JP2018097201A - Voice dialog device and voice dialog method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide conversation of a rich variation in a device which dialogs with a user.SOLUTION: A voice dialog device which dialogs with a user by voice comprises: voice acquisition means which acquires an utterance uttered by the user; a vocabulary database which stores a plurality of vocabulary data for generating a response sentence; and response generating means which generates the response sentence to the utterance on the basis of the stored vocabulary data. The vocabulary data are the data in which a first word is associated with a second word which is a word showing recognition of the user to the first word. The response generating means extracts the first word or the second word corresponding to an object mentioned in the utterance from the vocabulary data, to execute a first response generation to generate the response sentence using a word associated with the word.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an apparatus for interacting with a user by voice.

  Much research has been done to provide natural conversations in interactive robot systems. For example, Patent Document 1 discloses a response generation device for generating a response to an input utterance. The apparatus has a feature that, when there are a plurality of response sentence candidates for a certain topic, an appropriate response sentence is selected according to the abundance of topics and the degree of emotional brightness. According to this invention, a more appropriate response sentence can be generated.

JP 2007-219149 A

  In the invention according to Patent Document 1, a response can be generated only in the range of data accumulated in advance. In other words, it is not possible to increase the data at any time to enrich the content of the dialogue.

  On the other hand, a conversation system that actively increases the vocabulary that can be used by learning through conversation with a user is known. However, in such a system, when an unknown word appears in a conversation, the user can be asked and learned about the meaning of the word, but the conversation variation cannot be expanded based on the learning result. .

  The present invention has been made in consideration of the above-described problems, and an object of the present invention is to provide a conversation rich in variations in an apparatus that interacts with a user.

A voice interaction apparatus according to the present invention is a voice interaction apparatus that interacts with a user by voice,
A voice acquisition means for acquiring an utterance uttered by the user, a vocabulary database for storing a plurality of vocabulary data for generating a response sentence, and a response sentence for the utterance based on the stored vocabulary data Response generation means, and the vocabulary data is data in which a first word is associated with a second word that is a word representing the user's recognition of the first word, and the response The generation means extracts the first word or the second word corresponding to the object mentioned in the utterance from the vocabulary data, and generates a response sentence using the word associated with the word Response generation is performed.

The spoken dialogue apparatus according to the present invention generates a response sentence for an utterance made by a user, using vocabulary data registered in a vocabulary database. Vocabulary data is data in which a first word is associated with a second word, which is a word representing user recognition of the first word. The word representing the user's recognition is, for example, “a word representing the user's preference for the first word”, “a word describing the first word by the user”, “the user explained the concept of the first word Words ". As described above, the second word only needs to represent the user's recognition of the first word, and may not be unique. For example, when the first word is “ramen”, the second word may be a word representing a higher concept such as “noodles”, or a word representing preference such as “like”. Also good. For example, when the first word is “test”, the second word may be a word simply associated with the user, such as “tension”.
Note that the first and second words are not necessarily one word. For example, it may be a set of words (sentences) such as “tense”.

  The response generation means acquires a pair of words from the vocabulary data for the word corresponding to the word mentioned in the acquired utterance sentence, and generates a response sentence. For example, when the user makes an utterance referring to ramen, the word “ramen” may be searched from the first word, and a response sentence may be generated using the corresponding second word. Also, for example, when the user utters “tensioned”, the word “tension” is searched from the second word, and a response sentence is generated using the corresponding first word. Good.

  The first word and the second word are not necessarily one-to-one. For example, a plurality of second words may be associated with a specific first word, and a plurality of first words may be associated with a specific second word. According to such a configuration, it is possible to inflate variations of response sentences by following the correspondence between words.

  Further, the voice interaction apparatus according to the present invention asks the user a question asking about recognition of the object mentioned in the utterance, and generates or updates the vocabulary data based on an answer obtained from the user. The vocabulary collecting means may be further provided.

  In order to learn vocabulary data, the vocabulary collecting means may ask the user a question about recognition of a certain word. The content of the question may be something that asks the user's preference, for example, “Do you like XX?”, Or you may change the word to another word, such as “How does XX feel?” It may be what the user wants to describe. Also, it may ask the word concept itself, such as “What is XX?”.

