JP7297721B2 - Program, device, and method for creating dialogue scenario substituted with verbs according to character attributes - Google Patents

Description

The present invention relates to technology for creating a dialogue scenario for an agent that interacts with a user.

Dialogue systems that naturally interact with users using smartphones, tablet terminals, and robots have become widespread. A computer graphics agent appears on the display and interacts with the user by voice or text. The agent can be recognized by the user as a single person with whom the user can interact, and establishes a dialogue according to the user's situation, tastes and preferences, and emotions.
Agents have individual character attributes (age, gender, hometown, etc.) as well as person profiles. The user can feel a sense of familiarity by interacting with an agent having a character attribute that the user likes, for example.

FIG. 1 is an explanatory diagram showing the sense of discomfort that a user feels when interacting with an agent.

According to FIG. 1(a), an agent in her 20s with a female character attribute is having a conversation with a user as follows.
User: "What are you going to do for dinner tonight?"
Agent: "I'm going to eat fried rice."
User: "What kind of music do you often listen to?"
Agent: "Listen to KANA."

On the other hand, according to FIG. 1(b), a teenage male character attribute agent is having a dialogue with the user in the same manner using the same dialogue scenario as in FIG. 1(a).
User: "What are you going to do for dinner tonight?"
Agent: "I'm going to eat fried rice."
User: "What kind of music do you often listen to?"
Agent: "Listen to KANA."
In this case, the user feels uncomfortable with saying, for example, "I'll eat", even though he or she is interacting with an agent whose character attribute is a man in his teens. This is because teenagers and males tend to make rough utterances.

For this reason, agent dialogue scenarios generally need to be created manually according to character attributes. By manually creating an agent, it is possible to create an agent that feels individuality according to the character's attributes. However, manually creating dialogue scenarios for various agents poses a cost problem.

Conventionally, there is a technique for automatically rewriting each clause of an uttered sentence according to an agent with desired character attributes (see, for example, Patent Document 1). According to this technique, a "rewriting rule set" of manually created utterances is referred to according to character attributes such as age and gender.
For example, when creating a dialog scenario with a character attribute of a woman in her teens from a dialog scenario with a character attribute of a man in her 50s "I like golf", "na" is replaced with "yo". As a result, "I like golf" is created as a dialog scenario for the character attribute of a teenage female.
In general, golf is a popular sport among men in their fifties. For this reason, there is absolutely no sense of incongruity when a male agent in his fifties says, "I like golf." However, it feels strange for a teenage female agent to say, "I like golf." This is because it is not the content that a typical teenage woman utters.

On the other hand, there is a technique for replacing nouns included in spoken sentences of a dialogue scenario with nouns suitable for desired character attributes (see, for example, Non-Patent Document 1). According to this technique, a dialogue scenario matching the character attributes is created by executing the following steps.
(S1) Character attributes (for example, female in her twenties, male in her fifties, etc.) are specified.
(S2) For nouns included in the utterances of the character in the dialogue scenario, similar candidates of the nouns are extracted from Wikipedia (registered trademark).
(S3) From among the extracted similarity candidates, a message posted by an SNS user having the same character attribute as the designated character attribute is extracted.
(S4) One noun characteristically appearing in the posted text of the SNS user is determined.
(S5) Replace the noun included in the spoken sentence of the character in the dialogue scenario with the noun determined in S4.
For example, the utterance "I often listen to KANA" in a dialogue scenario with a character attribute of a woman in her 20s is rewritten as "I often listen to HAMADA" in a dialogue scenario with a character attribute of a man in her 50s. be able to.

JP 2013-196070 A

Shunichi Tahara et al., "Proposal and Evaluation of Automatic Substitution Method for Nouns Emphasizing Character in Dialogue Scenarios," DEIM 2019, [online], [Searched July 8, 2020], Internet <URL: https: //db-event.jpn.org/deim2019/post/papers/2.pdf＞ Wikipedia: Category "verb", [online], [searched on July 1, 2020], Internet <URL: https://ja.wikipedia.org/wiki/Category:%E5%8B%95%E8% A9%9E> Weblio Thesaurus, [online], [Searched on July 8, 2020], Internet <URL: https://thesaurus.weblio.jp/> Associative thesaurus, [online], [Searched on July 8, 2020], Internet <URL: https://renso-ruigo.com/word/> Thesaurus, [online], [Searched on July 8, 2020], Internet <URL: https://dictionary.goo.ne.jp/thsrs/>

