JP2019061050A - Interaction apparatus, interaction method, and program - Google Patents

Abstract

To improve a technology for bringing a sense of affinity to a user.SOLUTION: An interaction apparatus 100 includes a language analysis section 112, a history storage section 113, a system response sentence generation section 114, a system response sentence output section 115. The language analysis section 112 analyzes a user speech sentence to acquire user expression characteristics. The history storage section 113 stores the user speech sentence generated by a voice recognition section 111 and a history of user expression characteristics acquired by the language analysis section 112, as a user speech sentence history 121. The system response sentence generation section 114 generates a system response sentence on the basis of the user speech sentence history 121 stored by the history storage section 113. The system response sentence output section 115 outputs the system response sentence generated by the system response sentence generation section 114.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a technology for generating a response sentence when a robot or the like interacts with a user.

  Development of terminals and robots capable of interacting with users is in progress. And, when these terminals and robots interact with the user, technological development is also in progress to enable smooth interaction processing without getting tired of the user. For example, Patent Document 1 describes a dialog system capable of performing dialog processing that does not make the user bored by changing the topic, the word, and the like when the system generates a response sentence.

JP 2001-125900 A

  The dialog system described in Patent Document 1 provides a user-friendly dialog by optimizing the topic of the conversation at the time of response by using a word or the like frequently used by the user as an evaluation index. ing. However, since the sentence pattern to which the dialogue system responds is limited to one stored in advance in the pattern database, although the topic can be changed for each user, the sentence expression of the response sentence does not change. Therefore, there is a high possibility that the sentence expression will be decided, and it is considered difficult for the user to feel a sense of closeness. As described above, in the conventional dialog device, there is room for improvement in the technology for making the user feel familiar.

  The present invention has been made in view of the above-described circumstances, and an object thereof is to improve a technique for causing a user to feel familiarity.

In order to achieve the above object, the interactive device of the present invention is
A language analysis unit that language-analyzes a user's utterance sentence and acquires a user expression feature that is information indicating a feature of an expression included in the user's utterance sentence;
A history storage unit that stores the history of the user expression feature acquired by the language analysis unit as a user utterance sentence history in a storage unit;
A system response sentence generation unit that generates a system response sentence based on the user utterance sentence history stored in the history storage unit;
A system response sentence output unit that outputs a system response sentence generated by the system response sentence generation unit;
Equipped with

  According to the present invention, it is possible to improve the technology for making the user feel familiar.

It is a figure which shows the mode of the dialog of the dialog apparatus and user according to the embodiment of the present invention. It is a figure which shows the external appearance of the dialog apparatus which concerns on embodiment. It is a figure showing the composition of the dialog device concerning an embodiment. It is a figure which shows an example of the user speech sentence log memorize | stored in the memory | storage part of the dialog apparatus which concerns on embodiment. It is a figure which shows the example of the user speech sentence of each speech classification. It is a figure which shows an example of a result of the language analysis part which concerns on embodiment parsing-analyzed a user utterance sentence. It is a figure which shows an example of a result of the language analysis part which concerns on embodiment divided | segmented a predicate part into a self-supporting word part and an attachment word part. It is a flow chart which shows the processing procedure of the language analysis processing concerning an embodiment. It is a figure which shows an example of the positive / negative polarity word dictionary which concerns on embodiment. It is a figure which shows an example of the speech type determination rule which concerns on embodiment. It is a flow chart which shows a processing procedure of dialogue control processing concerning an embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals.

(Embodiment)
As shown in FIG. 1, the dialog device 100 according to the embodiment of the present invention is a robot capable of interacting with the user U by voice. As shown in FIG. 2, the interactive device 100 comprises a head 20 and a torso 30. The head 20 of the interactive device 100 is provided with a microphone 21, a camera 22, a speaker 23, and a sensor group 24.

  A plurality of microphones 21 are provided at the left and right of the head 20 and at the positions of the ears in the face of a person, and constitute an array microphone. The microphone 21 functions as a voice acquisition unit that acquires, as voice data, a voice uttered by the user U who is around the interactive device 100.

  The camera 22 is an imaging device provided at the center of the front of the head 20, at the position of the nose in the face of a person. The camera 22 functions as an image acquisition unit that acquires data (image data) of an image in front of the interactive device 100, and inputs the acquired image data to the control unit 110 described later. The interactive device 100 stores in advance the image data of the face of the user (or the feature amount of the image data) in the storage unit 120 described later. Then, the interactive device 100 can identify who the interactive partner is (user identification) by performing image recognition on the image data of the face acquired by the camera 22 during the interactive operation.

  The speaker 23 is provided on the lower side of the camera 22, at the position of the mouth in the face of a person. The speaker 23 functions as an audio output unit that outputs audio.

  The sensor group 24 is provided at the eye position in the face of a person. The sensor group 24 includes an acceleration sensor, an obstacle detection sensor, and the like, detects various physical quantities, and is used for attitude control of the interactive apparatus 100, collision avoidance, security assurance, and the like.