  The vocabulary collecting means may generate or update the vocabulary data based on the word when the utterance uttered by the user includes a word representing recognition of the target mentioned in the utterance. May be a feature.

  Thus, if the first word and the second word (that is, the word that the user is referring to and the word representing the user's perception) can be automatically extracted from the utterances made in the normal dialogue, Vocabulary data may be generated or updated.

  Further, the vocabulary database may store the plurality of vocabulary data in association with each user, and the response generation unit may use vocabulary data associated with the user who is interacting.

  By holding vocabulary data for each user and using the vocabulary data corresponding to the user during the conversation, personalized answering can be made.

  In addition, the second word is a word that represents a general concept of the first word, a word that the user describes the first word, or a word that represents the user's preference for the first word. One of them may be a feature.

  In this way, it is possible to widen variations of responses by connecting and storing related words.

The response generation means extracts the first word corresponding to the target mentioned in the utterance from the vocabulary data, and the second word associated with the first word is extracted. It is good also as extracting the related word which is other 1st words in common with, and performing the 2nd response production | generation which produces | generates a response sentence using the said related word.

  For different first words, the same second word may be associated. In such a case, another first word (related word) may be extracted via the second word and used to generate a response sentence. For example, if the words “ramen” and “noodles”, “udon” and “noodles” are stored in association with each other, and if an utterance mentioning ramen is made, the topic about udon will be shaken. It may be.

  In addition, in the second response generation, when there are a plurality of the second words related to the extracted first word, the response generation means is directly related to the related word in addition to the related word. The second word that is not attached may be further extracted, and a response sentence may be generated using the related word and the extracted second word.

  For example, when the words “ramen” and “noodles”, “ramen” and “like”, “udon” and “noodles” are stored in association with each other, and an utterance referring to ramen is made A response sentence such as “Do you like udon?” May be generated.

  Note that the present invention can be specified as a voice interactive device including at least a part of the above means. Further, it can be specified as a dialogue method performed by the voice dialogue apparatus. The above processes and means can be freely combined and implemented as long as no technical contradiction occurs.

  ADVANTAGE OF THE INVENTION According to this invention, in the apparatus which interacts with a user, a conversation with rich variation can be provided.

It is a system configuration figure of a dialog system concerning a first embodiment. 4 is a data flow diagram among the robot 10, the control device 20, and the server device 30. FIG. It is a flowchart figure of the process which the server apparatus 30 performs. It is an example of the word table in 1st embodiment. It is an example of the word table in 2nd embodiment.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.
The voice dialogue system according to the present embodiment is a system that performs dialogue with the user by acquiring voice uttered by the user, performing voice recognition, and generating a response sentence based on the recognition result.

(First embodiment)
<System configuration>
FIG. 1 is a system configuration diagram of the voice interaction system according to the first embodiment. The voice interaction system according to the present embodiment includes a robot 10, a control device 20, and a server device 30.

The robot 10 has a speaker, a microphone, a camera, and the like, and is a means for performing an interface with a user. The robot 10 may be a human type, a character type, or another shape.
The control device 20 is a device that issues a command to the robot 10. The server device 30 is a device that generates a response (response sentence) to be provided to the user in response to a request transmitted from the control device 20.
In the present embodiment, the robot 10 functions only as a user interface, and the server device 30 performs processing for controlling the entire system, such as recognition of utterance contents, generation of response sentences, and other processing. Further, the control device 20 performs a process that mediates between the robot 10 and the server device 30.

First, the robot 10 will be described.
The robot 10 includes a voice input unit 11, a short-range communication unit 12, and a voice output unit 13.

  The voice input unit 11 is means for acquiring voice uttered by the user. Specifically, sound is converted into an electric signal (hereinafter referred to as sound data) using a built-in microphone. The acquired voice data is transmitted to the control device 20 via the short-range communication unit 12 described later.

  The near field communication unit 12 is a unit that performs near field communication with the control device 20. In the present embodiment, the short-range communication unit 12 performs communication using the Bluetooth (registered trademark) standard. The short-range communication unit 12 stores information related to the control device 20 that is a pairing destination, and can be connected by simple processing. Note that the Bluetooth standard is also referred to as IEEE 802.15.1.

  The audio output unit 13 is means for outputting audio to be provided to the user. Specifically, voice data transmitted from the control device 20 is converted into voice using a built-in speaker.

