JP2007052043A - Voice interactive scenario generation method and system, and voice interactive scenario generation program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice interactive scenario generation method and a system, in which a voice interactive scenario for treating a plurality of subjects is simply generated. <P>SOLUTION: Information required for linking each original scenario from a resource prepared in the original scenario, information for determining to which original scenario the subject is transferred, and information for making voice recognition suitable for each original scenario in a state in which the subject is transferred, are extracted. A conformance information for conforming the original scenario to environment in which each scenario appropriately executes voice response is generated even when transferring to any original scenario from these information, and by holding the conformance information, the voice interactive scenario for treating the plurality of subjects is generated. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a voice conversation scenario creating method and apparatus for creating a voice conversation scenario capable of handling a plurality of topics, a voice conversation scenario creating program for realizing the apparatus by a computer, and a recording medium on which the program is recorded.

In recent years, a non-patent document 1 discloses a speech dialogue apparatus that achieves an object such as inputting a command related to a certain task through dialogue using a voice with a user by combining speech recognition and synthesized speech technologies. In addition, it is possible for human beings to allow words to be generated in items that are different from the items that the system requires the user to input at some point in the conversation, and to allow good grudges. In addition, a voice dialogue system capable of natural voice dialogue has been proposed.
Also, an animation agent character expressed as an anthropomorphic person is displayed on the screen as in Patent Document 1, and the user recognizes the voice spoken to speak to the agent, and the response is recorded in advance. Proposals have also been made for devices that advance conversations while notifying them by playing them back with synthetic sounds, displaying them as text on the screen, and displaying the nuances of the responses as animations of the agent. .

Furthermore, in recent years, it has become common to construct a speech dialogue system by describing a script language in a text file and a script language execution processing system reading a file in which the script language is described.
The text file in the script language includes a playback sentence when returning a response to the user, a branch rule for selecting the next response when there is a voice input or other input from the user, a voice recognition device, The minimum instructions for the speech synthesizer and other input devices are described. A text file in such a script language is called a voice dialogue scenario.

The execution processing system that has read the voice dialogue scenario executes voice dialogue by driving voice recognition, voice synthesis, and other input / output devices according to the description of the voice dialogue scenario. The detailed control contents of each device are incorporated in advance in the execution processing system in association with the script language command description. The combination of the voice dialogue scenario and its execution processing system has reduced the portability of the system, the construction cost, and the knowledge and experience required for the system creator.
On the other hand, assuming that the role of the dialogue system is a simple one-question-answer format, for each input, select an appropriate one from a plurality of dialogue systems and respond to the output of that system. Have also been proposed (Patent Documents 2 and 3).
JP 2004-295837 A JP 2004-240150 A JP 2004-240225 A Junichi Hirasawa, Shunichiro Yamamoto, Takaaki Hori, Katsutoshi Otsuki, “RexDialog, a spoken dialogue system for free speech for CTI”, Information Processing Society of Japan Research Report, 2003-SPL-47 (8), pp35-40.

However, even if the construction cost is reduced by the combination of the voice conversation scenario and its execution processing system, the construction cost may still be high. For example, there are cases where it is necessary to handle a plurality of topics and to accept words / phrases handled by all the topics at a time when voice input is possible during the execution of the voice dialogue.
For example, in the case of the city hall window general voice dialogue system, multiple topics such as moving-in / moving guidance, social insurance guidance, oversized garbage reception, etc. are handled, and voices are always input during the voice dialogue for all the topics. When creating a dialogue system that is capable of responding appropriately to input, there is a problem that the amount of description of the spoken dialogue system becomes enormous. If you always try to deal with all the topics, in the process of determining the next response content from the user's input utterance, input items necessary to achieve the topic purpose of each topic (this is called a slot) This is because it is necessary to inspect whether or not everything has changed due to the input from the previous utterance. In other words, in order to handle a plurality of topics and to always accept words and phrases input by each topic, it is necessary to describe at least a branching rule of the sum of the number of slots handled in each topic. This indicates that the description amount increases explosively when the number of topics is increased even a little.