  The head 20 and the torso 30 of the interactive device 100 are mutually connected by a neck joint 31 shown by a broken line as shown in FIG. The neck joint 31 includes a plurality of motors. By driving the plurality of motors by the control unit 110 described later, the head 20 of the interactive apparatus 100 can be rotated in three axes of the vertical direction, the horizontal direction, and the tilting direction. Thereby, interactive device 100 can perform, for example, nodding action.

  As shown in FIG. 2, the lower part of the trunk 30 of the dialogue apparatus 100 is provided with a foot part 32. The undercarriage 32 includes four wheels and a drive motor. Of the four wheels, two are disposed on the front side of the fuselage 30 as front wheels and the remaining two are disposed on the rear side of the fuselage 30 as rear wheels. As the wheel, for example, an omni wheel, a mechanum wheel or the like may be used. When the control unit 110 described later controls the drive motor to rotate the wheels, the interactive device 100 moves.

  Next, with reference to FIG. 3, the functional configuration of the interactive device 100 will be described. As shown in FIG. 3, the interactive device 100 includes an operation button 33, a control unit 110, and a storage unit 120 in addition to the configuration described above.

  Although not shown, the operation button 33 is provided at the back of the body 30. The operation buttons 33 are various buttons for operating the dialogue device 100. The operation button 33 includes a power button, a volume control button of the speaker 23, and the like.

  The control unit 110 is configured by a CPU (Central Processing Unit) or the like. The control unit 110 executes a program stored in the storage unit 120 to execute a speech recognition unit 111, a language analysis unit 112, a history storage unit 113, a system response sentence generation unit 114, a system response sentence output unit 115, and the like described later. It functions as the user identification unit 116. In addition, the control unit 110 has a clock function and a timer function, and can acquire current time (current date and time) and elapsed time.

  The storage unit 120 is configured by a read only memory (ROM), a random access memory (RAM), and the like, and stores programs executed by the CPU of the control unit 110, various data, and the like. In the storage unit 120, the history storage unit 113 stores a history of information related to the user utterance sentence as the user utterance sentence history 121.

  As shown in FIG. 4, the user utterance sentence history 121 includes information of “user”, “user utterance sentence”, “utterance type”, “end-of-sentence expression”, “postposition presence”, and “filler”. This indicates that the user indicated by the item of “user” has uttered the user utterance sentence indicated by the item of “user utterance sentence”. Also, the item "utterance type" has the utterance type of the user's utterance sentence, the item "statement end expression" has the sentence end expression of the user utterance sentence, and the item "possess / no existence" has the particle of the user's utterance sentence The presence of indicates that the filler in the user's utterance is stored in the item "filler". The utterance type, the sentence end expression, the presence or absence of a particle, and the filler will be described later.

  Next, each function realized by the control unit 110 will be described. As described above, the control unit 110 executes the program stored in the storage unit 120 to execute the speech recognition unit 111, the language analysis unit 112, the history storage unit 113, the system response sentence generation unit 114, and the system response sentence output. It functions as the unit 115 and the user identification unit 116. Further, the control unit 110 corresponds to the multithread function, and can execute a plurality of threads (flows of different processes) in parallel.

  The speech recognition unit 111 performs speech recognition on the speech data acquired by the speech acquisition unit (microphone 21), and acquires a user utterance sentence as text data.

  The language analysis unit 112 analyzes the language of the user utterance sentence (text data) acquired by the speech recognition unit 111 by language analysis processing described later. Specifically, the language analysis unit 112 performs morphological analysis and syntax analysis on the user utterance sentence to acquire part of speech information, modification information, and the like of each word in the user utterance sentence. Then, based on these, the language analysis unit 112 acquires information indicating the feature of the user's expression (user's expression feature) and determines the utterance type of the user's utterance sentence. The utterance type is, as shown in FIG. 5, information (utterance type information) indicating the type of the user's utterance text (assert, request, question, request, instruction, proposal, etc.). In addition, as the user expression features, there are sentence end expressions, presence or absence of particles, fillers and the like, which will be described later.

  The history storage unit 113 stores the user utterance sentence acquired by the speech recognition unit 111 and the user expression feature and the utterance type acquired or determined by the language analysis unit 112 as the user utterance sentence history 121 for each user. Store in The sentence end expression, the presence or absence of a particle, and the filler of the user expression feature will be described below.

  The sentence end expression is an expression at the end of the user utterance sentence. A method of extracting this sentence-end expression will be described using a user utterance sentence of the user A shown in FIG. 4 "I do not want to teach a recommended movie". When the language analysis unit 112 parses the user's utterance sentence, as shown in FIG. 6, it is possible to extract a predicate part (I want to be taught) along with the structure of the sentence. Then, as shown in FIG. 7, the language analysis unit 112 divides this predicate part into the independent word part and the adjunct part, so that this adjunct part can be extracted as the sentence end expression of the original user utterance sentence. .

  Here, the presence or absence of a particle refers to the presence or absence of a particle at the end of the dependency source according to the predicate portion of the user utterance sentence. Referring to FIG. 6, the dependency source for the predicate part (I want to be taught) is "Movie", and there is a case particle "" at the end, so in this example, "the presence or absence of a particle" "Is". As described above, “the presence or absence of a particle” can be acquired by the language analysis unit 112 analyzing the user's utterance sentence and the presence or absence of the end particle of the dependency source according to the predicate part of the user's utterance sentence.

