Apparatus and method for automatically learning user's utterance intention by social robot

The learning device and method enhance conversational systems by adaptively correcting identified user intentions in real-time, improving interaction accuracy and user satisfaction through adaptive learning.

KR103047771B1Active Publication Date: 2026-11-11KOREA INST OF ROBOT & CONVERGENCE
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
KR1020230090075
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2023-07-11
Publication Date
2026-11-11
Estimated Expiration
2043-07-11

AI Technical Summary

Technical Problem

Existing conversational systems struggle to accurately identify user speech intentions, leading to incorrect responses and dissatisfaction in interactions with robots.

Method used

A learning device and method that includes an acquisition unit to gather user responses, a determination unit to assess the correctness of identified intentions, and a processing unit to adaptively learn and correct responses based on additional user utterances.

Benefits of technology

Enables real-time identification and correction of incorrect user intentions, improving user satisfaction by ensuring accurate and adaptive responses, and providing training data for future machine learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112023076395458-PAT00001_ABST
    Figure 112023076395458-PAT00001_ABST
Patent Text Reader

Abstract

A learning device is provided. The learning device of the present invention may include: an acquisition unit that acquires a response from a user after a first action of a robot corresponding to a first utterance of a user; a determination unit that determines whether a first intention corresponding to the intention of the first utterance identified by the robot is an incorrect answer according to the response of the user; and a processing unit that reflects or learns the result of determining whether the first intention is an incorrect answer in a second action of the robot.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to a learning device and method for a social robot. Background Technology

[0003] Recently, the content industry for infants and toddlers utilizing humanoid companion robots has been growing, and user demand for such products is increasing. Notably, there are cases where various companies are developing content that allows elderly people living alone and young children to interact with speaker-type companion robots for home use, and are launching these products.

[0004] As a key foundational technology to realize these examples, it is necessary to first establish technology that listens to the user's utterance and identifies the intent behind it, in order to enable natural and continuous conversation with the user.

[0005] Korean Registered Patent Publication No. 2447513 discloses a technology that improves the performance of a conversational system in real time by automatically recognizing when the conversational system fails to understand a user's utterance and adding it as a new learning sentence. Prior art literature

[0007] Korean Registered Patent Publication No. 2447513 The problem to be solved

[0008] The present invention is intended to provide a learning device and method for automatically learning a user's speech intention. means of solving the problem

[0010] The learning device of the present invention may include: an acquisition unit that acquires a response from a user after a first action of a robot corresponding to a first utterance of a user; a determination unit that determines whether a first intention corresponding to the intention of the first utterance identified by the robot is incorrect according to the response of the user; and a processing unit that reflects or learns the result of determining whether the first intention is incorrect in a second action of the robot.

[0011] The learning method of the present invention may include: an acquisition step of acquiring a response from a user after a first action of a robot corresponding to a first utterance of a user; a determination step of determining whether a first intention corresponding to the intention of the first utterance identified by the robot is incorrect according to the response of the user; and a processing step of reflecting or learning the result of determining whether the first intention is incorrect in a second action of the robot.

[0012] At this time, the acquisition step may acquire the user's second utterance after the robot's first response corresponding to the first utterance.

[0013] The above determination step can determine whether the first intention, which formed the basis of the first answer, is incorrect through the analysis of the second utterance.

[0014] If the above processing step determines that the above first intention is incorrect, it can generate a question inquiring about the intention of the above first utterance.

[0015] A question inquiring about the intent of the first utterance may be included in the robot's second response to the second utterance.

[0016] The above acquisition step can acquire the user's third utterance after the above second answer.

[0017] When the third utterance is obtained, the processing step above can identify the second intention corresponding to the intention of the first utterance based on the content of the third utterance.

[0018] The above processing step can generate a question to verify whether the identified second intention is correct.

[0019] A question verifying whether the above second intention is correct can be applied in real time to the robot's third response to the above third utterance. Effects of the invention

[0021] The present invention can propose an algorithm that listens to a user's utterance and identifies the intent of the utterance, an algorithm that can derive the user's utterance intent that is not correctly identified by existing algorithms, and an algorithm that can automatically learn anew by applying the user's utterance intent to an existing learning model.

