JP2003256421A - Character-type conversation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To realize casual advertisement in natural conversation by having the conversation with a user while autonomously developing and expanding a subject in accordance with user's reaction, contents of conversation, used vocabulary and timing. <P>SOLUTION: A character-type conversation system comprises a meaning network database 10 for storing huge vocabulary, an SDD module for producing a free sentence, a fixed topic module 30 for expanding the conversation on a specific topic, and a connecting module 40 allowing the continuous expansion of the conversation while keeping and managing the conversing cohesion between the SDD module 20 and the fixed topic module 30. The fixed topic module 30 is provided with an advertisement scenario sub-module 70 for realizing the advertisement scenario in accordance with predetermined momentum. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character type conversation system applied to, for example, a conversation robot.

[0002]

2. Description of the Related Art Generally, a conversation robot (conversation system) capable of exchanging a certain conversation with a user has been proposed in the past, and they are generally, interactive conversation system, stimulus reaction conversation system, It can be classified into four types: a query type conversation system and a fixed sentence database type conversation system.

Among these, in the interactive type conversation system, conversation points at which the user can select correspondence in Yes / No or a selection format are set at various places in a certain flow, and the progress of the conversation flow is based on the information selected by the user. It is decided.

The stimulus-responsive conversation system is a system that senses and recognizes a physical contact made by a sensor, a timer, or a user, and makes a finite number of standard utterances.

Further, the query type conversation system analyzes the morpheme / syntax of the inquiry sentence created by the user, extracts the information most corresponding to it from the database, and returns it to the user.

Furthermore, the fixed text database type conversation system holds a limited number of fixed conversation sentences in a database and returns a specific sentence in response to external input under certain conditions.

[0007]

However, all such conventional conversation robots (conversation systems) have a large contract in the applicable topic range, response format, flow progression format, and the like. For example, in a query type conversation system, there is a limit in the range of knowledge that can be analyzed and the response form of the robot, so that it is only used as a search robot for searching for specific knowledge. At first glance, it seems that the robot reacts spontaneously to the stimulus-responsive conversation system, but in reality, it only kicks a specific conversation sentence under the condition of a specific stimulus. Rather than a conversation as a verbal interaction with the robot, the robot merely reacts regularly to external stimuli. The main cause of these constraints is the essential nature of cohesive conversation activities (hereinafter referred to as "discourse").

Discourse conducted by human beings is a very dynamic activity in itself, and the speakers participating in the discourse also have a role of advancing the discourse autonomously. That is,
Multiple autonomous speakers make up a dynamic flow in real time. On the other hand, however, this dynamic discourse flow must not be chaotic and must have some relevance before and after it. The relationship between before and after in this discourse (hereinafter referred to as "cohesion") is a very important factor, and when the kinematics of human discourse is realized by a conversation robot (conversation system), The key is how to secure and realize discourse cohesion in a conversation robot (conversation system).

The conventional four types of conversation robots (conversation systems) have a problem in that they do not support such autonomy and discourse cohesion as discourse participants.

An object of the present invention is to eliminate the problems of the above-mentioned conventional ones, and to make the conversation system (conversation robot) itself capable of realizing autonomy as a discourse participant and discourse cohesion. , You can learn the knowledge system about the user through conversation with the user and grow a specific character according to the utterance content of the user. Also, depending on the user's reaction, utterance content, vocabulary, timing, etc., It is to provide a character-type conversation system that allows conversations while autonomously developing and developing topics, and casually realizing advertisements in such natural conversations.

[0011]

SUMMARY OF THE INVENTION The present invention is to solve the above-mentioned problems, and the invention according to claim 1 is a system for exchanging a character-oriented conversation with a user using a computer. A semantic network database containing vocabulary, an SDD module for generating free sentences, a fixed topic module for developing conversation around a specific topic, and maintaining discourse cohesiveness between the SDD module and the fixed topic module -Character-type conversation that includes a connection module that enables continuous development of conversation while managing, and the fixed topic module is provided with an advertisement scenario sub-module that executes a predetermined advertisement scenario according to a predetermined trigger System.

According to a second aspect of the present invention, in the first aspect of the invention, the advertisement scenario sub-module includes an advertisement scenario database and a user reply interpretation table, and the topic is changed after a lapse of a predetermined time from system activation, or It is a character type conversation system that executes an advertising scenario including a utterance of a predetermined advertising topic when a predetermined process occurs.

