JP2008170817A - Interaction control device, interaction control method and interaction control program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an interaction control device capable of continuing natural interaction by timely correcting response to voice guide which is output in the past by a system. <P>SOLUTION: The interaction control device 100 for controlling interaction with a user, according to a scenario which is composed of nodes including a voice guide data and a voice recognition object word for defining user utterance in order to start outputting of the voice guide data, comprises: a voice recognition means 10 for recognizing the user utterance as a text data; a node recording means 12 for recording the node including the output voice guide data; a voice recognition object word acquiring means 13 for acquiring the voice recognition object word regarding the node which is recorded by the node recording means 10; and a next output determination means 14 for determining the voice guide data to be output next, based on the text data recognized by the voice recognition means 10 and the voice recognition object word acquired by the voice recognition object word acquiring means 13. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a dialog control apparatus that outputs guidance data according to a predetermined scenario in response to a user input, and more particularly to enable a dialog control apparatus to redo a response to guidance data output in the past in a timely manner. The present invention relates to a dialog control device, a dialog control method, and a dialog control program that can continue a natural dialog.

  2. Description of the Related Art Conventionally, there is known a simulated conversation system that recognizes a word input by a user and outputs voice guidance data corresponding to the recognized word by voice to continue a conversation with the user (see, for example, Patent Document 1). ).

  This simulated conversation system stores a conversation history of simulated conversations, and a given word is output as an output duration of a topic (consisting of a hierarchical group of questions) during voice output or as a response by a user. When the number of voice input, etc. satisfies a predetermined condition, voice guidance data of a topic different from the topic of the voice guidance data output immediately before is output regardless of the content of the most recent answer (voice input) by the user. To change the topic.

In addition, the simulated conversation system avoids the same question by not outputting the question that has already been output among the new topics after the topic change, but outputting the question that has not been output. To keep the conversation more natural while preventing it from becoming monotonous.
JP 2002-169804 A

  However, since the simulated conversation system described in Patent Document 1 changes the topic when a predetermined condition is satisfied, when the user desires another development for the question output in the immediately preceding topic, for example, Since the topic is changed even if it is desired to select an option different from the option already selected among the options presented in the question, the user cannot select the desired option.

  Also, since the questions that have already been output in the new topic after the topic change are not re-outputted, it is necessary to select an option that is different from the option already selected among the options presented in the previous question. However, the user cannot select a desired option.

  In view of the above, the present invention provides a dialog control device, a dialog control method, and a dialog capable of continuing a natural dialog by allowing a system to redo a response to guidance data output in the past in a timely manner. An object is to provide a control program.

  In order to achieve the above object, a dialogue control apparatus according to a first aspect of the present invention is a scenario comprising a node having guidance data and a recognition target word that defines a user input for starting output of the guidance data. A dialogue control apparatus for controlling a dialogue with a user along with a recognition means for recognizing user input as text data, a node recording means for recording a node having the output guide data, and the node recording means recorded Based on the recognition target word acquisition means for acquiring the recognition target word corresponding to the node, the text data recognized by the recognition means and the recognition target word acquired by the recognition target word acquisition means, the guidance data to be output next is determined. And a next output determination means.

  Further, the second invention is the dialogue control apparatus according to the first invention, wherein the next output determining means includes the recognition target word acquired by the recognition target word acquiring means in the text data recognized by the recognition means. When it is included, the guide data included in the node corresponding to the recognition target word is determined as guide data to be output next.

  Moreover, 3rd invention is the dialog control apparatus which concerns on 1st or 2nd invention, Comprising: The output of the said guidance data is an output by an audio | voice or a text, The said user input is an input by an audio | voice or a text It is characterized by that.

  According to a fourth aspect of the present invention, there is provided a dialogue control method for carrying out dialogue with a user in accordance with a scenario composed of nodes having guidance data and recognition target words for defining user input for starting output of the guidance data. An interactive control method for controlling, a recognition step for recognizing user input as text data, a node recording step for recording a node having the output guidance data, and a recognition target corresponding to the node recorded in the node recording step A recognition target word acquisition step for acquiring a word; a next output determination for determining next guidance data to be output based on the text data recognized in the recognition step and the recognition target word acquired in the recognition target word acquisition step And a step.

  The fifth invention is the dialogue control method according to the fourth invention, wherein the next output determining step includes the recognition target acquired in the recognition target word acquisition step in the text data recognized in the recognition step. When a word is included, the guide data included in the node corresponding to the recognition target word is determined as guide data to be output next.