  Next, the control device 20 will be described. The control device 20 is a device that mediates between the robot 10 and the server device 30, and is typically a small computer such as a mobile computer, a mobile phone, or a smartphone. The control device 20 can be configured as an information processing device having a CPU, a main storage device, and an auxiliary storage device. Each unit shown in FIG. 1 functions by loading a program stored in the auxiliary storage device into the main storage device and executing it by the CPU. Note that all or part of the illustrated functions may be executed using a circuit designed exclusively.

  The control device 20 includes a short-range communication unit 21, a control unit 22, and a communication unit 23.

  Since the functions of the short-range communication unit 21 are the same as those of the short-range communication unit 12 described above, detailed description thereof is omitted.

  The control unit 22 is a unit that acquires sound from the robot 10 and acquires a response to the acquired sound. Specifically, the voice acquired from the robot 10 is transmitted to the server device 30 (both described later) via the communication unit 23, and a corresponding response sentence is received from the server device 30. Further, the response sentence is converted into voice data by the voice synthesis function and transmitted to the robot 10. The voice transmitted to the robot 10 is provided to the user via the voice output unit 13. Thereby, the user can perform a conversation in a natural language.

  The communication unit 23 is a unit that communicates with the server device 30 by accessing the network via a communication line (for example, a wireless LAN or a mobile phone network).

The server device 30 is a device that generates a response sentence to be provided to the user after recognizing the transmitted voice, and includes a communication unit 31, a voice recognition unit 32, and a response generation unit 33.
Since the function of the communication unit 31 is the same as that of the communication unit 23 described above, detailed description thereof is omitted.

  The voice recognition unit 32 is a unit that performs voice recognition on the voice acquired by the voice input unit 11 of the robot and converts the voice into text. Speech recognition can be performed by known techniques. For example, the speech recognition unit 32 stores an acoustic model and a recognition dictionary, extracts features by comparing the acquired speech data with the acoustic model, and matches the extracted features with the recognition dictionary to generate speech. Recognize. The recognition result is transmitted to the response generation unit 33.

The response generation unit 33 is a unit that generates a response sentence to be provided to the user based on the text acquired from the voice recognition unit 32. The response sentence to be provided may be based on, for example, a dialogue scenario (dialog dictionary) stored in advance, or may be based on information obtained by searching a database or the web.
In the present embodiment, the response generation unit 33 has a table (word table) that is a dictionary for learning words, and can generate different response sentences according to learning results. Detailed processing contents will be described later.
The information acquired by the response generation unit 33 is transmitted to the control device 20 in a text format, then converted into synthesized speech, and output to the user via the robot 10.

  The server device 30 can also be configured as an information processing device having a CPU, a main storage device, and an auxiliary storage device. Each unit shown in FIG. 1 functions by loading a program stored in the auxiliary storage device into the main storage device and executing it by the CPU. Note that all or part of the illustrated functions may be executed using a circuit designed exclusively.

<Dialog flow>
Next, the processing performed by each unit shown in FIG. 1 and the data flow will be described with reference to FIG. 2 which is a flowchart for explaining the processing contents and the data flow.

  First, in step S11, the voice input unit 11 of the robot 10 acquires the voice spoken by the user through the microphone. The acquired voice is converted into voice data and transmitted to the control unit 22 of the control device 20 via the communication unit. In addition, the control unit 22 transmits the acquired voice data to the voice recognition unit 32 included in the server device 30.

Next, the voice recognition unit 32 performs voice recognition on the acquired voice data and converts it into text (step S12). The text obtained as a result of the speech recognition is transmitted to the response generation unit 33. Next, the response generation unit 33 generates a response based on the content of the utterance obtained from the user (step S13).
The generated response sentence is transmitted to the control device 20 and converted into voice data by the control unit 22 (step S14). Then, the audio data is transmitted to the robot 10 and output (reproduced) via the audio output unit 13 (step S15).

<Response generation method>
Next, a specific method for generating a response by the response generation unit 33 will be described. FIG. 3 is a flowchart showing in more detail the process in which the response generation unit 33 generates a response in step S13.