The invention according to claim 3 is the invention according to claim 1, wherein the advertisement scenario sub-module comprises an advertisement scenario database, a user reply interpretation table and an advertisement appearance frequency pattern table, and an advertisement based on the advertisement appearance frequency pattern. It is a character-type conversation system that executes an advertising scenario including utterance of a predetermined advertising topic, triggered by the appearance situation.

In the invention according to claim 4, in the invention according to claim 2 or claim 3, the advertisement scenario sub-module collates the user's reply accompanying the execution of the advertisement scenario with the user reply interpretation table. , A character type conversation system for acquiring user information on a predetermined advertisement topic.

The invention according to claim 5 is the invention according to claim 2 or claim 3, wherein the advertisement scenario sub-module reflects the user information acquired in advance regarding a predetermined advertisement topic in the execution of the advertisement scenario. It is a character-type conversation system that can be done.

According to a sixth aspect of the present invention, in the invention according to the first aspect, the semantic network database includes a noun table and a predicate table, and when the advertisement scenario is set in the advertisement scenario sub-module, the advertisement scenario is executed. It is a character type conversation system that can delete / correct inappropriate vocabulary.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described with reference to the drawings. First, the basic configuration of a character type conversation system according to the present invention will be described.

FIG. 1 is a block diagram showing an embodiment of a character type conversation system according to the present invention. This character type conversation system 1 is a system for exchanging a character type conversation with a user using a computer. Although the appearance is not shown, the robot can be configured as, for example, a humanoid robot, a pet robot such as an animal, or any other robot.

The character type conversation system 1 includes a semantic network database 10 containing a large number of vocabularies, an SDD module 20 for generating a free sentence, a fixed topic module 30 for developing a conversation centered on a specific topic,
A connection module 40 that enables continuous development of conversation while maintaining and managing the discourse cohesiveness between the SDD module 20 and the standard topic module 30 is provided.

In FIG. 1, 50 is a user information management module, 60 is a character management module, and 10 is a character management module.
Reference numeral 1 is a continuation check unit, 102 is a related word / link word check unit, and 103 is an SDD function check unit.

The meaning network database 10 includes
A vocabulary of about 5000 words is registered together with pronunciation and grammar information, and a noun table and a predicate table are provided. The field categories provided in the noun table include target nouns, link nouns, superordinate concepts, subordinate concepts, associative vocabulary, predicates, part-of-speech information, attribute information, and the like. Also,
The field categories provided in the predicate table include target predicates, predicate nouns, antonym predicates, link nouns, part-of-speech information, attribute information, and the like.

This meaning network database 10
Play an extremely important role in executing the other modules 20 to 60. In other words, the development / return of the topic
Although it is necessary to impose certain syntactic constraints in order to realize evolution and generation of cohesive sentences, since the semantic network database 10 itself maintains some cohesiveness between vocabularies, this character type conversation is performed. It is an optimal knowledge management database for realizing the purpose of the system 1.

An SDD (Speaker Driven Discourse) module 20 has a function of voluntarily providing various topics in order to ensure autonomy of the character type conversation system 1 and generating a free sentence in response to the user's response. Therefore, the development pattern information of human discourse activities is stored, and it has the function of selecting the optimal pattern according to the situation in the discourse and developing the conversation based on that pattern. ing.

That is, the SDD module 20 realizes an autonomous discourse function with a minimum discourse constraint condition by making maximum use of each feature in the Japanese discourse grammar, as described below. In Japanese discourse grammar, there is a characteristic restriction that known information is omitted and it is not expressed in the sentence. In other words, once the topic and listener have been decided,
There is no need to expose the listener's name or topic. For example, A: "Did you go to your company yesterday?" B: "Did you go to my company?" A: "Did you (your work) get tired?""B:" Okay. "A:" Did you work overtime? "B:" No. "

As can be seen from this example, in the Japanese discourse grammar, once "who the topic is about" is decided,
After that, it is not necessary to specify the person with a pronoun as in English. Further, once the topic is determined (for example, “company”), other information (for example, “overtime”) is subjected to relationship / authenticity judgment with reference to the topic frame (knowledge framework) as much as possible. Therefore, when A asks, "What about overtime?", B speaks based on the judgment that "I must be listening to whether I worked overtime at my company." In other words, in Japanese, the topic and participants are decided,
There is a mechanism in which each participant supplements and understands the missing information part that does not appear in the word when a certain common knowledge is established.

