JP7368335B2 - Programs, devices and methods for interacting with positive parroting responses - Google Patents

Description

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

``Conversation agent'' technology, which naturally interacts with users using smartphones and tablet devices, is becoming widespread. According to this technology, a computer graphics character on a display interacts with the user through voice and text. A character serving as a conversation agent can be recognized by the user as a person with whom he or she can have a conversation, and establishes a conversation according to the user's situation, hobbies, and emotions. Characters also have the same profiles as people (attributes such as age, gender, place of birth, etc.), and the user can interact with the characters with a sense of affinity.

A dialogue agent generally proceeds with a dialogue with a user based on a "dialogue scenario" registered in advance. At this time, if the user's utterance matches the dialogue scenario, an appropriate response can be given, but if not, the user often responds with a parroting response to the user's utterance. .
A parroting response sentence is a response sentence that is a part of the user's uttered sentence as it is. Users tend to have the illusion that the dialogue agent understands what they are saying by parroting responses. For this purpose, the user further attempts to utter a new utterance sentence. It becomes easier to elicit from the user utterances (for example, the user's own concerns) that can be used as a reference for counseling.

Conventionally, there is a technology in which a dialogue agent generates a response sentence depending on the presence or absence of a focus word included in a user's utterance (see, for example, Non-Patent Document 1). According to this technology, a focus word is extracted from a user's utterance using a learning model based on a plurality of features such as parts of speech and the order in which they appear. Depending on the presence or absence of the focal word, a fixed response sentence or a response sentence including the focal word is generated in a parrot-like manner.

FIG. 1 is an explanatory diagram showing a parrot-like response sentence of a dialogue agent.

According to FIG. 1, an interactive application is displayed on the display of a terminal 2, which is, for example, a smartphone. When an utterance is entered in the input form, a response from the dialogue agent character is displayed. Of course, the uttered sentence is not limited to text input into an input form, but may also be a voice collected by a microphone or a response sentence uttered from a speaker.
According to FIG. 1(a), the conversation agent is interacting with the user as follows.
User: "I ate gyoza today."
Dialogue agent: “Is it gyoza?”
In this case, the focus word "gyoza" is extracted from the user's utterance. A response sentence "Is it gyoza?" containing the focus word "gyoza" is generated and is sent back to the user.

Shinya Ishida et al., “Generation of listener responses that encourage speech for listening dialogue systems”, SIG-SLUD, 5(01), 1-6. (2016), [online], [September 3, 2020] Search], Internet <URL: https://jsai.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=1147&item_no=1&attribute_id=1&file_no=1>

AI and machine learning products, "sentiment analysis", [online], [searched on September 3, 2020], Internet < https://cloud.google.com/natural-language/docs/analyzing-sentiment?hl =ja＞

However, the user's utterances often include negative words (for example, words such as self-denial, suicidal thoughts, discrimination issues, etc.). According to the technique described in Non-Patent Document 1, a negative word included in a user's utterance is also extracted as a focus word, and a parrot-like response sentence including the focus word is returned.

According to FIG. 1(b), the conversation agent is interacting with the user as follows.
User: "I'm a bad person."
Dialogue agent: “Are you a bad person?”
When such a dialogue occurs, the user feels uncomfortable with the dialogue with the dialogue agent and feels even more negative. Furthermore, for example, it becomes difficult to elicit further utterances from the user, and as a result, trust and favorability towards the dialogue agent decrease.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a program, a device, and a method that prevent a negative dialogue from occurring due to a parrot-like response to a user.

According to the present invention, there is provided a program that causes a computer to function as a dialogue agent that responds with a parroting response sentence while replying to a user with a scenario response sentence in accordance with a dialogue scenario,
a simple sentence dividing means for dividing an uttered sentence from a user into a plurality of simple sentences;
an emotion polarity determination means for determining whether or not the emotion polarity is positive for each simple sentence;
Functions as a simple sentence determining means to determine one simple sentence with positive emotional polarity,
The dialog agent is characterized in that it causes the computer to function so as to reply with a parroting response sentence to the determined simple sentence.