In addition, as described in Patent Document 2 and Patent Document 3 described above, the role of the dialogue system is assumed to be a simple one-question-answer format, and an appropriate one is selected from a plurality of dialogue systems for each input. When the method of responding to the output of the system is taken,
Although this method can avoid an enormous amount of description, it cannot be applied to a dialogue system that requires several exchanges to achieve the purpose of each topic. was there.

  The present invention proposes a voice conversation scenario creation device that includes a plurality of original scenarios each corresponding to only one different topic, and creates a voice conversation scenario capable of dealing with a plurality of topics by connecting the plurality of original scenarios. The characteristic configuration is the information necessary to connect each original scenario from the resources provided in the original scenario, the information that determines which original scenario the topic will transition to, and the state in which the topic has transitioned However, information extraction means for extracting information for speech recognition of words related to topics handled in all original scenarios, and which original scenario is being executed no matter which original scenario is transitioned from the information extracted by this information extraction means Even if there is, it is possible to process the user's utterances on the topics handled by other original scenarios and generate alignment information to match the environment where voice response can be executed. Characterized in that it comprises an information creation unit.

  Further in the present invention, in the above-mentioned voice dialogue scenario creation device, the matching information creation means includes the name of each original scenario handled in the linked original scenario, the file name of the original scenario, the position of the file, and the original scenario to be executed first. Linked information creation means for creating linked information describing information, words input to input items necessary for achieving the purpose from the speech recognition resource in each original scenario, the name of each original scenario, and the original scenario A transition resource creating means for creating a transition resource for extracting a related word / phrase that recalls, appropriately recognizing these as a transition term phrase causing a topic transition, and outputting an understanding result, and a transition term phrase Is recognized, it confirms whether to transition to the original scenario corresponding to the recognition term phrase, displays the transition, and another original scenario that is connected Constituted by the creation Topic Transition dialog scenario means for creating a transition for dialog scenario to perform the operation to transfer control.

The matching information generating means used in the voice dialogue scenario creating apparatus of the present invention includes the dialogue information executable in the linked original scenario in addition to the linked information creating means, the transition resource creating means, and the transition scenario creating means. A navigation scenario creating means for displaying and guiding a voice dialogue procedure is provided.
The transition resource creating means used in the voice conversation scenario creating device of the present invention extracts a keyword belonging to a predetermined class from a keyword dictionary in resources provided in each original scenario, and attributes the keyword to the extracted keyword. A transition keyword dictionary creating means for creating a transition keyword collection to which information indicative of a scenario is added, and a scenario name keyword correspondence database that acts as a table indicating to which original scenario each keyword described in each original scenario belongs From the created scenario name keyword correspondence database creation means and the utterance understanding list described in each original scenario, the utterance understanding rules included in the rules of the classes included in the scenario keyword class list are extracted, and each extracted rule Attribute of class name as a keyword in the scenario A word that has been rewritten to the name with the information added and the slot name to which the distribution destination is assigned as the transition destination scenario related word, and the keyword in which the attribute information is added to the keyword in the scenario or the keyword that belongs to the transition destination scenario related word is voice-recognized If it appears in the sequence, the keyword is assigned to the destination scenario-related word slot, and if the keyword belonging to the scenario specification class appears in the speech-recognized word sequence, the keyword From the original scenario example list that created the language model used by each original scenario and the scenario utterance understanding rule creation means for creating an utterance understanding rule with the content that was assigned to the transition destination scenario slot, the scenario keyword class Extract example sentences containing classes in the list, and class names in each extracted example sentence Replace with the class name to which the information indicating the attribute of the keyword in the scenario is added, further extract the word included in the scenario related keyword list from each example sentence in the additional example sentence list, replace the word with the scenario related word class, A transition recognition language model / dictionary creation means for creating a transition language model and a transition recognition dictionary by writing each example sentence in the additional example sentence list after replacement for each word by a morpheme analysis means, It is characterized by comprising.

  According to the present invention, it is possible to create a voice conversation scenario corresponding to a plurality of topics with a very small amount of work. It also eliminates restrictions such as having to remember special commands for users to transition to another topic, accepts most keywords that can be handled on the system, and transition dialogs if necessary. Thus, it is possible to create a content that automatically shifts to, so that it is possible to provide a very convenient voice dialogue system.

