JP4826275B2 - Response generating apparatus, method, and program - Google Patents

Description

  The present invention relates to a response generation apparatus, method, and program, and more particularly to a response generation apparatus, method, and program for smoothly interacting with a user.

  2. Description of the Related Art Conventionally, an interactive system that interacts with a user in a natural manner and prevents the user from getting bored has been proposed. For this reason, it is necessary for the dialogue system not to speak one topic for a long time, or conversely change the topic one after another.

  Therefore, a robot that provides a user with a fun and natural conversation has been disclosed (see Patent Document 1). This robot includes a profile memory 121 that stores data (profile data) relating to user preferences. The robot 1 uses, as a filter, profile data stored in the profile memory 121 among the topic information provided from the server 101. Then, the robot 1 stores only the remaining information as a result of filtering by the filter 112B in the topic memory 76, and selects a topic using the information stored in the topic memory 76 when talking to the user. This makes it possible to select a topic that makes conversation fun.

In addition, a dialog control system that can improve entertainment is disclosed (see Patent Document 2). The dialogue control system has a robot and an information processing apparatus connected via a network. When a dialogue between the robot and the user is performed by word play (riddle riddle), the robot generates history data related to word play among the utterance contents of the user and transmits it to the information processing apparatus. The information processing apparatus selectively reads out the content data most suitable for the user from the storage means based on the history data and provides it to the original robot.
JP 2001-188787 A JP 2003-255991 A

  The technique of Patent Document 1 selects a topic based on profile data that is data relating to a user's taste, and is effective in selecting a rough topic. However, when there are a plurality of utterance candidates in the same topic, there is a problem that an effective utterance candidate cannot be selected.

  For example, “Actor C” is profiled, and as a speech candidate, “Actor C ate lunch”, “A movie starring actor C will be released”, “Actor C and actress A” If you think you are married, you cannot select which utterance candidate is more appropriate.

  Moreover, the technique of patent document 2 gives the difficulty level and the number of questions given to each riddle that is, for example, the utterance content. When this is applied to response selection, it is not practical in terms of the amount of description to describe an index such as difficulty level for every utterance individually. For example, using the above example, “Actor C ate lunch”, “A movie starring actor C will be released”, “Actor C and actress A got married.” When trying to describe an index in each of the above, the amount of description becomes enormous, which is not realistic.

  The present invention has been proposed to solve the above-described problems, and an object of the present invention is to provide a response generation apparatus, method, and program capable of smoothly performing a dialog with a user without incurring costs. To do.

In order to solve the above-described problem, the response generation apparatus of the present invention includes an input unit that inputs an utterance, a response generation unit that generates a plurality of responses to the utterance input to the input unit, and a topic for each concept. Concept memory means for storing abundance and positive / negative degree of emotion, context memory means for storing a concept included in the context of past dialogue, and a concept included in a response whose priority is to be determined are the context memory. A concept control means for controlling the degree of abundance of topics corresponding to the concept stored in the concept storage means when stored in the means, and a plurality of responses generated by the response generation means The abundance of topics corresponding to the concepts included in each response and the positive / negative degree of emotion are read from the concept storage means, and a plurality of responses are determined based on the read abundance of topics and the positive / negative degree of emotions. For each, as the priority is higher is greater richness of the topic, and, as the degree of positive or negative of the emotion is positive, so that the priority becomes higher, to determine the priority, the highest priority Response selection means for selecting a response, and response output means for outputting the response selected by the response selection means.

  The input means only needs to be able to input the utterance content, and it does not matter whether it is a voice. The response generation unit generates a plurality of responses to the utterance input to the input unit. The concept storage means stores the abundance of topics and the degree of emotional brightness for each concept. Here, the concept indicates a general meaning of a certain thing, and parts of speech such as nouns and verbs are not particularly limited. Topic abundance refers to an index that indicates the number of topics derived from the concept.

The concept control means lowers the degree of abundance of topics corresponding to the concept stored in the concept storage means when the concept included in the response whose priority is to be determined is stored in the context storage means. To control. The response selection means reads from the concept storage means the abundance of topics and the degree of emotional brightness corresponding to the concepts included in each of the generated responses. And a response selection means determines a priority for every some response based on the abundance of the read topic and the brightness level of an emotion, and selects a response with the highest priority. The selected response is output via response output means.

