JP4535804B2 - Spoken dialogue sequence state notation method, program, and spoken dialogue apparatus - Google Patents

Description

  The present invention relates to a voice interaction device, and more particularly to a method for expressing a state of a conversation sequence.

  In a speech dialogue system such as a conventional CTI (Computer Telephony Integration), there are many examples where the dialogue itself is described as a program. The feature is that various descriptions can be made by writing down the response process with a code for each voice input.

In VoiceXML (extensible Markup Language), a script is described for each state, and a browser analyzes the script (Non-Patent Document 1). VoiceXML aims to improve content productivity by using highly readable XML.
FIT (Information Science and Technology Forum) 2003, F-006, PP. 215-216

  For information access from a ubiquitous environment such as a voice portal or a car navigation device, a voice dialogue interface is often provided. In order to add various voice services in the future and to modify voice services according to the situation as needed, it is necessary to facilitate the creation and correction of voice conversation scenarios. However, with the conventional technology for describing state transitions, response and state transitions to voice input are coded as programs, and VoiceXML and the like write down scenarios for each state, so specialized programming skills are required. , It was difficult to estimate the effect of some modifications on the entire program (in order to make minor service modifications, it was necessary to modify many parts of the entire program).

  In addition, when used in a high-speed moving environment such as a car navigation device, the line between the center server and the client terminal may be disconnected. In the conventional telematics service using web, parameters such as user ID are retained by using cookie, but state transition is not managed, and when reconnecting, the main menu (top menu) is used again. It is necessary to follow. In addition, in voice dialogue systems such as CTI, the ease of description is enhanced by using VoiceXML, but since the XML file is still switched for each state, state restoration at reconnection is fully supported. It ’s hard to say.

  An object of the present invention is to provide a spoken dialogue sequence state notation method and a spoken dialogue apparatus in which scenarios can be easily added and modified.

  The present invention requires a state of a voice conversation sequence (speech exchange from the start of a voice conversation to a state where the purpose is achieved in order to use the voice conversation for a certain purpose) to achieve the purpose. The information is uniquely associated with the state of the slot in which the information is stored.

  Here, the state of the slot is “empty slot” in which no information is stored in the slot, and information is stored in the slot, but the information is not confirmed by the user. "Undetermined slot" which is the state, information stored in the slot is confirmed to the user, "confirmed slot" which is the confirmed state, when the user performs voice input simultaneously to a plurality of slots, When the user says that the recognition result is not correct, four states of “sequentially checking slot” in which information of each slot is being checked are taken.

  By expressing the voice interaction sequence in the slot state, it is possible to easily add or modify a scenario and to return the state at the time of reconnection.

  In the present invention, by expressing the state of the voice interaction sequence as a slot state of 2 bits (= 4 states), it is possible to describe the acceptability of the voice input by 1 bit, and the transition destination state is uniquely determined by the slot state. Therefore, it is excellent in completeness, and all the contents that can be described in the state transition diagram are normal systems, and contents that cannot be described are not accepted, so there is basically no concept of abnormal systems, and the construction and maintenance of dialog sequences is very easy. There is an effect to say.

  In addition, since this dialog sequence depends only on the slot state, the restart process can be performed simply by holding the dialog sequence name and the slot state, so there is no need for complicated restart processing or flow playback. There is an effect that it becomes easy to implement necessary and sufficient restart processing.

  Next, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram of a voice interactive apparatus according to an embodiment of the present invention. The voice interaction apparatus of this embodiment includes a voice input / output receiving unit 1, a data input / output receiving unit 2, a scenario DB 3, a recognition dictionary DB 4, a voice recognition unit 5, and a scenario analysis unit 6.

  The voice input / output reception unit 2 outputs the voice guidance output from the scenario analysis unit 6 to the user terminal device 7, and receives the voice input input from the terminal device 7 by the user to the voice guidance, and the scenario analysis unit 6 is output. The data input / output accepting unit 2 outputs the dialogue sequence name and slot state output from the scenario analysis unit 6 to the user terminal device 7, and the dialogue sequence name input from the terminal device 7 when the voice dialogue sequence is resumed. And the slot state are output to the scenario analysis unit 6. The scenario DB 3 stores, for each voice conversation sequence, a voice conversation scenario used when the scenario analysis unit 6 controls the voice conversation. The recognition dictionary DB 4 uses a recognition grammar used when the voice recognition unit 5 performs voice recognition. Is stored. The speech recognition unit 5 performs speech recognition processing on the speech output from the scenario analysis unit 6 using the speech recognition grammar specified by the scenario analysis unit 6 and outputs the recognition result to the scenario analysis unit 6. . The scenario analysis unit 6 outputs the voice guidance to the voice input / output receiving unit 1 and the dialog sequence name and the slot state (empty slot, unconfirmed slot, confirmed slot, sequentially confirmed slot) to the data input / output receiving unit 2. In response to the voice input of the user corresponding to the voice guidance from the voice input / output receiving unit 1, the voice recognition unit 5 is instructed to perform voice recognition processing with an appropriate voice recognition grammar, and according to the voice recognition result Then, the state of the voice dialogue scenario in the scenario DB 3 is changed, and at the time of resuming the voice dialogue, the voice dialogue is resumed by the dialogue sequence name and the slot state output from the data input / output accepting unit 2.