The voice dialogue scenario creation method and the voice dialogue creation device according to the present invention can be configured by hardware. However, in order to carry out more simply, the voice dialogue scenario creation program proposed in the present invention is installed in a computer, and the voice is created in the computer. The best mode is a mode of functioning as a dialogue scenario creation device.
In order for a computer to function as a voice conversation scenario creation device according to the present invention, information necessary for linking each original scenario from resources provided in the original scenario in the computer by a voice conversation scenario creation program installed in the computer, which Information that determines whether to link topics to the original scenario, information extraction means that extracts information for performing speech recognition suitable for each original scenario in the state where the topics have transitioned, and which information from the information extracted by this information extraction means A matching information generation unit that generates matching information for matching with an environment in which each original scenario can properly execute a voice response even when transitioning to the original scenario is constructed, and the matching information generation unit generates matching information, By holding this consistency information as a resource, each corresponds to only one different topic A plurality of original scenarios, can function as a possible speech dialog scenario corresponding to a plurality of topics.

The outline of the voice dialogue scenario creating apparatus according to the present invention will be described with reference to FIG. In the figure, reference numeral 100 denotes an original scenario group composed of original scenarios 100A, 100B,... 100N corresponding to only one different topic. Each of the original scenarios 100A, 100B,... 100N includes resources 101A, 101B,... 101N that store the names of the original scenarios, speech recognition / utterance understanding resources used in each of the original scenarios 100A, 100B,.
Reference numeral 200 denotes information extraction means. This information extraction means 200 extracts information necessary for connecting these original scenarios 100A, 100B,... 100N from the respective resources 101A, 101B,. Information necessary to link the original scenarios 100A, 100B,... 100N includes, for example, the names of the original scenarios 101A, 101B,... 101N, the file name of each original scenario, the location of the file, and the original scenario to be executed first. Information to be specified, etc.

The information extracted by the information extraction unit 200 is transferred to the matching information generation unit 300, the matching information generation unit 300 generates the matching information 400, and the matching information 400 is stored in the matching information storage unit 500.
The voice conversation scenario executing means 600 uses a plurality of resources 101A, 101B,... 101N of the original scenarios 100A, 100B,... 100N stored in the original scenario group 100 and the matching information 400 stored in the matching information storage unit 500. It functions as a spoken dialogue system corresponding to the topic.

Details of each part will be described in order with reference to FIG. Each original scenario 100A, 100B,... 100N includes scenario file names 101A-1, 101B-1,..., Keyword dictionaries 101A-2, 101B-2, etc., utterance understanding rule lists 101A-3, 101B-3,. Example sentence lists 101A-4, 101B-4,.
Here, the keyword dictionaries 101A-2, 101B-2,... Indicate keyword dictionaries that are input to input items (hereinafter referred to as slots) necessary for achieving the purpose used in each of the original scenarios 101A, 101B,. . The utterance understanding restriction lists 101A-3, 101B-3,... Are a collection of utterance understanding rules for extracting the keywords to be put into the slots from the speech-recognized word strings and allocating them to the slots. 101A-4, 101B-4,... Are examples of spoken sentences used for learning when creating a class language model and a word dictionary to be used for speech recognition when each original scenario is executed.

1 extracts scenario file name 102, scenario keyword class list 103, scenario related keyword list 104, and additional example sentence list as information to be prepared at the time of contact from these original scenarios 100A, 100B,. 105 and the contents explanation 106 of each scenario are extracted. At the same time, these pieces of information are input to the matching information generating means 300 shown in FIG.
In the embodiment shown in FIG. 3, the matching information generating unit 300 is configured by a transition resource creating unit 301, a navigation scenario creating unit 302, a transition scenario creating unit 303, and a linked information creating unit 304. .

The transition resource creation unit 301 generates a transition resource 401 that forms part of the matching information 400. The content of the transition resource 401 includes a transition keyword dictionary 401-1, a transition utterance understanding rule 401-2, a transition language model 401-3, and a transition recognition dictionary 401-4. Further, the transition resource creation means 301 also creates a scenario name keyword correspondence DB (database) 401-5.
The navigation scenario generation unit 303 generates a transition scenario 404, and the connection information generation unit 304 generates connection information 405.