[0022] According to the present invention, the intent of a user's first utterance can be automatically identified in the first instance. Subsequently, whether the intent of the utterance identified in the first instance is incorrect can be checked through additional utterances by the user. Furthermore, if the intent of the utterance identified in the first instance is determined to be incorrect, a question inquiring directly about the intent of the utterance can be performed by a robot in the second instance. The content of the user's utterance newly obtained through the question can be identified as the intent of the utterance in the second instance. The intent identified in the second instance regarding the user's first utterance can be used as training data for machine learning together with the first utterance.

[0023] According to the learning device and method of the present invention, a new, unregistered second intention regarding a user's first utterance can be directly obtained through the user and machine learned. In particular, according to the present invention, since the conversation of a robot can be continued using the newly identified second intention while conversing with the user, the user's satisfaction with the conversational robot can be significantly improved. Brief explanation of the drawing

[0025] Figure 1 is a schematic diagram showing a learning device. Figure 2 is a schematic diagram showing the operation of a learning device. Figure 3 is a flowchart illustrating the retraining process of the model of understanding. Figure 4 is a flowchart illustrating the learning method of the present invention. FIG. 5 is a drawing showing a computing device according to an embodiment of the present invention. Specific details for implementing the invention

[0026] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0027] In this specification, redundant descriptions of identical components are omitted.

[0028] Furthermore, when a component is described in this specification as being 'connected' or 'connected' to another component, it should be understood that it may be directly connected to or connected to the other component, or that there may be other components in between. On the other hand, when a component is described in this specification as being 'directly connected' or 'directly connected' to another component, it should be understood that there are no other components in between.

[0029] Furthermore, the terms used in this specification are used merely to describe specific embodiments and are not intended to limit the invention.

[0030] Additionally, in this specification, singular expressions may include plural expressions unless the context clearly indicates otherwise.

[0031] Furthermore, in this specification, terms such as 'comprising' or 'having' are intended merely to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not excluding in advance the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0032] Additionally, in this specification, the term "and / or" includes a combination of the plurality of described items or any of the plurality of described items. In this specification, "A or B" may include "A," "B," or "both A and B."

[0033] In addition, detailed descriptions of known functions and configurations that may obscure the essence of the invention will be omitted in this specification.

[0034] The learning device (100) of the present invention may be installed on a robot (200) or on a controller that controls the robot (200). Alternatively, the learning device (100) may be installed on a server that communicates with the robot (200).

[0035] For interaction through voice recognition, the robot (200) may include a collection unit (210), a first conversion unit (230), an extraction unit (250), an answer unit (270), a second conversion unit (280), and an output unit (290). In some cases, a learning device (100) may be added to the robot (200).

[0036] The collection unit (210) can collect the user's voice and the user's speech (data). The collection unit (210) may include a microphone.

[0037] The first conversion unit (230) can convert the user's speech corresponding to voice data into a text format.

[0038] The extraction unit (250) can analyze the user's utterance converted into text format and extract the user's intention through the analysis.

[0039] The answer unit (270) can generate an answer corresponding to the user's intent extracted by the extraction unit (250) or extract it from a database.

[0040] The second conversion unit (280) can convert the text-formatted answer content generated or extracted through the answer unit (270) into a voice format.

[0041] The output unit (290) can output the answer content converted into a voice format to the user. The output unit (290) may include a speaker.

[0042] The learning device (100) can be used to improve the performance of speech recognition. The performance of speech recognition can be determined by whether the user's speech intent, that is, the intention the user intends to express through speech, can be accurately grasped.

[0043] The intention that the user intends to express through the first utterance, that is, the intention of the first utterance, may vary depending on the situation, even if the content of the utterance is the same.

[0044] For example, if a user says "it hurts," the place of pain could be the mind or a part of the body. Whether the body or the mind is hurting may vary depending on the user's language habits or the situation they are currently in.

[0045] Therefore, it is difficult to uniformly determine the intent of the utterance content for all users, and it is better to determine it adaptively for each user. However, at the time of product shipment or at the initial stage when the user's habits cannot be determined, the robot (200) needs to determine whether the user is mentally ill or physically ill as the intent of the utterance and perform a response accordingly. Subsequently, if the response to the user is determined to be incorrect (wrong answer), a process of correcting the response needs to be performed. A learning device (100) may be used to correct such wrong answers.