However, with only noun replacement as in the technology described in Non-Patent Document 1, the user may feel uncomfortable with the utterance sentence of the dialogue scenario.
For example, "KANA" is an artist popular among women in their twenties, and "HAMADA" is an artist popular among men in their fifties.
At this time, it is assumed that the utterance sentence of the agent with the character attribute of a woman in her twenties is "my KANA CD was stolen". It is assumed that the noun included in this utterance is replaced with the utterance sentence of a male character in his fifties by the technique described in Non-Patent Document 1. Here, "KANA" included in the utterance sentence of the female character in her twenties is replaced with "HAMADA" suitable for the male character in her fifties. Then, it will be replaced with the utterance sentence "My HAMADA CD was stolen."
However, the verb "pakuru" in the rewritten utterance sentence is a word often used by young people, and there is a sense of incongruity as an utterance that characterizes a man in his fifties. That is, since the verbs included in the spoken sentence are not replaced with verbs that match the character's attributes, the character's expression remains unnatural.

On the other hand, the inventors of the present application thought that simply replacing nouns and phrases in a dialogue scenario would not represent the character's characteristics based on the character's attributes. In fact, I thought that the ``verb'', which is reminiscent of what the person based on the character attribute actually uttered, might be the most representative of the character's characteristics based on the character attribute.

Also, for example, Wikipedia (registered trademark) pages are mostly nouns, and there are few articles related to verbs. Therefore, there is also the problem that there are few substitution candidates for verbs. For example, looking at the category "verb", there are only 47 verb articles. Therefore, it is extremely difficult to replace the verbs included in the utterance sentences of the dialogue scenario with verbs corresponding to the character attributes, simply referring to the verbs on Wikipedia.

SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a program, apparatus, and method for creating a dialogue scenario in which verbs are substituted according to character attributes.

According to the program of the present invention,
dialogue scenario storage means for accumulating dialogue scenarios for each character attribute;
Posted text accumulation means for accumulating a plurality of posted texts from a plurality of users for each character attribute;
sememe determining means for determining sememes in utterance verbs described in an utterance sentence of a dialogue scenario based on the first character attribute, and for associating a plurality of synonymous verbs having the same sememe;
Characteristic synonymous verb selection means for selecting, as a characteristic synonymous verb, a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe, using a user's posted text based on the second character attribute. and,
Making the computer function as verb replacement means for replacing the said utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with a characteristic synonym verb to make the utterance sentence of the dialogue scenario of the second character attribute. characterized by

According to another embodiment of the program of the present invention,
The semantic determining means is
When a plurality of sememes are determined for an utterance verb described in an utterance sentence of a dialogue scenario based on the first character attribute,
Using a synonym corpus, for each sememe, searching for a plurality of synonymous verbs that have the same sememe,
Counting the appearance frequency of each synonymous verb for each sememe in the user's posted text based on the first character attribute,
It is also preferred to have the computer function to determine the sememe with the highest frequency of occurrence of each synonym.

According to another embodiment of the program of the present invention,
The semantic determining means is
part-of-speech extracting means for extracting nouns and verbs that form a dependency relation from an utterance sentence of a dialogue scenario based on the first character attribute;
Synonymous verb search means for searching for synonymous verbs similar to the dependent verb for each sememe using the synonym corpus;
Co-occurrence frequency for searching the frequency of co-occurrence between a dependency noun and each synonymous verb from the user's posted text based on the first character attribute, and determining the sememe with the highest co-occurrence frequency for each synonym verb It is also preferable to have a computer function as a search means.