  First, in step S21, an object referred to by the user (an object that is a topic. Hereinafter, a word representing the object is referred to as an object word) is determined. The target word can be determined, for example, by performing morphological analysis on the text output by the speech recognition unit 32 and analyzing a plurality of obtained words. For example, when the user makes an utterance such as “I ate ramen for lunch today”, it can be determined that the user refers to ramen (or lunch).

Step S22 will be described later.
In step S23, it is determined whether or not the target word has been learned by referring to the stored word table. FIG. 4 is an example of a word table stored in the response generation unit 33. In the present embodiment, the first word and the second word are stored in the word table in association with each other. The first word is a learned word, and the second word is a word that represents user recognition of the first word.
Examples of the word representing user recognition include the following. Each of the first word and the second word may be a single word or a set of words.
(A) A word that describes the first word (a word associated with the first word)
For example, the first word is described by another word, such as “smooth” for “udon”. Further, for example, it may be a word simply associated with the first word, such as “tense” for “test”. In this embodiment, all are called adjectives.
(B) A word representing the superordinate concept of the first word For example, “ramen” and “udon” both have a superordinate concept of “noodles”, and therefore can be represented by the word “noodles” in both. .
(C) A word representing the user's preference for the first word. For example, “like”, “dislike”, “somewhat like”, “very dislike” (hereinafter referred to as preference expression).
In the example of FIG. 4A, the word “test” and the word “pounding” are associated with the user whose user ID is U001. Here, the description will be continued by taking as an example a situation where the word corresponding to “ramen” determined in step S21 does not exist in the table (not learned).

When the word referred to by the user is an unlearned word, the process proceeds to step S24 to generate a question for the user. In the present embodiment, the question generated in step S24 is one of the following three types. Any type of question may be used.
(1) A question that allows a user to describe a word. For example, it is a question such as "What does XX feel?"
(2) Question that asks the general concept of words For example, a question such as “What is XX?”.
(3) Question that asks the user's preference for words For example, a question such as “Do you like OO?”.

In step S25, the generated question is output and an answer to the question is obtained from the user. Here, an answer is acquired from the user after the processing in steps S13 to S15 is temporarily advanced. That is, the flow shown in FIG. 2 is executed once again, and the process returns to step S13 again. According to such processing, it is possible to obtain from the user one of a preference expression, an adjective, and a superordinate concept corresponding to the target word.
When the type of question is (1) described above, since the adjective is included in the response to the question, it is extracted. For example, an adjective can be acquired by extracting a phrase whose word sequence matches the following pattern from utterances performed by the user.

For example, it is assumed that the user answers “smoothly feel like ramen” and the analysis “smoothly (adverb) + do (verb)” is made. In this case, the word “smoothly” is registered as the second word.
FIG. 4B shows an example in which a record (vocabulary data) is added based on the user's recognition that “ramen slippery”.

When the type of the question is (2) described above, the response to the question includes a word that represents the general concept of the target word, and is extracted.
For example, the user's answer can be changed by various patterns such as “XX”, “XX”, “XX”, “XX”, and “XX”. What is necessary is just to extract the part used as a direct answer by the normalized expression (regular expression).

When the type of the question is (3) described above, the response to the question includes a preference expression by the user, and is extracted.
For example, it may be determined whether the question is affirmed or denied, such as “Yes” or “No”, or an absolute expression such as “Like” or “I don't like” may be extracted. Also, negative expressions and double negative expressions (not like, not dislike etc.) may be considered.

  When the first word and the second word corresponding to the first word are acquired, these words are set and recorded in the word table (step S26).

Note that learning may be possible without asking the user questions.
In the present embodiment, in step S22, it is determined whether or not learning is possible only with the content of the acquired utterance. For example, if the utterance contains any of the preferred expressions, adjectives, or superordinate concepts, such as “I like ramen”, it is possible to learn without asking additional questions. The process proceeds to step S26. In this case, it is possible to learn by associating “ramen” with “like”. If learning is not possible, the process proceeds to step S23.

  In step S27, a response sentence is generated using the learning result (that is, using information recorded in the word table). The same applies when it is determined in step S23 that learning has been completed.

In step S27, the target word is searched from the word table, and when the target word exists, a paired word (may be the first word or the second word) is extracted. And used to generate response sentences.
For example, in the case where the information shown in FIG. 4A is recorded, and the user's utterance content is “I am thrilled now”, “I am thrilled (second word)”. The corresponding word “test (first word)” is extracted, and for example, a response sentence such as “is it a test?” Is generated.
In addition, when the user refers to “test”, the word “exciting (second word)” corresponding to “test (first word)” is extracted, and a response sentence such as “exciting” is added. It may be generated.