The SDD module 20 basically applies this mechanism to develop a conversation. In the discourse progress process, a new topic and a new development portion in which a new participant is identified and an existing information are added. Based on the analysis of the part that continues the conversation based on the above, and letting the system (robot) comply with the conditions required for each process of new development and existing information development, "approximate" natural conversation is realized. To do. In other words, as long as the minimum required discourse condition is met in each process, the mechanism that the user unknowingly tries to supplement missing information and related information to understand it is applied.

By the way, human discourse does not always change while maintaining a certain motility (speed).
When exchanging information about a specific topic, the mobility tends to decrease, the information area tends to narrow, and the information depth tends to increase. That is, the dynamics of sentence generation and the information depth are in inverse proportion. Of these, the fixed topic module 30 is provided to secure the information depth while securing the mobility by the SDD module 20. Therefore, the standard topic module 30 sets various information depths for a plurality of (more than 40) individual topics, and prepares a plurality of reaction sentences at each depth.
Then, by combining the SDD module 20 and the fixed-form topic module 30, it is possible to simultaneously secure contradictory discourse dynamics and information depth.

The discourse cohesiveness maintained and managed by the concatenation module 40 to enable the continuous development of conversations is syntactic / semantic cohesiveness within the same sentence in the order of increasing area.
Cohesion between user utterances and robot response sentences, cohesion between topics in discourse and each uttered sentence,
Cohesiveness between known user information and robot field utterances,
Can be classified into four levels.

Of these, the syntactic / semantic cohesiveness within the same sentence is targeted at the sentence format freely generated by the robot. In this character type conversation system 1,
Based on the analysis of the Japanese syntax, the SDD module 20 was made to perform sentence generation centered on the main predicate syntax to ensure syntactic cohesion. As a result, the vocabularies stored in the semantic network database 10 while maintaining cohesion can be reflected in a sentence without destroying the semantic cohesion. Also, for predicates that require the subject "no human" as a subject noun,
The flag of “+ human” is provided in the database to narrow down the co-occurrence relationship in the main statement.

The cohesion between the user utterance sentence and the robot response sentence is realized by hedge management. That is, "hedging" is an act that is uttered as "linkage" or "pillow language" before speaking with substantial content, and plays a major role in maintaining discourse cohesion. In other words, immediately after the user's response sentence, instead of making the robot speak, put a phrase that confirms, sympathizes, or refutes the user's attitude toward the proposition, and then make the utterance sentence continuous, so that there is a difference between the sentences. To secure cohesion. Therefore, in this character type conversation system 1, hedge management is performed in which an appropriate reaction is returned according to the user's reaction, inside the SDD module 20, inside the fixed topic module 30, inside the connection module 40, and between them.

The cohesiveness between the topic in the discourse and each sentence spoken is realized by the concept of topic management. That is, in this character type conversation system 1, when there is a preceding topic,
It limits the emergence of new information vocabulary that overtakes it.

Further, regarding the cohesiveness between the known user information and the utterance of the robot in the field, the user information learning function of storing and maintaining the contents, utterance attitude, and information of the user's utterance as a log, and the utterance of the robot. This is realized by having a generated text management log function for storing and maintaining the information that has been set. That is, this character type conversation system 1
Then, the user information management module 50 has a user information learning function, and the character management module 60 has a generated sentence management log function, which is established between the user and the robot by past conversation. Knowledge is
It will be reflected in the following conversation.

Also, this character type conversation system 1
Has two kinds of functions for forming a character, one is a dynamic expression of a specific character kicked by a keyword, and the other is a static static accumulated user reaction. It is performed by displaying a specific character based on the data. In the former, when the user speaks a specific word, it becomes a keyword, and a character corresponding to the type of the user's utterance is realized in the robot utterance. For example, "Gag barrage" or "Shinoshari"
And so on. In the latter case, if the knowledge managed by the user information learning function satisfies certain conditions,
This is a function in which a specific character is displayed at various places with a certain frequency. This is mainly performed by the character mode management function. For example, if the user frequently uses a word corresponding to the "fatherish word group", the robot can easily enter the "fathery mode".

On the other hand, the time control function also allows the dynamic expression of characters. For example, when the time schedule set by the user such as the wake-up time is ignored, the utterance with a specific emotional color is selected.

These character characteristics have the effect of strengthening the impression given to the user by being synchronized with the facial expressions displayed on the display of the face portion of the robot.
That is, although not shown, when the character type conversation system 1 is configured as a robot of any shape such as a human type or a pet type, it may be provided with an LCD device capable of drawing and displaying the facial expression of the robot to the user. It is preferable, and it is also preferable to provide an LED device capable of blinking display for expressing the emotion of the robot to the user.