  The filler is an interjection included in the user's utterance sentence, such as "uh," "well." In this case, we will also include the throw particle in the filler for "... ne", "... saa", etc. In the example of FIG. 6, since a filler does not exist, a "filler" becomes "absent" in this example. As another example, if the user utterance sentence of the user C shown in FIG. 4 "A-, a recommended movie, will teach you", since the "A-" at the beginning is an interjection, it is extracted as a filler. In this manner, the “filler” can be extracted as an interjection or an interjection particle included in the user's utterance sentence by the language analysis unit 112 analyzing the user's utterance sentence to obtain part-of-speech information.

  The system response sentence generation unit 114 generates a system response sentence of a standard expression for the user utterance sentence based on the user utterance sentence and the language analysis result of the user utterance sentence. Then, the system response sentence generation unit 114 converts the system response sentence based on the user expression feature acquired from the user utterance sentence history 121 stored in the history storage unit 113. The system response sentence generation unit 114 may generate a system response sentence based on the user expression feature from the beginning without generating a system response sentence of a standard expression.

  A method will be described in which the system response sentence generator 114 converts the system response sentence based on the user expression feature of the user currently interacting. In this process, the system response sentence generation unit 114 generates a sentence end expression having a high appearance frequency and a particle having a high appearance frequency (an end particle of a dependency source related to the predicate part) in the past user utterance sentences. System response sentences are converted on the basis of usage patterns (particle use patterns) and frequently filed fillers. Here, the particle use pattern used when obtaining the appearance frequency in the present embodiment is a pattern using particles (pattern using particles with particles) and a pattern not using particles (no particles for simplicity). And two types of particle usage pattern).

  First, among the data included in the user utterance sentence history 121, the system response sentence generation unit 114 targets all data of the user who is currently interacting, and ends all sentences stored in the item "end-of-sentence expression". In the expression, the sentence end expression having the appearance frequency equal to or more than the sentence end expression reference frequency (for example, 0.2) is extracted as the frequent appearance end expression. The appearance frequency is obtained as the number of the sentence end expressions in the data of the user / the total number of data. Here, when a plurality of sentence end expressions are equal to or higher than the sentence end expression reference frequency, the sentence end expression having the highest appearance frequency among them may be used as the frequent sentence end expression, or randomly selected from sentence end expressions greater than the sentence end expression reference frequency. It is also possible to select one sentence end expression and use it as a frequent sentence end expression.

  The sentence end expression reference frequency does not have to be a fixed value. For example, when the total number of data in the user utterance sentence history 121 of the user is n, the number of data increases, such as setting the sentence end expression reference frequency to 2 / n (however, 1 when n = 1). According to the above, the sentence end expression reference frequency may be reduced. Further, the system response sentence generation unit 114 may extract not the sentence end expression having the appearance frequency equal to or more than the sentence end expression reference frequency but the sentence end expression having the highest appearance frequency as the frequent appearance end expression.

  Then, the system response sentence generation unit 114 converts the system end sentence expression of the system response sentence into the frequently appearing sentence end expression extracted according to the appearance frequency as described above.

  Next, the system response sentence generation unit 114 sets the appearance frequency to the particle reference frequency (this value is, for example, 0. 0) for the data of the user currently in the dialogue among the data included in the user utterance sentence history 121. 7) Use patterns of particles that are set to a value of 0.5 or more), such as 7 or more, as frequent occurrence particle use patterns. The appearance frequency is obtained as the number of the particle use patterns in the data of the user / the total number of data.

  Specifically, the system response sentence generation unit 114 is stored in the item “presence or absence of particles” for the data of the user who is currently interacting among the data included in the user utterance sentence history 121. Find the number of "Yes" and the number of "None". Here, assuming that the number of "present" is m and the number of "absent" is n, the frequency of occurrence of particle use patterns with particles is m / (m + n), and the frequency of occurrence of particle use patterns without particles is n It can be expressed as / (m + n). Here, if n / (m + n) is higher than the reference frequency of the particle, the particle use pattern of no particle is extracted as the frequent use particle use pattern, and the system response sentence generation unit 114 Delete the last particle of the relevant dependency source).

  Also, if m / (m + n) is higher than the reference frequency, the particle use pattern with particles is extracted as the frequent use particle use pattern, and the system response sentence generation unit 114 does not delete the particles from the system response sentence (system Leave the particle of the response sentence as it is). Also, if both m / (m + n) and n / (m + n) are less than the reference frequency of the particle, the user's particle use pattern is determined to be a use pattern in which presence or absence of particle is mixed, and the system response sentence The generation unit 114 deletes the particles (the particles at the end of the dependency source related to the predicate part) included in the system response sentence with a probability of 50%. However, this probability need not be limited to 50%, and may be deleted, for example, with the probability of n / m. By deleting with the probability of n / m, it is possible to make a system response statement that more reflects the frequency of particle use by the user.