Although FIG. 4A illustrates an adjective, the second word may be a preference expression. For example, when “ramen” and “like” are related, and the user mentions “ramen”, a response sentence such as “I like ramen” may be generated. In addition, when the user mentions “what he / she likes”, a response sentence such as “which do you like ramen?” May be generated.
Similarly, the second word may be a superordinate concept. For example, when “ramen” and “noodles” are related, and the user refers to “ramen”, a response sentence such as “noodles” may be generated. In addition, when the user refers to “noodles”, a response sentence such as “ramen or kadane” may be generated.

  In addition, since it will become unnatural if a response is produced | generated using the information learned immediately before when it changes from step S26 to step S27, it is preferable not to use the information just learned.

  As described, in the first embodiment, the first word and the second word representing the user's recognition for the first word are stored in association with each other, and a response sentence is generated by referring to each other. Thereby, a topic can be produced | generated based on a user's recognition, and the variation of a response can be enriched.

(Second embodiment)
In the first embodiment, the response sentence is generated by mutually referring to the first word and the second word. On the other hand, in the second embodiment, when the same word is related to a plurality of first words as the second word, another first word related to the same word (related word) ) Is extracted and used for the response.

  For example, in the example of FIG. 4C, the word “test” and the word “interview” are both associated with the word “pounding”. In such a case, when the user utters the content of “I have an interview”, the related word “test” is extracted via the second word, and “excitingly like an interview and a test. May be generated.

  4C, the word “ramen” and the word “tsukemen” are both associated with the word “smooth”. In such a case, when the user utters content such as “Would you like to eat ramen?”, The word “Tsukemen” is extracted via the second word, and a response sentence “Tsukemen is good!” May be generated.

  As described above, in the second embodiment, other related words that are common to the second word are extracted and used to generate a response sentence. Since the related word is a word for which the user recognizes certain relevance, it is possible to present a topic that is natural to the user.

(Third embodiment)
In the first and second embodiments, only one second word is defined. In contrast, the third embodiment is an embodiment in which the second word is held by a plurality of fields for each type.

  Since the configuration of the voice interaction system according to the third embodiment is the same as that of the first embodiment, description thereof will be omitted, and only differences in data and processing used will be described.

  FIG. 5 is an example of a word table stored in the response generation unit 33 in the third embodiment. In the present embodiment, the second word defined in the word table is represented by three of “taste expression”, “adjective word”, and “superordinate concept”.

  Further, in the third embodiment, when the target word is acquired in the dialogue, as in the second embodiment, other words having the same second word (preference, adjective, or superordinate concept) are common. Are extracted as related words and used to generate response sentences.

Processing of the response generation unit 33 in the third embodiment will be described with reference to FIG. Here, it is assumed that the user has made an utterance that “a new udon shop has been made” and it is determined in step S21 that the device refers to “udon”. Also, it is assumed that the response generation unit 33 stores the word table shown in FIG.
In this case, steps S22 and S23 are negative.

In the third embodiment, when adding vocabulary data, what is extracted based on the user's utterance (whether it is a preference expression, an adjective, or a superordinate concept) is determined and appropriate This is different from the second embodiment in that it is stored in a simple field.
In the third embodiment, steps S24 and S25 may be repeated a plurality of times to ask a plurality of types of questions. For example, you may make it ask a question about a superordinate concept and a taste twice.
In this example, as a result of asking “What is udon?”, An answer such as “It ’s noodles” is obtained, and as a result of asking “Do you like udon?” It is assumed that an answer has been obtained. As a result, the word table is as shown in FIG.

In the third embodiment, in step S27, as in the second embodiment, the first word associated with the same word among the second words (preference expression, adjective, or superordinate concept) is associated. A word is extracted as a related word and used for a response.
For example, when an utterance referring to “Udon” is made, the word “Soba” having the same superordinate concept is acquired, and a response sentence having a topic about soba can be generated. For example, a response sentence such as “Udonka. Noodles are good for soba” may be generated.