Hereinafter, the operation of each unit will be described. Here, if necessary, the file name may include a network address of a computer holding the file, a location in a storage device in the computer, and the like. The same applies to the file name.
In each of the utterance understanding rule lists 101A-3, 101B-3,..., A symbol indicating the attribute (hereinafter referred to as a class) of a word to be extracted as a keyword to be input to the slot and the keyword at the position of the class. An instruction as to whether or not to distribute is shown in pairs. In order to perform the allocation to the slot with higher accuracy, information on the word strings before and after the appearance of the class symbol may be added. For example, from the utterance "From" Yokohama Station "to" Shinagawa Station ", the rules for allocating" Yokohama Station "to" Departure Station Slot "and" Shinagawa Station "to" Destination Station Slot "are as follows: become.

Example of utterance comprehension rules: “From <Station name>: From station slot to <Station name>: Arrival station slot”
In this utterance understanding rule example, “<station name>” indicates that a word belonging to the station name class appears, and “: departure station slot” and “: arrival station slot” are words included in the immediately preceding class symbol, respectively. Is assigned to each slot.
The keyword is composed of three pieces of information including the keyword notation, the class in the class language model, and the semantic feature of the keyword.
In addition, in the original scenarios 100A, 100B,..., Parameters related to timbre and speech speed when using voice synthesis when performing a voice response, and agents are displayed on the screen in order to express the individuality of each scenario agent. A description for designating image data or moving image data is also included.

  In order to connect scenarios, in addition to the above, in addition to the above, a scenario name which is a name of each original scenario to be connected (for example, “garbage separation guide”) and a keyword dictionary 101A- of the original scenarios 100A, 100B. 2, 101B-2... Are keywords that characterize the scenario (for example, if the scenario name is “garbage separation guidance”, the name of each garbage, a word such as garbage classification type) is called a scenario keyword, and The scenario keyword class list 103 is a list of classes to which each scenario keyword belongs.

A list of scenario-related keywords that are not included in the above scenario keywords, but the topics handled by the scenario and words that are strongly related to the topic name (such as "How to throw away garbage") are called scenario-related keywords. Scenario-related keyword list 104,
Example sentences of an utterance sentence in which a scenario-related word registered in the scenario-related word list appears, an additional example sentence list 104 that collects example sentences of an utterance sentence in which a scenario name appears, a topic handled in the scenario file, an outline thereof, and the like Are prepared for each original scenario 100A, 100B... 100N.

Also, in the linked original scenario, the name of the starting scenario indicating the scenario name from which each of the original scenarios before linking or the partial scenario of the navigation scenario 403 created in this embodiment should be started. 406 (see FIG. 2) and a navigation scenario name 407 (see FIG. 2), which is a scenario name (for example, “XX city general guidance”, etc.) to be given to the navigation scenario 403 as a generic name of the connected scenarios, are prepared. To do.
In the creation of the linked dialogue scenario, the scenario file name of each original scenario that constitutes each partial scenario in the linked original scenarios 100A, 100B,... 100N, the scenario name assigned to the scenario, and the start scenario When the linked information creation unit 304 reads the name, the file name of the navigation scenario 403 output from the navigation scenario creation unit 302, the scenario name assigned to the scenario, and the execution of the transition scenario are started The connection information 405 that lists the scenario name or scenario file name of the partial scenario to be executed first is output.

  Next, FIG. 4 shows an example of the configuration of the transition resource creation unit 301. The transition resource creation means 301 receives the keyword dictionary extracted by the information extraction means 200, the utterance understanding rule list, the original scenario example sentence list, the additional example sentence list, and the navigation scenario name, and uses these as transition resource creation means. The transition scenario keyword dictionary creation means 310 arranged in 301 outputs the transition keyword dictionary 401-1, the scenario name keyword correspondence DB creation means 320 outputs the scenario name keyword correspondence DB 401-5, and the transition utterance. The understanding rule creation means 330 outputs the transition utterance understanding rules 401-2, and the transition recognition language model / dictionary creation means 340 outputs the transition language model 401-3 and the transition word dictionary 401-4, respectively. To do.