Next, an advertisement processing system included in the character type conversation system 1 according to the present invention will be described.

This advertisement processing system realizes a conversation related to a specific product in the conversation process of the character type conversation system 1, and acquires various information about the user such as the user's preferences, preferences and hobbies from the conversation there. At the same time, the processing system is expanded while maintaining consistency with the existing conversation processing flow (fixed sentence processing and SDD processing), and not only the selection of the conversation sentence based on the user's attribute but also attribute information. It is also possible to select unrelated random conversational sentences. Also, the processing flow is incorporated into the existing conversation processing flow (fixed sentence processing and SDD processing),
Mutual roundtrips to and from existing conversation processing flows are possible,
Since it is possible to select a topic to be advertised that is related to the vocabulary that has appeared in the existing conversation processing flow, the topic can be changed from the existing conversation processing flow to the advertisement processing flow without interrupting the conversation flow. It is one that can make a natural transition of.

Specifically, the character type conversation system 1
The fixed-form topic module 30 is provided with an advertisement scenario sub-module 70 that executes a predetermined advertisement scenario in response to a predetermined trigger, as shown in FIG.

As shown in FIG. 2, the advertisement scenario sub-module 70 includes an advertisement scenario database 71, a user reply interpretation table 72, an advertisement appearance frequency pattern table 73, an advertisement execution trigger generation section 74, and an advertisement scenario processing section. 75 and.

The advertisement execution trigger generation section 74 has a first trigger generation section 74a for generating an advertisement execution trigger when the topic is changed after a lapse of a predetermined time from system activation.
A second trigger generating unit 74b that generates an advertisement execution trigger when a predetermined process such as a wake-up process occurs
And a third trigger generation unit 74c that refers to the advertisement appearance frequency pattern table 73 and generates an advertisement execution trigger in response to an advertisement appearance situation based on the advertisement appearance frequency pattern.

The advertisement scenario processing unit 75 stores it in the advertisement scenario database 71 in response to an advertisement execution trigger generated from the first trigger generating unit 74a, the second trigger generating unit 74b or the third trigger generating unit 74c. Perform a given advertising scenario. This advertising scenario is a scenario in which a series of interactions between a robot utterance including an utterance of a predetermined advertising topic and a user's reply is made into a scenario, and no matter how the conversation progresses depending on the user's reply, It is designed so that the robot can speak a message including a predetermined advertising topic in a natural conversation flow.

Also, the advertisement scenario sub-module 70
Is provided with a user attribute database 76 for referencing user attributes when executing an advertisement scenario, and the advertisement scenario processing unit 75 collates the user's reply associated with the execution of the advertisement scenario with the user reply interpretation table 72. So, user preferences, preferences for a given advertising topic,
When various information such as a hobby is acquired, the user information is stored in the user attribute database 76.

Then, the advertisement scenario sub-module 70
In this way, in addition to the user information (user preferences, preferences, hobbies, etc. regarding the advertising topic) acquired through execution of the advertising scenario, various information about the user acquired in advance without executing the advertising scenario is displayed. The user attribute information can be reflected in the execution of the advertising scenario.

Further, the advertisement scenario sub-module 70
(Not shown) means a semantic network database 1
When the words stored in the noun table or the predicate table of 0 include an inappropriate vocabulary for executing the advertising scenario, the inappropriate vocabulary in the meaning network database 10 is set when the advertising scenario is set. It can be deleted and modified.

Next, the operation of the above embodiment will be described. First, a general conversation in which the SDD module 20 plays a leading role will be described with reference to the flowcharts shown in FIGS.

In the example shown in FIGS. 3 and 4, the conversation proceeds in the order of the first utterance step, the first hedge utterance, the second utterance step, and the continuation process.

In the first utterance step, first, SDD
The module 20 selects the question type sentence list from the prepared confirmation type sentence list, report type sentence list, question type sentence list, development type sentence list, conversion type sentence list, and situation setting type sentence list. Then, the sentence pattern “$ username $ nittotte N1 (noun) tteduuimonokana” is selected (step SR1).