According to another embodiment of the program of the present invention,
It further functions as a desire expression determination means for determining the presence or absence of a desire expression for each simple sentence,
It is also preferable that the simple sentence determining means causes the computer to function so as to determine one simple sentence in which the emotional polarity is positive and the expression of desire is present.

According to another embodiment of the program of the present invention,
It further functions as a behavioral expression determining means for determining the presence or absence of a behavioral expression for each simple sentence,
It is also preferable that the simple sentence determining means causes the computer to function so as to determine one simple sentence in which the emotional polarity is positive, the expression of desire is present, and the expression of action is present.

According to another embodiment of the program of the present invention,
further functioning as a response history recording means for recording whether or not the response sentence that the dialogue agent replied to the user is a parroting response sentence;
When the dialogue agent attempts to reply to the user with a parroting response sentence, if the conversation agent has responded with a parroting response sentence immediately before being recorded in the response history recording means, the conversation agent is configured to respond with a parroting response sentence. It is also preferable to make the computer functional.

According to another embodiment of the program of the present invention,
It further functions as a response history recording means for recording response sentences that the dialogue agent replies to the user,
If the noun and predicate included in the parroting response sentence that the dialog agent attempts to reply to the user are the same as the noun and predicate included in the response sentence that was answered immediately before and recorded in the response history recording means, It is also preferable for the computer to function so as to respond with a complimentary response sentence.

According to another embodiment of the program of the present invention,
It further functions as a response history recording means for recording the response text and response date and time that the dialog agent replied to the user,
If the nouns and predicates included in the parroting-like response sentence that it attempts to reply to the user are the same as the nouns and predicates included in the response sentence recorded in the response history recording means, the dialogue agent It is also preferable that the computer function so as to respond with a response sentence and to also respond with a response sentence that reminds the user of the same parroting response sentence that was answered at the past response date and time of the response sentence.

According to another embodiment of the program of the present invention,
further functions as a noun history recording means for recording nouns included in response sentences sent by the dialogue agent to the user for a predetermined period;
It is also preferable that the dialog agent causes the computer to function so as to reply with a parroting response sentence when a noun included in the parroting response sentence to be replied to the user is recorded in the noun history recording means.

According to another embodiment of the program of the present invention,
If there are only simple sentences whose emotional polarity is negative according to the simple sentence determination method,
It is also preferable that the dialog agent function so as to respond with a scenario response sentence with positive emotional polarity, without responding with a reciprocal response sentence.

According to another embodiment of the program of the present invention,
If there are only simple sentences whose emotional polarity is neutral according to the simple sentence determination means, or if there are only neutral and negative simple sentences,
It is also preferable that the dialogue agent function to respond with a parroting response sentence regarding the final neutral sentence.

According to the present invention, there is provided a dialogue device having a dialogue agent that responds with a parroting response sentence while replying to a user with a scenario response sentence in accordance with a dialogue scenario,
a simple sentence dividing means for dividing an uttered sentence from a user into a plurality of simple sentences;
an emotion polarity determination means for determining whether or not the emotion polarity is positive for each simple sentence;
a simple sentence determining means for determining one simple sentence with positive emotional polarity,
The dialog agent is characterized in that it responds to the determined simple sentence with a parrot-like response sentence.

According to the present invention, there is provided a dialogue method for a device having a dialogue agent that responds with a parroting response sentence while replying to a user with a scenario response sentence in accordance with a dialogue scenario,
The device is
a first step of dividing the utterance from the user into a plurality of simple sentences;
a second step of determining whether the emotional polarity is positive or not for each simple sentence;
a third step of determining one simple sentence with positive emotional polarity;
The dialog agent is characterized in that it responds to the determined simple sentence with a parrot-like response sentence.

According to the program, device, and method of the present invention, it is possible to prevent a negative dialogue from occurring by replying to the user with a parrot-like response sentence. In particular, dialogue can be promoted through positive, parroting responses.