According to another embodiment of the program of the present invention,
The characteristic synonym verb selection means is
As described below, using Akaike's Information Criterion (AIC), the evaluation value V(P1,P2) of the synonymous verb P2 for the character attribute P1 is calculated, and the evaluation value V(P1,P2) is equal to or greater than a predetermined threshold value, or the similar verb P2, whose evaluation value V(P1, P2) is a predetermined number in descending order, is detected.
p11: Number of posts related to synonym P2 by users with character attribute P1
p12: Number of posts by users with character attribute P1 that do not have synonym P2
p21: Number of posts related to synonym P2 by users other than character attribute P1
p22: Number of posts by users other than character attribute P1 that have a synonym P2
p11 + p12 + p21 + p22 = number of posts posted by all users
MLL_IM(P1,P2) = (p11+p12) log(p11+p12)
+(p11+p21) log(p11+p21)
+(p21+p22) log(p21+p22)
+(p12+p22) log(p12+p22)-2N logN
MLL_DM(P1,P2) = p11 log p11 + p12 log p12 + p21 log p21 + p22 log p22 - N log N
However, N = p11 + p12 + p21 + p22
AIC_IM(P1, P2) = -2 × MLL_IM(P1, P2) + 2 × 2
AIC_IM(P1,P2): Dependent AIC of a pair of character attribute P1 and synonym P2
AIC_DM(P1, P2) = -2 × MLL_DM(P1, P2) + 2 × 3
AIC_DM(P1,P2): Independent AIC for a pair of character attribute P1 and synonym P2
V(P1,P2) = AIC_IM(P1,P2) - AIC_DM(P1,P2)
It is also preferable to have the computer function as

According to another embodiment of the program of the present invention,
Posted texts are texts posted by unspecified users on SNS (Social Networking Service) or blogs,
It is also preferred to have the computer function that character attributes are classifying attributes based on age, gender, hometown, or any combination thereof.

According to another embodiment of the program of the present invention,
Dialogue means for an agent to interact with a user using a dialogue scenario;
During the dialogue with the user through the dialogue means, when or after the utterance sentence replaced with the characteristic synonymous verb in the dialogue scenario is uttered to the user, does the utterance sentence make the user feel uncomfortable? It is also possible to make the computer further function as speech verb correction means for re-substituting the replaced feature synonymous verbs back to the original speech verbs when it is determined that the user feels uncomfortable. preferable.

According to the dialogue scenario creation device of the present invention,
dialogue scenario storage means for accumulating dialogue scenarios for each character attribute;
Posted text accumulation means for accumulating a plurality of posted texts from a plurality of users for each character attribute;
sememe determining means for determining sememes in utterance verbs described in an utterance sentence of a dialogue scenario based on the first character attribute, and for associating a plurality of synonymous verbs having the same sememe;
Characteristic synonymous verb selection means for selecting, as a characteristic synonymous verb, a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe, using a user's posted text based on the second character attribute. and,
a verb replacement means for replacing the utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with a characteristic synonym verb to make the utterance sentence of the dialogue scenario of the second character attribute. and

According to the dialogue scenario creation method of the present invention,
The device
dialogue scenario storage means for accumulating dialogue scenarios for each character attribute;
a posted message accumulating means for accumulating a plurality of posted messages from a plurality of users for each character attribute;
a first step of determining a semantic verb in an utterance verb described in an uttered sentence of a dialogue scenario based on the first character attribute and associating a plurality of synonymous verbs having the same semantic;
a second step of selecting, as a feature synonymous verb, a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe, using the user's posted text based on the second character attribute;
and executing a third step of replacing the said utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with the characteristic synonym verb to make the utterance sentence of the dialogue scenario of the second character attribute. characterized by

According to the program, apparatus, and method of the present invention, it is possible to create a dialogue scenario in which verbs are substituted according to character attributes.

FIG. 11 is an explanatory diagram showing a feeling of discomfort that a user has due to interaction with an agent; 1 is a functional configuration diagram of a dialogue scenario creation device according to the present invention; FIG. FIG. 4 is an explanatory diagram of a dialogue scenario accumulation unit and a posted message accumulation unit according to the present invention; 4 is a functional configuration diagram of a semantic determination unit in the present invention; FIG. FIG. 10 is an explanatory diagram of a feature-synonymous verb selection unit in the present invention; FIG. 4 is an explanatory diagram of a verb replacement section in the present invention; FIG. 4 is an illustration representing a dialogue scenario replaced by the present invention; FIG. 3 is an illustration representing an interface for correcting verbs that have been substituted according to the present invention;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 2 is a functional configuration diagram of the dialogue scenario creation device according to the present invention.

The dialogue scenario creation device 1 of the present invention automatically creates a dialogue scenario according to the character attributes of the agent.
According to FIG. 2, the dialogue scenario creation device 1 includes a dialogue unit 100, a dialogue scenario storage unit 101, a posted sentence storage unit 102, a semantic determination unit 11, a characteristic synonym verb selection unit 12, and a verb replacement unit. It has a section 13 and a speech verb correction section 14 . These functional components are implemented by executing a program that causes a computer installed in the device to function. In addition, the flow of processing of these functional components can also be understood as a dialogue scenario creation method.