  Therefore, the above invention generates a plurality of responses to the input utterance and prioritizes each response based on the abundance of topics corresponding to the concepts included in each of the responses and the degree of emotional brightness. The user's utterance is developed, the system shows interest in the user's utterance, expresses emotion, and synchronizes with the user's emotion Or a response to follow when the user's utterance content is negative. Note that the present invention can also be applied to a response generation method and a program.

  According to the present invention, it is possible to smoothly perform a dialogue with a user without incurring costs.

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

[First Embodiment]
FIG. 1 is a block diagram showing a configuration of a response generation apparatus according to the first embodiment of the present invention.

  The response generation apparatus according to the first embodiment includes a microphone 11 that generates an audio signal from a user utterance, a recognition unit 12 that recognizes an audio signal output from the microphone 11, an analysis unit 13 that analyzes the user utterance, A response generation unit 14 that generates utterance candidates based on the analysis result, a response selection unit 15 that selects one response utterance from a plurality of utterance candidates, a concept database 16, a speaker 21 that outputs a response utterance as a voice, and a response And a display 22 for outputting the speech as an image.

  FIG. 2 is a block diagram illustrating a configuration of the analysis unit 13. The analysis unit 13 includes a user utterance intention analyzer 31 that analyzes a user utterance intention, and an utterance-utterance intention correspondence table database 32 that stores an utterance-utterance intention correspondence table.

  FIG. 3 is a diagram showing a configuration of an utterance-utterance intention correspondence table. The utterance-utterance intention correspondence table defines the relationship between user utterances and their intentions. For example, a user utterance such as "Good morning", "Hello" is intended to "greeting". User utterances such as “what is“ (noun) ”” and “what is“ (noun) ”” are intended to be “information request”. User utterances such as “Uh?” And “What did you say?” Are intended to be “questions”.

  FIG. 4 is a block diagram illustrating a configuration of the response generation unit 14. Hereinafter, the utterance of the apparatus in response to the user utterance is referred to as “system utterance”. The response generation unit 14 includes a system utterance intention selector 41, an utterance intention correspondence table database 42, a related information search engine 43, a related information database 44, a response generator 45, and an utterance database 46. .

  FIG. 5 is a diagram showing the configuration of the utterance intention correspondence table stored in the utterance intention correspondence table database 42. The utterance intention correspondence table defines the intention of the system utterance with respect to the user utterance. For example, when the user utterance intention is “greeting”, the system utterance intention is “repeating user utterance”. Further, when the user utterance intention is “information transmission”, the system utterance intention is any one of “information transmission”, “consideration”, and “evaluation / impression”.

  FIG. 6 is a diagram showing the configuration of the related information database 44. The related information database is a group of related terms that are terms related to each other. For example, related terms of “mountain” include “meadow” and “ranch”. Related words of “party” include “feast”, “dress”, and “fun”.

  The response generator 45 generates one or more utterance candidates using information stored in the utterance database 46 based on the system utterance intention. Here, the utterance database 46 also stores various information for generating utterance candidates from related words and information on the utterances.

  For example, when the system utterance intention is “information transmission”, the response generator 45 generates utterance candidates using related words. For example, when the system utterance intention is “conflict”, the response generator 45 randomly selects one of the plurality of conflict utterances shown in FIG. 7 and outputs the selected one as an utterance candidate.

  FIG. 8 is a diagram showing the configuration of the concept database 16. The related information database defines “abundance of topics” and “emotions” for each concept (word). “Topic abundance” refers to an index indicating the number of topics derived from the concept, such as a situation in which the number of hits is large in a search engine. In this embodiment, “buzz richness” is expressed in three stages of “large”, “medium”, and “small”. “Emotion” represents the degree of positive or negative emotion. In this embodiment, for example, “positive”, “negative”, and “neutral” are distinguished. Yes.

  The response generation apparatus configured as described above can realize, for example, the following interactive example 1 with the user by executing the first response generation routine.