[0046] FIG. 1 is a schematic diagram showing a learning device (100).

[0047] The learning device (100) illustrated in FIG. 1 may include an acquisition unit (110), a discrimination unit (150), and a processing unit (170).

[0048] The acquisition unit (110) can acquire the user's response after the first action of the robot (200) corresponding to the user's first utterance. When the learning device (100) is mounted on the robot (200), the acquisition unit (110) can be formed integrally with the collection unit (210).

[0049] In response to a user's first utterance, the robot (200) can respond to and interact with the user through actions such as voice or gesture. At this time, the robot (200) can primarily take a primary action set by default in response to the first utterance.

[0050] If the first action is the correct answer to the first utterance, the user will perform the next utterance that follows the first utterance. However, if the first action is not the answer the user wanted, but something else, that is, an incorrect answer that misunderstood the intent of the first utterance, the user may not be able to continue the conversation and may show an unpleasant reaction to the incorrect answer. At this time, in order to relieve the user's unpleasantness and to allow the robot (200) to continue interacting with the user, the discrimination unit (150) can determine whether the first action is an error.

[0051] For example, the discrimination unit (150) can determine whether the first intention initially identified by the robot (200) is incorrect based on the user's response. At this time, the first intention may correspond to the intention of the first utterance. If the first intention is incorrectly identified, the action or first answer of the robot (200) generated based on the first intention will also inevitably be incorrect.

[0052] The determination unit (150) can determine whether the first intention is incorrect through a second utterance corresponding to the user's additional utterance regarding the first answer and the action of the robot (200). For example, if the first intention is incorrect, the robot's (200) first answer to the first utterance is bound to be nonsensical from the user's perspective. A user who hears a nonsensical answer may perform a counter-utterance such as "What did you say?" In this way, the incorrectness of the first intention can be determined through the analysis of the user's additional utterance (second utterance) regarding the first answer.

[0053] Alternatively, the discrimination unit (150) may photograph the user's face and analyze facial expressions through the analysis of the captured image. The discrimination unit (150) may determine whether there is an error in the action of the robot (200) or whether the first intention is incorrect through facial expressions.

[0054] The processing unit (170) can reflect the result of determining whether the first intention is incorrect in the second action of the robot (200) or learn it.

[0055] For example, the processing unit (170) can reflect the result of the first intention's incorrect answer determination in the answer immediately following the user's second utterance used for the first intention's incorrect answer determination (second answer).

[0056] Alternatively, the processing unit (170) can machine learn that at least the first intention is not the correct answer. When the learning device (100) is mounted on the robot (200) and the processing unit (170) generates a second answer, the processing unit (170) may be formed integrally with the answer unit (270).

[0057] FIG. 2 is a schematic diagram showing the operation of a learning device (100).

[0058] The operation of the learning device (100) may require a user's response, such as a second utterance by the user. Therefore, the operation of the learning device (100) requires a preceding operation of a robot (200) capable of eliciting a user's response.

[0059] As shown in FIG. 2, the robot (200) can convert a user's first utterance into a text format that can be processed by the learning device (100) using a first converter such as STT (Speech To Text) (S 10).

[0060] For example, the user's first utterance may be 'I feel heartbroken,' and the intention of the first utterance may be 'I am sad.' In this case, the robot (200) may only be able to receive the content of the utterance 'I feel heartbroken,' and the intention of the first utterance 'I am sad' may not be received from the user.

[0061] The extraction unit (250) of the robot (200) can identify and extract the intention of the first utterance as ‘sad’ (correct response) or identify and extract it as ‘part of the body hurts’ (incorrect response) (S 20) by using only ‘heart hurts’.

[0062] Accordingly, the answer unit (270) of the robot (200) can perform a first answer to a correct answer such as 'My heart hurts too' (S 31). Alternatively, the answer unit (270) of the robot (200) can perform a first answer to an incorrect answer such as 'Do you want to go to the hospital?' (S 32).