The dialogue scenario creation device 1 of the present invention is
Determining sememes related to verbs included in the utterance sentence of the first dialogue scenario by the agent of the first character attribute (sememe determining unit 11),
selecting a synonymous verb that characterizes the second character attribute of the second dialogue scenario to be created from among a plurality of synonymous verbs present in the semantic (characteristic synonymous verb selection unit 12);
Replaces the verb included in the spoken sentence of the first dialogue scenario with a synonymous verb corresponding to the attribute of the second character (verb replacement unit 13)
By doing so, a second dialogue scenario is automatically created for the agent of the second character attribute.

[Dialogue unit 100]
The dialog unit 100 activates an agent that interacts with the terminal 2 owned by the user by voice or text via the user interface. The user can select an agent according to desired character attributes (age, sex, hometown, etc.). The agent refers to the dialog scenario storage unit 101 and interacts with the user using a dialog scenario corresponding to the character's attributes. The agent can be recognized by the user as a person based on its character attributes.

[Dialogue scenario accumulation unit 101]
The dialogue scenario accumulation unit 101 accumulates dialogue scenarios for each character attribute.
The dialogue scenario storage unit 101 constructs a dialogue scenario from received sentences and uttered sentences. A received sentence is assumed to be a text that a user would utter, and an uttered sentence is a text uttered by an agent in response to the received sentence.
Also, the character attribute of the agent is, for example, a classification attribute based on age, gender, hometown, or a combination thereof.

FIG. 3 is an explanatory diagram of a dialogue scenario storage unit and a posted text storage unit according to the present invention.

According to FIG. 3A, the dialogue scenario of the first character attribute accumulated in the dialogue scenario accumulation unit 101 is shown. Here, the first character attribute is, for example, female in her twenties.
(Received sentence) (Spoken sentence)
・・・・・・・・・・・・ ・・・・・・・・
"What are you going to do for dinner?""I'm going to eat fried rice."
・・・・・・・・・・・・ ・・・・・・・・
The verb in the utterance sentence of the agent is "eat", and the way of speaking assumes the first character attribute "twenties, female". In order to adjust the utterance sentences of subtle verbs according to the character attributes, according to the prior art, they were created manually.
According to the embodiment of the present invention, the second dialogue scenario with the second character attribute "teenage male" is automatically created from the first dialogue scenario with the first character attribute "twenties female".

[Posted text accumulation unit 102]
The posted message accumulation unit 102 accumulates a plurality of posted messages from a plurality of users for each character attribute (user attribute, demographic).
Posted texts may be texts posted to, for example, SNS (Social Networking Service) or blogs, and may be transmitted by an unspecified number of third parties or by a specific person.

According to FIG. 3B, for example, the following posted sentences are listed for each character attribute.
Posted text based on the first character attribute (20s, female)
"I can eat fried rice."
・・・・
"Eating fried rice seems to be fine."
"Let's have some fried rice."
Posted text based on the second character attribute (teens, male)
・・・・
"Let's eat"
・・・・
"Rice, eat and be full"
"May I have some fried rice?"
Posted sentences may have a masculine tone, a feminine tone, a youthful tone, an elderly tone, or a dialect-like tone. These affect the verbs of the posted text according to the character attributes.

In order to classify an unspecified number of posted texts by character attribute, it is possible to use a technique of estimating character attributes such as gender, age, and hometown for each posted text, as in Patent Document 1, for example. The poster's profile (character attribute) is estimated using the interaction relationship between posters that can be acquired from the SNS site server.
The character attribute of the posted message is the same classification attribute as the character attribute of the dialogue scenario. Character attributes particularly affect verbs (eg, endings) in spoken sentences.

The posted text with the first character attribute is posted by users (unspecified number of people or a specific person) with the same or similar character attribute as the first dialogue scenario to be replaced.
The posted message with the second character attribute is a posted message posted by users (unspecified number of people or a specific person) with the same or similar character attribute as the second dialogue scenario to be replaced.

[Sememe determining unit 11]
A semantic determination unit 11 determines a "semantic verb" in an "utterance verb" described in an utterance sentence of a dialogue scenario based on the first character attribute, and associates a plurality of "synonymous verbs" having the same semantic. . Here, if a plurality of semantics are determined for the "utterance verb" described in the utterance sentence of the dialogue scenario based on the first character attribute, one sememe is determined.