(Dialogue example 1)
User utterance 1: Hello.
System utterance 2: Hello.
User utterance 3: I went to A's house yesterday.
System utterance 4: So then?
User utterance 5: It was very feasting.
System utterance 6: Was it a party?
User utterance 7: Yeah, it was a birthday party.
System utterance 8: Hey, awesome.

  FIG. 9 is a flowchart showing a first response generation routine for generating a response to the user utterance.

  In step S1, each part of the response generation device is initialized, and the process proceeds to step S2. In step S2, the microphone 11 generates an audio signal corresponding to the user utterance, and the recognition unit 12 recognizes the audio signal, and proceeds to step S3.

  In step S3, the user utterance intention analyzer 31 of the analysis unit 13 refers to the utterance-utterance intention correspondence table database 32, estimates the utterance intention based on the user utterance recognized by the recognition unit 12, and performs step S4. Proceed to For example, the user utterance intention analyzer 31 estimates “greeting” for the user utterance 1 and “information transmission” for the user utterance 3.

  In step S4, the system utterance intention selector 41 of the response generation unit 14 refers to the utterance intention correspondence table database 42, estimates the system utterance intention corresponding to the user utterance intention estimated in step S3, and performs step S5. Proceed to For example, the system utterance intention selector 41 estimates “return of user utterance” for “greeting”, and “information transmission”, “consideration”, “evaluation / impression” for “information transmission”. Any one is estimated at random.

  In step S5, the user utterance intention analyzer 31 determines whether or not the system utterance intention estimated in step S4 is “information transmission”. If the determination is affirmative, the system utterance intention is supplied to the related information search engine 43. If the determination is negative, the system utterance intention is supplied to the response generator 45, and the process proceeds to step S6.

  In step S6, the related information search engine 43 refers to the related information database 44 and searches for and extracts related words related to the words included in the user utterance. For example, in the case of user utterance 5, party, dress, and fun are searched for as related words of “feast”. Then, the response generator 45 refers to the utterance database 46, generates utterance candidates using the words and the related words, and proceeds to step S7. As utterance candidates, for example, “Did you have a party?”, “Did you wear a dress?”, “Did you have a feast?”, Or “Is it fun?” Are generated. Here, the system utterance “Did you have a party?” Or “Did you wear a dress?” Is a response with the role of developing the user utterance “I was very frustrated.” The system utterance “Did you have a feast?” Is a response that has a role to show interest in the immediately preceding user utterance. Furthermore, the system utterance “Is it fun?” Is a response to synchronize with the emotion of the previous user.

  In step S7, the response selection unit 15 determines the priority for each of the utterance candidates generated in step S6. Specifically, the response selection unit 15 executes the following priority determination routine.

  FIG. 10 is a flowchart showing a priority determination routine. In step S21, the response selection unit 15 initializes the priority and proceeds to step S22.

  In step S <b> 22, the response selection unit 15 searches the concept database 16 and extracts “topic richness” and “emotion” for the concept (word) included in the utterance candidate generated by the response generation unit 14. Then, the process proceeds to step S23.

  In step S <b> 23, the response selection unit 15 determines the priority of the concept included in the utterance candidate based on “topic richness”. Specifically, it is determined whether the “abundance of topics” is “Large”, “Medium”, or “Small”. When “Large”, the priority is +3, and when “Middle”, the priority is When the degree is +2 or “small”, the priority is determined as +1 and the process proceeds to step S24.

  In step S24, the response selection unit 15 determines the priority of the concept included in the utterance candidate based on “emotion”. Specifically, it is determined whether “emotion” is “positive”, “negative”, or “neutral”. When “positive”, the priority is +3, and when it is “negative”, the priority is −3. , “Neutral”, the priority is determined to be +0, and the process proceeds to Step S25.

  In step S25, the response selection unit 15 adds priorities based on “topic richness” and “emotion” for each utterance candidate, and selects the utterance candidate with the highest priority. The priority of “Is it a party?” Is 6 (abundance of topics 3, emotion 3), the priority of “Did you wear a dress?” 2 (abundance of topics 2, emotion 0), “I enjoyed it. The priority of “?” Is 4 (buzz abundance 1, emotion 3). Therefore, “Is it a party?” With the highest priority is selected. When this routine ends, the process proceeds to step S8 in FIG.