[First Embodiment]
Scenario Description Example FIG. 2 shows a scenario description example of the present invention.

  Hereinafter, it demonstrates along a flow.

  The initial state is a state in which the number of slots X that can accept a scenario is arranged. In the scenario of FIG. 2, since the number of store type slots is 2, the initial state is XX. The voice input / output receiving unit 1 performs a reading process for the user using the character string defined in the Guidance of the scenario description as guidance. In this state, 1 is entered in the scenario description column for inputs that can be accepted, and 0 is entered in the scenario description column for inputs that cannot be accepted.

For example, in the scenario of FIG. 2, it is assumed that “Japanese food”, “Western food”, and “Chinese food” are acceptable as the types of shops. In that case, the comparison between the input that can be accepted by this scenario and the utterance example is as follows.
SLa: Japanese food SLa SLa: Japanese food or Western food SLa NOT SLa: Chinese food, not Western food SLA: Western food, ANY: Anything, YES: Yes NO: No Initial state XX
In the initial state XX, the voice input / output receiving unit 1 plays the guidance “Is the store input?” And Sec1_XX. bnf is set in the voice recognition unit 5 to wait for user input. Note that Sec1_XX. bnf is a file name of a file in which vocabulary that can be recognized is described in BNF grammar, for example, as shown in FIG. In the scenario shown in FIG. 2, the input that can be accepted in the initial state XX is an input in which 1 is entered in the column. In this case, (SLa | SLa SLa | ANY) can be accepted. Note that the input “return” is analyzed and processed in the scenario analysis unit 6 without being described in the scenario because the previous state is stacked and held in the scenario analysis unit 6.

  When “Japanese food” is input in the initial state XX, Sec1_XX. In accordance with bnf, the value {“FOOD_Japanese food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Then, the slot is filled from the front of the slot X by the content of the return value from the voice recognition unit 5 (excluding the category “FOOD_”). In this case, since all slots are empty, “Japanese food” is set in the first slot X. Here, “Japanese food” in the slot X is an unconfirmed input state in which the user is not confirmed in consideration of misrecognition of voice recognition or the like. Accordingly, the first slot X becomes the slot Y, and the second slot X does not change, so the state transitions to the state YX.

  In addition, when “Japanese or Western food” is input in the initial state XX, Sec1_XX. According to bnf, the value {“FOOD_Japanese food”, “OR”, “FOOD_Western food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there are inputs to the two slots, and considers that [SLa SLa] has been input. Then, the slot is filled from the front of the slot X. In this case, since all slots are empty, “Japanese food” is set in the first slot X, and “Western food” is set in the second slot X. Here, “Japanese food” in slot X and “Western food” in slot X are unconfirmed input states that have not been confirmed by the user. Therefore, the first slot X becomes slot Y, the second slot X becomes slot Y, and the state transitions to YY.

  Further, when “anything” is input in the initial state XX, Sec1_XX. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Then, the slot is filled from the front of the slot X. In this case, since all slots are empty, “ANY” is set in the first slot X. Here, “ANY” in the slot X is an unconfirmed input state. Accordingly, the first slot X becomes the slot Y, and the second slot X does not change, so the state transitions to the slot YX.

State YX
In the state YX, one slot is filled with unconfirmed information. Therefore, a dialogue is performed in which unconfirmed information is confirmed and confirmed. The voice input / output receiving unit 1 plays a guidance such as “Are you sure you want to input Japanese food?”, And Sec1_YX. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state YX is an input in which 1 is entered in the column. In this case, (SLa | SLa SLa | SLa NOT SLa | NOT SLa | ANY | YES | NO) is accepted. It becomes possible.

  When “Western food” is input in the state YX, Sec1_YX. In accordance with bnf, the value {“FOOD_Western food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is YX, the slot is filled from the front of the indeterminate or empty slot. In this case, since the first slot Y is unconfirmed and an input “Western food” different from the value “Japanese food” already in the slot Y is accepted, the content of the slot Y is set to “Western food”. Accordingly, the slot state becomes YX, and the guidance returns to the same state YX as “Are you sure you want to input Western food?”.

  When “Japanese food” is input in the state YX, Sec1_YX. In accordance with bnf, the value {“FOOD_Japanese food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is YX, the slot is filled from the front of the indeterminate or empty slot. In this case, since the first slot Y is unconfirmed and the same input as the value “Japanese food” already in the slot Y is accepted, the contents of the slot Y are set to “Japanese food” as they are, and the same state YX is set. The guidance “Return input is Japanese food?” Will be repeated.

  When “Western food or Chinese” is input in the state YX, Sec1_YX. According to bnf, the value {“FOOD_Western food”, “OR”, “FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there are inputs to the two slots, and considers that [SLa SLa] has been input. Since the current slot state is YX, the slot is filled from the front of the indeterminate or empty slot. In this case, since the first slot Y is unconfirmed, and the value “Japanese food” different from the input “Western food” and “Chinese food” is set, the content of the first slot Y is “Western food”. Since the second slot X is reset, the contents of the slot X are set to “Chinese”. Therefore, the first slot Y becomes slot Y, the second slot X becomes slot Y, and the state transitions to YY.