A "sememe" is a concept-based unit that makes up the meaning of a word. In linguistics, we distinguish between ``the meaning of a sentence'' and ``the meaning of a word''. Sememes are elements that make up the "meaning" of a word, along with phonemes and morphemes.

FIG. 4 is a functional configuration diagram of a semantic determination unit in the present invention.

According to FIG. 4, the sememe determining unit 11 inputs the dialogue scenario of the first character attribute, and outputs a sememe table including the determined semantics and synonyms.
For example, assume that a dialog scenario of a female agent in her twenties is input as a dialog scenario of the input first character attribute.
(utterance sentence)
"I will eat fried rice"

According to FIG. 4, the semantic determination unit 11 has a part-of-speech extraction unit 111 , a synonym search unit 112 , and a co-occurrence frequency search unit 113 .

<Part-of-speech extraction unit 111>
The part-of-speech extraction unit 111 extracts nouns and verbs that form a dependency relation from the utterance sentence of the dialogue scenario based on the first character attribute.
The part-of-speech extraction unit 111 may use, for example, the Japanese dependency analyzer CaboCha. When a sentence is input to such an analyzer, nouns and verbs are output.
According to FIG. 4, the part-of-speech extraction unit 111 extracts nouns and verbs as follows.
(spoken sentence) (noun) (verb)
"I will eat fried rice""Friedrice""I will eat"
The part-of-speech extraction unit 111 outputs the verb "eat" to the synonym search unit 112, and outputs the noun "fried rice".

<Synonymous verb search unit 112>
The synonym verb search unit 112 uses the "synonym corpus" to search for "synonym verbs" similar to the dependent verb for each sememe.

The synonym corpus is composed of sememes and synonyms, and includes the following, for example.
・Weblio thesaurus (see Non-Patent Document 3)
・ Associative thesaurus (see Non-Patent Document 4)
・Thesaurus dictionary (see Non-Patent Document 5)
For example, when the verb “eat” is input, the synonym search unit 112 searches for one or more synonyms similar to the verb “eat”.
At this time, the verb ``eat'' has the meaning of ``to eat,'' such as ``eat rice,''``fish,'' and ``eat vegetables.'' There is a semantic element of "supporting oneself", such as "do what you like and make a living". A plurality of synonymous verbs are retrieved for each sememe.
(verb) (semine) (synonym)
"eat""have a meal" eat
eat
have a meal
to drink
live on one's own
live on
Eat The plural synonyms ``eat'', ``eat,''``eat'' and ``drink'' equate the seme ``eat''. In addition, plural synonyms ``live'', ``live life'' and ``eat'' have the same sememe ``self-support''.

The synonymous verb search unit 112 derives all semantics for the input verb, and searches for all synonymous verbs for each sememe. The retrieved sememe table is output to co-occurrence frequency retrieval section 113 .

<Co-occurrence frequency search unit 113>
The co-occurrence frequency search unit 113 searches for “the frequency of co-occurrence between the dependent noun and each synonymous verb” from the user's posted text based on the first character attribute, and finds the frequency of each synonym verb for each sememe. Count the frequency of occurrence. Then, the sememe with the highest co-occurrence frequency of synonymous verbs is determined.

According to FIG. 4, the co-occurrence frequency search unit 113 receives, for example, the semantic table of the verb “eat” from the synonym search unit 112 and the noun “fried rice” from the part-of-speech extraction unit 111 . Also, the co-occurrence frequency search unit 113 refers to the posted message storage unit 102 and acquires posted messages of the first character attribute.
Here, the co-occurrence frequency search unit 113 determines sememes with a high co-occurrence frequency between the noun “fried rice” and each synonymous verb in the sememe table.

According to FIG. 4, it is assumed that there are eight posted texts with the first character attribute in the posted text storage unit 102 . Among them, in the following three posted sentences, the noun “fried rice” and a synonymous verb that is synonymous with the semantic “eat” co-occur.
(posted text) (noun) (synonymous verb)
"I'll eat fried rice.""Friedrice."
"You look healthy after eating fried rice""Friedrice""Eat"
"Let's eat fried rice""Friedrice""Eat"
According to FIG. 4, synonymous verbs with a high co-occurrence frequency for the noun "fried rice" include "eat" and "eat". These synonymous verbs correspond to the semantic "to eat". Then, when the semantics "eat" and "support yourself" are compared, the sememe with the highest co-occurrence frequency is "eat".
That is, the co-occurrence frequency search unit 113 determines the sememe "to eat" and outputs a sememe table in which the synonymous verbs "eat", "eat", "eat", and "kakeru" are associated with each other.