  The sixth invention is the dialogue control method according to the fourth or fifth invention, wherein the output of the guidance data is output by voice or text, and the user input is input by voice or text. It is characterized by being.

  The seventh invention is a dialogue control program for causing a computer to execute the dialogue control method according to any of the fourth to sixth inventions.

  With the above-described means, the present invention provides a dialog control device, a dialog control method, and a dialog control capable of continuing a natural dialog by allowing a system to repeat a response to guidance data output in the past in a timely manner. A program can be provided.

  Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings.

(1) Outline of Embodiment A dialogue control apparatus according to the present invention is an apparatus that provides information required by a user while outputting voice guidance according to a scenario. For example, gourmet introduction and shopping introduction in a car navigation system It is used to move between domains such as sightseeing spot introduction, audio operation, device operation, etc., and to control input operations within each domain.

  A “scenario” is a collection of information (hereinafter referred to as “nodes”) composed of voice guidance data and a speech recognition target word that defines a transition condition for transition to the next voice guidance data. The voice guidance order is determined by this data structure.

  The “voice recognition target word” is a word registered in advance as a word expected to be uttered by the user, and may be a keyword or a key phrase. When the dialogue control device recognizes that a predetermined speech recognition target word is uttered by the user, the dialogue control device outputs the voice guidance data of the predetermined node assuming that the transition condition to the predetermined node is satisfied.

  “Transition condition” refers to a condition for transition to a predetermined node. For example, a predetermined speech recognition target word is recognized as a user utterance, or may be set based on external information or a user profile. Good. The “external information” when the interactive control device is mounted on the vehicle includes the open / close state of the window, the engine start / stop state, the remaining fuel amount, the brake usage state, the vehicle speed, the accelerator opening, the air conditioner set temperature, The “user profile” when there is information such as the in-vehicle radio usage status and the dialogue control device is installed in the vehicle includes the driver's gender, family structure, birthday, hobbies, and interval for taking a break when driving for a long time. , Vehicle usage frequency (every morning and evening, morning and evening weekdays, weekends, once a month, every day in the evening, etc.), favorite food, favorite music, etc.

  Each scenario has specific contents such as a scenario output during driving on an expressway, a scenario output when approaching a registered facility, and the root node (the node at the top of the tree structure). ), Branch nodes (referred to as nodes at the branch point of the tree structure), and leaf nodes (referred to as nodes at the end of the tree structure).

  Note that when the transition condition is satisfied (for example, a case where a speech recognition target word is included in the user utterance), a predetermined process such as voice guidance at a predetermined node is started. This is called “ignition”, and the transition condition to the node is called “ignition condition”.

  In addition, it is determined whether or not the ignition condition of the node is satisfied, and when the ignition condition is satisfied, a predetermined process such as voice guidance is started, and when the ignition condition is not satisfied, the determination process is periodically performed. Control that is repeatedly executed is called "ignition control".

  FIG. 2 is a diagram showing a tree structure of a scenario, and shows that scenario A has a root node N1, branch nodes N2, N3, and leaf nodes N4, N5, N6, N7.

  FIG. 3 is a diagram illustrating a configuration example of nodes, and each node includes a voice guidance data unit 50 and a voice recognition target word unit 51. FIG. 4 is a diagram illustrating a configuration example of the speech recognition target word unit 51, and the speech recognition target word unit 51 holds a correspondence relationship between each speech recognition target word and the destination node.

  For example, the speech recognition target words for the voice guidance data “Let's guide sightseeing? There are specialties, specialties, and highlights, but which is better?” Stored in the voice guidance data section 50 of the node N3 are “specialties”, “ “Specialties”, “Highlights”, “I don't want to hear”, etc. are registered, and the dialog control device recognizes that “specialties” are included in the user's utterance, so that the transition condition to the node N5 is satisfied To fire the node N5.

  In addition, the dialog control device records information about nodes that have been ignited in the past as a dialog history, and when recognizing a user's utterance, the speech recognition target word of each of the nodes recorded in the dialog history is Speech recognition target words of nodes that were acquired in the order of new records and fired in the past (the speech recognition target words of the most recent node are designated as “current speech recognition target words” and fired in the past other than the current speech recognition target words It is determined whether or not the speech recognition target word of the selected node is included in the user's utterance as “existing speech recognition target word”.

  When the current speech recognition target word is included in the user's utterance, the dialogue control device fires a node corresponding to the current speech recognition target word and outputs a voice guidance, but the current speech recognition is included in the user's utterance. When the target word is not included, the dialogue control apparatus determines whether the already-recognized speech recognition target word is included in the user's utterance, and when the already-spoken speech recognition target word is included in the user's utterance, A node corresponding to the speech recognition target word is fired.