  In addition, when “Japanese or Western food” is input in the state YX, Sec1_YX. According to bnf, the value {“FOOD_Japanese food”, “OR”, “FOOD_Western food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there are inputs to the two slots, and considers that [SLa SLa] has been input. Since the current slot state is YX, the slot is filled from the front of the indeterminate or empty slot. In this case, since the first slot Y is unconfirmed and the same value as the input “Japanese food” is set, the content of the slot Y is held as “Japanese food”, and the second slot X is empty. , The contents of slot X are set to “Western food”. Therefore, slot Y becomes slot Y, slot X becomes slot Y, and transitions to state YY.

  In the state YX, when “Western food instead of Japanese food” is input, Sec1_YX. In accordance with bnf, the values {“FOOD_Japanese food”, “NOT”, “FOOD_Western food”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [SLa NOT SLa] has been input. Since the current slot state is YX, the value “Japanese food” of the undetermined slot Y is referred to and reset to “Western food” which is the corrected input value. Therefore, the slot Y becomes the slot Y, the slot X remains as it is, and a transition is made to the state YX.

  In the state YX, when “Western food is not so” is entered, Sec1_YX. In accordance with bnf, the values {“NOT”, “FOOD_Western food”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [NOT SLa] has been input. Since the current slot state is YX, the value “Japanese food” of the undetermined slot Y is referred to and reset to “Western food” which is the corrected input value. Therefore, the slot Y becomes the slot Y, the slot X remains as it is, and a transition is made to the state YX.

  When “anything” is input in the state YX, Sec1_YX. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is YX, the value “Japanese food” of the undetermined slot Y is referred to, and since it is different from the input value “ANY”, it is reset to the corrected input value “ANY”. Therefore, the slot Y becomes the slot Y, the slot X remains as it is, and a transition is made to the state YX.

  If “Yes” is input in the state YX, Sec1_YX. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is YX, the value of the undetermined slot Y becomes the confirmed slot X, the slot X remains as it is, and the slot state becomes ZX. Here, since the decision slot = 1 falls below the number of slots = 2, the sequence shifts to a sequence for confirming whether only this input is necessary. This is a state ZXC because it is confirmed whether or not the transition to ZX is allowed (confirmation). Therefore, when “Yes” is input in the state YX, the state transitions to the state ZXC. However, if the accepted Y is “ANY”, the slot state is set to ZZ = {“ANY”, “ANY”}, and Sec1_ZZ. Run script.

  When “No” is input in the state YX, Sec1_YX. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is YX, the value of the undetermined slot Y becomes the empty slot X, the second slot X remains as it is, and a transition is made to the state XX.

State YY
In the state YY, the two slots are filled with unconfirmed information. Therefore, a dialogue is performed in which unconfirmed information is confirmed and confirmed. The voice input / output reception unit 1 circulates guidance such as “Do you want to input Japanese and Western foods?”, Sec1_YY. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state YY is an input in which 1 is entered in the column. In this case, (SLa | SLa SLa | SLa NOT SLa | NOT SLa | ANY | YES | NO) is accepted. It becomes possible. Here, since the acceptable input in the state YY and the acceptable input in the state YX are completely the same, the recognition grammar is Sec1_YX. It is the same as bnf.

  When “Chinese” is input in the state YY, Sec1_YY. In accordance with bnf, the value {“FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is YY, the values of the first Y and the second Y which are unconfirmed slots are deleted, and the slot is filled from the front with the input value “Chinese”. Therefore, the state transitions to the state YX.

  When “Western food or Chinese” is input in the state YY, Sec1_YY. According to bnf, the value {“FOOD_Western food”, “OR”, “FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there are inputs to the two slots, and considers that [SLa SLa] has been input. Since the current slot state is YY, the values of the first Y and the second Y, which are indeterminate slots, are deleted, and the slots are filled from the front with the input values “Western food” and “Chinese food”. Therefore, the state transitions to the same state YY.

  When “Chinese food is not Japanese food” is input in the state YY, Sec1_YY. In accordance with bnf, the values {“FOOD_Japanese food”, “NOT”, “FOOD_Chinese”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [SLa NOT SLa] has been input. Since the current slot state is YY, the first and second Y values which are unconfirmed slots are compared with the value of the input to be corrected “Japanese food”, and the value of the corresponding first slot Y is set to “ Change to “Chinese”. The value of the second slot Y is not changed. Therefore, the state transitions to the same state YY.

  If “Chinese but not so” is input in the state YY, Sec1_YY. In accordance with bnf, the values {“NOT”, “FOOD_Chinese”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input and considers that [NOT SLa] has been input. Since the current slot state is YY, the values of the first Y and the second Y which are unconfirmed slots are deleted, and the slot is filled from the front with the input value “Chinese”. Therefore, the slot state becomes YX, and the state changes to YX.

  When “anything” is input in the state YY, Sec1_YY. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is YY, the values of the first Y and the second Y which are unconfirmed slots are deleted, and the slot is filled from the front with the input value “ANY”. Therefore, the slot state becomes YX, and the state changes to YX.