It is an explanatory diagram showing a parrot-like response sentence of a dialogue agent. 1 is a basic functional configuration diagram of an interaction device according to the present invention. FIG. 3 is an explanatory diagram showing the flow of creating a response sentence in the functional configuration of FIG. 2; FIG. 3 is an explanatory diagram showing a parrot-like response sentence in the dialog agent of the present invention. FIG. 3 is a functional configuration diagram with additional functions to the dialogue device according to the present invention. It is an explanatory diagram showing processing of a simple sentence decision part. It is an explanatory diagram showing a parrot-like response sentence in other embodiments. It is an explanatory diagram showing a parrot-like response sentence using a noun history record part.

Hereinafter, embodiments of the present invention will be described in detail using the drawings.

FIG. 2 is a basic functional configuration diagram of the dialog device according to the present invention.

According to FIG. 2, the dialogue device 1 of the present invention includes a dialogue agent 10 that responds with a "parroting response sentence" while replying to a user with a "scenario response sentence" in accordance with a dialogue scenario. .
Furthermore, the dialogue device 1 further includes a simple sentence segmentation section 11 , an emotional polarity determination section 120 , and a simple sentence determination section 13 . These functional components are realized by executing a program that causes a computer installed in the device to function. Furthermore, the processing flow of these functional components can also be understood as a dialog method.

According to FIG. 2, the dialogue device 1 exchanges utterances and response sentences with the terminal 2 via the network. For example, a dialogue application is installed on the terminal 2, and transmits sentences uttered by the user to the dialogue device 1, and clearly indicates the response sentences received from the dialogue device 1 to the user.
Of course, the entire dialogue device 1 may be configured as a dialogue application and installed on the terminal 2.

[Simple sentence division part 11]
The simple sentence dividing unit 11 receives an uttered sentence from the user and divides it into a plurality of "simple sentences." Since the user's utterances include a mixture of multiple sentences, complex sentences, and simple sentences, all of them are divided into simple sentences. The plurality of divided simple sentences are output to the emotion polarity determining section 120.

The simple sentence division unit 11 executes the following two processes.
(Process 1) Using a morphological analyzer (for example, MeCab (registered trademark)), the user's utterance is divided into morphemes, and "predicates" such as verbs and adjectives are extracted.
(Process 2) If the uttered sentence includes a conjunctive particle, predicates extracted before and after the conjunctive particle are extracted. Then, a phrase related to the predicate is extracted as a "simple sentence".

FIG. 3 is an explanatory diagram showing the flow of creating a response sentence in the functional configuration of FIG. 2.

Assume that the following user's utterance is input to the simple sentence segmentation unit 11.
``I'm doing some cleaning on my day off, so I can't go to the movie theater. I'd like to watch a video at home.''

This uttered sentence consisting of a plurality of sentences is divided into simple sentences by the following steps.
(s1) A period "." is detected by morphological analysis and divided into multiple sentences.
1st sentence: "I won't be able to go to the movie theater because I'll be doing a lot of cleaning on my day off." (compound sentence)
Second sentence: "I want to watch the video at home."
(s2) For the first sentence (complex sentence), the conjunctive particle "node" is detected by morphological analysis. The independent verb ``suru'' and the independent verb ``go'' are included before and after the conjunctive particle.
First sentence: "I'm going to... go..." (complex sentence)
Then, as shown below, the sentence is divided into a phrase related to the independent verb "suru" and a phrase related to the independent verb "go" as simple sentences.
Simple sentence: “I do a lot of cleaning on my days off.”
Simple sentence: “I can’t go to the movie theater.”
(s3) Regarding the second sentence, no conjunctive particle is detected by morphological analysis. It also includes the auxiliary verb ``tai''.
Second sentence: “I want to...”
Then, as shown below, phrases related to the independent verb "shitai" are extracted as simple sentences.
Simple sentence: “I want to watch a video at home.”

[Emotional polarity determination unit 120]
The emotional polarity determination unit 120 determines the “emotional polarity” of each simple sentence output from the simple sentence segmentation unit 11. Then, the emotional polarity of each simple sentence is output to the simple sentence determination unit 13.
The emotional polarity is determined as positive/negative/neutral. Specifically, there are a method of using a determiner such as Support Vector Machine, and a method of using Cloud Natural Language API provided by Google (registered trademark) (for example, see Non-Patent Document 2).

