commands executed by the processing target are stored in advance, and the information representing the correspondence relationship and the information necessary for the dialogue with the user are stored. Register the command in the dialog scenario,
Dialog scenario generation method for performing any one or more of the above. A dialogue scenario required for a natural language dialogue system that consists of an information processing unit, an information storage unit, and an information input / output unit, and acts as an operation target device or computer program based on a user request in natural language A program for a dialogue scenario generation system for generating
An operation target related document describing functions and operation methods of the operation target stored in the information storage unit in a natural language, and a document structure analysis rule necessary for analyzing the structure of the operation target related document; , Using an expression analysis rule for extracting information necessary for the dialog with the user and a dialog scenario generation rule that is a rule for generating the dialog scenario,
In the information processing unit,
Analyzing the structure of the operation target related document based on the document structure analysis rule, and selecting a necessary item from the operation target related document;
Analyzing the selected item on the basis of the expression analysis rule, and extracting information necessary for the dialogue with the user;
Using the information input / output unit to prompt the developer of the natural language dialogue system to input information and obtaining information from the developer;
Generating the dialogue scenario based on the extracted information, the information obtained from the developer, and the dialogue scenario generation rule;
Interactive scenario generation program that executes In the dialogue scenario generation program according to claim 14,
The document structure analysis rule is:
A heading at the beginning of a chapter or section,
Bulleted items,
Items that describe the operating procedure,
Including information related to the format in the operation target related document and information related to the hierarchical relationship between the items.
In the information processing unit,
An interactive scenario in which, when selecting a necessary item from the operation target related document based on the document structure analysis rule, an item corresponding to the format and a hierarchical relationship between items in the operation target related document are selected together. Generation program. In the dialogue scenario generation program according to claim 14,
The dialog scenario generation rule is:
Rules for generating a response sentence that is a message for asking a question, confirmation, explanation, etc. to the user from the natural language dialogue system;
Rules for generating, modifying and adding dialogue steps which are units of interaction in dialogue between the user and the natural language dialogue system;
Rules for generating a speech recognition grammar, which is information for recognizing speech input by the user,
Including one or more of
In the information processing unit,
Based on the dialogue scenario generation rule and the information acquired from the developer, from the information necessary for the dialogue with the extracted user,
Response sentence generation,
Creating, modifying and adding interactive steps,
Generation of grammar for speech recognition,
A dialogue scenario generation program for executing one or more of the above. In the dialogue scenario generation program according to claim 16,
In the information processing unit,
Information necessary for dialogue with the user is presented to the developer, and the developer inputs a command to be executed by the operation target with respect to the presented information, and the entered command is displayed in the dialogue scenario. Register with
The information necessary for the dialogue with the user and the information representing the correspondence relationship between the commands executed by the processing target are stored in advance, and the information representing the correspondence relationship and the information necessary for the dialogue with the user are stored. Register the command in the dialog scenario,
A dialogue scenario generation program for executing one or more of the above.

Images