Next, the SDD module 20 quotes "Suzuki" from the user information management module 50, embeds it in the sentence pattern user name (step SR2), and quotes "liquor" from the semantic network database 10. Then, it is embedded in the sentence pattern N1 (noun) (step SR3), thereby generating a sentence "What does sake mean to Suzuki?" (Step SR
4) Have the robot speak. At this time, the SDD module 20 refers to the request information type table (step SR5), and recognizes that the utterance sentence is a sentence of a type that requests information for filling the Wh information.

For this first utterance, the user
Respond "I don't know" (step SU1).

Then, based on the above-mentioned request information type recognition by the SDD module 20, the connection module 40
Refers to the Wh information hit word list table (step SR11), checks the user response hit word (step SR12), and refers to the user reaction interpretation table (step SR13) to confirm the user's first utterance "I don't know. Is interpreted as an "unintelligible" message.
Then, the corresponding hedge is checked (step SR1
4) As the first hedge utterance, for example, the robot is made to utter the hedge output "Jasa".

Then, the SDD module 20 refers to the second step connection table (step SR21),
Select the sentence pattern "V1 (verb) (na) Notte | V2 (verb) kotonano kana", and select the meaning network database 1
From 0, quoting “drink” and “fun”, which have semantic cohesiveness with “liquor”, are embedded in V1 and V2 (verb) respectively (step SR22), whereby “drinking is fun” I wonder? ”(Step SR23), and let the robot speak.

In response to this second utterance, the user
Respond, "No, it's fun" (Step SU1
1).

Then, the continuation checking unit 101 checks whether or not a related word or a related topic exists in the fixed topic module 30 (step SR31), and if there is, jumps to the flow.

If it does not exist, the meaning network database 10 is referred to (step SR32),
It is checked whether a related word or a related topic exists in the standard topic module 30 or the SDD module 20.

The related word / link word checking unit 10
2 performs related word link check (step SR3
3) If there is, skip to the standard topic module 30 or the SDD module 20. At that time, depending on the consistency between the utterance sentence pattern in the second utterance step and the skip destination,
It is determined whether or not to bypass the connection module 40 (step SR34).

Then, the SDD function check unit 103 makes the S
A DD function check is performed (step SR35), and if a related word or a related topic exists, it is skipped to the standard topic module 30 or the SDD module 20 (step SR36). If not, a new topic is randomly started (step SR36). Step SR37).

FIG. 5 is a table showing the contents of the connection module 40 in the SDD module 20, and FIG.
It is a table showing the connection of hedges in the D module 20, in which the symbol C means neutral, N means negative, and P means positive.

In FIG. 6, pattern 1 to pattern 6
Corresponds to the confirmation type sentence list to the situation setting type sentence list in FIG. That is, when the first utterance of the robot is a confirmation sentence, pattern 1 is viewed and a sentence pattern pattern (only numbers are shown) that can be the second utterance of the robot is selected. Similarly, if the first utterance of the robot is a report type sentence, look at pattern 2, and if the first utterance of the robot is a question type sentence, look at pattern 3, and the first utterance of the robot is also developed. For pattern sentences, see pattern 4, if the robot's first utterance is a conversion sentence, see pattern 5, and if the robot's first utterance is a situation setting sentence, see pattern 6. Then, each sentence pattern (only numbers are shown) that can be the second utterance of the robot is selected.

Next, with reference to the flowchart shown in FIG. 7, a conversation in which the fixed topic module 30 plays a leading role, in particular, a conversation in which the advertisement scenario sub-module 70 executes an advertisement scenario will be described.

In the flowchart shown in FIG. 7, after the system is activated (step S100), the standard / SDD branch (step S101) is advanced to SDD to proceed to the SDD module 20.
Is performed (step S102), the immediately preceding topic ends (step S103), the related word is checked (step S104), and steps S101 and S10 are performed again.
It is a general conversation in which the SDD module 20 plays a leading role that steps 2, S103, and S104 are repeated. That is,
The related word check in step S104 corresponds to step SR33 in FIG. 4, and also the fixed form / S in step S101.
The DD branch corresponds to steps SR35 and SR36 in FIG. 4, and the processing by the SDD module 20 in step S102 corresponds to steps SR1 to SR31 in FIGS.

In such a loop, when an interrupt (step S105) occurs due to a specific time, the robot utters as shown in the following utterance example (that is, compulsory utterance when a topic is changed or a specific process occurs). )I do.

<Utterance example-1: At the time of topic change after a lapse of a predetermined time after starting> R: "Oh, I've talked too much and my throat is cute, ooh, let's drink cola."