According to FIG. 3, the emotional polarity of each simple sentence is determined as follows.
“I do a lot of cleaning on my days off” -> Neutral “I can’t go to the movie theater” -> Negative “I want to watch videos at home” -> Positive

[Single sentence determination unit 13]
The simple sentence determining unit 13 receives the emotional polarity of each simple sentence from the emotional polarity determining unit 120 and determines one simple sentence with positive emotional polarity. The determined simple sentence is output to the dialogue agent 10.

According to the uttered sentence of the first example (FIG. 3), the following simple sentence is determined.
The utterance in the first example is, ``I have to do some cleaning on my day off, so I can't go to the movie theater.I can't go to the movies at home.''
I want to see it. ”
“I do a lot of cleaning on my day off” -> Neutral “I can’t go to the movie theater” -> Negative “I want to watch videos at home” -> ★Positive The determined simple sentence is “I want to watch videos at home.” Become.

According to the uttered sentence of the second example, the following simple sentence is determined.
Second example utterance: “I have plans to meet a friend and go to my favorite restaurant.I also want to watch a movie.
is. ”
"I have plans to meet a friend"-> Neutral "I'm going to my favorite store"-> ★Positive "I also want to see a movie"-> ★Positive Here, two simple sentences are determined to be positive. In this case, the simple sentence determining unit 13 determines the following simple sentence, "I also want to go to the movie theater."

[Dialogue agent 10]
The dialogue agent 10 generally proceeds with a dialogue with a user based on a "dialogue scenario" registered in advance. A dialogue scenario is composed of a sequence of uttered sentences and response sentences. The uttered sentence is text that is assumed to be uttered by the user, and the response sentence is text that the dialog agent replies to the uttered sentence.
The dialogue agent 10 can respond with an appropriate response sentence if the user's utterance matches the dialogue scenario, but if it does not, it can respond with a parroting response to the user's utterance. , waits for the next utterance from the user.

The dialogue agent 10 responds to the simple sentence determined by the simple sentence determination unit 13 with a parroting response sentence. A parroting response sentence basically consists of a "predicate" included in a simple sentence and a "noun" related to the predicate. Furthermore, auxiliary verbs and particles such as ``dane'' and ``nonone'' are added immediately after the verb.
Since the simple sentence determined by the simple sentence determining unit 13 has a positive emotional expression, the parroting response sentence will also be positive.

According to FIG. 3, the simple sentence determination unit 13 has determined the simple sentence ``I'd like to watch a video at home.'' The dialogue agent 10 includes the verb "I want to see" and "video (wo)" related to "I want to see" included in the simple sentence, and adds the auxiliary verb and particle "nonone" immediately after the verb. do. As a result, a parroting response sentence is generated, ``I guess you want to watch the video at home.''
The parroting response sentence is then sent to the user's terminal 2.

FIG. 4 is an explanatory diagram showing a parrot-like response sentence in the dialog agent of the present invention.
According to FIG. 4, the conversation agent is interacting with the user as follows.
User's utterance: ``I have to do some deep cleaning on my day off, so I can't go to the movie theater.
I want to watch videos at home. ”
Dialogue agent's response: "You want to watch the video at home, right?"

<Other first embodiment of dialogue agent>
There are cases where the simple sentence determining unit 13 determines that the emotional polarity is "only simple sentences that are negative."
When a user responds in a supportive manner to a negative simple sentence in a user's utterance, the user's desire to interact may decrease as described below.
User's utterance: "I'm a bad person."
Dialogue agent's response: "Yes, yes."
In other words, when the dialogue agent 10 responds with a supportive response to a user's negative utterance (for example, a utterance that is self-denying, suicidal, or discriminatory), it may end up affirming the user's negative utterance. become.

In this case, the dialog agent 10 registers in advance a scenario response sentence with positive emotional polarity. Then, without replying with a conciliatory response sentence, a scenario-like response sentence with positive emotional polarity is replied.
User's utterance: "I'm a bad person."
Dialogue agent's response: "That's not the case."