  Next, among the data included in the user utterance sentence history 121, the system response sentence generation unit 114 targets all data of the user currently interacting, and stores all the data stored in the “filler” item. Among the above, fillers having an appearance frequency equal to or higher than the filler reference frequency (for example, 0.1) are extracted as frequent fillers. When a plurality of fillers are extracted, the most frequently occurring filler may be used as the frequent filler. Then, when the frequent filler is extracted, the system response sentence generator 114 inserts the frequent filler into the system response sentence.

  The filler reference frequency does not have to be a fixed value. For example, when the total number of data in the user utterance sentence history 121 of the user is n, the filler reference frequency may be decreased as the number of data increases, such as setting the filler reference frequency to 1 / n. . Further, the system response sentence generation unit 114 may extract not the filler whose appearance frequency is the filler reference frequency or more but the filler having the highest appearance frequency as a frequent appearance filler.

  As described above, by converting the system response sentence, the system response sentence generation unit 114 performs system response of the user's expression characteristic based on the three viewpoints of the user expression feature (the sentence end expression, the particle use pattern, and the filler). It can be reflected in the statement.

  The system response sentence output unit 115 synthesizes the voice of the system response sentence generated and converted by the system response sentence generation unit 114, and outputs the voice of the system response sentence from the speaker 23.

  The user identification unit 116 performs image recognition on the image data of the face of the user acquired by the camera 22 to identify the current conversation partner. Note that the user identification unit 116 may identify the conversation partner by recognizing the feature of the voice from the voice data acquired by the microphone 21. Alternatively, the other party of interaction may be identified using both the recognition result of the image data acquired by the camera 22 and the recognition result of the audio data acquired by the microphone 21.

  The functional configuration of the dialog device 100 has been described above. Next, language analysis processing by the language analysis unit 112 will be described with reference to FIG. This process is performed each time voice recognition of a voice uttered by the user is performed to acquire text data.

  First, the language analysis unit 112 performs morphological analysis and syntax analysis on the user utterance sentence to acquire part-of-speech information, dependency information, and the like of each word in the user utterance sentence (step S101). Next, the language analysis unit 112 extracts the annex part of the predicate part of the syntactically analyzed user utterance sentence as the sentence end expression of the user utterance sentence (step S102). Then, the language analysis unit 112 acquires the presence or absence of the last particle of the dependency source related to the predicate part of the user utterance sentence (step S103). Furthermore, the language analysis unit 112 extracts a filler from the user utterance sentence based on the part-of-speech information (step S104). By the processing from step S101 to step S104, the user expression feature included in the user utterance text is acquired.

  Next, the language analysis unit 112 adds positive / negative polarity information to the analysis result acquired in step S101 based on the positive / negative polarity word dictionary (step S105). Here, as shown in FIG. 9, a positive / negative polarity word dictionary is a phrase that indicates a positive expression (eg, favorite (verb), delicious (adjective), beautiful (adjective verb), etc.) and negative expression. This is a word dictionary in which words (for example, dislike (verbs), bad taste (adjectives), unskilled (adjective verbs), etc.) expressing.

  If the analysis result acquired in step S101 by the language analysis unit 112 includes a term representing a positive expression registered in the positive / negative polarity word dictionary, “positive” is added to the analysis result as positive / negative polarity information. Be done. In addition, if the analysis result acquired by the language analysis unit 112 in step S101 includes a term representing a negative expression registered in the positive / negative polarity word dictionary, “negative” as positive / negative polarity information in the analysis result Is granted. If there is no word registered in the positive / negative polarity word dictionary in the analysis result acquired by the language analysis unit 112 in step S101, positive / negative polarity information is not attached to the analysis result.

  Then, the language analysis unit 112 applies the utterance type determination rule to the analysis result, determines the utterance type (step S106), and ends the process. Here, as shown in FIG. 10, the utterance type determination rule is a rule for determining the utterance type of the user utterance sentence based on the part of speech, inflected form, positive / negative polarity information, etc. at the end of the user utterance sentence. In addition, when it does not correspond to any rule, the utterance type of the said user utterance sentence is determined to be "others." However, the utterance type determination rule shown in FIG. 10 is merely an example. Also, the types of utterance types are not limited to those shown in FIG. For example, "greeting" ( "Hello" or the like, the speech of greeting), "consent" ( "I agree" or the like, speech indicating that you have agreed to the utterance of the interlocutor), "thanks" ( "Thank you", etc. It is also possible to add to the types of utterance types such as “speaking for expressing gratitude”, “apologize” (“sorry,” uttering expressing apology, etc.).

  By the language analysis processing described above, acquisition of the user expression feature included in the user utterance text and determination of the utterance type of the user utterance text are performed. Next, the dialogue control process executed by the control unit 110 of the dialogue apparatus 100 will be described with reference to FIG. This process is started when the dialog device 100 is activated.

  First, the user identification unit 116 performs image recognition on the image data acquired by the camera 22 to identify the user (step S201). In the case where image recognition can not be performed, for example, when a face is not shown in the image data acquired by the camera 22, the user who has been identified immediately before is regarded as the user who is currently interacting.