  In addition, when the node corresponding to the speech recognition target word has already been ignited within a predetermined time, the dialogue control apparatus may prohibit the ignition of the node.

  As a result, the dialog control device shifts to an option that was not selected in a question by the dialog control device at a certain point in time by a user utterance such as “What is XX of the day?” It is possible to flexibly cope with a case where the user desires, and a natural conversation can be continued.

(2) Details of Embodiment FIG. 1 is a diagram showing a configuration example of a dialogue control apparatus according to the present invention. A dialogue control apparatus 100 includes a control unit 1, a voice acquisition unit 2, a storage unit 3, and a voice output unit 4. The voice guidance data stored in the storage unit 3 is read in accordance with a predetermined scenario, and the voice guidance data is output from the voice output unit 4.

  The control unit 1 is a computer including a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and the like, and includes a voice recognition unit 10, a data providing unit 11, a node recording unit 12, and a voice recognition unit. Programs corresponding to the target word acquisition means 13 and the next output determination means 14 are stored in the ROM, and these programs are expanded on the RAM to cause the CPU to execute corresponding processing.

  The voice acquisition unit 2 is a device for acquiring a voice uttered by a user. For example, when the dialogue control apparatus 100 is mounted on a vehicle, the voice acquisition unit 2 has directivity so that the voice of a driver driving the vehicle can be reliably recognized. A microphone is used.

  In addition, the voice acquisition unit 2 may have an input sound detection function, and may acquire a sound other than stationary noise as a voice generated by the user's utterance. Also good. In addition, the voice acquisition unit 2 may have an utterance detection function, and may detect and acquire only a human utterance portion of input voice. The voice acquisition unit 2 outputs the acquired user voice to the control unit 1.

  The storage unit 3 is a device for storing various types of information required by the dialogue control apparatus 100. For example, a scenario database 30 that systematically configures a plurality of scenarios having the data structure shown in FIGS. And a hard disk for storing a dialogue history database 31 to be described later.

  The voice output unit 4 is a device for outputting voice guidance data by voice, and is, for example, an in-vehicle speaker that outputs voice guidance data output from the control unit 1 by voice.

  Next, various units included in the control unit 1 will be described in detail.

  The voice recognition means 10 is means for recognizing a user utterance as text data. For example, the voice recognition means 10 converts the user utterance acquired via the voice acquisition unit 2 into text data.

  The voice recognition means 10 not only recognizes user utterances as text data, but also recognizes the reliability of the recognition result, the length of the user utterance (number of words or time), the utterance speed of the user utterance or the change in the utterance speed, or the user Features such as the utterance volume of an utterance may be detected and stored in association with text data. This is because the user can more accurately recognize the intention put into the utterance.

  The data providing unit 11 is a unit for providing voice guidance data according to the scenario. For example, the voice providing data of a node that satisfies the firing condition is output from the voice output unit 4.

  The node recording unit 12 is a unit for recording a node having the output voice guidance data as a dialogue history. For example, as shown in FIG. 5, the firing history number indicating the firing order, the number of the fired node And the time when the node was ignited are recorded in the dialogue history database 31 in the storage unit 3 in association with each other.

  Note that the node recording unit 12 may record a concept keyword representing voice guidance data included in each node, or a subkeyword or subkey phrase included in the concept, in the dialogue history database.

  The speech recognition target word acquisition unit 13 is a unit for acquiring the speech recognition target word that the fired node has. For example, the speech recognition target word acquisition unit 13 refers to the speech recognition target word unit 51 of the node fired most recently. Get the current speech recognition target word.

  Further, the speech recognition target word acquisition means 13 may refer to the dialogue history database 31 to acquire the already-speech recognition target words that the nodes fired in the past have.

  The next output determination unit 14 is a unit for determining the voice guidance to be output next, and includes, for example, a domain determination unit 140, a similar scenario search unit 141, a topic conversion unit 142, a dialogue end unit 143, and the like. The voice guidance to be output next is determined based on the utterance content and the utterance timing.

  The domain determination means 140 is a means for determining whether or not a domain related to the user's utterance content (hereinafter referred to as “related domain”) exists. Based on this, the existence of the related domain is determined.

  “Domain” means a set of information gathered with common contents. For example, there are a gourmet introduction domain, a shopping introduction domain, a sightseeing spot introduction domain, an audio operation domain, and a device operation domain in a car navigation system. The domain has a concept keyword that characterizes itself (for example, there is a sport), and the concept keyword further includes a subkeyword or a subkey phrase included in the concept (for example, baseball, soccer, volleyball, etc.). ).