<Other second embodiment of dialogue agent>
There are cases where the emotional polarity determined by the simple sentence determining unit 13 is "only neutral simple sentences" or "only neutral and negative simple sentences".
In this case, the dialog agent 10 responds with a parroting response sentence regarding the final neutral simple sentence.
User's utterance: "I'm going to do some deep cleaning on my day off and then watch a video at home."
Emotional polarity of a simple sentence: “I will do some deep cleaning on my day off” -> Neutral
"I watch videos at home"-> Neutral Dialogue agent's response: "So I watch videos at home."

FIG. 5 is a functional configuration diagram with additional functions to the dialogue device according to the present invention.
FIG. 6 is an explanatory diagram showing the processing of the simple sentence determination unit.

According to FIG. 5, compared to FIG. 2, it includes a desire expression determination section 121, a behavioral expression determination section 122, a response history recording section 14, a noun history recording section 15, and a negativity analysis section 16.

[Desire expression determination unit 121]
The desire expression determining unit 121 determines whether or not a desire expression is present in each simple sentence divided by the simple sentence dividing unit 11 .
The desire expression determination unit 121 morphologically analyzes a simple sentence, for example, and determines whether it contains a desire expression such as the auxiliary verb "want" or the adjectives "want" or "want."

According to the uttered sentence of the first example, the following simple sentence is determined.
The utterance in the first example is, ``I have to do some cleaning on my day off, so I can't go to the movie theater.I can't go to the movies at home.''
I want to see it. ”
＜Single sentence＞ -> ＜Expression of desire＞
“I do a lot of cleaning on my day off” -> None “I can’t go to the movie theater” -> None “I want to watch videos at home” -> Yes (auxiliary verb “I want”)
According to the uttered sentence of the second example, the following simple sentence is determined.
Second example utterance: “I have plans to meet a friend and go to my favorite restaurant.I also want to watch a movie.
is. ”
＜Single sentence＞ -> ＜Expression of desire＞
“I have plans to meet a friend” -> None “I’m going to my favorite restaurant” -> None “I also want to see a movie” -> Yes (auxiliary verb “want”)
At this time, the simple sentence determining unit 13 determines one simple sentence in which the emotional polarity is positive and the expression of desire is present.

[Behavioral expression determination unit 122]
The behavioral expression determining unit 122 determines whether or not there is a behavioral expression for each simple sentence divided by the simple sentence dividing unit 11 .

According to the uttered sentence of the second example, the following simple sentence is determined.
Second example utterance: “I have plans to meet a friend and go to my favorite restaurant.I also want to watch a movie.
is. ”
<Single sentence>-><Behavioralexpression>
"I have plans to meet a friend"-> None "I'm going to my favorite restaurant"-> Yes (verb "go")
“I also want to watch a movie” -> Yes (verb “watch”)
At this time, the simple sentence determination unit 13 determines one simple sentence in which the emotional polarity is positive, the desire expression is present, and the action expression is present.

[Response history recording unit 14]
Regarding the response history recording section 14, there are the following three embodiments.
<First embodiment: Preventing responses to continuous parroting response sentences>
<Second embodiment: Preventing responses from parroting response sentences with the same content>
<Third embodiment: Reply that alludes to past parroting response sentences>

<First embodiment: Preventing responses to continuous parroting response sentences>
If the user continuously responds with parroting responses during a dialogue, the user may simply repeat the words included in the user's utterances, giving the impression that the user lacks initiative and is just annoying. As a result, it is expected that the possibility that the user will feel uncomfortable will increase.

FIG. 7 is an explanatory diagram showing a parrot-like response sentence in another embodiment.

According to FIG. 7(a), the conversation agent 10 interacts with the user as follows, for example.
User's utterance: ``I have plans to meet a friend and will go to my favorite store.
I also want to watch movies. ”
Dialogue agent's response: "You also want to see a movie."
User's utterance: "I want to see a new movie."
Dialogue agent's response: "So you want to see the new movie?"
A user :"···"

For this purpose, the response history recording unit 14 records whether the response sentence that the dialogue agent 10 has replied to the user is "a parrot-like response sentence or not."
Then, when attempting to reply to the user with a parroting response sentence, if the conversation agent 10 has responded with a parroting response sentence immediately before being recorded in the response history recording unit 14, the dialogue agent 10 returns the parroting response sentence to the user. reply. That is, immediately after replying with a parroting response sentence, the user is prevented from continuously replying with a parroting response. Then, respond with a conciliatory response such as, ``I see,'' or ``Yeah, yeah.'' Conversely, if the previous response recorded in the response history recording unit 14 is not a parroting response, a parroting response may be returned.