  In step S8, the response selection unit 15 determines whether or not there are a plurality of utterance candidates having the same priority. If the determination is affirmative, the process proceeds to step S9. If the determination is negative, the process proceeds to step S12.

  In step S9, the response selection unit 15 randomly selects any one of the utterance candidates having the same priority, and proceeds to step S12 with the selected one as the response utterance.

  On the other hand, in step S10, the response generator 45 of the response generation unit 14 generates an utterance candidate and proceeds to step S11. Here, for example, when the system utterance intention is “repeating user utterance”, the same utterance as the input user utterance is generated. Also, when the system utterance intention is “conversation”, one randomly selected from the plural connotations utterances shown in FIG. 7 is output as the utterance candidate. When the system utterance intention is “evaluation / impression”, utterance candidates related to evaluation / impression of the user utterance are generated. For example, an input user's utterance may be recognized, an emotion that the recognized user's utterance may mean is estimated, and a response utterance that represents the same emotion as the estimated emotion may be generated. As a result, it is possible to make evaluations and impressions in synchronization with the user's emotions. As another application example, when the user's emotion is negative, it may be possible to follow the user utterance by generating an utterance having a positive content. The system utterance 4 in the dialogue example 2 is an utterance having a role of following a negative user utterance (user utterance 3).

(Dialogue example 2)
User utterance 1: I went to A's house yesterday.
System utterance 2: So then?
User utterance 3: It was very feasting, but it was boring because there were many people who didn't know it.
System utterance 4: But if you have a feast, you can enjoy it alone?
User utterance 5: Well.

  Further, when the system utterance intention is “request information transmission”, utterance candidates that request desired information are generated. When the system utterance intention is “question”, the previous system utterance is generated.

  In step S11, when there are a plurality of utterance candidates, the response selection unit 15 selects one at random, sets the selected one as a response utterance, and proceeds to step S12.

  In step S12, the response selection unit 15 supplies the response utterance selected in any of steps S7, S9, and S11 as a system utterance to the speaker 21 and the display 22, and proceeds to step S13. As a result, for example, for the user utterance 5, the system utterance 6 “Is it a party?” Is output as a sound and further output as an image.

  In step S14, it is determined whether or not there is a user utterance input. If the determination is affirmative, the process returns to step S2, and if the determination is negative, this routine is terminated.

  As described above, the response generation apparatus according to the first embodiment assigns a priority to each utterance candidate based on “buzz abundance” and “emotion” corresponding to words included in each of a plurality of utterance candidates. Decide and output the system utterance with the highest priority. From this, the response utterance device develops the user's utterance, the system shows interest in the user's utterance, expresses the emotion, synchronizes with the user's emotion, or the user's utterance content is negative You can respond to follow up. Therefore, since it responds to a user's utterance while deeply digging out a concept that enables a rich conversation and a pleasant conversation, a pleasant conversation with the user can be continued for a long time.

[Second Embodiment]
Next, a second embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected about the site | part same as 1st Embodiment, and the detailed description is abbreviate | omitted.

  FIG. 11 is a block diagram illustrating a configuration of a response generation apparatus according to the second embodiment. In the second embodiment, an analysis unit 13A is used instead of the analysis unit 13 shown in FIG. 1, and a response generation unit 14A is used instead of the response generation unit 14. The analysis unit 13A extracts a predicate and a case element from the user utterance recognized by the recognition unit 12. The response generation unit 14A generates one or more utterance candidates based on the predicate and the case element extracted by the analysis unit 13A.

  FIG. 12 is a block diagram illustrating a configuration of the analysis unit 13A. The analysis unit 13A includes a morphological analyzer 51 that performs morphological analysis of user utterances, a dependency analyzer 52 that analyzes the dependency (modification relationship) of each clause, a case analyzer 53 that extracts predicates and case elements, An indispensable case completion engine 54 that complements the indispensable case for the extracted predicate, a case dictionary 55 that defines the relationship between the predicate and the indispensable case (case element), a predicate relation analyzer 56 that analyzes the relationship between the predicates, And a predicate relation database 57 for defining a relation between predicates.