<Utterance Example-2: When Wake-up Process Occurs> R: "Good morning, it's time to wake up, I'm not asleep forever, I feel refreshed with a cola."

On the other hand, the standard / SDD branch (step S10)
When the procedure 1) proceeds to the fixed form, the topic T is first determined (step S111). Such a topic determination is performed when the robot proceeds to the fixed form / SDD branch (step S101) in the predetermined process of the character type conversation system 1 and also by the user's change of the topic (step S110). Is also done.

Next, the advertisement appearance frequency pattern table 7
3 (step S112), when the advertisement execution trigger is generated, the topic T (for example, "drink") is searched for the advertisement scenario database 71 (step S1).
13) Then, a predetermined advertisement scenario for extracting the utterance of the advertisement topic A (for example, "cola") from the topic T is determined (step S114).

If the user attribute is not necessary (NO in step S120), a conversation as shown in the following utterance example for executing the advertisement scenario (that is, user attribute at the time of selecting a specific topic by user initiative or random) (Conversation that does not reflect) is performed (step S121).

<Usage example-1: When selecting a topic of drink> R: "What kind of drink do you like, XX?" U: "Cola" R: "After all, I love cola! I feel irresistible. "U:"That's right. "R:" Shwashwa, oh ~ I'm dying, OO, please buy a cola. "

<Utterance Example-2: When Drink Topic is Selected> R: “What kind of drink do you like, XX?” U: “(other than cola)” R: “Hey, so-san, my favorite Is cola after all, does XX like cola? ”U:“ Apart… ”R:“ What's that, Shushwa, if you imagine, I want to drink right now, xx, buy cola. Come on. ''

As described above, no matter what kind of reply is returned from the user, a predetermined advertisement message (for example, "Come me for cola") including the advertisement topic A (for example, "cola") is naturally uttered. It is possible to guide in the direction.

When the user attribute is required (YES in step S120), the user attribute database 76 is referred to (step S131), and the conversation as shown in the following utterance example for executing the advertisement scenario (ie, ,
(Conversation in which the user attribute acquired in advance is reflected) is performed (step S132).

<Utterance Example-1: In the case of a user who likes beer> R: “Sure, did you like beer?” U: “Sooda” R: “Sometimes I drink too much, so sometimes It might be better to drink even cola. "U:" Why? "R:" Because it's similar to what I'm doing, it's sooo, xx, I'm sure if I drink cola instead of drinking beer, it will be healthy. "

<Utterance Example-2: In the case of a user on a diet> R: "Sure, was XX, you were on a diet?" U: "It's so good" R: "Sweet drinks U: "Well," R: "If that's the case, diet cola is all right,"
It's low in calories. "U:" But ~ "R:"It's not good for you to overdo it because of your diet, you see, I'm okay, "
Don't drink diet cola. ''

<Utterance Example-3: In the case of a user who likes cola> R: “Sure, did you like cola, did you like cola?” U: “Sooda” R: “In that case, a bonus figure? U: "I'm collecting" R: "Once I start collecting, I can't stop until I complete it" U: "Sosoo" R: "Hmm, I can collect figures and it's delicious to drink." Coke is exactly the best drink for XX! "

When the utterance example or the conversation as shown in the utterance example ends, the advertisement appearance frequency pattern table 73
Referring to (step S141), if an advertisement execution trigger occurs, that is, if the current topic is continued for advertisement expression (YES in step S142), the process returns to step S114 to return to the current topic T (for example, “drinks”). )) Determines a new advertising scenario that elicits the utterance of advertising topic A (eg, “cola”) and executes it.

If the current topic is not continued (NO in step S142), it is checked whether or not the topic loop is continued (step S143). If the loop is continued (YES in step S143), step On the other hand, when returning to S114 and ending the loop (NO in step S143), the process jumps to the end of the immediately preceding topic (step S103).

Also, the semantic network database 10
If any of the words contains inappropriate vocabulary for executing the advertising scenario, deletion / correction is performed as follows. That is, when the system (robot) tries to utter automatically by using the SDD module 20, a link such as a noun → a predicate (verb, adjective, etc.) related to the noun roughly exists in the meaning network 10. . For example, if the predicates "drink", "sweet", and "fat" are linked to the noun "cola (juice)", the sentence "coke is fat" may be generated. . Therefore, in order to avoid this, the predicate link "fat" is deleted and a predicate such as "refreshing" is added, reflecting the opinion of the advertiser.