The transition keyword dictionary creation means 310 generates the transition keyword dictionary 401-1 using the transition keyword dictionary creation method. FIG. 5 shows the flow.
In the transition keyword dictionary creation method, the procedure of scenario keyword extraction step S5-1, scenario keyword class reassignment step S5-2, scenario related word class reassignment step S5-3, scenario name keyword generation step S5-4 is performed. A step (S5-5) of repeating for all the original scenarios to be connected and writing a list of keywords obtained by those procedures to a file or the like.

In the scenario keyword extraction step S5-1, keywords belonging to the class indicated in the scenario keyword class list are extracted from the keyword dictionary 101A-2, 101B-2,.
Next, in the in-scenario keyword class reassignment step S5-2, the class names of all keywords extracted in the in-scenario keyword extraction step S5-1 are replaced with “in-scenario keyword-d”. Note that d is information indicating the attribute of the keyword in the scenario, and is the scenario name of the original scenario being processed at that time. If the scenario name is “garbage separation guidance”, the class name is replaced with “keyword in scenario-garbage separation guidance”.

Next, in the scenario related word class reassignment step S5-3, each keyword in the scenario related keyword list is read, and all of the keyword classes are replaced with “scenario related word-d”. d is the name of the scenario currently being processed as in the intra-scenario keyword class reassignment step S5-2.
Further, in the scenario name keyword generation step S5-4, a keyword is generated with the notation and the semantic feature as the scenario name and the class as the “scenario designation” class.
Execute the above for each original scenario, and generate the scenario name keyword for the navigation scenario name as well as the transition keyword dictionary creation method, then reassign the keyword class within the scenario, reassign the scenario related term class, and the scenario name Each keyword is rewritten by keyword generation, and all the generated keywords are written to a file or the like as transition keywords.

The scenario name keyword correspondence DB (database) creation means 320 creates a scenario name keyword correspondence DB-401 that acts as a table indicating to which original scenario a certain keyword belongs, using the scenario name keyword correspondence DB creation method. FIG. 6 shows a flow of a scenario name keyword correspondence DB creation method.
In the scenario name keyword correspondence DB creation method, the transition keyword dictionary 401-1 created by the transition keyword dictionary creation means 310 is read. Then, in the scenario name-notation acquisition step S6-1, keywords belonging to the “in-scenario keyword-d” class and the “scenario-related word-d” class are acquired. d is the name of each original scenario to be connected. Then, for each acquired keyword, the scenario name portion is acquired from the notation and the class name.

The correspondence between the extracted keyword description and the scenario name is registered as one record in the DB in DB registration processing step S6-2. The database to be registered may be a general relational database system accessed via the SQL language or the like, and a scenario name corresponding to the notation is listed on one line with a delimiter character such as a comma (,). It may be in the form of a text file.
The transition utterance understanding rule creation means 330 generates a transition utterance understanding rule 401-2 by using the transition utterance understanding rule creation method. FIG. 7 shows the flow of the transition utterance understanding rule creation method.

The utterance comprehension rule creation method for transition includes each of the four means of understanding rule extraction step S7-1, understanding rule rewriting step S7-2, rule generation 1 step S7-2, and rule generation 2 step S7-4. It is repeated for the original scenario, and the generated rules are written to a file or the like.
In the understanding rule extraction step S7-2, classes included in the scenario keyword class list 103 are included in the rules from the utterance understanding rule lists 101A-3, 101B-3 (see FIG. 2) of the original scenarios 100A and 100B. Extract utterance comprehension rules.

In the understanding rule rewriting step S7-2, the class name of each rule extracted in the understanding rule extracting step S7-1 is set to “in-scenario keyword-d” (d is the original scenario name currently being processed), and the allocation slot name is changed. Rewrite each in the “Transition destination scenario related word” slot.
In rule generation 1 step S7-3, if a keyword belonging to “in-scenario keyword-d” or “scenario-related word-d” appears in the speech-recognized word string, the keyword is placed in the “transition destination scenario-related word” slot. Generate an utterance comprehension rule with the content of “Sort by”.

In the rule generation step S7-4, if a keyword belonging to the “scenario designation” class appears in the speech-recognized word string, an utterance comprehension rule with the content that the keyword is assigned to the “transition destination scenario” slot is generated. To do.
The above is executed for each original scenario, and further, rule generation 2 step S7-4 is executed for the navigation scenario name, then, understanding rule rewriting step S7-2, rule generation 1 step S7-3, rule generation 2 step S7- The utterance comprehension rules rewritten and generated in step 4 are written out to a file or the like, and set as transition utterance comprehension rules 401-2.