  Next, the control unit 110 determines whether to respond to the user (step S202). This determination can be made from various points of view. For example, if the dialog device 100 has not yet heard the user's voice, it is difficult to respond to the user. In this case, it is determined that the user does not respond. In addition, since the normal dialogue responds alternately with the dialogue partner, it may be determined that it does not respond this time if it is determined to respond the previous time, and it may be determined to respond this time if it is determined not to respond the previous time.

  If the control unit 110 determines not to respond (step S202; No), the voice recognition unit 111 performs voice recognition on the voice data acquired from the microphone 21 to obtain text data of the user's utterance sentence (step S211). Step S211 is also referred to as speech recognition step.

  Then, the language analysis unit 112 language-analyzes the text data acquired in step S211 by the language analysis processing (FIG. 8) described above, and acquires the user expression feature included in the user utterance text and the utterance type of the user utterance text The determination is made (step S212). Step S212 is also called a language analysis step.

  Then, the history storage unit 113 stores the user name identified in step S201, the user utterance sentence acquired in step S211, the utterance type determined in step S212, and the acquired user expression feature (text end expression, presence or absence of a particle, filler), It is added to the user utterance sentence history 121 and stored in the storage unit 120 (step S213). Step S213 is also called a history storage step. Then, the process returns to step S201.

  On the other hand, when it is determined in step S202 that the control unit 110 responds (step S202; Yes), the system response sentence generation unit 114 generates a system response sentence for the immediately preceding user utterance sentence (step S221). The system response statement generated here is a system response statement of a standard expression, and can be generated by a conventional technique. The immediately preceding user utterance sentence can also be acquired from the user utterance sentence history 121, but generation of a system response sentence usually requires a language analysis result of the user utterance sentence. Therefore, control unit 110 stores the result of language analysis performed by language analysis unit 112 in step S212 on the immediately preceding user's utterance in storage unit 120 so that system response sentence generation unit 114 can use it in step S221. It is desirable to keep

  Then, the system response sentence generation unit 114 converts the system response sentence generated in step S221 into a system response sentence using an expression corresponding to a user expression feature included in the user utterance sentence history 121 stored in the storage unit 120. It converts (step S222). Step S222 is also called a system response sentence generation step.

  For example, it is assumed that the standard system response sentence generated in step S221 is "Tell me your favorite food". Then, assuming that the conversation partner is user A, the sentence in the sentence end is "I want to ask you," there is a particle, there is no filler, so the system response sentence, "I want to teach you favorite food Converted to Also, assuming that the conversation partner is user C, the sentence at the end of the sentence is "Takeru", there is no particle, and the filler is "Ah," so the system response sentence is "Ah, your favorite food, tell me Converted to

  Then, the system response sentence output unit 115 performs speech synthesis on the system response sentence generated or converted by the system response sentence generation unit 114, and outputs the synthesized speech from the speaker 23 (step S223). Step S223 is also called a system response sentence output step. Then, the process returns to step S201.

  By the dialog control process described above, the dialog device 100 can generate a system response sentence reflecting the feature of the user's expression. Therefore, the system response sentence issued from the dialogue apparatus 100 becomes a system reply sentence imitating the user's whistling for each user, and the dialogue apparatus 100 can realize the dialogue in which the user feels attachment or closeness. . Thus, the dialog device 100 can improve the technology for making the user feel familiar.

(Modification 1)
In the above-described embodiment, the system response sentence generation unit 114 has the highest appearance frequency among all sentence end expressions stored in the item “end of sentence expression” among the data included in the user utterance sentence history 121. Sentence endings are extracted as frequent ending sentences. However, since the sentence end expression also changes depending on the utterance type, the appearance frequency in consideration of the utterance type may be used. Such a modified example 1 will be described.

  In the first modification, the system response sentence generation unit 114 first narrows down only the data of the user currently interacting with among the data included in the user utterance sentence history 121. Next, the system response sentence generation unit 114 determines the utterance type of the system response sentence generated in step S221 of the dialogue control process (FIG. 11), and the utterance type of the data included in the user utterance sentence history 121. Narrows down to the same data as the utterance type of the system response sentence. Then, the system response sentence generation unit 114 extracts the sentence end expression having the highest appearance frequency as the frequent appearance end expression from among the sentence end expressions stored in the item “state end expression” of the data thus narrowed down.

  As described above, by extracting the frequently appearing sentence end expression in consideration of the utterance type, in the first modification, it is possible to generate a system response sentence in which the user's expression is closely imitated for each utterance type. It is possible to realize a dialogue that makes the user feel more attached and familiar. As described above, the dialog device according to the first modification can further improve the technique for making the user feel familiar.

(Modification 2)
In the above-described embodiment, the system response sentence spoken by the dialog device 100 is a system response sentence imitating the user's whistling by using the user expression feature of the user in the dialogue. However, it is not limited to this. Since some users do not like being able to imitate their own whistle, the system response sentence may be converted into a representation opposite to the user representation feature. Such a modification 2 will be described.

  In the second modification, the system response sentence generation unit 114 searches for a sentence end expression different from the frequently appearing sentence end expression of the user in the dialogue from the user utterance sentence history 121 of another user and searches for the sentence end expression of the system response sentence. Convert to a sentence end expression. Note that if the system response statement generated at step 221 of the dialog control process (FIG. 11) is different from the frequent statement end expression of the user in the dialog, the sentence statement is not converted (standard expression You may leave it as it is.