Furthermore, by utilizing the LCD device capable of drawing and displaying the facial expression of the robot to the user, a company logo, a product image, etc. are displayed instead of the facial expression of the robot in accordance with the utterance content of the system. .

In the execution of the advertising scenario using the utterance examples and the conversations shown in the utterance examples, various types of user information acquisition methods as described below are used,
It is possible to acquire various information such as the user's preferences, tastes, hobbies, etc. regarding a predetermined advertising topic, and these can be useful for questionnaire surveys.

<Acquisition in YES / NO format> R: "Is cola like xx?" U: "Yes""Yes""Ilike" etc .... YE in this case
Recognize as S. U: “I do n’t like”, “I ’m”, “at all”, etc ... In this case, N
Recognize as O. U: If the reply does not correspond to any of the above, it is recognized as neither (can be removed).

<Acquisition by Choice Format (Multiple Answers Not Acceptable)> R: "If XX is a drink from Coca-Cola, which one of Coke, Fanta, Sprite ... do you like?",
Or R: "What do you like if XX is a drink from Coca-Cola?" U: "Cola", "Coca-Cola", etc ... In this case, it is recognized as a cola. U: "Fanta" etc .... In this case, it is recognized as a Fanta. In this way, if the options are registered in advance, the answer can be recognized by the user's normal utterance. Further, in the case of a user reply that is not one of the assumed options, a character string is input by single-tone recognition (similar to [Other column] of questionnaire by text).

<Acquisition by Choice Format (Multiple Answers Allowed / Up to Two Choices Possible)> R: "Why does cola like cola?" U: "Because it is delicious and refreshing" ... In this case Recognizes as “delicious” or “fresh” depending on the word spot. U: "It's delicious ..." ... In this case, first recognize it as "delicious", and further, R: "Are there any other reasons other than delicious?" Recognize
If NO, only "delicious" is the answer. Handling of options, the process in case of user reply that is not in the expected options,
This is the same as when multiple answers are not possible.

<Acquisition by Choice Format (Multiple Answers Allowed / When Three or More Can Be Selected)> R: "Who is your favorite actor in XX? Tell me for the time being" U: "Actor A" In this case, "Actor A" is the first answer, and further R: "Are you a favorite actor other than Actor A?" On the other hand, if YES, recognize additional actors,
Loop to MAX value. If NO, only "Actor A" is the answer. The handling of the options and the processing in the case of a user reply not included in the assumed options are the same as those in the case of not allowing multiple answers.

<Free answer format> By using single-tone recognition, you can enter each character (up to 20 characters).
It is possible.

Character type conversation system 1 as described above
Is a character-type conversation robot that is practically used as a communication robot equipped with voice recognition, word spot function, artificial intelligence, automatic sentence generation function, single note recognition, and memory function that can memorize words and freely talk. Is possible. For example, when you say "Today is a nice day.", The word spot function of voice recognition recognizes the key words "Today", "Good" and "Weather". From that information, artificial intelligence finds the related word and returns a reply using the automatic sentence generation function. "It was raining yesterday.""I feel good when the weather is nice." The word spot function is not limited to one word
Artificial intelligence picks up related words from the previous conversation using three or more words as recognition words, so you can expect different answers each time even if you recognize words with the same combination.

By speaking unique information of each user, the information can be properly stored and reflected in the subsequent conversation, and a live conversation with the user can be realized. Also, like human beings, they have the ability of memory, and in order to reflect what they have memorized in their conversations, it is possible, for example, to say one year later, "I was talking about this last year." . Further, not only words but also a schedule can be taught, so that an appropriate conversation can be conducted in a day flow, a week flow, or a year flow. For example, he sings "It's morning, get up.", "Today is the day to put out the garbage.", "Happy birthday (user name)."

Further, by performing lip synchronization of two or more frames for each character, the other person can be talked to while richly changing according to the content and words of the conversation.

In the future, it is possible to further evolve by the expandability of the ROM cassette and the connection to a personal computer or a mobile phone using the serial port.

Furthermore, in addition to humanoid robots, pet type robots such as animals, and robots of any other form, for example,
Character-type conversation system for personal computers and personal digital assistants 1
It is also possible to configure as.