  Here, there are nine types of Japanese cases: Ga, W, D, Kara, To, De, F, Made, and Yori ("Basic Japanese Grammar-Revised Version-" “Kuroshio Publishing, p. 74-80). The case element is a noun phrase that comes immediately before the case, and is included in a clause that modifies the “predicate”. A “predicate” is a word representing a movement or state expressed in an utterance, and a phrase whose part of speech begins with a verb, an adjective, or a sa-noun.

  FIG. 13 is a diagram illustrating the configuration of the case dictionary 55. The case dictionary 55 shows the relationship between the predicate and the essential case, that is, the relationship between the predicate and the case that co-occurs with it. The case dictionary 55 defines that “create” co-occurs with “people”, “things”, and “people”, and “dexterity” co-occurs with “people”. Here, “person” indicates a noun relating to a person, and “thing” indicates a noun relating to an object.

  FIG. 14 is a diagram showing the configuration of the predicate relation database 57. The predicate relationship database 57 represents the relationship between predicate 1 and predicate 2. For example, “dexterity” is the “reason” for “making”. “Going” is “going” and “equal”. “Check” is a “means” for “knowing”. Conversely, “know” is the “result” of “examine”.

  The response generation device configured as described above can realize, for example, the following interaction example 3 with the user by executing the second response generation routine.

(Dialogue example 3)
User utterance 1: Dad was very dexterous and made all the living sideboards.
System utterance 2: Did you make a sideboard?
User utterance 3: Yes. Or make a home theater system in the living room.
System utterance 4: Dad made it?
User utterance 5: Yes.
System utterance 6: So then?
FIG. 15 is a flowchart showing a second response generation routine for generating a response to the user utterance. Here, the same processes as those in the first response generation routine are denoted by the same reference numerals. First, the case where the user utterance 1 “Dad was very dexterous and made all the sideboards in the living room” will be described as an example. Similarly to the first response generation routine, after the processes of steps S1 and S2 are executed, the process proceeds to step S31.

  In step S31, the analysis unit 13A analyzes the user utterance recognized by the recognition unit 12. At this time, each part which comprises the analysis part 13A shown in FIG. 12 performs the following process.

  First, the morpheme analyzer 51 performs morpheme analysis and outputs morpheme information such as parts of speech from the user utterance. For example, as shown in FIG. 16, the morphological analyzer 51 includes “dad (noun)”, “ha (particle)”, “very (adjective)”,..., “Making (verb)”, “ta (auxiliary verb). ) "," No (particle) ".

  The dependency analyzer 52 analyzes the dependency (modification relationship) of each clause using the analysis result of the morphological analyzer 51 as follows. For example, as shown in FIG. 17, “dad” and “very” relate to “dexterity”. “Living” is related to “Sideboard”. “Descriptive”, “Sideboard” and “All” are related to “I made”.

  The case analyzer 53 extracts predicates and case elements from the analysis result of the dependency analyzer 52. Specifically, the case analyzer 53 extracts from the analysis result of the dependency analyzer 52, as shown in FIG. 18, a predicate “dexterity” and a “dad” who modifies it. . The case analyzer 53 further extracts another predicate “Make” and “Sideboard wo” that modifies it.

  The essential case complementing engine 54 collates the analysis result of the case analyzer 53 with the case dictionary 55, and complements and outputs the essential case that is the case in which the predicate co-occurs. For example, according to the case dictionary 55 of FIG. 13, “Make” has a relationship that co-occurs with “Human Ga”, “Things”, and “H”. On the other hand, in the analysis result shown in FIG. 18, “Make” already has a modification relationship with “Sideboard”, but has no modification relationship with other cases.

  Therefore, the essential case complementing engine 54 supplements other cases (ga case, d case) that co-occur in “make”. However, in the analysis result shown in FIG. 18, there is a ga (father ga), but there is no d. Therefore, in the present embodiment, as shown in FIG. 19, the essential case complementing engine 54 does not complement “done” with respect to “make”, but complements only the case (dad ’s case).