The transition recognition language model / dictionary creation means 340 creates the transition language model 401-3 and the transition recognition dictionary 401-4 using the transition recognition language model / dictionary creation method. FIG. 8 shows the flow of the transition recognition language model / dictionary creation method.
The transition recognition language model / dictionary creation method includes an example sentence extraction step S8-1, an example sentence class conversion step S8-2, a keyword / class replacement step S8-3, and a morpheme analysis step for each original scenario to be connected. Step S8-4 is performed, and it includes a language model calculation / dictionary creation step S8-5 and a process step S8-6 for writing them out.

In the example sentence extraction step S8-1, in the scenario keyword class list from the original scenario example sentence list (the morpheme analysis is performed and the position where the keyword is entered is replaced by the class name) that has generated the language model used by each original scenario. Extract example sentences that contain classes in
In the example sentence class step S8-2, this time, the additional example sentence list 105 (FIG. 2) is read. If each example sentence in the additional example sentence list 105 includes a word included in the scenario-related keyword list 104, the word part is “ Replace with the “scenario-related word-d” class.
In the keyword-class conversion step S8-3, this time, the additional example sentence list 105 (FIG. 2) is read, and if each example sentence in the additional example sentence list 105 has a word included in the scenario-related keyword list 104, the word portion Is replaced with the “scenario-related word-d” class.

In the morpheme analysis step S8-4, each example sentence in the additional example sentence list after the replacement of the keyword with the class in the keyword-class replacement step S8-3 is divided for each word by using the morpheme analyzer, and the reading pseudonym Shake.
In the language model calculation / dictionary generation step S8-5, the example sentence class conversion step S8-2, the morpheme-analyzed example sentence created in the morpheme analysis step S8-4, and the transition keyword dictionary as information in the in-class word dictionary The language model and dictionary are created using the language model / dictionary creation method. For this, the methods disclosed in Japanese Patent Application Laid-Open Nos. 2004-69858 and 2004-53745 can be used.

The calculated language model and the generated dictionary are saved in a file or the like, respectively, and are used as a transition language model 401-3 and a transition recognition dictionary 401-4.
The navigation scenario creation means 302 (see FIG. 2) includes a navigation scenario name 403, a scenario name assigned to each original scenario, and a navigation scenario 403 that is an interactive scenario that plays an overall role in the connected interactive scenario. Is generated.
The navigating scenario 403 is an interactive scenario described to explain what kind of topic the system can perform. When generating a scenario for navigation, a flow that can be described without specifying each original scenario name to be linked is saved in advance as a template, and the navigation scenario creating means 302 is executed, When the name of the original scenario to be connected is determined, it can be realized by adding a necessary information for completion in the template and creating a complete scenario file.

  Moreover, the transition scenario generation unit 303 reads each original scenario name and the navigation scenario name, and generates a transition scenario. For example, if the user indicates one of the linked original scenarios, that is, if a keyword belonging to the scenario-designated class is entered in the “Transition Destination Scenario” slot, the original scenarios indicated by the scenario names are linked. It responds by enumerating the scenario name and prompts the user to speak. In addition, when there are a plurality of scenarios that are candidates for the transition destination, that is, as a result of searching the scenario name keyword correspondence DB 402 from the keyword notation input in the “transition destination scenario related word” slot, a plurality of corresponding scenario names are obtained. In this case, the dialogue scenario is described so as to conduct a dialogue such as confirming which one to transition to.

As described above, a linked dialogue scenario can be created.
In the dialogue scenario execution means 600 (see FIG. 1), the connection information is read and the original scenario scenario file (one of the original scenarios or the navigation scenario) to be executed first is read and executed.
If the scenario to be executed first is one of the original scenarios, the scenario for transition and the resource for transition are read at the same time. Perform utterance comprehension. When the utterance comprehension result is obtained from the transition resource and the recognition result from the original scenario source is not obtained or the recognition result from the original scenario source is obtained, the likelihood is higher than the result of the transition resource. If the state becomes extremely small, a response action based on the transition scenario is executed. When execution of another original scenario or navigation scenario is requested from the transition scenario based on the result of the dialog by the transition scenario, the dialog scenario executing means 600 reads and executes the scenario. When the navigation scenario is executed, the speech recognition / utterance understanding process is performed only for the transition resource. The navigating scenario also requests the dialog scenario executing means 600 to execute one of the original scenarios or the end of the dialog according to the result of the dialog, and the dialog scenario executing means 600 that received the request executes the original scenario. And system termination processing.

