JP2010157081A - Response generation device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To naturally dialogize by generating a suitable sentence corresponding to an input content from users. <P>SOLUTION: A first user's deliverance input from a microphone 12 is sound recognized, a state discrimination part 22 discriminates whether or not a state is included in the first user's deliverance based on analysis result morphologically analyzed for the sound-recognized first user's deliverance, then when the state is included, an emotional polarity estimation part 24 estimates an emotional polarity expressed by the first user's deliverance, and a question generation part 26 generates and outputs a question sentence questioning an emotion to a user. An emotional polarity extraction part 28 extracts the emotional polarity from analysis result of a second user's deliverance to the question sentence, and a polarity conformity discrimination part 30 discriminates whether or not the estimated emotional polarity and the extracted emotional polarity conform. A response generation part 32 generates a response sentence indicating "agreement" when the emotional polarity conforms and generates a response sentence indicating "surprise" or selects a response sentence from formula response sentences of supportive response or promotion when it does not conform. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a response generation device and a program, and more particularly, to a response generation device and a program for performing a smooth dialogue with a user.

Conventionally, a concept is extracted from an input user utterance, a plurality of response sentences using the extracted concept or its related words are generated, and based on “abundance of topics” and “emotion” for each predetermined concept Thus, there has been proposed a response generation device that determines and outputs a response sentence having a high priority from a plurality of generated response sentences (see, for example, Patent Document 1).
JP 2007-219149 A

  However, since the response generation device of Patent Document 1 generates a response sentence using the concept included in the user utterance or its related word, even if the utterance contents are the same even if the utterance is uttered with different intentions Has the problem that the same response sentence will be generated, and natural dialogue may not be possible. For example, with respect to the following Dialogue Example 1 and Dialogue Example 2, in the response generation device of Patent Document 1, based on “It was fun” of the user utterance 2, in both cases of Dialogue Example 1 and Dialogue Example 2, for example, “It was fun. A response sentence such as “Tadaa” will be generated.

(Dialogue example 1)
User utterance 1: I went to an amusement park.
System response: How was it?
User utterance 2: It was fun.
(Dialogue example 2)
User utterance 1: It was an athletic meet in the rain.
System response: How was it?
User utterance 2: It was fun.

  The present invention has been made to solve the above problem, and provides a response generation apparatus and program capable of generating an appropriate response sentence corresponding to the input content from the user and performing a natural conversation. The purpose is to do.

  In order to achieve the above object, a response generation apparatus according to the present invention includes: an input unit that inputs an input sentence from a user; and an input sentence that is input by the input unit before outputting a question sentence. As an input sentence, based on an analysis result obtained by analyzing the structure of the first input sentence, it is determined whether or not the first input sentence includes a word indicating a situation or a combination of a word and a word Control is performed so as to output a question sentence prepared in advance for asking the user about emotions when it is determined by the determination means and the situation determination means that the first input sentence includes a situation. When the first control means and the situation determination means determine that the first input sentence includes a situation, the first input is based on the analysis result of the first input sentence. Estimating means for estimating the emotion polarity represented by the sentence; An analysis result obtained by analyzing the structure of the second input sentence using the input sentence input from the user by the input means as a second input sentence for the question sentence controlled and output by the first control means From the extraction means for extracting the emotion polarity of the second input sentence, the emotion polarity estimated by the estimation means and the emotion polarity extracted by the extraction means are prepared in advance. A first response sentence is generated and output, and if they do not match, a second response means prepared to generate and output a second response sentence prepared in advance is included. .