  In addition, each domain has a scenario for providing related information (for example, a tourism guide scenario in the tourist destination introduction domain) and a scenario for executing a related operation (for example, a volume adjustment scenario in the audio operation domain). ).

  Further, the dialogue control apparatus 100 may be provided with a thesaurus to determine the relation between the user's utterance content and the domain, and the user's utterance content using the TF-IDF (Term Frequency-Inverse Document Frequency) method. And the relationship between the domain and the domain may be determined.

  The similar scenario search unit 141 is a unit for searching for a scenario whose content is similar to the content of the user's utterance or a scenario whose content is similar to the content of the fired scenario (hereinafter referred to as “similar scenario”). .

  The similar scenario search means 141 includes, for example, a phrase included in the user's utterance, or a speech recognition target word included in a node in a scenario fired recently, a phrase included in the voice output data, and each scenario to be searched A similar scenario is searched from all scenarios based on the speech recognition target word of the node and the phrase included in the speech output data.

  Each scenario may have a concept keyword that characterizes itself (for example, there is a tourist guide) as well as a domain, and further includes a subkeyword or subkey phrase included in the concept (for example, a hot spring, an observation deck). And souvenir shops, etc.).

  Further, the dialog control apparatus 100 may include a thesaurus for determining the similarity between the user's utterance content or the content of the fired scenario and the content of the scenario to be searched, and uses the TF-IDF method. Thus, the degree of similarity may be determined.

  The similar scenario search means 141 is a voice recognition means only when the user's utterance length (one utterance length or the total utterance length in each scenario) is a predetermined time or more. A similar scenario may be searched by extracting an independent word from the text data recognized by 10.

  When the user's utterance length is long, it can be assumed that the user is active in the dialog with the dialog control apparatus 100, and by igniting a similar scenario, the dialog control apparatus 100 can continue the natural dialog with the user. Because you can.

  In addition, when the user's utterance length is short, it can be assumed that the user is reluctant to interact with the dialog control apparatus 100, and by preventing the similar scenario from being ignited, the dialog control apparatus 100 allows a topic of low interest to be discussed. This is because it is possible to prevent the user from feeling uncomfortable.

  The topic conversion unit 142 is a unit for converting a topic, and for example, selects a specific scenario from a scenario group that can be ignited using a random number.

  The dialog ending unit 143 is a unit that ends the dialog. For example, the dialog ending unit 143 outputs an end notification message that clearly indicates the end of the dialog to the user.

  Next, the flow of processing for determining the voice guidance to be output next by the next output determining means 14 will be described with reference to FIG. FIG. 6 is a flowchart showing the flow of this process.

  First, when the voice acquisition unit 2 detects a user utterance, the dialogue control apparatus 100 determines whether or not the voice recognition unit 10 has recognized (step S1).

  When the user utterance can be recognized as text data (YES in step S1), the dialogue control apparatus 100 refers to the dialogue history database 31 recorded by the node recording unit 12 by the voice recognition target word acquisition unit 13 and fires in the past. The speech recognition target word possessed by the node is acquired, and it is determined whether or not the speech recognition target word is included in the text data (step S2).

  It should be noted that the speech recognition target acquisition means 13 has a phrase such as “Yes”, “No”, “Any”, “Any” or “It” for past nodes other than the last fired node. Is not acquired as a speech recognition target word. This is because it cannot be specified which node the answer is for.

  When the text recognition target word is included in the text data (YES in step S2), the dialogue control apparatus 100 causes the next output determination unit 14 to execute a node transition process described later (step S3).

  When the text recognition target word is not included in the text data (NO in step S2), the dialogue control apparatus 100 searches for a similar scenario by the similar scenario search unit 141 in the next output determination unit 14 (step S4).

  When a similar scenario can be searched (YES in step S4), the dialogue control apparatus 100 causes the data providing unit 11 to output voice guidance data included in the root node of the similar scenario from the voice output unit 4 (step S5).

  When the similar scenario cannot be searched (NO in step S4), the dialogue control apparatus 100 determines whether or not there is a domain related to the user's utterance content by the domain determination unit 140 in the next output determination unit 14 (step S4). S6).

  When the related domain exists (YES in step S6), the dialogue control apparatus 100 causes the data providing unit 11 to output the voice guidance data of the scenario root node in the related domain from the voice output unit 4 (step S7). ), If the related domain does not exist (NO in step S6), the next output determining unit 14 is caused to execute a dialogue continuation possibility determination process described later (step S8).