According to FIG. 7(b), the conversation agent 10 interacts with the user as follows, for example.
User's utterance: ``I have plans to meet a friend and will go to my favorite store.
I also want to watch movies. ”
Dialogue agent's response: "You also want to see a movie."
*Recorded "Parrot-like response included" in the response history recording section 14 User's utterance: "I want to see the new movie."
*It is determined that "Parrot-like response sentence included" was recorded immediately before Dialogue agent's response sentence: "I see."
*Record "No parroting response" in the response history recording section 14.

<Second embodiment: Preventing responses from parroting response sentences with the same content>
During a dialogue, if the user utters a sentence containing the same word many times in a predetermined period (for example, one hour), the dialogue agent 10 also quotes the same word many times and makes a parroting response sentence. I will reply. In that case as well, the user may be left with the impression that the user lacks independence and is just noisy. As a result, it is expected that the user will become bored with the dialogue with the dialogue agent.

To this end, the response history recording unit 14 records response sentences that the conversation agent has replied to the user.
At this time, the dialogue agent 10 compares the nouns and predicates included in the parroting response sentence that it attempts to reply to the user with the nouns and predicates included in the response sentence that was answered immediately before and that is recorded in the response history recording unit 14. If they are the same, reply with a conciliatory response.

For example, the interaction agent 10 interacts with the user as follows.
User's utterance: ``I have plans to meet a friend and will go to my favorite store.
I also want to watch movies. ”
Dialogue agent's response: "You also want to see a movie."
*Record the response sentence ``I also want to see a movie.'' in the response history recording section 14 User's utterance: ``I want to see a new movie.''
*Noun “movie” and predicate “watch” in the response sentence recorded in the response history recording unit 14
and the noun ``image'' in the response sentence that is being attempted as a parroting response sentence.
Determined to be the same as the predicate “picture” and “watch” Dialogue agent’s response: “I see.”
*Record the response sentence “I see” in the response history recording section 14.

Here, it is preferable that the response history recording section 14 maintains records only for a predetermined period (for example, one hour). That is, the record may be deleted, for example, one hour after the response sentence is recorded. In this case, the conversation agent 10 can respond with a parroting response, ``So you want to watch a movie,'' after one hour has elapsed, for example.

<Third embodiment: Reply that alludes to past parroting response sentences>
During a dialogue, if the other party remembers what the user has said in the past, the user's satisfaction with the dialogue can be increased. If the parroting-like response sentences that the dialogue agent 10 has responded to in the past and the parroting-like response sentences that it is currently trying to reply to have the same content, a statement that hints to the user that it remembers what the user has said in the past (recalling) It is also preferable to add a response sentence that causes For example, it is also preferable to respond with a response sentence, ``You said that before, right!''.

The response history recording unit 14 records the response text and response date and time that the conversation agent replied to the user.
At this time, if the noun and predicate included in the parroting response sentence that the dialog agent 10 attempts to reply to the user are the same as the noun and predicate included in the response sentence recorded in the response history recording unit 14, In addition to replying with the parroting-like response sentence, it also replies with "a response sentence that reminds us of the same parroting-like response sentence that was answered at the past response date and time of the response sentence."

It is assumed that the response history recording unit 14 records, for example, as follows.
July 7, 2020 15:45:10: “You want to watch a movie, right?”
On top of that, it is assumed that currently (July 31, 2020), the conversation agent 10 is attempting to respond with a parroting response sentence, "So you want to watch a movie."
Here, the noun "movie" and the predicate "I want to see" included in the parroting response sentence to be replied to the user, and the noun "movie" and the predicate "I want to see" included in the response sentence recorded in the response history recording unit 14. "I want to see it" are the same. At this time, reply with a response sentence as shown below.
Dialogue agent's response: "I want to see a movie, you said that before.
right"
``You said that before, right?'' is a response sentence that reminds the user of the same parroting response in the past. Such an additional response sentence can give the user the illusion that the user has remembered the message.