[0063] Subsequently, the user's second utterance regarding the first response can be obtained through the acquisition unit (110). Specifically, the acquisition unit (110) can obtain the user's second utterance after the robot (200)'s first response corresponding to the first utterance. The acquisition unit (110) can obtain the second utterance during the conversation between the robot (200) and the user.

[0064] The discrimination unit (150) can determine whether the first intention identified by the robot (200) is incorrect based on user reactions such as a second utterance. In other words, the discrimination unit (150) can determine whether the first intention that formed the basis of the first answer is incorrect through the analysis of the second utterance. For example, if the discrimination unit (150) derives a user reaction to the correct intention, it can determine that the first intention is correct (reaction identification) (S 40). Afterward, the robot (200) can proceed with a preset next first scenario of interaction (S 51). If the first intention is correct, the robot (200) can operate similarly to a state where the learning device (100) is excluded.

[0065] The subsequent process of acquiring a second utterance and determining whether the first intention is incorrect may correspond to the operation of the learning device (100) (S 500).

[0066] The discrimination unit (150) can determine the first intention as an incorrect answer when the user’s response to the incorrect intention, for example, when the user’s second utterance asks back the first answer, such as ‘What?’ or ‘What do you mean?’ (response identification) (S 60).

[0067] If the first intention is determined to be incorrect, the processing unit (170) can generate a question asking for the correct intention of the first utterance, for example, 'Then, what does 'heart hurts' mean?'. The question generated by the processing unit (170) can be included in the second answer of the robot (200) to the second utterance (inducing a correct response) (S 70).

[0068] The acquisition unit (110) can acquire the user's third utterance after the second response (S 80). For example, 'It hurts my heart' can be the third utterance.

[0069] When the processing unit (170) obtains the third utterance, it can identify the second intention, 'sad,' which corresponds to the correct intention of the first utterance, based on the content of the third utterance. Afterward, the robot (200) can execute a pre-set second scenario of interaction (S 52). The second scenario may include an utterance based on the second intention identified by the user through the third utterance.

[0070] It may be difficult for the robot (200) to extract a second intention for a first utterance in a state where it has not been previously learned. However, according to the learning device (100), a second intention corresponding to the correct intention of the first utterance can be identified in real time on-site through a second answer that directly asks the user a question. Accordingly, according to the learning device (100) of the present invention, user-adaptive intention identification is identified in real time on-site during conversation, and the identification result can be used as training data for future machine learning.

[0071] The second intention identified by the processing unit (170) can be applied in real time to the robot (200)'s third response (which may be included in the second scenario) to the third utterance. For example, the third response may include asking the cause of the second intention, suggesting a way to resolve the second intention, or presenting an empathetic response. For example, the third response may include content such as "Why are you sad?", "Shall we play a game to forget the sadness?", or "My heart hurts too."

[0072] The processing unit (170) can modify the existing first intention matched to the first utterance into a second intention adaptedly identified by the user, and train a machine learning model 'ML model' using the modified training data (S 90). At this time, the training data may include a data set in which the first utterance is the problem and the second intention is the correct answer.

[0073] The processing unit (170) can decompose the text string of the user's third utterance into morpheme units and extract a string excluding words without feature points, such as particles. Then, the processing unit (170) can learn the string excluding words without feature points together with the utterance intention tag.

[0074] For example, if the first utterance is 'Nice to meet you,' the first utterance can be converted into a text string 'User said(string)' through the first conversion unit (230). The extraction unit (250) can extract the intention 'Greeting' through the analysis of the string. Depending on the extracted intention, the response unit (270) can generate 'Nice to meet you too!' and simultaneously take an additional gesture 'Motion.greeting' to express greeting. The 'Nice to meet you too!' generated by the response unit (270) is converted into a voice, speech format 'tts.wav' through the second conversion unit (280), and the voice 'Nice to meet you too!' is output through the robot (200), and the motion or gesture 'Waving hand happily' can be output.

[0075] Afterward, regarding the second utterance of a user for which the correct intention was not obtained by the learned model, the processing unit (170) can identify the correct second intention for the first utterance through an algorithm that directly inquires about the correct intention. The second intention identified directly through the user is added to the utterance intention database (DB) and can be newly applied as training data for machine learning.

[0076] Each element of the speech intent database can be composed of 'Tag', 'Patterns', and 'Responses'.