  When “Yes” is input in the state YY, Sec1_YY. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is YY, the first unconfirmed slot Y becomes the confirmed slot Z, the second unconfirmed slot Y becomes the confirmed slot Z, and the state corresponds to the state ZZ. Here, since the number of slots equal to the number of slots = 2 has become a definite slot, the purpose of this interactive sequence of filling the slot is achieved. Therefore, Sec1_ZZ. The process described in the script is executed.

  When “No” is input in the state YY, Sec1_YY. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is YY and there are a plurality of unconfirmed slots, the process proceeds to the sequential confirmation process. Therefore, the sequential confirmation state is defined as W, and the state transitions to the state WW.

State WW
In the state WW, two slots are filled with necessary sequential confirmation information. Therefore, a dialogue is performed in which unconfirmed information is sequentially confirmed and confirmed. The voice input / output reception unit 1 sends a guidance such as “Are you sure you want to use Japanese food for the first input?”, And Sec1_WW. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state WW is an input in which 1 is entered in the column. In this case, (SLa | SLa SLa | SLa NOT SLa | NOT SLa | ANY | YES | NO) is accepted. It becomes possible. Here, since the acceptable input in the state WW and the acceptable input in the state YX are completely the same, the recognition grammar is Sec1_YX. It is the same as bnf.

  When “Chinese” is input in the state WW, Sec1_WW. In accordance with bnf, the value {“FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is WW, the value “Japanese food” of the first unconfirmed slot W being confirmed is reset by the input value “Chinese food”. Since “Chinese Chinese” is an indeterminate state, the sequence proceeds to the confirmation sequence again, and the state transitions to the state WW.

  When “Western food or Chinese” is input in the state WW, Sec1_WW. According to bnf, the value {“FOOD_Western food”, “OR”, “FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there are inputs to the two slots, and considers that [SLa SLa] has been input. Since the current slot state is WW, the values of the first unconfirmed slot W and the second unconfirmed slot W being sequentially confirmed are deleted, and the slots are filled from the front with the input values “Western food” and “Chinese food”. . Accordingly, the state transitions to the state YY.

  In the state WW, when “Chinese instead of Japanese” is input, Sec1_WW. In accordance with bnf, the values {“FOOD_Japanese food”, “NOT”, “FOOD_Chinese”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [SLa NOT SLa] has been input. Since the current slot state is WW, the value of the first confirmed slot W that is not being sequentially confirmed is changed to “Chinese”. Since “Chinese Chinese” is an indeterminate state, the sequence proceeds to the confirmation sequence again, and the state transitions to the state WW.

  When “Chinese but not so” is input in the state WW, Sec1_WW. In accordance with bnf, the values {“NOT”, “FOOD_Chinese”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input and considers that [NOT SLa] has been input. Since the current slot state is WW, the value of the first unconfirmed slot W being sequentially confirmed is deleted, and the slot W is filled with the input value “Chinese”. Since “Chinese Chinese” is an indeterminate state, the sequence proceeds to the confirmation sequence again, and the state transitions to the state WW.

  When “anything” is input in the state WW, Sec1_WW. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is WW, the first unconfirmed slot W being sequentially confirmed is reset by the input value “ANY”. Therefore, the slot state becomes WW, and a transition is made to the state WW.

  When “Yes” is input in the state WW, Sec1_WW. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is WW, the first unconfirmed slot W that is being sequentially confirmed becomes the confirmed slot Z and transitions to the state ZW.

  When “No” is input in the state WW, Sec1_WW. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is WW, the first unconfirmed slot W being sequentially confirmed becomes the empty slot X and the slot state becomes XW, but the confirmation information (Z) → unconfirmed information (Y, W) In accordance with the rule that slots are sorted in the order of empty (X), X and W are interchanged, the slot state becomes WX, and the state changes to WX.

State WX
In the state WX, one slot is sequentially confirmed. The voice input / output reception unit 1 plays a guidance such as “Are you sure you want to input Western food?”, And Sec1_XW. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario, the input that can be accepted in the state WX is an input in which 1 is entered in the column. In this case, (SLa | SLa SLa | SLa NOT SLa | NOT SLa | ANY | YES | NO) can be accepted. Here, since the acceptable input in the state WX and the acceptable input in the state YX are completely the same, the recognition grammar is Sec1_YX. It is the same as bnf.

  When “Chinese” is input in the state WX, Sec1_XW. In accordance with bnf, the value {“FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is WX, the value “Western food” of the unconfirmed slot W being confirmed is changed to the input value “Chinese”. Therefore, the slot state becomes YX, and the state changes to YX.

  When “Western food or Chinese” is input in the state WX, Sec1_XW. According to bnf, the value {“FOOD_Western food”, “OR”, “FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there are inputs to the two slots, and considers that [SLa SLa] has been input. Since the current slot state is WX, the value of the undetermined slot W during the sequential confirmation is deleted, and the slot is filled from the front with the input values “Western food” and “Chinese”. Accordingly, the state transitions to the state YY.

  When “Chinese instead of Western food” is input in the state WX, Sec1_XW. According to bnf, the value {“FOOD_Western food”, “NOT”, “FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [SLa NOT SLa] has been input. Since the current slot state is WX, the value of the undetermined slot W during the sequential confirmation is compared with the input “Western food” and changed to “Chinese”. Therefore, the slot state becomes YX, and the state changes to YX.