  Further, the response generation program according to the present invention uses the input sentence input by the input means for inputting the input sentence from the user as the first input sentence before the computer outputs the question sentence. A situation determination means for determining whether or not the first input sentence includes a word or a combination of words and words based on an analysis result obtained by analyzing a structure of the input sentence; and the situation determination means The first control means for controlling to output a question sentence prepared in advance for asking the user for emotion when it is determined that a situation is included in the first input sentence, When it is determined that a situation is included in the first input sentence by the situation determination unit, the emotion polarity represented by the first input sentence is estimated based on the analysis result of the first input sentence. Estimating means and the first control means From the analysis result obtained by analyzing the structure of the second input sentence using the input sentence input from the user by the input means as the second input sentence for the question sentence controlled and output, When the extraction means for extracting the emotion polarity of the two input sentences matches the emotion polarity estimated by the estimation means and the emotion polarity extracted by the extraction means, a first response prepared in advance This is a program for functioning as a second control means for generating and outputting a sentence and controlling so as to generate and output a second response sentence prepared in advance if they do not match.

  According to the response generation device and the program according to the present invention, the situation determination unit uses the input sentence input by the input unit that inputs the input sentence from the user before outputting the question sentence as the first input sentence. Based on the analysis result obtained by analyzing the structure of one input sentence, it is determined whether or not the first input sentence includes a word representing a situation or a combination of a word and a word. “Situation” refers to an action, event, or event associated with some emotion.

  The first control means outputs a question sentence prepared in advance for asking the user about emotions when the situation judgment means judges that the situation is included in the first input sentence. Control. With respect to this question sentence, the input sentence input from the user by the input means is used as the second input sentence, and the extraction means analyzes the emotion of the second input sentence from the analysis result obtained by analyzing the structure of the second input sentence. Extract polarity. In addition, when the situation determination unit determines that the situation is included in the first input sentence, the estimation unit determines the emotion polarity represented by the first input sentence based on the analysis result of the first input sentence. Is estimated. When the second control means matches the emotion polarity estimated by the estimation means with the emotion polarity extracted by the extraction means, a first response sentence prepared in advance is generated and output. If they do not match, control is performed so as to generate and output a second response sentence prepared in advance.

  Thus, the response sentence can be made different depending on whether or not the emotion polarity estimated from the first input sentence matches the emotion polarity extracted from the second input sentence. An appropriate response sentence corresponding to the content can be generated and a natural conversation can be performed.

  In addition, the first response sentence is a response sentence indicating consent to the second input sentence, and the second response sentence is a response sentence indicating surprise to the second input sentence, It can be a response sentence that prompts the user to input an input sentence, or a response sentence of a conflict. In this way, when the emotion polarity estimated from the first input sentence matches the emotion polarity extracted from the second input sentence, a response sentence indicating consent is output and the emotion polarities do not match. In some cases, a natural conversation can be performed by outputting a response sentence indicating surprise. Further, if the emotion polarities do not match, for example, it is assumed that there is a possibility that there is a misrecognition of voice recognition or an input error of an input sentence by the user, and based on an analysis result that may be misrecognized. Instead of generating a response sentence, a response sentence that prompts the user to input an input sentence or a response sentence that is compatible can be output without breaking the dialog.

  Further, the response generation apparatus of the present invention provides a response sentence that prompts the user to input an input sentence or a response of a conflict when the situation determination means determines that the first input sentence does not include a situation A third control means for controlling to output a sentence can be included. When the situation is not included in the first input sentence, the emotional polarity cannot be estimated. Therefore, by outputting a response sentence that prompts the user to input the input sentence or a response sentence of the conflict, the dialogue is performed. It is possible to proceed without failing.

  As described above, according to the response generation device and the program of the present invention, it is possible to generate an appropriate response sentence corresponding to the input content from the user and perform a natural conversation.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the present embodiment, a case will be described in which the present invention is applied to a response generation apparatus that executes a predetermined process and outputs a voice by using an utterance from a user as an input.

  As illustrated in FIG. 1, the response generation device 10 according to the first exemplary embodiment includes a microphone 12 that collects user utterances to generate an audio signal, a speaker 14 that outputs audio, a microphone 12, and a speaker 14. And a computer 16 that executes a predetermined process for generating an appropriate response sentence.

  The computer 16 includes a CPU that controls the entire response generation apparatus 10, a ROM as a storage medium that stores various programs such as a response generation program to be described later, a RAM that temporarily stores data as a work area, and a storage unit that stores various information. As an HDD, an I / O (input / output) port, and a bus connecting them. A microphone 12 and a speaker 14 are connected to the I / O port.