  Note that “dexterity” has a co-occurrence with “human ga” according to the case dictionary 55 of FIG. 13, but in the analysis result shown in FIG. Already qualified. Therefore, the essential case supplement engine 54 does not need to supplement “dexterity”.

  The predicate relationship analyzer 56 refers to the predicate relationship database 57 and obtains the relationship between the predicates included in the complement result of the mandatory case complement engine 54. Here, as shown in FIG. 20, the predicate relationship analyzer 56 obtains that there is a “reason” relationship between “dexterous (present)” and “create (past)”. When the above process ends, the process proceeds to step S32 in FIG.

  In step S32, the response generation unit 14 determines whether or not a predicate or a case element exists in the user utterance using the analysis result of the analysis unit 13. If the determination is affirmative, the response generation unit 14 proceeds to step S33. When it is determined (when there is no predicate or case element), the process proceeds to step S10.

  In step S33, as shown in FIG. 21, the response generation unit 14A shown in FIG. 11 a) confirms the spoken case element (confirmation of the case element), and b) asks for the omitted case element. (Question of abbreviated elements), c) why the predicate was made, when and where the question was asked (question of the predicate), d) confirmation of the relationship between the predicates (relationship between the predicates), 4 Types of utterance candidates are generated, and the process proceeds to step S34.

  As the utterance candidates, for user utterance 1, in the case of a), for example, “Dad is dexterous?”, “Dad made?”, “Did you make a sideboard?” Are generated. In the case of b), for example, “who made it?” is generated. In the case of c), for example, “Why is it dexterous?”, “When is it dexterous?”, “When did you make it?”, “Where did you make it?” are generated. In the case of d), “Did you make dexterity?” is generated. Here, “Why is it dexterous?” “When did you make it?” “Where did you make it?” Utterances that expand user utterances, “Dad is dexterous?”, “Who made it?” "Is it dexterous?" Is an utterance that shows interest in user utterances, and "Dad did you make it?" Or "Did you use a sideboard?"

  In step S34, the response generation unit 14 determines whether there are a plurality of utterance candidates. If the determination is affirmative, the process proceeds to step S7. If the determination is negative, the process proceeds to step S12. When there is one utterance candidate, that utterance candidate is used as a response utterance in the steps described later.

  In step S7, the response selection unit 15 determines the priority for each of the utterance candidates generated in step S6, as in the first embodiment. As a result, as shown in FIG. 22, priority 6 (the topic abundance 3, emotion 3) for “Dad is dexterous” and priority 11 for “Dad made?” A priority of 12 (buzz abundance 6, emotion 6), etc. are determined for (buzz abundance 5, emotion 6) and “Did you make a sideboard?”. Then, the highest priority “Did you make a sideboard?” Is selected. If there are a plurality of priorities having the same value, one utterance candidate is selected at random (steps S8 and S9).

  As described above, the response generation apparatus according to the second embodiment is based on “topic richness” and “emotion” corresponding to words included in a plurality of utterance candidates, as in the first embodiment. The priority is determined for each utterance candidate. As a result, the response utterance device develops the user's utterance, the system shows interest in the user's utterance, expresses the emotion, synchronizes with the user's emotion, or the user's utterance content is negative You can respond like you follow. Therefore, it is possible to respond to user utterances so that topics are rich and a pleasant conversation is possible.

  Furthermore, since the said response production | generation apparatus extracts the predicate and a case element from a user utterance, and produces | generates the response which confirms the extracted predicate or case element, it can respond without preparing a dialogue plan beforehand. In addition, as long as there is a user utterance, the response generation device generates a response that digs deeply into the user utterance, so that the conversation can be smoothly performed. Furthermore, by generating a response from the predicate or case element included in the user utterance, the response generation device does not require a plan or knowledge about a specific topic in the response generation process, and can respond to any topic.

[Third Embodiment]
Next, a third embodiment of the present invention will be described. In addition, the same code | symbol is attached | subjected to the site | part same as embodiment mentioned above, and a different part from 1st Embodiment is mainly demonstrated.