  When “Chinese but not so” is input in the state WX, Sec1_XW. In accordance with bnf, the values {“NOT”, “FOOD_Chinese”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input and considers that [NOT SLa] has been input. Since the current slot state is WX, the value of the unconfirmed slot W during the sequential confirmation is deleted, and the slot W is filled with the input value “Chinese”. Therefore, the slot state becomes YX, and the state changes to YX.

  When “anything” is input in the state WX, Sec1_XW. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is WX, the value of the undetermined slot W being sequentially confirmed is deleted, and the slot is filled with the input value “ANY”. Therefore, the slot state becomes YX, and the state changes to YX.

  When “Yes” is input in the state WX, Sec1_XW. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is WX, the unconfirmed slot W being sequentially confirmed becomes the confirmed slot Z, and the slot state becomes ZX. Here, the number of slots that are less than the two slots that can be input is determined, and therefore, a sequence for confirming whether or not only this input is necessary, as in the case of inputting “Yes” in the state YX. Transition to. This is a state ZXC because it is confirmed whether or not the transition to ZX is allowed (confirmation). Therefore, when “Yes” is input in the state WX, the state transitions to the state ZXC.

  When “No” is input in the state WX, Sec1_XW. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is WX, the unconfirmed slot W that is being sequentially confirmed becomes the empty slot X, and transitions to the state XX.

State ZW
In the state ZW, it becomes a state in which confirmation is sequentially performed for one slot. The voice input / output receiving unit 1 circulates guidance such as “Are you sure you want Western food for the next input?”, And Sec1_ZW. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state ZW is an input in which 1 is entered in the column. In this case, (SLa | SLa NOT SLa | NOT SLa | ANY | YES | NO) is acceptable. .

  When “Chinese” is input in the state ZW, Sec1_ZW. In accordance with bnf, the value {“FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is ZW, the value “Western food” of the unconfirmed slot W being confirmed is changed to the input value “Chinese”. Therefore, the state changes to the state ZW.

  When “Chinese instead of Western food” is input in the state ZW, Sec1_ZW. According to bnf, the value {“FOOD_Western food”, “NOT”, “FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [SLa NOT SLa] has been input. Since the current slot state is ZW, the value of the unconfirmed slot W during the sequential confirmation is compared with the input “Western food” and is changed to “Chinese”. Therefore, the state changes to the state ZW.

  When “Chinese but not so” is input in the state ZW, Sec1_ZW. In accordance with bnf, the values {“NOT”, “FOOD_Chinese”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input and considers that [NOT SLa] has been input. Since the current slot state is ZW, the value of the unconfirmed slot W during the sequential confirmation is deleted, and the slot W is filled with the input value “Chinese”. Therefore, the state is changed to ZW.

  When “anything” is input in the state ZW, Sec1_XW. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is ZW, the value of the undetermined slot W during the sequential confirmation is deleted, and the slot W is filled with the input value “ANY”. Therefore, the state is changed to ZW.

  When “Yes” is input in the state ZW, Sec1_ZW. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is ZW, the unconfirmed slot W being sequentially confirmed becomes the confirmed slot Z, the slot state becomes ZZ, and the state corresponds to the state ZZ. Here, since the number of slots equal to the number of slots = 2 has become a definite slot, the purpose of this interactive sequence of filling the slot is achieved. Therefore, Sec1_ZZ. The process described in the script is executed.

  When “No” is input in the state ZW, Sec1_ZW. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is ZW, the unconfirmed slot W that is being sequentially confirmed becomes the empty slot X. Therefore, the slot state is ZX, and the number of slots less than that is determined for two input possible slot numbers. Therefore, as in the case where “Yes” is input in the state YX, the process proceeds to a sequence for confirming whether only this input is necessary. This is a state ZXC because it is confirmed whether or not the transition to ZX is allowed. Therefore, when “Yes” is input in the state XW, the state transitions to the state ZXC.

State ZXC
In the state ZXC, when slots less than the number of slots that can be input are confirmed, it is confirmed whether or not the present dialogue sequence is completed only by the confirmed information. The voice input / output receiving unit 1 circulates guidance such as “Is it OK to input only Japanese food?”, And Sec1_ZXC. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state ZXC is an input in which 1 is entered in the column, and in this case, (YES | NO) is acceptable.

  When “Yes” is input in the state ZXC, Sec1_ZXC. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is ZXC, it is considered that the purpose of this dialogue sequence has been achieved, and Sec1_ZZ. The process described in the script is executed.

  When “NO” is input in the state ZXC, Sec1_ZXC. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is ZXC, the process shifts to a sequence for determining an undetermined empty slot X. That is, when “No” is input in the state ZXC, the state ZX is transitioned to.

State ZX
In the state ZX, when slots less than the number of slots that can be input are determined, information is input to the remaining empty slots. The voice input / output receiving unit 1 sends a guidance such as “What is the other shop type?”, And Sec1_ZX. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state ZX is an input in which 1 is entered in the column. In this case, (SLa | ANY) is acceptable.