[Characteristic synonym selection unit 12]
The characteristic synonymous verb selection unit 12 uses the user's posted text based on the second character attribute to select a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe as a characteristic synonymous verb. Select as

The second character attribute is based on the newly created second dialogue scenario agent. The agent may remind the user of a specific person (for example, an entertainer or celebrity). In that case, as the user's posted message based on the second character attribute, it is possible to use the posted message that the specific person has posted in the past. As a result, it is possible to automatically create a second dialogue scenario based on a specific person from a general first dialogue scenario based on an unspecified person.

FIG. 5 is an explanatory diagram of the characteristic synonym verb selection unit in the present invention.

According to FIG. 5, the sememe table is input from the sememe determining unit 11, and the posted message of the second character attribute based on the dialogue scenario to be created is referred to.
Here, assume that a male in his teens is set as the second character attribute, and a dialogue scenario for that agent is created.
The characteristic synonymous verb selection unit 12 searches the frequency of occurrence of the following plural synonymous verbs in the same sememe using the user's posted text based on the second character attribute.
Significant ``to have a meal''
Synonymous verbs
Then, the characteristic synonymous verb selection unit 12 searches for each of the synonymous verbs posted sentences with the second character attribute in which the synonymous verb is described, and selects the synonymous verb with the highest appearance frequency as the characteristic Select as

The feature synonymous verb selection unit 12 calculates the evaluation value V(P1,P2) of the synonymous verb P2 with respect to the character attribute P1 using Akaike's Information Criterion (AIC) as follows, A similar verb P2 whose evaluation value V(P1, P2) is equal to or greater than a predetermined threshold value, or whose evaluation value V(P1, P2) is a predetermined number in descending order is set as a detection target.
p11: Number of posts related to synonym P2 by users with character attribute P1
p12: Number of posts by users with character attribute P1 that do not have synonym P2
p21: Number of posts related to synonym P2 by users other than character attribute P1
p22: Number of posts by users other than character attribute P1 that have a synonym P2
p11 + p12 + p21 + p22 = number of posts posted by all users
MLL_IM(P1,P2) = (p11+p12) log(p11+p12)
+(p11+p21) log(p11+p21)
+(p21+p22) log(p21+p22)
+(p12+p22) log(p12+p22)-2N log N
MLL_DM(P1,P2) = p11 log p11 + p12 log p12 + p21 log p21 + p22 log p22 - N log N
However, N = p11 + p12 + p21 + p22
AIC_IM(P1, P2) = -2 × MLL_IM(P1, P2) + 2 × 2
AIC_IM(P1,P2): Dependent AIC of a pair of character attribute P1 and synonym P2
AIC_DM(P1, P2) = -2 × MLL_DM(P1, P2) + 2 × 3
AIC_DM(P1,P2): Independent AIC for a pair of character attribute P1 and synonym P2
V(P1,P2) = AIC_IM(P1,P2) - AIC_DM(P1,P2)

The characteristic synonymous verb selection unit 12 selects one synonymous verb according to the Akaike information criterion. Here, the top several synonymous verbs may be selected, and one synonymous verb may be randomly selected from among them.

According to FIG. 5, when the Akaike information criterion for each synonymous verb is calculated for men in their teens with the character attribute, it can be understood that the synonymous verb "eat" has a high appearance frequency.
Finally, the characteristic synonymous verb selection unit 12 outputs the verb “eat” and the synonymous verb “eat” to the verb replacement unit 13 .

[Verb replacement unit 13]
The verb replacement unit 13 replaces the utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with the characteristic synonym verb to obtain the utterance sentence of the dialogue scenario of the second character attribute.

FIG. 6 is an explanatory diagram of the verb replacement part in the present invention.