  FIG. 23 is a block diagram illustrating a configuration of a response generation apparatus according to the third embodiment. The response generation apparatus is obtained by adding a concept control unit 17 to the configuration shown in FIG. In the present embodiment, the configuration shown in FIG. 23 will be described as an example, but the concept control unit 17 may be added to the configuration shown in FIG.

  The concept control unit 17 updates “abundance of topics” in the concept database 16 corresponding to the concept before the response selection unit 15 determines the priority based on the concept included in the utterance candidate.

  Specifically, the concept control unit 17 includes a context information database 61 that stores context information, a user information database 62 that stores user information, and a concept database content determination unit that determines the contents stored in the concept database 16. 63.

  The context information database 61 stores N concepts (words) included in past user utterances and system utterances. Therefore, when the dialogue is continued and the number of concepts stored in the context information database 61 exceeds N, the oldest concepts are deleted in order. For this reason, only the latest N concepts are stored in the context information database 61.

  As shown in FIG. 24, the user information database 62 stores initial values of the individual concept database 16 reflecting user preferences and the like as user information. Therefore, the user information database 62 is referred to only once in the dialog with the user.

  FIG. 25 is a flowchart showing a concept database update routine. The concept database content determination unit 63 sequentially updates the concept database 16 so that the content of the dialog with the user is not biased by executing this routine.

  In step S41, the concept database content determination unit 63 determines whether or not the user information database 62 has been referred to after starting the dialogue. If the determination is affirmative, the process proceeds to step S45. If the determination is negative, the process proceeds to step S42.

  In step S42, the concept database content determination unit 63 refers to the user information database 62 and proceeds to step S43.

  In step S43, the concept database content determination unit 63 determines whether or not the concept (concept included in the utterance candidate) exists in the user information database 62. If the determination is affirmative, the process proceeds to step S44. If so, go to Step S45.

  In step S44, the concept database content determination unit 63 reads “buzz abundance” and “emotion” corresponding to the concept from the user information database 62, and uses the “buzz abundance” and “emotion” as a concept database. The concept database 16 is initialized by writing to 16, and the process proceeds to step S45.

  In step S45, the concept database content determination unit 63 determines whether or not the concept exists in the context information database 61. If the determination is affirmative, the process proceeds to step S46. If the determination is negative, the process proceeds to step S47.

  In step S <b> 46, the concept database content determination unit 63 performs an update process on the concept database 16 to lower the “buzz richness” of the concept by one rank. Therefore, the concept “richness of topics” included in recent user utterances and system utterances is lowered by one. And the priority determination process of step S7 mentioned above is performed.

  The response selection unit 15 determines the priority for each utterance candidate using the concept database 16 updated to the latest data. Therefore, when the concept included in the utterance candidate is the same as the concept included in the recent user utterance and the system utterance, the response selection unit 15 determines a low priority for the utterance candidate. For this reason, the utterance candidate is not selected as a response utterance, and another utterance candidate is selected.

  As described above, when the concept included in the utterance candidate is the same as the concept included in the recent user utterance and the system utterance, the response generating apparatus according to the third embodiment corresponds to the concept “abundance of topics”. The concept database 16 is updated so as to lower the rank. Therefore, since the response generation apparatus determines a low priority for the utterance candidate using the concept database 16, an utterance candidate including a concept different from the recently used concept is output as a response utterance. For this reason, since the response generation device can output a response utterance of a different concept each time it responds to the user utterance, the content of the dialog with the user can be smoothly advanced.

  Note that the present invention is not limited to the above-described embodiment, and it is needless to say that the present invention can also be applied to a design modified within the scope of the claims.

  For example, although the speaker 21 and the display 22 are used as means for presenting a response utterance, a robot apparatus capable of expressing the response content may be used. In addition, the response generation device according to each embodiment may include only one of the speaker 21 and the display 22.

  Furthermore, the response generation device is not limited to the above-described configuration, and a keyboard may be used instead of the microphone 11 and the recognition unit 12. That is, the input unit for inputting the user utterance is not particularly limited.