  When “Chinese” is input in the state ZX, Sec1_ZX. In accordance with bnf, the value {“FOOD_Chinese”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is ZX, the input value “Chinese” is stored in the empty slot X. Therefore, the state transitions to the state ZY.

  When “anything” is input in the state ZX, Sec1_ZX. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is ZX, the value of the empty slot X is deleted, and the slot is filled with the input value “ANY”. Therefore, the state transitions to the state ZY.

State ZY
In the state ZY, when slots less than the number of slots that can be input are determined, the information input to the remaining empty slots is checked. The voice input / output reception unit 1 plays a guidance such as “Are you sure you want to use Chinese?”, And Sec1_ZY. bnf is set in the voice recognition unit 5 to wait for user input. In the scenario shown in FIG. 2, the input that can be accepted in the state ZY is an input in which 1 is entered in the column. In this case, (SLa | SLa NOT SLa | NOT SLa | ANY | YES | NO) is acceptable. .

  When “Western food” is input in the state ZY, Sec1_ZY. In accordance with bnf, the value {“FOOD_Western food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [SLa] has been input. Since the current slot state is ZY, the value of the undetermined slot Y is changed to the input value “Western food”. Therefore, the state transitions to the same state ZY.

  When “Western food instead of Chinese” is input in the state ZY, Sec1_ZY. According to bnf, the value {“FOOD_Chinese”, “NOT”, “FOOD_Western food”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [SLa NOT SLa] has been input. Since the current slot state is ZY, the value of the unconfirmed slot Y corresponding to the correction object “Chinese Chinese” is changed to the input value “Western food”. Therefore, the state transitions to the same state ZY.

  When “Western food is not so” is input in the state ZY, Sec1_ZY. In accordance with bnf, the values {“NOT”, “FOOD_Western food”} are returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a correction input for one slot, and considers that [NOT SLa] has been input. Since the current slot state is ZY, the value of the undetermined slot Y is changed to the input value “Western food”. Therefore, the state transitions to the same state ZY.

  When “anything” is input in the state ZY, Sec1_ZY. In accordance with bnf, the value {“ANY”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot, and considers that [ANY] has been input. Since the current slot state is ZY, the value of the indeterminate slot Y is deleted, and the slot is filled with the input value “ANY”. Therefore, the state transitions to the same state ZY.

When “Yes” is input in the state ZY, Sec1_ZY. In accordance with bnf, the value {“YES”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an acceptance intention input for the confirmation guidance, and considers that [YES] has been input. Since the current slot state is ZY, the slot state becomes ZZ, and it is considered that the purpose of this dialogue sequence has been achieved, and Sec1_ZZ. The process described in the script is executed. However, different processing is performed when the accepted Y is as follows.
When Y = Z, the slot contents are merged, so the slot state is ZX. Therefore, a slot less than the number of slots that can be input is determined, and the state transits to the state ZXC.
・ When Y = “ANY”, the content of the confirmed Z conflicts with the content of the confirmed Z. However, when the confirmed content is prioritized, and Z = “Japanese food”, the slot state is changed to ZZ = {“Japanese food”, ” ''} And Sec1_ZZ. Run script.

  If “NO” is input in the state ZY, Sec1_ZY. In accordance with bnf, the value {“NO”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is a rejection intention input for the confirmation guidance, and considers that [NO] has been input. Since the current slot state is ZY, the undetermined slot Y becomes the empty slot X. Therefore, a slot less than the number of slots that can be input is determined, and the state transits to the state ZXC.

Sec1_ZZ. About Sec1_ZZ. The script is a script that is executed when the dialogue sequence Sec1 is completed. Here, a search processing execution method and a dialog sequence name to be started after the search are described. Sec1_ZZ. An example of script description is shown in FIG. For example, when “Japanese food” and “Western food” are confirmed in the dialogue sequence (Sec1), SQL of “select ^* from table where SLa =“ Japanese food ”| SLa =“ Western food ”” is called as DB access, The returned array @Result is passed to the next interactive sequence (Sec2), and Sec2 is activated.

[Second Embodiment]
An embodiment of the disconnection / resumption process will be described.

  Since the dialogue sequence of the present invention depends only on the slot state, the interruption / return operation can be easily performed by designating the slot state.

  For example, in the state YY, when {"Japanese food", "Western food"} is input and the confirmation guidance is interrupted ("Are you sure you want Japanese food and Western food?"), The terminal device 7 displays the dialogue sequence name Sec1. And state YY and {“Japanese food”, “Western food”} are stored. When the connection is resumed, the terminal device 7 transmits the dialogue sequence name Sec1 and the state YY and {“Japanese food”, “Western food”} to the voice interactive device. The scenario analysis unit 6 of the voice interaction device can return to the state before disconnection by calling and executing the interaction sequence Sec1 with the state YY and the slot value {“Japanese food”, “Western food”}.

[Third Embodiment]
An embodiment in the case where there are multiple types of slots will be described.

  For example, in the parking lot search, consider a case where a search such as “500 yen / H or less within 800 m” is performed using the distance from the current location and the fee. In this case, since the distance and the charge are input in different categories, different input IDs, SLa and SLb are assigned to the respective categories, and the number of slots available for each category is designated. In the example scenario shown in FIG. 6, SLa {DISTANCE} = 1 and SLb {PRICE} = 1. In the case of a plurality of categories and a plurality of slots, SLa = 2, SLb = 1, and SLc = 3, and the initial state is expressed as XXXXXX.