  Next, the flow of node migration processing will be described with reference to FIG. FIG. 7 is a flowchart showing the flow of node migration processing.

  First, the next output determination unit 14 refers to the dialogue history database 31 to calculate the elapsed time after firing of a node having a speech recognition target word included in the text data (hereinafter referred to as “corresponding node”) ( Step S11).

  When there are a plurality of corresponding nodes, the next output determining unit 14 gives priority to the larger (newer) firing history number. This is because it is considered that the user often wants to return to the previous choice or topic.

  Further, the next output determining unit 14 gives priority to the lower node when there are a plurality of corresponding nodes in the same scenario. This is because the node at the lower level of the scenario presents a more specific question than the node at the higher level, and it is considered that the voice guidance expected by the user can be provided more quickly.

  However, if the user's utterance includes a phrase specifying a date such as “yesterday” or “noon”, priority may be given to a corresponding node having an ignition history that matches the date and time.

  When the elapsed time after firing of the corresponding node is less than a threshold value T1 (for example, 5 minutes) (YES in step S11), the next output determining unit 14 refers to the dialogue history database 31 and shifts indicated by the speech recognition target word. It is determined whether or not the previous node (hereinafter referred to as “corresponding node”) has been ignited (step S12), and if the corresponding node has been ignited (YES in step S12), a dialog continuation continuity determination process described later. Is executed (step S13).

  If the corresponding node has not been ignited (NO in step S12), the next output determining unit 14 determines the corresponding node as a node to be output next (step S14), and the voice providing data that the corresponding node has by the data providing unit 11 Is output from the audio output unit 4.

  On the other hand, when the elapsed time after firing of the corresponding node is equal to or greater than a threshold T1 (for example, 5 minutes) (NO in step S11), the next output determining means 14 is “XX (voice recognition target word included in user utterance)”. Confirmation message such as “Is it OK?”, “XX (speech recognition target word included in user utterance) and △△ (another speech recognition target word) is XX?” (Step) S15) After obtaining the user's confirmation (for example, the case where the user utterance “Yes” is acquired) (YES in step S16), the corresponding node is determined as the node to be output next (step S14).

  When the user confirmation is not obtained (for example, the case where the user utterance “No” is acquired) (NO in Step S16), the next output determination unit 14 executes a dialog continuation continuity determination process described later ( Step S17).

  This is because, since the predetermined time has elapsed, the user may not correctly remember the contents of questions and options output by the dialog control apparatus 100.

  In this case, even if the corresponding node has been fired, the next output determining unit 14 may cause the voice output unit 4 to output the voice guidance data of the corresponding node.

  This is because, since the predetermined time has elapsed, even if the voice guidance data of the corresponding node is output again, the user does not feel bothered.

  The next output determining unit 14 refers to the thesaurus stored in the storage unit 3 to obtain a word similar to the speech recognition target word (hereinafter referred to as “similar target word”), and the similar target word is text data. May be output as a confirmation message such as “Are you sure you want to use the speech recognition target word corresponding to the similar target word included in the user utterance?”.

  In addition, the next output determination unit 14 may execute the node migration process when the user's confirmation is obtained (for example, when the user utterance “Yes” is acquired). This is because the user may not accurately remember the options presented by the dialog control apparatus 100 in the past.

  The dialog control apparatus 100 changes the control method based on the elapsed time after firing, but may change the control method based on how many dialog histories (number of nodes) are traced back.

  Next, the flow of the process for determining whether or not to continue the conversation will be described with reference to FIG. FIG. 8 is a flowchart showing the flow of the process for determining whether or not to continue the conversation.

  First, the next output determination unit 14 determines whether or not there is a scenario that satisfies the ignition condition by the ignition control (step S21).

  When there is a scenario that can be ignited (YES in step S21), the next output determining unit 14 determines whether or not there is a similar scenario similar to the scenario fired most recently by the similar scenario search unit 141 ( Step S22).

  If there is a similar scenario similar to the most recently fired scenario, the dialogue control apparatus 100 causes the data providing means 11 to output the voice guidance data of the root node of the similar scenario from the voice output unit 4 (step S23). ).

  When there is no similar scenario similar to the scenario fired most recently, the dialog control apparatus 100 uses the topic conversion unit 142 of the next output determination unit 14 to select a specific scenario from the scenario group that can be fired using a random number. Then, the voice providing data of the route node of the selected scenario is output from the voice output unit 4 by the data providing means 11 (step S24).