It is also preferable to provide diversity to the additional response sentences depending on the response date and time recorded in the response history recording unit 14 from the current date and time.
If a response sentence with the same content as the parroting response sentence to which the user is attempting to respond was recorded yesterday in the response history recording unit 14, the response sentence ``You said that yesterday too, right?'' can be replied.
Also, if the information was recorded 2 days to 1 week ago, you can respond with the response, ``I remember saying that the other day.''
Furthermore, if the information was recorded on the previous Wednesday, it is also possible to reply with the response sentence, ``If I remember correctly, you said that on Wednesday as well.''

[Noun history recording unit 15]
The noun history recording unit 15 records for a predetermined period the "nouns" included in the response sentences sent by the dialogue agent 10 to the user.
The noun history recording unit 15 deletes nouns for which a predetermined period (for example, 300 seconds) has elapsed since the time of recording.
If the noun included in the parroting response sentence that is intended to reply to the user is recorded in the noun history recording unit 15, the dialog agent 10 responds with a parroting response sentence. In other words, a parrot-like response sentence that includes a noun that has been quoted within a predetermined period is not responded to.

FIG. 8 is an explanatory diagram showing a parroting response sentence using the noun history recording unit.

(Time t) The nouns "movie,""park,""restaurant," and "game" are recorded in the noun history recording unit 15 based on the past parroting response sentences. Further, the elapsed time is recorded for each noun, and the elapsed time is incremented by, for example, one second.
At this point, the interaction agent 10 attempts to interact with the user, for example as follows.
User's utterance: "I want to do some cleaning and play some games on my day off."
Emotional polarity of a simple sentence: “I will do some deep cleaning on my day off” -> Neutral
"I want to play games."-> Positive Dialogue agent's response: "You want to play games."
Here, the noun "game" in the parroting response sentence "I want to play a game, right?" is recorded in the noun history recording section 15. To this end, only respond with a conciliatory response, as shown below.
Dialogue agent's response: "I see."

(Time t+40 seconds)
Forty seconds after time t, the elapsed time for each noun in the noun history recording unit 15 is also incremented by 40 seconds. Here, the noun "game" is deleted because the predetermined period of 300 seconds has been reached.
At this point, the interaction agent 10 interacts with the user, for example, as follows.
User's utterance: "I want to do some cleaning and play some games on my day off."
Emotional polarity of a simple sentence: “I will do some deep cleaning on my day off” -> Neutral
"I want to play games."-> Positive Dialogue agent's response: "You want to play games."

[Negativity analysis section 16]
The negativity analysis unit 16 analyzes the negativity of a predetermined condition for a simple sentence whose emotional polarity is negative. If the negativity satisfies a predetermined condition, the application is notified to that effect.
Here, the "negativity" in a user's utterance is determined to be a highly dangerous utterance, such as strong self-denial, suicidal thoughts, or racist thoughts. This can be determined by, for example, using a machine learning model such as a support vector machine to previously learn highly negative utterances and less negative utterances as training data.
When the dialogue agent 10 determines that the utterance is highly negative, it responds with a scenario-like response sentence that strongly denies it, such as "I don't think so."

As described in detail above, according to the program, device, and method of the present invention, it is possible to prevent a negative dialogue from occurring due to a parrot-like response to the user. In particular, dialogue can be promoted through positive, parroting responses.

Regarding the various embodiments of the present invention described above, various changes, modifications, and omissions within the scope of the technical idea and viewpoint of the present invention can be easily made by those skilled in the art. The above description is merely an example and is not intended to be limiting in any way. The invention is limited only by the claims and their equivalents.