  A part of the state transition table example in the parking lot search is shown in FIG. Since the distance and fee can be input to slots of 0 or more and less than 1, respectively, the initial state is XX.

State XX
In the initial state XX, the guidance “Distance or fee is” is played, and Sec2_XX. bnf is set in the voice recognition unit 5 to wait for user input.

  When “within 800 m” is input in the initial state XX, Sec2_XX. In accordance with bnf, the value {“DISTANCE_800m”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot related to the distance, and considers that [SLa] has been input. Then, the slot is filled from the front of the slot X. In this case, “800 m” is set in the first slot X related to the distance, and the slot state is Y. Therefore, the state transitions to the state YX.

  When “500 yen or less” is input in the initial state XX, Sec2_XX. In accordance with bnf, the value {“PRICE_500 yen”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for one slot related to the charge, and considers that [SLb] has been input. Then, the slot is filled from the front of the second slot X related to the fee. In this case, “800 m” is set in the slot Xb, and the slot state is Y. Therefore, the state transitions to the state XY.

  In the initial state XX, when “500 yen or less within 800 m” is input, Sec2_XX. In accordance with bnf, the value {“DISTANCE_800 m”, “PRICE_500 yen”} is returned from the speech recognition unit 5 to the scenario analysis unit 6. The scenario analysis unit 6 determines that there is an input for each slot regarding the distance and the charge, and considers that [SLa SLb] has been input. Then, the input value “800 m” related to the distance is filled from the front of the first slot X related to the distance, and the input value “500 yen” related to the charge is filled from the front of the second slot X related to the charge. Therefore, the state transitions to the state YY. Subsequent processing rules follow the same rules as in the first embodiment.

[Fourth Embodiment]
Acceptable inputs are set for each scenario, and in addition to those shown in FIGS. 2 and 6, the following inputs are also conceivable.
・ [SLa SLa NOT SLa SLa]
In the YY state in which {"Japanese food", "Western food"} is in two slots, the input "Chinese and Italian food, not Japanese food and Western food" corresponds to this.
・ [NOT SLa SLa]
In the YY state in which {"Japanese food", "Western food"} is contained in the two slots, the input “Chinese or Italian food instead” corresponds to this.

[Fifth Embodiment]
The scenario examples shown in FIGS. 2 and 6 include a confirmation sequence such as “Are you sure you want Japanese food?” Or “Are you sure you want only Western food?” However, depending on the situation in which the scenario is used, it may be necessary to skip these confirmation sequences. In that case, a Skip attribute is assigned to the state shown in FIG. The scenario analysis unit 6 determines the slot Y as Z for the state having the Skip attribute, and unconditionally shifts the state having C like ZXC to ZX.

[Sixth Embodiment]
In the scenario example shown in FIG. 2, in a state such as ZXC in which the number of slots smaller than the number of slots that can be input is confirmed and whether to proceed to the next processing such as search using only the confirmed slots, only “Slot [1]” According to the guidance “OK”, the scenario is considered to be complete if it is [YES], and transitions to a state (ZX) in which an insufficient slot is input if it is [NO]. However, when a guidance such as “Do you want to input another condition” is set, the meanings of [YES] and [NO] are opposite. Therefore, ZXD, which is the opposite destination of ZXC with respect to [YES] and [NO], is defined.

  When Y is confirmed by inputting [YES] in YX or the like, the scenario analysis unit 6 first searches for the state ZXC, and if not found, searches for the state ZXD. In the state ZXD, the [YES] input means a transition to the state ZX, and the [NO] input means that the process proceeds to the next process as in the state ZZ.

  In addition to what is implemented by dedicated hardware, the present invention records a program for realizing the function on a computer-readable recording medium, and the program recorded on the recording medium is stored in a computer system. It may be read and executed. The computer-readable recording medium refers to a recording medium such as a flexible disk, a magneto-optical disk, and a CD-ROM, and a storage device such as a hard disk device built in a computer system. Furthermore, a computer-readable recording medium is a server that dynamically holds a program (transmission medium or transmission wave) for a short period of time, as in the case of transmitting a program via the Internet, and a server in that case. Some of them hold programs for a certain period of time, such as volatile memory inside computer systems.

It is a block diagram of the voice interactive apparatus of one Embodiment of this invention. It is a figure which shows the example of the state transition table of a meal search. Sec1_XX. It is a figure which shows the example of description of bnf. Sec1_YX. It is a figure which shows the example of description of bnf. Sec1_ZZ. It is a figure which shows the example of description of script. It is a figure which shows the example of the state transition table of a parking lot search. Sec2_XX. It is a figure which shows the example of description of bnf. It is a figure which shows the example of the scenario containing a skip.