  On the other hand, if there is no scenario that can be ignited (NO in step S21), the dialogue control apparatus 100 terminates the dialogue such as “please call me again when necessary” by the dialogue termination means 143 of the next output determination means 14. Is output (step S25).

  Next, a flow in which the dialogue control apparatus 100 controls the dialogue will be described using an example of dialogue between the dialogue control apparatus 100 and the user.

  When recognizing that the vehicle has reached a predetermined point based on position information output by GPS (Global Positioning System), the dialogue control device 100 ignites a tourist guidance scenario A (see FIG. 2), and data providing means 11 Thus, the voice guidance data of the nodes N1 and N2 of the tourist guidance scenario A is sequentially output from the voice output unit 4, and the dialogue with the user is continued.

  After that, the dialogue control apparatus 100 causes the voice output unit 4 to output the voice guidance data of the node N3 of the tourist guidance scenario A "Tourist guidance? Should there be information on specialties, special products, and highlights?

  The node recording unit 12 records the fired nodes N1, N2, and N3 in the dialogue history database 31 in the firing order, and records the flow of dialogue between the user and the dialogue control device 100 (see FIG. 5). .

  In the speech recognition target word part 51 of the node N3, speech recognition target words “specialties”, “specialties”, and “highlights” are registered as transition conditions to the nodes N5, N6, and N7, respectively. “I don't want to hear” is registered as a condition for ending the dialogue (see FIG. 4).

  At this time, when the user utters “special”, the voice recognition means 10 converts the user utterance into text data, and the voice recognition target word acquisition means 13 reads the voice recognition target word portion 51 of the nodes N1, N2, and N3. , The speech recognition target words possessed by each of the nodes N1, N2, and N3 are acquired.

  Thereafter, the next output determining means 14 detects that the current speech recognition target word “specialty” in the node 3 is included in the user utterance, and refers to the speech recognition target word portion 51 (see FIG. 4) of the node 3. Thus, the node N5 associated with the current speech recognition target word “specialty” is determined as a node having voice guidance data to be output next.

  Thereafter, the dialogue control apparatus 100 causes the voice output unit 4 to output the voice guidance data “the famous product around here is eel” possessed by the node N5. The node recording unit 12 adds the fired node N5 to the dialogue history database 31.

  In the speech recognition target word part 51 of the node N5, the speech recognition target words “I want to eat eel” and “Iiie eel” are registered as a transition condition to the succeeding node, and the speech recognition target word “So, I don't know. ”,“ I don't like eels ”, etc. are registered as conditions for terminating the dialogue.

  At this time, when the user utters “Yes, I did not know”, the voice recognition means 10 converts the user utterance into text data, and the voice recognition target word acquisition means 13 reads the nodes N1, N2, N3, N5. The speech recognition target words included in each of the nodes N1, N2, N3, and N5 are acquired with reference to the respective speech recognition target word portions 51 of the nodes.

  Thereafter, the next output determination means 14 detects that the current speech recognition target word “Yes, I did not know” in the node 5 is included in the user utterance, and refers to the speech recognition target word unit 51 of the node 5. End the conversation.

  As described above, how the dialog control apparatus 100 continues the dialogue according to the user's utterance contents when the user resumes the utterance after once ending the dialogue will be described below.

  First, a case will be described in which the user utters “I want to eat a little bit early, but I want to eat.”

  In this case, the next output determination unit 14 refers to the dialogue history database 31 and recognizes that the current speech recognition target word “I want to eat eel” of the node N5 fired immediately before is included in the user utterance.

  The subsequent node corresponding to the current speech recognition target word “I want to eat eel” has not fired in the past, and the next output determining means 14 determines that the voice guidance data of the subsequent node is to be output next.

  If the subsequent node has fired in the past, the next output determination means 14 may change the topic by outputting a topic change message such as “There is such an interesting story” so as not to repeat the same question, You may explicitly end the dialogue by outputting an end notification message such as “My tourist guide is over. I will study more.”

  Further, when the subsequent node has fired in the past, the next output determining means 14 may fire a similar scenario. By firing a similar scenario, you can avoid giving users the impression that they always ask the same question, and they can provide a completely different topic so as not to confuse the user. It is.

  Next, a case where the user utters “Is it a little early but will I have dinner?” Will be described.

  In this case, the next output determining unit 14 refers to the dialogue history database 31 and recognizes that the speech recognition target word of the node fired in the past is not included in the user utterance.