1 Dialogue device 10 Dialogue agent 11 Simple sentence segmentation unit 120 Emotion polarity determination unit 121 Desire expression determination unit 122 Behavioral expression determination unit 13 Simple sentence determination unit 14 Response history recording unit 15 Noun history recording unit 16 Negativity analysis unit 2 Terminal

Claims (11)

A program that causes a computer to function as a dialogue agent that responds with parroting response sentences while replying to a user with scenario-like response sentences in accordance with a dialogue scenario, the program comprising:
a simple sentence dividing means for dividing an uttered sentence from a user into a plurality of simple sentences;
an emotion polarity determination means for determining whether or not the emotion polarity is positive for each simple sentence;
Functions as a simple sentence determining means to determine one simple sentence with positive emotional polarity,
A dialogue agent is a program that causes a computer to respond to a determined simple sentence with a parrot-like response sentence. It further functions as a desire expression determination means for determining the presence or absence of a desire expression for each simple sentence,
2. The program according to claim 1, wherein the simple sentence determining means causes the computer to function to determine one simple sentence that has positive emotional polarity and expresses a desire. It further functions as a behavioral expression determining means for determining the presence or absence of a behavioral expression for each simple sentence,
3. The simple sentence determining means causes the computer to function to determine one simple sentence in which the emotional polarity is positive, the expression of desire is present, and the expression of action is present. program. further functioning as a response history recording means for recording whether or not the response sentence that the dialogue agent replied to the user is a parroting response sentence;
When the dialogue agent attempts to reply to the user with a parroting response sentence, if the conversation agent has responded with a parroting response sentence immediately before being recorded in the response history recording means, the conversation agent is configured to respond with a parroting response sentence. 4. The program according to claim 1, which causes a computer to function. It further functions as a response history recording means for recording response sentences that the dialogue agent replies to the user,
If the noun and predicate included in the parroting response sentence that the dialog agent attempts to reply to the user are the same as the noun and predicate included in the response sentence that was answered immediately before and recorded in the response history recording means, The program according to any one of claims 1 to 3, which causes a computer to function so as to respond with a response sentence. It further functions as a response history recording means for recording the response text and response date and time that the dialog agent replied to the user,
If the nouns and predicates included in the parroting-like response sentence that it attempts to reply to the user are the same as the nouns and predicates included in the response sentence recorded in the response history recording means, the dialogue agent Claims 1 to 3, characterized in that the computer is made to function so as to reply with a response sentence and also to reply with a response sentence that reminds the user of a past response date and time of the same response sentence. The program described in any one of the above. further functions as a noun history recording means for recording nouns included in response sentences sent by the dialogue agent to the user for a predetermined period;
The dialogue agent is characterized in that, when a noun included in a parroting response sentence that is attempted to reply to the user is recorded in the noun history recording means, it causes the computer to function to reply with a parroting response sentence. The program according to any one of claims 1 to 3. If there are only simple sentences whose emotional polarity is negative according to the simple sentence determination method,
The program according to any one of claims 1 to 3, characterized in that the dialogue agent functions to respond with a scenario response sentence in which the emotional polarity is positive, without responding with a mutual response sentence. . If there are only simple sentences whose emotional polarity is neutral according to the simple sentence determination means, or if there are only neutral and negative simple sentences,
4. The program according to claim 1, wherein the dialogue agent functions to respond with a parroting response sentence regarding the last simple sentence that is neutral. A dialogue device having a dialogue agent that responds with a parrot-like response sentence while replying to a user with a scenario-like response sentence in accordance with a dialogue scenario,
a simple sentence dividing means for dividing an uttered sentence from a user into a plurality of simple sentences;
an emotion polarity determination means for determining whether or not the emotion polarity is positive for each simple sentence;
a simple sentence determining means for determining one simple sentence with positive emotional polarity,
A dialogue device characterized in that the dialogue agent replies with a parrot-like response sentence to the determined simple sentence. A dialogue method for a device having a dialogue agent that responds with a parroting response sentence while replying to a user with a scenario response sentence in accordance with a dialogue scenario,
The device is
a first step of dividing the utterance from the user into a plurality of simple sentences;
a second step of determining whether the emotional polarity is positive or not for each simple sentence;
a third step of determining one simple sentence with positive emotional polarity;
A dialogue method for a device, characterized in that the dialogue agent replies with a parrot-like response sentence to the determined simple sentence.

Images