Explanation of symbols

1 Voice input / output reception part 2 Data input / output reception part 3 Scenario DB
4 recognition dictionary DB
5 Speech recognition unit 6 Scenario analysis unit 7 Terminal device

Claims (9)

A voice dialogue sequence state expression method performed in a voice dialogue device,
Assign slots for each piece of information needed to achieve a given goal in a voice conversation with the user,
Information is stored in each slot according to each voice input from the start of voice dialogue to the achievement of a predetermined purpose,
Information is empty slots is a state that is not stored, but information is stored, the information is unconfirmed slot is in a state that is not determined to confirm the user information is stored,該情 When there is a confirmed slot and a plurality of unconfirmed slots in which the information has been confirmed and confirmed to the user, and the user confirmation result of whether or not the information stored in each unconfirmed slot is correct , is set for each of the undetermined slot, by managing the status of each slot in the four states of the sequential verify the slot is being confirmed information of the undetermined slot, the voice interaction sequence status the expressed in association with the unique and slot state,
For each sequentially confirmed slot set for each unconfirmed slot, if the information of the sequentially confirmed slot is confirmed and confirmed to the user, the status from the sequentially confirmed slot to the confirmed slot When the correction information of the information of the slot being sequentially confirmed is input and the correction information is a plurality of other information, a plurality of slots being sequentially confirmed in which the other information is stored separately , Each state is changed to the unconfirmed slot, and when the correction information is one piece of other information, the other information is stored in the sequential confirmation slot, and the state waiting for confirmation by the user is maintained.
Spoken dialogue sequence state notation method.   The method according to claim 1, wherein the slot state further includes a control state code for confirming to the user whether or not to proceed to the next sequence with only the confirmed slots when slots less than the number of slots that can be input are confirmed. . The state of the voice interaction sequence is represented by a plurality of states defined by combinations of two or more slot states among the four slot states of the empty slot, the confirmed slot, the unconfirmed slot, and the sequentially confirmed slot. For each of the states, according to the state transition table in which acceptable or unacceptable is set for a plurality of defined inputs that respectively define a plurality of voice input patterns that can be input by the user in the voice interaction sequence. The method according to claim 1 or 2, wherein the state of the voice interaction sequence is expressed in association with the slot state .   The method according to claim 3, wherein the predefined input is a single input, an OR input, an AND input, a partial designation correction input, a correction input, an ANY input, a YES input, and a NO input for the same type of slot.   The method according to claim 3, wherein the predefined input is a single input, an OR input, an AND input, a partial designation correction input, a correction input, an ANY input, a YES input, and a NO input for different types of slots.   The method according to any one of claims 1 to 5, wherein a transition destination of a voice interaction sequence is uniquely determined according to a slot state changed with respect to an acceptable input. When processing according to the voice interaction sequence is interrupted, a slot state indicating the state of the voice interaction sequence at the time of the interruption is held, and when the processing is resumed, the state at the time of interruption is referred to the held slot state The method according to claim 1, wherein the method returns to A voice interaction device,
A voice input / output receiving unit that outputs voice guidance to the user and receives voice input in response to the voice guidance from the user;
A data input / output receiving unit for outputting the input dialog sequence name and slot state to the user and receiving the dialog sequence name and slot state input by the user when the voice dialog sequence is resumed;
A scenario DB that stores voice conversation scenarios used for controlling the voice conversation for each voice conversation sequence;
A recognition dictionary DB that stores speech recognition grammar;
A speech recognition unit that performs speech recognition processing on the input speech using the designated speech recognition grammar stored in the recognition dictionary DB and outputs the speech recognition result;
Interaction sequence name and slot status output to the data input receiving unit, the input to the voice output unit in response to the voice guidance, the voice recognition unit on the speech input of the user, and instructs the voice The speech recognition process is performed with the recognition grammar, and the state of the speech dialogue scenario stored in the scenario DB is changed according to the speech recognition result, and is output from the data input / output receiving unit when the speech dialogue is resumed. resuming the scenario analyzer voice conversation by interaction sequence name and slot state, has,
The scenario analysis unit
Slots are assigned to each piece of information necessary to achieve a predetermined purpose in a voice dialogue with the user, and information is stored in each slot according to each voice input from the start of the voice dialogue to the achievement of the predetermined goal. And
Information is empty slots is a state that is not stored, but information is stored, the information is unconfirmed slot is in a state that is not determined to confirm the user information is stored,該情 When there is a confirmed slot and a plurality of unconfirmed slots in which the information has been confirmed and confirmed to the user, and the user confirmation result of whether or not the information stored in each unconfirmed slot is correct By managing the state of each of the slots in the four states of the sequentially confirming slot, which is set for each of the unconfirmed slots and confirming the information of the unconfirmed slot , Express the state uniquely associated with the slot state ,
For each sequentially confirmed slot set for each unconfirmed slot, if the information of the sequentially confirmed slot is confirmed and confirmed to the user, the status from the sequentially confirmed slot to the confirmed slot When the correction information of the information of the slot being sequentially confirmed is input and the correction information is a plurality of other information, a plurality of slots being sequentially confirmed in which the other information is stored separately , Each state is changed to the unconfirmed slot, and when the correction information is one piece of other information, the other information is stored in the sequential confirmation slot, and the state waiting for confirmation by the user is maintained. Spoken dialogue device.   A program for executing the spoken dialogue sequence state notation method according to any one of claims 1 to 7 on a computer.

Images