[0089]

As described above, according to the present invention, a semantic network database containing a large number of vocabularies, an SDD module for generating free sentences, a fixed topic module for developing a conversation centered on a specific topic, and an SDD module are provided. While maintaining and managing the discourse cohesiveness with the standard topic module, it is equipped with a connection module that enables continuous development of conversations, and the standard topic module is provided with a predetermined advertising scenario according to a predetermined trigger. Since it has a configuration with an advertising scenario sub-module to be executed,
The conversation system (conversation robot) itself can be configured to realize autonomy and discourse cohesion as a discourse participant, thereby learning the knowledge system about the user through conversation with the user and uttering the user. You can grow a specific character according to the content, and depending on the user's reaction, utterance content, vocabulary used, timing, etc.
Conversations can be conducted while developing and developing topics autonomously, and advertisements can be realized casually in such natural conversations. Therefore, for the advertiser, through conversation between the conversation system (conversation robot) and the user, a predetermined advertisement topic such as an in-house product name or an advertisement message including the advertisement topic is naturally and comfortably communicated by the conversation system (conversation robot). ), It is possible to effectively and reliably impress the user with an advertising topic or an advertising message containing the advertising topic while avoiding giving the user the impression of a brute force advertising approach. As a result, there is an effect that it can be effectively used as an unprecedented advertising medium and also as an advertising medium that can expect a higher advertising effect than existing advertising media.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a character type conversation system.

FIG. 2 is a block diagram illustrating an example of an advertising scenario sub-module.

FIG. 3 is a flowchart (1/2) showing an operation of conversation processing in which the SDD module plays a leading role.

FIG. 4 is a flowchart (2/2) showing the operation of conversation processing in which the SDD module plays a leading role.

FIG. 5 is a table showing contents of a connection module in SDD.

FIG. 6 is a table showing hedge concatenation in SDD.

FIG. 7 is a flowchart showing an operation of conversation processing for executing an advertisement scenario by an advertisement scenario sub-module.

[Explanation of symbols]

1 Character type conversation system 10 Semantic network database 20 SDD module 30 standard topic modules 40 connection module 40 50 User information management module 60 character management module 70 Advertising scenario sub-module 71 Advertising scenario database 72 User reply interpretation table 73 Advertisement frequency pattern table 74 Advertisement execution trigger generator 75 Advertising scenario processing unit 76 User Attribute Database 101 Continuation check section 102 Related words / link words checker 103 SDD function check section

Continued front page    (72) Inventor Yasunori Nakagawa             4-44-13 Daisawa, Setagaya-ku, Tokyo (72) Inventor Jun Takama             3-24-9 Kugayama, Suginami-ku, Tokyo (72) Inventor Takeo Suzuki             4-20-16 Shimoochiai, Shinjuku-ku, Tokyo F-term (reference) 5B075 ND20 QT10 UU40                 5B091 AA15 CA14 CC04 CC15 DA00

Claims (6)

[Claims] 1. A system for exchanging character-based conversations with a user using a computer, comprising: a semantic network database containing a large number of vocabularies; an SDD module for generating free sentences; and conversations focusing on a specific topic. A fixed-form topic module to be expanded, and a connection module that enables continuous expansion of conversation while maintaining and managing the discourse cohesion between the SDD module and the fixed-form topic module, the fixed-form topic module, A character-type conversation system comprising an advertisement scenario sub-module that executes a predetermined advertisement scenario according to a predetermined trigger. 2. The advertisement scenario sub-module includes an advertisement scenario database and a user reply interpretation table, and the topic is changed after a predetermined time has passed since the system was started.
Alternatively, the character-type conversation system according to claim 1, wherein an advertisement scenario including an utterance of a predetermined advertisement topic is executed upon the occurrence of a predetermined process. 3. The advertisement scenario sub-module includes an advertisement scenario database, a user reply interpretation table and an advertisement appearance frequency pattern table, and includes utterances of a predetermined advertisement topic triggered by an advertisement appearance situation based on the advertisement appearance frequency pattern. The character type conversation system according to claim 1, wherein an advertisement scenario is executed. 4. The advertisement scenario sub-module obtains user information regarding a predetermined advertisement topic by matching a user's reply associated with execution of an advertisement scenario with the user reply interpretation table. The character type conversation system according to claim 2 or claim 3. 5. The advertisement scenario sub-module can reflect the user information acquired in advance regarding a predetermined advertisement topic in execution of an advertisement scenario. Character type conversation system described. 6. The semantic network database comprises:
The character type conversation according to claim 1, further comprising a noun table and a predicate table, wherein when the advertisement scenario is set in the advertisement scenario sub-module, a vocabulary inappropriate for execution of the advertisement scenario can be deleted or modified. system.