  By the way, there are cases in which any one of the preference expression, adjective, and superordinate concept is common and the others are contrary to each other for the plurality of first words. This will be described.

  For example, in the example of FIG. 5B, some of the related second words are contrary to each other, like udon, but dislikes the same noodle soba. Therefore, in step S27, a response sentence that discusses the contradiction may be generated. For example, a response sentence such as “I like udon but don't like soba” can be generated.

  In the third embodiment, as described above, the second word is subdivided and stored and used to generate a response sentence. According to this configuration, more words can be extracted from the vocabulary database. Moreover, when some of the second words are contradictory, the topic used in the dialogue can be further increased, for example, by pointing out this.

(Modification)
The above embodiment is merely an example, and the present invention can be implemented with appropriate modifications within a range not departing from the gist thereof.

  For example, in the description of the embodiment, the server device 30 performs voice recognition, but the control device 20 may have a means for performing voice recognition. In the description of the embodiment, the server apparatus 30 generates a response sentence. However, the control apparatus 20 may have a means for generating a response sentence. Further, the robot 10 may perform all processes without using the control device 20 and the server device 30.

DESCRIPTION OF SYMBOLS 10 ... Robot 11 ... Voice input part 12, 21 ... Short-range communication part 13 ... Voice output part 14 ... Operation control part 20 ... Control apparatus 22 ... Control part 23, DESCRIPTION OF SYMBOLS 31 ... Communication part 30 ... Server apparatus 32 ... Voice recognition part 33 ... Response generation part

Claims (9)

A voice interaction device that interacts with a user by voice,
Voice acquisition means for acquiring a speech uttered by the user;
A vocabulary database for storing a plurality of vocabulary data for generating a response sentence;
Response generating means for generating a response sentence to the utterance based on the stored vocabulary data,
The vocabulary data is data in which a first word is associated with a second word that is a word representing the user's recognition of the first word;
The response generation means extracts the first word or the second word corresponding to the object mentioned in the utterance from the vocabulary data, and generates a response sentence using a word associated with the word. One response generation,
Spoken dialogue device. Lexical data collection means for asking the user a question asking about recognition of the object mentioned in the utterance and generating or updating the vocabulary data based on an answer obtained from the user;
The voice interactive apparatus according to claim 1. The vocabulary collecting means generates or updates the vocabulary data based on the word when the utterance uttered by the user includes a word representing recognition of the object mentioned in the utterance.
The voice interactive apparatus according to claim 2. The vocabulary database stores the plurality of vocabulary data in association with each user;
The response generating means uses vocabulary data associated with a user who is interacting;
The voice interactive apparatus according to claim 2 or 3. The second word is a word that represents a superordinate concept of the first word, a word that the user describes the first word, or a word that represents the user's preference for the first word Either
The voice interactive apparatus according to claim 1. The response generation means extracts the first word corresponding to the object mentioned in the utterance from the vocabulary data, and the related second word is common to the extracted first word Extracting a related word that is another first word to perform a second response generation that generates a response sentence using the related word,
The voice interactive apparatus according to claim 1. In the second response generation, when there are a plurality of the second words related to the extracted first word, the response generation means is directly related to the related word in addition to the related word. Further extracting the second word that is not, and generating a response sentence using the related word and the extracted second word;
The voice interactive apparatus according to claim 6. A dialogue method performed by a voice dialogue device that dialogues with a user by voice,
A voice acquisition step of acquiring a speech uttered by the user;
A response generation step of generating a response sentence for the utterance based on a plurality of vocabulary data for generating a response sentence,
The vocabulary data is data in which a first word is associated with a second word that is a word representing the user's recognition of the first word;
In the response generation step, the first word or the second word corresponding to the object mentioned in the utterance is extracted from the vocabulary data, and a response sentence is generated using a word associated with the word. One response generation,
How to interact. A voice interaction device that interacts with a user by voice,
Voice acquisition means for acquiring a speech uttered by the user;
Response generating means for generating a response sentence for the utterance based on a plurality of vocabulary data for generating a response sentence,
The vocabulary data is data in which a first word is associated with a second word that is a word representing the user's recognition of the first word;
The response generation means extracts the first word or the second word corresponding to the object mentioned in the utterance from the vocabulary data, and generates a response sentence using a word associated with the word. One response generation,
Spoken dialogue device.

Images