  Then, the system response sentence generation unit 114 uses the particles in the usage pattern reverse to the usage pattern of the user's particles in the dialog. That is, of the data included in the user utterance sentence history 121, the system response sentence generation unit 114 is “presence” stored in the item “presence / absence of particles” for the data of the user who is currently interacting. Find the number of and the number of "none". Here, as described above, the number of “present” is represented by m, and the number of “absent” is represented by n. Then, if n / (m + n) is higher than the reference frequency of particles, the particle use pattern without particles is the user's particle use pattern, and conversely, the system response sentence generation unit 114 Do not delete the last particle of the dependency source (retain the particles of the system response sentence).

  If m / (m + n) is higher than the reference frequency of the particles, the system response sentence generation unit 114 deletes the particles from the system response sentence because the particle use pattern with particles is the user's particle use pattern. If both m / (m + n) and n / (m + n) are less than the reference frequency of particles, it is determined that the user's particle use pattern is a use pattern in which presence or absence of particles is mixed. In this case, m and n are compared, and if n is larger (since the user often omits particles), the system response sentence generator 114 generates a particle (a predicate part) included in the system response sentence. Do not delete the last particle of the dependency source (retain the particles of the system response sentence). Also, if m is larger (because the user often does not omit particles), the system response sentence generator 114 generates a particle (a particle at the end of the dependency source of the predicate part) contained in the system response sentence. Delete).

  Then, the system response sentence generation unit 114 sets all data stored in the item of “filler” for the data of the user who is currently interacting among the data included in the user utterance sentence history 121. Examine. Then, if the appearance frequency of the filler is smaller than the filler reference frequency (for example, 0.1), the filler (for example, "Ano", "Eor", etc.) is inserted into the system utterance text. The filler to be inserted here may be selected from the fillers included in the user utterance sentence history 121 of another user.

  As described above, by converting the system utterance sentence into a different expression from the user expression feature, a system utterance sentence in which the user's mouthpiece is contrasted (the user's mouthpiece is not reflected) is generated. be able to. Therefore, in the second modification, it is possible to generate a system response sentence that is not disliked by a user who does not like being imitated by his own whistle, and a sense of closeness is also made to a user who does not want You can realize the dialogue that makes Therefore, the dialog device according to the second modification can improve the technique for making the user feel familiar.

  Although the expression used for the system response sentence generated by the dialog device according to the second modification is different from the expression of the user expression feature, it is different from the expression of the user expression feature, and thus the different expression is selected. It can be said that it is a kind of expression corresponding to the user expression feature.

(Modification 3)
In the above-described embodiment, the system response sentence generation unit 114 targets the data of the user currently in the dialogue among the data included in the user utterance sentence history 121, the frequently appearing sentence end expression, the presence or absence of the particle, the frequent appearance filler You are getting However, it is not necessary to limit to such processing, and for example, the system response sentence may be converted based on the user expression feature of the user with the largest number of user utterances. Such a modification 3 will be described.

  In the third modification, the control unit 110 also functions as an utterance number counting unit that counts the number of utterances of the user. The utterance number counting unit refers to the “user” item of the user utterance sentence history 121 stored in the storage unit 120, and counts the number of user utterance sentences of each user. Then, the user with the largest number of user utterance sentences is selected as the representative user.

  Then, the system response sentence generation unit 114 targets the data of the user who is currently interacting among the data included in the user utterance sentence history 121 when acquiring the frequent appearing sentence end expression, the presence or absence of the particle, and the frequent appearing filler. Instead, the data of the representative user selected by the speech counting unit is targeted.

  As described above, the dialog device according to the third modification has the most frequent conversation frequency by acquiring the frequently used sentence end expression of the representative user, presence or absence of particles, and frequently used fillers and converting the system response sentence using them It will respond to the user with a system response statement that mimics the user's whistle. Then, it is possible to inspire the user's willingness to interact with the user who wants to imitate his / her own whistling than others, and can make each user feel a sense of closeness. Therefore, the dialogue apparatus concerning modification 3 can improve the art for making a user feel a sense of closeness.

(Other modifications)
In addition, this invention is not limited to the above-mentioned embodiment, A various change is possible. For example, in the above-mentioned embodiment, although a user expression feature was caught from three viewpoints of a sentence end expression, the existence of a particle, and a filler, it is not restricted to this. For example, first-person expressions ("I", "I", etc.), second-person expressions ("you", "you", etc.), etc. may also be used as user expression features. By doing this, the user's expression can be reflected more, and the interactive device can make the user feel more familiar.

  In addition, when performing the process of step S221 of the above-mentioned dialogue control process (FIG. 11), the process of step S222 is also performed together to directly convert the system response sentence of the standard expression without directly following the procedure. A system response sentence reflecting the user expression feature of the user may be generated.