According to FIG. 6, the verb replacement unit 13 replaces the verb "eat" with the synonym "eat" in the dialogue scenario of the first character attribute (twenties, female), and the second character attribute (10 (1980s, 1980s, male) dialogue scenario.
Spoken sentence of the dialogue text of the first character attribute: "Let's eat fried rice."
Spoken sentence of the dialogue text of the second character attribute: "Let's eat fried rice"

As described above, by repeating the semantic determination unit 11, the feature synonym selection unit 12, and the verb replacement unit 13, a dialogue scenario with a desired character attribute is finally selected from dialogue scenarios with different character attributes. can be created automatically.

FIG. 7 is an illustration representing a dialogue scenario replaced by the present invention.

According to FIG. 7, a dialogue scenario with a character attribute of a man in his twenties is created from the dialogue scenario with a character attribute of a woman in his twenties as shown in FIG. 1(a).
User: "What are you going to do for dinner tonight?"
Agent: "I'm going to eat fried rice."
The user is interacting with an agent whose character attribute is a man in his teens, and does not feel uncomfortable with the utterance of "I'll eat it." This is because teenagers and males tend to make rough utterances.

Here, from the user's point of view, there are cases where the uttered sentences of the dialogue scenario in which the verbs have been replaced have an unnatural sense of incongruity in Japanese. For example, assume that there are the following utterance sentences in a dialogue scenario.
"I want to eat fried rice"-> Verb: "eat", function word: "tai"
When the synonymous verb "eat" to be replaced is selected, the following utterance sentence is created.
"I want to eat fried rice"->"I want to eat fried rice"
In this way, since the verbs are replaced without considering the conjugation of the verbs, a sense of incongruity arises in the Japanese language.

FIG. 8 is an illustration representing an interface for correcting a substituted verb according to the present invention.

[Speech verb correction unit 14]
The utterance verb correction unit 14, during the dialog with the user in the dialog unit 100, when or after the utterance sentence replaced with the characteristic synonym verb in the dialog scenario is uttered to the user, An inquiry is made by text or voice as to whether or not the uttered sentence causes a sense of incongruity. At this time, the utterance verb correction unit 14 transmits inquiry information to the user interface of the terminal 2 via the dialogue unit 100 .
According to FIG. 8, two inquiry buttons are displayed on the interactive application of the terminal 2 as follows.
"There is a sense of incongruity in the answer"
"There is no sense of incongruity in the answer"

When it is determined that the user feels something wrong, the user taps “There is something wrong with the response”. This tap information is transmitted from the terminal 2 to the dialogue scenario creation device 1 . The tap information is recognized by the utterance verb correction unit 14, and the replaced feature synonymous verb is replaced again so as to restore the original utterance verb.
In this way, the agent can judge the correctness of the verb substitution by interacting with the user.

As described in detail above, according to the program, device and method of the present invention, it is possible to create a dialogue scenario in which verbs are substituted according to character attributes. As a result, it is possible to automatically create a dialogue scenario that does not feel out of place as a character attribute.

For the various embodiments of the present invention described above, various changes, modifications and omissions within the spirit and scope of the present invention can be easily made by those skilled in the art. The foregoing description is exemplary only and is not intended to be limiting. The invention is to be limited only as limited by the claims and the equivalents thereof.

1 Dialogue scenario creation device 100 Dialogue unit 101 Dialogue scenario storage unit 102 Posted sentence storage unit 11 Semantic determination unit 111 Part of speech extraction unit 112 Synonymous verb search unit 113 Co-occurrence frequency search unit 12 Feature synonymous verb selection unit 13 Verb replacement unit 14 speech verb correction unit 2 terminal

Claims (8)