[0077] 'Tag' can be set as the name of the utterance intention, such as 'Greeting', 'Ready', or 'Goodbye', representing the user's intention implied by the utterance.

[0078] 'Patterns' are the content of utterances from which the user's intent can be extracted; in the case of the 'Greeting' Tag, they can be set to the content of general utterances corresponding to the Tag, such as 'hi', 'hello', or 'nice to meet you'.

[0079] 'Responses' are the robot's (200) responses to the user's speech, and in the case of the 'Greeting' Tag, they can be set as 'hi, I'm XX', 'hello, call me XX', 'nice to meet you too'.

[0080] If the user performs a second utterance indicating that the first intention is incorrect, it may mean that the prediction accuracy for all Tags defined in the utterance intention database is lower than the set value. In this case, the processing unit (170) can directly obtain the correct second intention of the first utterance through an algorithm that asks for the intention of the first utterance in order to correct the classification error of the Tag defined in the utterance intention database.

[0081] When the first intention is incorrect, the processing unit (170) can create a 'misunderstand' Tag in the existing user speech intention database to obtain a second intention directly from the user, and learn an initial machine learning model (understanding model) by inputting expected patterns and expected responses that correspond to the tag.

[0082] The processing unit (170) can configure the correct second intention tag (Tag) and the first utterance content directly presented by the user through the third utterance to fit the format of the user utterance intention database, and then add, modify, and save them to the existing user utterance intention database.

[0083] The processing unit (170) can interact with the user again through a new recognition model after training a machine learning model on the modified user speech intent database.

[0084] According to the above description, the processing unit (170) can learn the intention of the first utterance as a second intention based on the third utterance. The identification model machine-learned by the processing unit (170) can be installed in the robot (200). The identification model can provide the robot (200) with a second intention, rather than the first intention, for the re-acquired first utterance. However, in some cases, if "my heart hurts" is based on pain in a specific body part, the "my heart hurts" may mean "pain in a specific body part." In this case, the first intention identified in the initial stage may correspond to the correct intention of the first utterance. Nevertheless, the robot (200) may perform an answer based on the second intention through prior learning. In this case, if the answer is determined to be incorrect, it may be desirable to present an answer based on the first intention instead of the second answer asking for the user's intention.

[0085] Therefore, if the identification model determines that the second intention is incorrect through the user's utterance regarding the robot's (200) answer based on the second intention, it can immediately provide the first intention to the robot (200). Upon receiving the first intention, the robot (200) can perform an answer based on the first intention instead of asking a question inquiring about the intention of the first utterance.

[0086] The learning device (100) may present a response (third response) that matches the newly extracted intent to the same user utterance (first utterance) as before to check whether the incorrectly identified utterance intent has been correctly reflected, and may resume the development of the scenario. If the third response corresponds to the correct intent, the operation after the third response may follow the S 40 and S 50 routes of FIG. 2.

[0087] Figure 3 is a flowchart illustrating the retraining process of the model of understanding.

[0088] The identification model learned by the processing unit (170) can be formed as a fully connected neural network consisting of 8 nodes and 2 hidden layers, and can be formed with a simple structure compared to other machine learning models.

[0089] First, the extraction unit can extract the user's conversational intent by creating an understanding model or using a pre-generated understanding model (s 21) (S 20).

[0090] The discrimination unit identifies the user's response and, depending on the user's response, can determine whether the previously extracted intention is correct or incorrect (S 40, S 60).

[0091] If it is determined to be an incorrect response, the processing unit (170) can generate a question to obtain the correct intention from the user and convert the correct intention obtained from the user into a DB format of the user's utterance (S 91). The processing unit (170) can store and register the correct intention converted into a DB format along with the user's first utterance in the intention extraction DB (S 92). The processing unit (170) can retrain a learning model, that is, a grasping model, using the first utterance stored in the intention extraction DB and the correct intention obtained from the user (S 93).

[0092] Meanwhile, in order to process gestures among user actions, the processing unit (170) can distinguish the robot (200)'s response to the user's speech into multiple categories.

[0093] The processing unit (170) can provide the robot (200) with a gesture to be taken according to each answer.