  Further, in the above-described embodiment, the interactive device 100 includes the user identification unit 116, performs user identification with the camera 22 or the microphone 21, and stores the user utterance sentence history 121 for each user. However, when it is assumed that the interactive device is used exclusively for a specific user, the user identification unit 116 need not be provided. In this case, the camera 22 is also unnecessary, and there is no need to provide an item of “user” in the user utterance sentence history 121.

  Moreover, in the above-mentioned embodiment, although the user is identified each time in step S201 of the dialog control process (FIG. 11), the present invention is not limited to this. Immediately after the dialog device 100 is activated, the user identification process corresponding to only step S201 of the dialogue control process (FIG. 11) is performed to identify the dialogue user, and then the dialogue control process omitting step S201 is performed. Also good. Then, in the user identification process, the user may have the face pointed at the camera 22 by the interactive device 100 uttering “please show me a face”. After startup, the dialog device 100 may utter “Tell me your name” and use the microphone 21 to identify the user by voice recognition without using the camera 22.

  Further, in the case where the user can not be identified in any of the user identification processes, the interactive device 100 defines a user name such as “default user”, for example, and the user name of the unidentifiable user is “ It may be treated as "default user" or the like.

  Further, in the above-described embodiment, the interactive device 100 is a movable robot, and various operations can be performed by including the sensor group 24, the neck joint 31, the foot portion 32, and the like. However, even if the dialog device can not perform various operations such as movement, it can only interact with the user. In this case, the sensor group 24, the neck joint 31, and the foot portion 32 are unnecessary.

  Each function of dialogue device 100 can also be implemented by a computer such as a normal PC (Personal Computer). Specifically, in the above embodiment, it has been described that a program such as dialog control processing performed by the dialog device 100 is stored in advance in the ROM of the storage unit 120. However, the program is distributed by being stored in a computer-readable recording medium such as a flexible disk, a compact disc read only memory (CD-ROM), a digital versatile disc (DVD) and a magneto-optical disc (MO). May be read and installed in a computer to configure a computer capable of realizing the functions described above.

  Although the preferred embodiments of the present invention have been described above, the present invention is not limited to the specific embodiments, and the present invention includes the invention described in the claims and the equivalents thereof. Be In the following, the invention described in the original claims of the present application is appended.

(Supplementary Note 1)
A language analysis unit that language-analyzes a user's utterance sentence and acquires a user expression feature that is information indicating a feature of an expression included in the user's utterance sentence;
A history storage unit that stores the history of the user expression feature acquired by the language analysis unit as a user utterance sentence history in a storage unit;
A system response sentence generation unit that generates a system response sentence based on the user utterance sentence history stored in the history storage unit;
A system response sentence output unit that outputs a system response sentence generated by the system response sentence generation unit;
Interactive device comprising:

(Supplementary Note 2)
The system response sentence generation unit generates a system response sentence using an expression corresponding to the user expression feature whose appearance frequency is equal to or higher than a reference frequency in the user utterance sentence history stored in the history storage unit.
The interactive apparatus as described in appendix 1.

(Supplementary Note 3)
An expression corresponding to the user expression feature is the same expression as the user expression feature.
The interactive apparatus as described in appendix 2.

(Supplementary Note 4)
An expression corresponding to the user expression feature is an expression that contrasts with the user expression feature.
The interactive apparatus as described in appendix 2.

(Supplementary Note 5)
The language analysis unit further acquires an utterance type, which is information indicating the type of the user utterance text,
The history storage unit stores the history of the user utterance sentence, the user expression feature, and the utterance type acquired by the language analysis unit in the storage unit as a user utterance sentence history.
The system response sentence generation unit calculates the appearance frequency of the user expression feature for each of the utterance types from the user utterance sentence history stored in the history storage unit, and the user expression in the utterance types acquired by the language analysis unit A system response sentence is generated using an expression corresponding to the user expression feature whose appearance frequency is equal to or higher than a reference frequency.
The interactive device according to any one of appendices 2 to 4.

(Supplementary Note 6)
It further comprises a user identification unit for identifying the user,
The history storage unit stores a user utterance sentence history in the storage unit for each user identified by the user identification unit.
The system response sentence generation unit corresponds to the user expression feature whose appearance frequency of the user expression feature of the user identified by the user identification unit is equal to or higher than a reference frequency in the user utterance sentence history stored in the history storage unit. Create a system response statement using the
The interactive device according to any one of appendices 2 to 5.

(Appendix 7)
The system further comprises an utterance number counting unit that counts the number of utterances of each user,
The system response sentence generation unit is configured such that, among the user utterance sentence histories stored in the history storage unit, the appearance frequency of the user expression feature of the user with the largest number of utterances counted by the utterance number counting unit is a reference frequency or more Generating a system response sentence using an expression corresponding to the user expression feature,
The interactive apparatus as described in appendix 6.

(Supplementary Note 8)
The user expression feature is at least one of a sentence end expression of the user utterance sentence, a particle use pattern of the user utterance sentence, and a filler included in the user utterance sentence.
The interactive device according to any one of appendices 1-7.

(Appendix 9)
A dialog method for outputting a system response sentence generated using a characteristic sentence end expression, a particle use pattern or an expression corresponding to a filler that frequently appears in the user's utterance sentence based on a history of a user's utterance sentence uttered by the user .