It is a block diagram which shows the structure of the response generation apparatus which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of an analysis part. It is a figure which shows the structure of an utterance-utterance intention correspondence table | surface. It is a block diagram which shows the structure of a response production | generation part. It is a figure which shows the structure of the speech intention corresponding table memorize | stored in the speech intention corresponding table database. It is a figure which shows the structure of a related information database. It is a figure which shows a mutual talk. It is a figure which shows the structure of a concept database. It is a flowchart which shows the 1st response production | generation routine which produces | generates the response with respect to a user utterance. It is a flowchart which shows a priority determination routine. It is a block diagram which shows the structure of the response generation apparatus which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the structure of an analysis part. It is a figure which shows the structure of a case dictionary. It is a figure which shows the structure of a predicate relation database. It is a flowchart which shows the 1st response production | generation routine which produces | generates the response with respect to a user utterance. It is a figure which shows the morpheme information output from a morphological analyzer. It is a figure which shows the relationship of the dependency of a phrase. It is a figure which shows the predicate extracted by the case analyzer and the case element corresponding to it. It is a figure which shows the state by which the case element of ga case which is an essential case with respect to "make" was complemented. It is a figure which shows the relationship between predicates. It is a figure which shows the kind of speech candidate. It is a figure which shows the priority of each utterance candidate. It is a block diagram which shows the structure of the response generation apparatus which concerns on the 3rd Embodiment of this invention. It is a figure which shows the structure of a user information database. It is a flowchart which shows a concept database update routine.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Microphone 12 Recognition part 13, 13A Analysis part 14, 14A Response generation part 15 Response selection part 16 Concept database 17 Concept control part 17
21 speaker 22 display 31 user utterance intention analyzer 32 utterance-utterance intention correspondence table database

Claims (4)

An input means for inputting an utterance;
Response generating means for generating a plurality of responses to the utterance input to the input means;
For each concept, a concept storage means for storing the abundance of topics and the degree of positive / negative emotions,
A context storage means for storing concepts included in the context of past dialogue;
When the concept included in the response whose priority is to be determined is stored in the context storage unit, the degree of abundance of topics corresponding to the concept stored in the concept storage unit is reduced. Conceptual control means to control;
The abundance of topics corresponding to the concepts included in each of the responses generated by the response generation means and the degree of positive / negative of emotion are read out from the concept storage means, and the abundance of topics read and the degree of positive / negative of emotions are read. On the basis of a plurality of responses, the priority is determined so that the priority is higher as the abundance of the topic is larger and the priority is higher as the positive / negative degree of the emotion is positive. Response selection means for selecting a response with the highest priority;
Response output means for outputting the response selected by the response selection means;
A response generation device. The response generation unit extracts a concept from an utterance input to the input unit, generates a related word related to the extracted concept, and generates a plurality of responses using the extracted concept or the related word. Item 2. The response generation device according to Item 1. Enter the utterance,
Generating a plurality of responses to the input utterance;
When the concept included in the response whose priority is to be determined is stored in the context storage means that stores the concept included in the context of the past dialogue, the abundance of topics and the degree of positive / negative emotion for each concept Control to reduce the degree of abundance of topics corresponding to the concept stored in the concept storage means for storing
From the concept storage means, read out the abundance of topics corresponding to the concepts included in each of the generated responses, the degree of positive or negative emotion,
Based on the read abundance of topics and the degree of positive / negative of emotion, for each of a plurality of responses, the greater the abundance of the topic, the higher the priority, and the more positive the degree of positive / negative of the emotion, Decide the priority so that the priority is high,
Select the response with the highest priority,
A response generation method for outputting the selected response. On the computer,
Enter the speech,
Generating a plurality of responses to the input utterance,
When the concept included in the response whose priority is to be determined is stored in the context storage means that stores the concept included in the context of the past dialogue, the abundance of topics and the degree of positive / negative emotion for each concept Control to reduce the degree of abundance of topics corresponding to the concept stored in the concept storage means for storing
From the concept memory means, richness of the topic that corresponds to the concepts included in each of the plurality of response the generated and read out the degree of positive or negative emotions,
Based on the read abundance of topics and the degree of positive / negative of emotion, for each of a plurality of responses, the greater the abundance of the topic, the higher the priority, and the more positive the degree of positive / negative of the emotion, as priority is higher, to determine the priority,
Select highest priority response,
Output the selected response
A response generator to make things happen.

Images