[0094] The processing unit (170) can set multiple basic gestures that can be taken for a new answer in preparation for the case where the robot (200) performs a new answer different from the previously registered answer.

[0095] The processing unit (170) may additionally provide suggestion information to the robot (200) to induce the robot (200) to randomly perform one of a plurality of basic gestures when performing a new answer.

[0096] When the robot (200) performs a non-existent answer, such as a third answer based on the user's third utterance, the answer may not have a specific gesture set. According to the present embodiment, additional suggestion information is provided to the robot (200), and the robot (200) can use the suggestion information to randomly perform one of a plurality of basic gestures for the non-existent answer. Accordingly, the interaction of the robot (200) with the user is improved, and a positive effect on forming a bond with the robot (200) can be expected.

[0097] Figure 4 is a flowchart illustrating the learning method of the present invention.

[0098] The learning method of Fig. 4 can be performed by the learning device (100) illustrated in Fig. 1.

[0099] The learning method may include an acquisition step (S 510), a discrimination step (S 520), and a processing step (S 530).

[0100] The acquisition step (S 510) can acquire the user's response after the first action of the robot (200) corresponding to the user's first utterance. The acquisition step (S 510) can be performed by the acquisition unit (110).

[0101] The acquisition step (S 510) can acquire the user's second utterance after the first response of the robot (200) corresponding to the first utterance.

[0102] The acquisition step (S 510) can acquire the user's third utterance after the robot's (200) second response.

[0103] The determination step (S 520) can determine whether the first intention corresponding to the intention of the first utterance identified by the robot (200) is incorrect, depending on the user's response. The determination step (S 520) can be performed by the determination unit (150).

[0104] The determination step (S 520) can determine whether the first intention that formed the basis of the first answer is incorrect through the analysis of the second utterance.

[0105] The processing step (S 530) can reflect the result of determining whether the first intention is incorrect in the second action of the robot (200) or learn it. The processing step (S 530) can be performed by the processing unit (170).

[0106] If the first intention is determined to be incorrect in the processing step (S 530), a question inquiring about the intention of the first utterance may be generated. At this time, the question inquiring about the intention of the first utterance may be included in the second answer of the robot (200) to the second utterance.

[0107] When the third utterance is obtained, the processing step (S 530) can identify the second intention corresponding to the intention of the first utterance based on the content of the third utterance.

[0108] The processing step (S 530) can generate a question to verify whether the second intention is correct. At this time, the question to verify whether the second intention is correct can be applied in real time to the third answer of the robot (200) to the third utterance.

[0109] FIG. 5 is a drawing showing a computing device according to an embodiment of the present invention. The computing device (TN100) of FIG. 5 may be a device described in the present specification (e.g., a learning device (100), etc.).

[0110] In the embodiment of FIG. 6, the computing device (TN100) may include at least one processor (TN110), a transceiver (TN120), and a memory (TN130). Additionally, the computing device (TN100) may further include a storage device (TN140), an input interface device (TN150), an output interface device (TN160), etc. The components included in the computing device (TN100) may be connected by a bus (TN170) to communicate with each other.

[0111] The processor (TN110) can execute a program command stored in at least one of the memory (TN130) and the storage device (TN140). The processor (TN110) may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. The processor (TN110) may be configured to implement the procedures, functions, and methods described in relation to embodiments of the present invention. The processor (TN110) can control each component of the computing device (TN100).

[0112] Each of the memory (TN130) and the storage device (TN140) can store various information related to the operation of the processor (TN110). Each of the memory (TN130) and the storage device (TN140) may be composed of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory (TN130) may be composed of at least one of read-only memory (ROM) and random access memory (RAM).

[0113] The transmitting and receiving device (TN120) can transmit or receive wired or wireless signals. The transmitting and receiving device (TN120) can be connected to a network to perform communication.

[0114] Meanwhile, embodiments of the present invention are not limited to being implemented only through the apparatus and / or methods described so far, but may also be implemented through a program that realizes a function corresponding to the configuration of the embodiments of the present invention or a recording medium on which such program is recorded, and such implementation can be easily achieved by a person skilled in the art to which the present invention belongs based on the description of the embodiments described above.