  In addition, when the computer 16 is described by functional blocks divided for each function realizing means determined based on hardware and software, the voice signal input from the microphone 12 is recognized as shown in FIG. A situation is included in the user utterance based on the analysis result of the language analysis unit 20 that performs morphological analysis on the character string information indicating the user utterance that has been voice-recognized using a general morphological analyzer. When the situation is included in the user utterance, when the situation is included in the user utterance, when the situation is included in the user utterance, the emotion polarity estimation section 24 that estimates the emotion polarity represented by the user utterance In addition, a question generation unit 26 that generates a question sentence that asks the user for emotions, and an emotion that extracts emotion polarities from the analysis result obtained by analyzing the user utterance for the question sentence by the language analysis unit 20 Of the polarity matching determination unit 30 and the polarity matching determination unit 30 for determining whether or not the emotion polarity estimated by the sex extraction unit 28 and the emotion polarity estimation unit 24 and the emotion polarity extracted by the emotion polarity extraction unit 28 match. Based on the determination result, a response generation unit 32 that generates different response sentences, a standard response part 34 that selects a standard response sentence when a situation is not included in the user utterance, a question generation unit 26, and a response generation unit 32 , And an output unit 36 that converts the response sentence generated or selected by the standard response unit 34 into an audio signal and outputs it from the speaker 14.

  In this embodiment, as described later, since it is necessary to distinguish between a user utterance input before the output of the question sentence and a user utterance input after the output, the former is referred to as “first user utterance”. The latter is referred to as “second user utterance”.

  The situation determination unit 22 determines whether the analysis result of the language analysis unit 20 for the first user utterance includes a word indicating the situation or a combination of the word and the word. “Situation” is an action, event, or event associated with some emotion. Here, (i) “verb” is included in the analysis result, and (ii) “situation noun +”. If "" is included, it is determined that "Situation" is included. “Situation noun” is a noun indicating an event such as “rain” or “athletic meet”. Situation nouns are determined in advance by creating a situation noun dictionary as shown in FIG. 2 and referring to this situation noun dictionary at the time of determination.

  The emotion polarity estimation unit 24 estimates whether the emotion polarity represented by the first user utterance is positive or negative when the situation determination unit 22 determines that a situation is included in the first user utterance. The estimation of the emotion polarity is performed by, for example, learning the learning data with known emotion polarity using the SVM (Support Vector Machine) method to construct an emotion polarity model, and analyzing the result of the emotion polarity model and the first user utterance By comparing with.

  The question generation unit 26 generates a question sentence that asks the user about emotions when the situation determination unit 22 determines that a situation is included in the first user utterance. The question sentence is a basic question sentence, for example, "What do you think?" Or "What do you feel?" Prepared in advance, and selected from this basic question sentence based on the analysis result of user utterance A question sentence is generated by correcting the tense and expression (such as polite expression or broken expression) of one sentence. Also, a question sentence example as shown in FIG. 3 may be created in advance and selected randomly from this. The generated question sentence is converted into an audio signal by the output unit 36 and output from the speaker 14.

  The emotion polarity extraction unit 28 extracts the emotion polarity from the analysis result obtained by analyzing the second user utterance input after the question sentence generated by the question generation unit 26 is output by the language analysis unit 20. Since the question sentence generated by the question generation unit 26 is a question sentence for asking the user's emotion, the second user utterance input here is considered to contain a word directly representing the emotion. It is done. Therefore, the emotion polarity is extracted with reference to the emotion polarity dictionary as shown in FIG. 4 for the word representing the emotion included in the second user utterance. The emotion polarity dictionary uses a technique described in non-patent literature “Emotion polarity extraction of words by spin model” (Takamura et al., Journal of Information Processing Society of Japan, vol. 47, No. 2, pp. 627-637, 2006). Create in advance.