dialogue scenario storage means for accumulating dialogue scenarios for each character attribute;
Posted text accumulation means for accumulating a plurality of posted texts from a plurality of users for each character attribute;
sememe determining means for determining sememes in utterance verbs described in an utterance sentence of a dialogue scenario based on the first character attribute, and for associating a plurality of synonymous verbs having the same sememe;
Characteristic synonymous verb selection means for selecting, as a characteristic synonymous verb, a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe, using a user's posted text based on the second character attribute. and,
Making the computer function as verb replacement means for replacing the said utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with a characteristic synonym verb to make the utterance sentence of the dialogue scenario of the second character attribute. A program characterized by The semantic determining means is
When a plurality of sememes are determined for an utterance verb described in an utterance sentence of a dialogue scenario based on the first character attribute,
Using a synonym corpus, for each sememe, searching for a plurality of synonymous verbs that have the same sememe,
Counting the appearance frequency of each synonymous verb for each sememe in the user's posted text based on the first character attribute,
2. The program according to claim 1, which causes the computer to determine the sememe with the highest frequency of occurrence of each synonymous verb. The semantic determining means is
part-of-speech extracting means for extracting nouns and verbs that form a dependency relation from an utterance sentence of a dialogue scenario based on the first character attribute;
Synonymous verb search means for searching for synonymous verbs similar to the dependent verb for each sememe using the synonym corpus;
Co-occurrence frequency for searching the frequency of co-occurrence between a dependency noun and each synonymous verb from the user's posted text based on the first character attribute, and determining the sememe with the highest co-occurrence frequency for each synonym verb 3. The program according to claim 1, which causes a computer to function as search means. The characteristic synonym verb selection means is
As described below, using Akaike's Information Criterion (AIC), the evaluation value V(P1,P2) of the synonymous verb P2 for the character attribute P1 is calculated, and the evaluation value V(P1,P2) is equal to or greater than a predetermined threshold value, or the similar verb P2, whose evaluation value V(P1, P2) is a predetermined number in descending order, is detected.
p11: Number of posts related to synonym P2 by users with character attribute P1
p12: Number of posts by users with character attribute P1 that do not have synonym P2
p21: Number of posts related to synonym P2 by users other than character attribute P1
p22: Number of posts by users other than character attribute P1 that have a synonym P2
p11 + p12 + p21 + p22 = number of posts posted by all users
MLL_IM(P1,P2) = (p11+p12) log(p11+p12)
+(p11+p21) log(p11+p21)
+(p21+p22) log(p21+p22)
+(p12+p22) log(p12+p22)-2N logN
MLL_DM(P1,P2) = p11 log p11 + p12 log p12 + p21 log p21 + p22 log p22 - N log N
However, N = p11 + p12 + p21 + p22
AIC_IM(P1, P2) = -2 × MLL_IM(P1, P2) + 2 × 2
AIC_IM(P1,P2): Dependent AIC of a pair of character attribute P1 and synonym P2
AIC_DM(P1, P2) = -2 × MLL_DM(P1, P2) + 2 × 3
AIC_DM(P1,P2): Independent AIC for a pair of character attribute P1 and synonym P2
V(P1,P2) = AIC_IM(P1,P2) - AIC_DM(P1,P2)
4. The program according to any one of claims 1 to 3, which causes a computer to function as a computer program. Posted texts are texts posted by unspecified users on SNS (Social Networking Service) or blogs,
5. A program according to any one of claims 1 to 4, characterized in that the character attributes cause the computer to function as classification attributes based on age, gender, hometown, or a combination thereof. Dialogue means for an agent to interact with a user using a dialogue scenario;
During the dialogue with the user through the dialogue means, when or after the utterance sentence replaced with the characteristic synonymous verb in the dialogue scenario is uttered to the user, does the utterance sentence make the user feel uncomfortable? Inquiring by text or voice about whether the 6. A program according to any one of claims 1 to 5. dialogue scenario storage means for accumulating dialogue scenarios for each character attribute;
Posted text accumulation means for accumulating a plurality of posted texts from a plurality of users for each character attribute;
sememe determining means for determining sememes in utterance verbs described in an utterance sentence of a dialogue scenario based on the first character attribute, and for associating a plurality of synonymous verbs having the same sememe;
Characteristic synonymous verb selection means for selecting, as a characteristic synonymous verb, a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe, using a user's posted text based on the second character attribute. and,
a verb replacement means for replacing the utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with a characteristic synonym verb to make the utterance sentence of the dialogue scenario of the second character attribute. dialogue scenario creation device. A device interaction scenario creation method comprising:
The device
dialogue scenario storage means for accumulating dialogue scenarios for each character attribute;
a posted message accumulating means for accumulating a plurality of posted messages from a plurality of users for each character attribute;
a first step of determining a semantic verb in an utterance verb described in an uttered sentence of a dialogue scenario based on the first character attribute and associating a plurality of synonymous verbs having the same semantic;
a second step of selecting, as a feature synonymous verb, a synonymous verb with the highest appearance frequency among a plurality of synonymous verbs in the same sememe, using the user's posted text based on the second character attribute;
and executing a third step of replacing the said utterance verb described in the utterance sentence of the dialogue scenario based on the first character attribute with the characteristic synonym verb to make the utterance sentence of the dialogue scenario of the second character attribute. A dialog scenario creation method characterized by:

Images