[0115] Although embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by a person skilled in the art using the basic concept of the present invention as defined in the following claims also fall within the scope of the present invention. Explanation of the symbols

[0117] 100...learning device 110...acquisition unit 150...Discrimination Unit 170...Processing Unit 200...Robot 210...Collection Department 230...1st conversion unit 250...extraction unit 270...Answer section 280...Second conversion section 290...Output section

Claims

Claim 1 The apparatus comprises: an acquisition unit that acquires the user's response after a robot's first action corresponding to the user's first utterance; a determination unit that determines whether a first intention corresponding to the intention of the first utterance, which the robot has primarily identified, is an incorrect answer according to the user's response; and a processing unit that reflects the result of determining whether the first intention is incorrect in the robot's second action. The acquisition unit acquires the user's second utterance after the robot's first response corresponding to the first utterance, and the determination unit determines whether the first intention that formed the basis of the first response is incorrect through the analysis of the second utterance. If the first intention is determined to be incorrect, the processing unit generates a question inquiring about the intention of the first utterance, and the question is included in the robot's second response to the second utterance. The acquisition unit acquires the user's third utterance after the second response, and if the third utterance is acquired, the processing unit generates a second corresponding to the intention of the first utterance based on the content of the third utterance. A learning device that secondarily identifies an intention, wherein the second intention identified by the processing unit is applied in real time to the robot's third response to the third utterance, wherein the processing unit learns the intention of the first utterance as the second intention, wherein the identification model machine-learned by the processing unit is loaded onto the robot, wherein the identification model provides the robot with the second intention, rather than the first intention, for the re-acquired first utterance, and wherein if the second intention is determined to be an incorrect answer through the user's utterance regarding the robot's response based on the second intention, the identification model immediately provides the first intention to the robot, and wherein the robot, having received the first intention, performs an answer based on the first intention instead of a question inquiring about the intention of the first utterance. Claim 2 delete Claim 3 delete Claim 4 A learning device according to claim 1, wherein the processing unit classifies the robot's response to the user's utterance into multiple categories, the processing unit provides the robot with a gesture to be taken according to each of the responses, the processing unit sets multiple basic gestures that can be taken for a new response in preparation for the case where the robot performs a new response different from a previously registered response, and the processing unit additionally provides the robot with suggestion information that induces the robot to randomly perform one of the multiple basic gestures when performing a new response. Claim 5 A learning method performed by a learning device comprises: an acquisition step of acquiring a user's response after a first action of a robot corresponding to a user's first utterance; a determination step of determining whether a first intention corresponding to the intention of the first utterance identified by the robot is an incorrect answer according to the user's response; and a processing step of reflecting or learning the result of determining whether the first intention is an incorrect answer in a second action of the robot.The method includes, wherein the acquisition step acquires the user's second utterance after the robot's first response corresponding to the first utterance, the discrimination step determines whether the first intention forming the basis of the first response is incorrect through the analysis of the second utterance, the processing step generates a question inquiring about the intention of the first utterance if the first intention is determined to be incorrect, and the question inquiring about the intention of the first utterance is included in the robot's second response to the second utterance, the acquisition step acquires the user's third utterance after the second response, the processing step secondarily identifies a second intention corresponding to the intention of the first utterance based on the content of the third utterance if the third utterance is acquired, the processing step generates a question verifying whether the identified second intention is correct, and the question verifying whether the second intention is correct is applied in real time to the robot's third response to the third utterance, the processing step learns the intention of the first utterance as the second intention, and the machine learned in the processing step A learning method in which an understanding model is mounted on the robot, the understanding model is configured to provide the robot with the second intention, rather than the first intention, regarding the first utterance that is re-acquired, and if the second intention is determined to be incorrect through the user's utterance regarding the robot's answer based on the second intention, the understanding model is configured to immediately provide the first intention to the robot, and the robot, having received the first intention, performs an answer based on the first intention instead of a question inquiring about the intention of the first utterance.

Citation Information

Patent Citations

  • Synchronization of humanoid robot gestures and voice

    JP2014504959A

  • Apparatus and method for processing user''s locution, and dialog management apparatus

    KR1020170033722A

  • Dialogue system and method for controlling the same

    KR1020210130465A