(Supplementary Note 10)
A language analysis step of linguistically analyzing a user's utterance sentence to acquire a user expression feature that is information indicating a feature of an expression included in the user's utterance sentence;
A history storing step of storing the history of the user expression feature acquired in the language analysis step as a user utterance sentence history in a storage unit;
A system response sentence generation step of generating a system response sentence based on the user utterance sentence history stored in the history storage step;
A system response sentence output step for outputting the system response sentence generated in the system response sentence generation step;
Interactive method including:

(Supplementary Note 11)
On the computer
A language analysis step of linguistically analyzing a user's utterance sentence to acquire a user expression feature which is information indicating a feature of an expression included in the user's utterance sentence;
A history storage step of storing the history of the user expression feature acquired in the language analysis step as a user utterance sentence history in a storage unit;
A system response sentence generation step of generating a system response sentence based on the user utterance sentence history stored in the history storage step;
A system response sentence output step of outputting the system response sentence generated in the system response sentence generation step;
A program to run a program.

Reference Signs List 20 head 21 microphone 22 camera 23 speaker 24 sensor group 30 body 31 neck joint 32 leg portion 33 operation button 100 dialog device 110 control unit 111 voice recognition unit 112 language analysis unit 113 history storage unit 114 system response sentence generation unit 115 system response sentence output unit 116 user identification unit 120 storage unit 121 user utterance Statement history, U ... User

Claims (11)

A language analysis unit that language-analyzes a user's utterance sentence and acquires a user expression feature that is information indicating a feature of an expression included in the user's utterance sentence;
A history storage unit that stores the history of the user expression feature acquired by the language analysis unit as a user utterance sentence history in a storage unit;
A system response sentence generation unit that generates a system response sentence based on the user utterance sentence history stored in the history storage unit;
A system response sentence output unit that outputs a system response sentence generated by the system response sentence generation unit;
Interactive device comprising: The system response sentence generation unit generates a system response sentence using an expression corresponding to the user expression feature whose appearance frequency is equal to or higher than a reference frequency in the user utterance sentence history stored in the history storage unit.
An interactive device according to claim 1. An expression corresponding to the user expression feature is the same expression as the user expression feature.
An interactive device according to claim 2. An expression corresponding to the user expression feature is an expression that contrasts with the user expression feature.
An interactive device according to claim 2. The language analysis unit further acquires an utterance type, which is information indicating the type of the user utterance text,
The history storage unit stores the history of the user utterance sentence, the user expression feature, and the utterance type acquired by the language analysis unit in the storage unit as a user utterance sentence history.
The system response sentence generation unit calculates the appearance frequency of the user expression feature for each of the utterance types from the user utterance sentence history stored in the history storage unit, and the user expression in the utterance types acquired by the language analysis unit A system response sentence is generated using an expression corresponding to the user expression feature whose appearance frequency is equal to or higher than a reference frequency.
The interactive device according to any one of claims 2 to 4. It further comprises a user identification unit for identifying the user,
The history storage unit stores a user utterance sentence history in the storage unit for each user identified by the user identification unit.
The system response sentence generation unit corresponds to the user expression feature whose appearance frequency of the user expression feature of the user identified by the user identification unit is equal to or higher than a reference frequency in the user utterance sentence history stored in the history storage unit. Create a system response statement using the
The interactive device according to any one of claims 2 to 5. The system further comprises an utterance number counting unit that counts the number of utterances of each user,
The system response sentence generation unit is configured such that, among the user utterance sentence histories stored in the history storage unit, the appearance frequency of the user expression feature of the user with the largest number of utterances counted by the utterance number counting unit is a reference frequency or more Generating a system response sentence using an expression corresponding to the user expression feature,
The interactive device according to claim 6. The user expression feature is at least one of a sentence end expression of the user utterance sentence, a particle use pattern of the user utterance sentence, and a filler included in the user utterance sentence.
The interactive device according to any one of claims 1 to 7.   A dialog method for outputting a system response sentence generated using a characteristic sentence end expression, a particle use pattern or an expression corresponding to a filler that frequently appears in the user's utterance sentence based on a history of a user's utterance sentence uttered by the user . A language analysis step of linguistically analyzing a user's utterance sentence to acquire a user expression feature that is information indicating a feature of an expression included in the user's utterance sentence;
A history storing step of storing the history of the user expression feature acquired in the language analysis step as a user utterance sentence history in a storage unit;
A system response sentence generation step of generating a system response sentence based on the user utterance sentence history stored in the history storage step;
A system response sentence output step for outputting the system response sentence generated in the system response sentence generation step;
Interactive method including: On the computer
A language analysis step of linguistically analyzing a user's utterance sentence to acquire a user expression feature which is information indicating a feature of an expression included in the user's utterance sentence;
A history storage step of storing the history of the user expression feature acquired in the language analysis step as a user utterance sentence history in a storage unit;
A system response sentence generation step of generating a system response sentence based on the user utterance sentence history stored in the history storage step;
A system response sentence output step of outputting the system response sentence generated in the system response sentence generation step;
A program to run a program.

Images