In this embodiment, information that has to be newly created manually as input for connection of the dialogue system only needs to enumerate names and keywords, and the creation cost is all that is to be handled by the dialogue system. There are far fewer rules to wait for the topic slot compared to each cost (> 2 times the number of slots).
From the generation process described above, the transition resource can recognize and understand the keywords handled by the original scenario to be linked and the keywords associated with the original scenario.
When an utterance understanding result is obtained from the transition resource, a transition scenario or a navigation scenario can be used to transition to a dialogue scenario that deals with another topic.

  Although the voice conversation scenario creation device described above can be realized by hardware, in practice, the voice dialogue scenario creation program according to the present invention is installed in a computer and the computer functions as a voice dialogue scenario creation device. The form is the most feasible. The voice interaction scenario creation program according to the present invention is described in a computer-readable program language, and is recorded on a recording medium such as a CD-ROM or a magnetic disk that can be read by the computer. The computer is installed from these recording media or through a communication line, and is decrypted by a CPU provided in the computer to execute a voice dialogue scenario creation operation.

  The voice dialogue scenario creating method and apparatus according to the present invention is utilized in a program creation field for constructing a voice dialogue system.

The block diagram for demonstrating the outline | summary of the voice dialogue scenario creation method and apparatus by this invention. The block diagram for demonstrating one Example of the voice dialogue scenario creation apparatus by this invention. FIG. 3 is a block diagram for explaining the continuation of FIG. 2. The block diagram for demonstrating the detail of the resource creation means for transition used for the Example of this invention. 6 is a flowchart for explaining the operation of the transition resource creating unit shown in FIG. The flowchart for demonstrating operation | movement of the scenario name keyword corresponding | compatible database preparation means used for the Example of this invention. The flowchart for demonstrating operation | movement of the utterance understanding rule preparation means for a transition used for the Example of this invention. The flowchart for demonstrating operation | movement of the recognition language model for a transition and the recognition dictionary preparation means used for the Example by this invention.

Explanation of symbols

100 original scenario group 401-1 transition keyword dictionary 100A to 100N original scenario 401-2 transition utterance understanding rules 101A to 101N resources 401-3 transition language model 101A-1 scenario file name 401-4 transition recognition dictionary 101A- 2 Keyword dictionary 401-5 Scenario keyword correspondence DB
101A-3 Utterance comprehension rule list 403 Navigating scenario 101A-4 Original scenario example list 404 Transition scenario
102 Scenario file name 405 Link information
103 Scenario Keyword List
104 Scenario-related keyword list
105 List of additional example sentences
106 Content of each scenario
200 Information extraction means
300 Consistency information creation means
301 Transition resource creation means
302 Dialogue scenario
303 Transition scenario creation means
304 Link information creation means

Claims (8)