  Therefore, the next output determining unit 14 refers to the thesaurus stored in the storage unit 3 to obtain a similar target word similar to the speech recognition target word, and is similar to the current speech recognition target word “I want to eat eel” at the node N5. It recognizes that the similar object word “dinner” registered as being included is included in the text data, and determines that the voice guidance data of the subsequent node is to be output next.

  Next, a case where the user utters “Oh, I forgot to buy souvenirs. What was the special product?” Will be described.

  In this case, the next output determining unit 14 refers to the dialogue history database 31 and recognizes that the already-recognized speech recognition target word “specialized product” of the node N3 fired two nodes before is included in the user utterance.

  The node N6 corresponding to the existing speech recognition target word “special product” has not been ignited in the past, and the next output determining means 14 uses the voice guidance data held by the node N6 corresponding to the existing speech recognition target word “special product”. Next, it is determined to be output.

  Even if the node N6 has fired in the past, if the predetermined time has elapsed since the firing, the next output determining means 14 causes the voice guidance data of the node N6 to be output next. You may make it determine as. This is because when a predetermined time has passed since the ignition, the user is not uncomfortable even if the same question is repeated.

  Next, a case where the user speaks “Oh, I forgot to buy souvenirs” will be described.

  In this case, the next output determination unit 14 refers to the dialogue history database 31 and recognizes that the speech recognition target word of the node that has fired in the past is not included in the user utterance.

  Therefore, the next output determining unit 14 refers to the thesaurus stored in the storage unit 3 to acquire a similar target word similar to the voice recognition target word, and the already-spoken voice recognition target word of the node N3 fired two nodes before. Recognizing that the similar target word “souvenir” registered as being similar to “special product” is included in the text data, and following the voice guidance data possessed by the node N6 corresponding to the existing speech recognition target word “special product” Is determined to be output.

  Next, a case where the user utters “I feel like I have heard that there is a famous hot spring around here” will be described.

  In this case, the next output determining unit 14 refers to the dialogue history database 31 and recognizes that the speech recognition target word and the similar target word of the node fired in the past are not included in the user utterance.

  Therefore, the next output determination unit 14 uses the similar scenario search unit 141 to determine the independent words “so”, “say”, “this”, “around”, “famous”, “hot spring”, “al”, “listen”, “Koto”, “Aru”, “Ki”, “Sue”, “Bad” are extracted, and the sub-keyword “Hot Spring” associated with the independent keyword “Hot Spring” and the concept keyword “Tourist Guide” of the Tourist Guide Scenario A ", The tourist guide scenario A is extracted as a related scenario.

  As a result, the next output determining means 14 determines that the voice guidance data of the route node N1 of the sightseeing guide scenario A is to be output next.

  Next, the case where the user utters “Tell me the result of yesterday's baseball” will be described.

  In this case, the next output determining unit 14 refers to the dialogue history database 31 and recognizes that the speech recognition target word and the similar target word of the node fired in the past are not included in the user utterance. Recognize that it does not exist.

  Therefore, the next output determination means 14 causes the domain determination means 140 to extract the independent words “so”, “say”, “yesterday”, “baseball”, “result”, “teach”, and the independent word “baseball”. Since the sub-keyword “baseball” associated with the concept keyword “sports” of the sports domain matches, it is determined that the user utterance is related to the sports domain.

  Thereby, the next output determination means 14 determines as what outputs the voice guidance data which the route node of the scenario in a sports domain has next.

  Next, a case where the user utters “Oh, I am getting sleepy” will be described.

  In this case, the next output determining unit 14 refers to the dialogue history database 31 and recognizes that the speech recognition target word and the similar target word of the node fired in the past are not included in the user utterance, and the similar Recognize that there are no scenarios or related domains.

  Therefore, the next output determining means 14 determines whether or not there is a scenario that satisfies the ignition condition by the ignition control. If there is a scenario that can be ignited, the next voice output data that the root node of the scenario has is output. Decide what to do.

  On the other hand, if there is no scenario that can be ignited, the next output determining means 14 determines that the end notification message is to be output next.

  Next, a case where the user's speech is detected but cannot be recognized as text data will be described.

  In this case, the next output determining unit 14 determines whether or not there is a scenario that satisfies the ignition condition by the ignition control. If there is a similar scenario similar to the most recently fired scenario, the voice guidance data held by the root node of the similar scenario is Is determined to be output.

  When there is a scenario that can be ignited but there is no similar scenario, the next output determination means 14 outputs the voice guidance data that the root node of the scenario that can be ignited randomly using a random number next outputs. decide.

  If there is no scenario that can be ignited, the next output determining means 14 determines that the end notification message is to be output next.