A method for creating a voice conversation scenario comprising a plurality of original scenarios each corresponding to only one different topic, and creating a voice conversation scenario capable of corresponding to a plurality of topics by connecting the plurality of original scenarios.
Information necessary for linking each original scenario from the resources provided in the above original scenario, information for determining which original scenario the topic is to be transitioned to, and voice recognition suitable for each original scenario in the state where the topic is transitioned Extract information to be performed, generate matching information to match the environment in which each original scenario can properly execute voice response regardless of which original scenario transitions from these information, and retain this matching information A voice conversation scenario creation method characterized by creating a voice conversation scenario capable of dealing with a plurality of topics. In the voice interaction scenario creating method according to claim 1,
The above alignment information is
The name of each original scenario handled in the linked original scenario, the file name of the original scenario, the location of the file, the linkage information describing the information of the original scenario to be executed first,
Extract the words that are input to the input items necessary to achieve the objective from the speech recognition resources in each original scenario, the name of each original scenario, and related phrases that recall the original scenario, and transition these to the topic Resources for transition to properly recognize speech as a transition term phrase that causes
When a transition term phrase is recognized, it confirms whether the transition to the original scenario corresponding to the recognition term phrase is displayed, displays the transition, and executes an operation to transfer control to another connected original scenario. And a transition dialogue scenario for
A method for creating a voice dialogue scenario, comprising: A voice conversation scenario creation device that includes a plurality of original scenarios each corresponding to only one different topic, and creates a voice conversation scenario capable of supporting a plurality of topics by connecting the plurality of original scenarios.
Information necessary for linking each original scenario from the resources provided in the above original scenario, information for determining which original scenario the topic should transition to, and speech recognition suitable for each original scenario in the state where the topic has transitioned Information extracting means for extracting information to be performed;
A consistency information creating means for processing and uttering a user's utterance on a topic handled by another original scenario regardless of which original scenario is being executed from the information extracted by the information extracting means,
A voice dialogue scenario creating device comprising: In the voice dialogue scenario creation device according to claim 3,
The matching information creation means
A concatenation information creation means for creating concatenation information describing the name of each original scenario handled in the concatenated original scenario, the file name of the original scenario, the location of the file, and the information of the original scenario to be executed first;
Extract the words that are input to the input items necessary to achieve the objective from the speech recognition resources in each original scenario, the name of each original scenario, and related phrases that recall the original scenario, and transition these to the topic A transition resource creating means for creating a transition resource for appropriately recognizing and outputting an understanding result as a transition term phrase that causes
When a transition term phrase is recognized, it confirms whether the transition to the original scenario corresponding to the recognition term phrase is displayed, displays the transition, and executes an operation to transfer control to another connected original scenario. A transition scenario creating means for creating a transition scenario for
A voice dialogue scenario creation device characterized by comprising: In the voice dialogue scenario creation device according to claim 3 or 4,
In addition to the connection information creation unit, the transition resource creation unit, and the transition scenario creation unit, the matching information generation unit includes:
A spoken dialogue scenario creating device comprising: a navigation scenario creating means for displaying dialogue contents executable in the linked original scenario and guiding a voice dialogue procedure. The voice conversation scenario creation device according to any one of claims 3 to 5,
The transition resource creation means
A transition keyword dictionary that extracts keywords belonging to a predetermined class from a keyword dictionary in resources provided in each original scenario, and creates a transition keyword collection in which information indicating the attribute of the keyword is added to the extracted keywords. Creating means;
Scenario name keyword correspondence database creation means for creating a scenario name keyword correspondence database that acts as a table indicating which original scenario each keyword described in each original scenario belongs;
From the list of utterance understanding rules described in each original scenario, utterance understanding rules included in the rules of the classes included in the scenario keyword class list are extracted, and the class name of each extracted rule is used as a keyword in the scenario. Rewrite the name to which the attribute information is added and the distribution destination slot name to the transition destination scenario related word, respectively, and the keyword with the attribute information added to the keyword in the scenario or the keyword belonging to the transition destination scenario related word is spoken If it appears in the recognized word string, the utterance comprehension rule with the content that the keyword should be assigned to the transition destination scenario-related word slot, and the keyword belonging to the scenario specification class appears in the voice-recognized word string , An utterance comprehension rule with the content that the keyword is assigned to the destination scenario slot Transition for speech understanding rule generation means for generating,
Extract example sentences including classes in the scenario keyword class list from the original scenario example sentence list that generated the language model used by each original scenario, and use the class name in each extracted example sentence to search for keywords in the scenario. Replaced with the class name to which the information to be provided is added, further extracted the keywords included in the scenario-related keyword list from each example sentence in the additional example sentence list, replaced the keyword class with the scenario-related word class, and additional example sentences A transition recognition language model / dictionary creation means for creating a transition language model and a transition recognition dictionary from each of the example sentences in the list by morphological analysis means for each word;
A voice conversation scenario creation device characterized by comprising:   7. A spoken dialogue scenario creation program that is described in a program language that can be decoded by a computer and that causes the computer to function as at least a voice dialogue scenario creation device according to claim 3.   A recording medium comprising a recording medium readable by a computer, wherein the voice conversation scenario creating program according to claim 7 is recorded on the recording medium.

Images