  With the above configuration, the dialogue control apparatus 100 records a node that has already been ignited in a scenario composed of a plurality of nodes, and the speech recognition target word of the already ignited node is included in the user utterance. Since the node below the node is fired, the voice guidance expected by the user can be provided promptly and in a timely manner, and a natural conversation can be continued.

  Further, by removing the restriction that only the question presented by the dialog control device 100 can be answered only by the user's utterance and allowing the dialog control device 100 to directly respond to the question presented in the past, a natural dialogue can be performed. It is possible to continue, enhance the user's familiarity with the dialog control apparatus 100, and promote the continuous use of the dialog control apparatus 100.

  Although the preferred embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the present invention. Can be added.

  For example, in the above-described embodiment, each node has information on the transition condition to the lower node (ignition condition for firing the lower node) and the speech recognition target words constituting the transition condition. You may make it have the speech recognition target word which comprises its own ignition conditions and the ignition conditions.

  In addition, the dialogue control apparatus 100 is configured based on voice input / output, such as recognizing a user's utterance while outputting voice guidance according to a scenario and determining voice guidance to be output next. A configuration based on input / output by text, such as determining text data to be displayed next by recognizing text input by a user via a keyboard while displaying text data on a display device such as a display, etc. Also good.

  In this case, the scenario is composed of nodes having text guidance data and a recognition target word that defines a transition condition for transitioning to the next text guidance data, and the recognition target word is a word that is expected to be input by the user. Registered in advance.

  The voice acquisition unit 2 is replaced with a text input unit such as a keyboard, a mouse, or a touch panel, and the voice output unit 4 is replaced with a text output unit such as a liquid crystal display.

  The input may be configured based on voice and the output based on text, or the input may be configured based on text and the output based on speech.

It is a figure which shows the structural example of a dialogue control apparatus. It is a figure which shows the tree structure of a scenario. It is a figure which shows the structural example of a node. It is a figure which shows the structural example of a speech recognition object word part. It is a figure which shows the structural example of a dialogue history database. It is a flowchart which shows the flow of the process which determines the voice guidance which a next output determination means outputs next. It is a flowchart which shows the flow of a node transfer process. It is a flowchart which shows the flow of a conversation continuation possibility determination process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Control part 2 Voice acquisition part 3 Storage | storage part 4 Voice output part 10 Voice recognition means 11 Data provision means 12 Node recording means 13 Voice recognition object word acquisition means 14 Secondary output determination means 30 Scenario database 31 Dialog history database 50 Voice guidance data part 51 Speech recognition target word part 100 Dialog control device 140 Domain determination means 141 Similar scenario search means 142 Topic conversion means 143 Dialog termination means N1 to N7 nodes

Claims (7)

A dialogue control device for controlling dialogue with a user in accordance with a scenario composed of nodes having guidance data and recognition target words for defining user input for starting output of the guidance data,
Recognition means for recognizing user input as text data;
Node recording means for recording a node having the output guidance data;
Recognition target word acquisition means for acquiring a recognition target word corresponding to the node recorded by the node recording means;
Next output determining means for determining guide data to be output next based on the text data recognized by the recognition means and the recognition target word acquired by the recognition target word acquiring means;
A dialogue control apparatus comprising: The next output determining means, when the recognition target word acquired by the recognition target word acquisition means is included in the text data recognized by the recognition means, next outputs guidance data possessed by a node corresponding to the recognition target word Decide as guidance data to be
The dialogue control apparatus according to claim 1, wherein The output of the guidance data is output by voice or text,
The user input is input by voice or text.
The dialogue control apparatus according to claim 1 or 2, wherein A dialogue control method for controlling dialogue with a user in accordance with a scenario composed of nodes having guidance data and recognition target words for defining user input for starting output of the guidance data,
A recognition step for recognizing user input as text data;
A node recording step for recording a node having the output guidance data;
A recognition target word acquisition step of acquiring a recognition target word corresponding to the node recorded in the node recording step;
A next output determination step of determining guidance data to be output next based on the text data recognized in the recognition step and the recognition target word acquired in the recognition target word acquisition step;
A dialogue control method comprising: In the next output determining step, when the recognition target word acquired in the recognition target word acquisition step is included in the text data recognized in the recognition step, the guidance data included in the node corresponding to the recognition target word is Determined as guidance data to be output to
The dialogue control method according to claim 4, wherein: The output of the guidance data is output by voice or text,
The user input is input by voice or text.
6. The dialogue control method according to claim 4 or 5, wherein   A dialogue control program for causing a computer to execute the dialogue control method according to any one of claims 4 to 6.

Images