  If the polarity match determination unit 30 determines that the emotion polarity estimated by the emotion polarity estimation unit 24 matches the emotion polarity extracted by the emotion polarity extraction unit 28, the response generation unit 32 determines “agree”. A response sentence including the nuance of “surprise” is generated when it is determined that they do not match. The generated response sentence is converted into an audio signal by the output unit 36 and output from the speaker 14.

  For example, a response sentence example and a response sentence example format as shown in FIG. 5 are prepared in advance, and the response sentence including the “agreement” nuance is selected at random from the response sentence example or generated using the response sentence example format. To do. The response sentence example format can be, for example, a format like “After all (user's emotion)”, and expresses the emotion extracted from the second user utterance in the part of “(user's emotion)” A response sentence can be generated by inserting a word.

  For example, a response sentence example and a response sentence example format as shown in FIG. 6 are prepared in advance, and a response sentence including “surprise” nuances is randomly selected from the response sentence example or generated using the response sentence example format. To do. The response sentence example format can be, for example, a format such as “E? (User's emotion)”, and a word representing the emotion extracted from the second user utterance in the “(user's emotion)” portion. Can be inserted to generate a response sentence. In addition, the response sentence selected and generated in this manner may be combined with a response sentence that asks for a reason, for example, “why (user's emotion)?”.

  When the situation determination unit 22 determines that the situation is not included in the first user utterance, the standard response unit 34 randomly selects a standard response sentence from predetermined fixed response sentences. The selected standard response sentence is converted into an audio signal by the output unit 36 and output from the speaker 14. As the standard response sentence, for example, as shown in FIG. 7, a response sentence that prompts the user to speak or a response sentence that is compatible is determined in advance.

  Next, a response generation processing routine in the response generation device 10 of the first exemplary embodiment will be described with reference to FIG. This routine is performed by the CPU executing a response generation program stored in the ROM.

  In step 100, it is determined whether or not the first user utterance has been input from the microphone 12. If the first user utterance has been input, the process proceeds to step 102. Repeat the determination in this step. Here, it is assumed that the first user utterance “It was an athletic meet in the rain” was input.

  In step 102, the input voice signal indicating the first user utterance is recognized as voice string information, and morphological analysis is performed on the string information.

  Next, in step 104, it is determined whether or not “situation” is included in the first user utterance based on the analysis result of the morphological analysis. If “situation” is included, the process proceeds to step 106; otherwise, the process proceeds to step 122. Here, “Sports” defined as a situational noun in the situational noun dictionary shown in FIG. 2 and “da” following the situational noun are included. (Ii) “Situation noun +” ” ”Is satisfied, it is determined that“ Situation ”is included, and the process proceeds to Step 106.

  In step 106, the emotion polarity of the first user utterance is estimated based on the analysis result and a predetermined emotion polarity model. Here, it is assumed that the emotional polarity is estimated as “negative”.

  Next, in step 108, a response sentence for asking the user about emotions is generated, converted into an audio signal, and output. Here, it is assumed that “What do you think?” Is selected and output from the question sentence example shown in FIG.

  Next, in step 110, it is determined whether or not a second user utterance has been input from the microphone 12. If a user utterance has been input, the process proceeds to step 112, and an audio signal indicating the input user utterance is obtained. Speech recognition is performed to obtain character string information, and morphological analysis is performed on the character string information. If not input, the determination in this step is repeated until input. Here, it is assumed that the second user utterance “It was fun” was input.

  Next, in step 114, the emotion polarity represented by the second user utterance is extracted based on the analysis result and the emotion polarity dictionary shown in FIG. Here, the emotion polarity “positive” is extracted from “I enjoyed” included in the analysis result with reference to the emotion polarity dictionary.

  Next, in step 116, it is determined whether or not the emotion polarity estimated in step 106 matches the emotion polarity extracted in step 114. If they match, the process proceeds to step 118, and if they do not match, the process proceeds to step 120. Here, since the emotional polarity estimated in step 106 is “negative” and the emotional polarity extracted in step 114 is “positive” and does not match, the result is negative and the process proceeds to step 120.

  In step 120, a response sentence including the nuance of “surprise” is generated. For example, the response sentence example format “E? (User's emotion)” is selected from the response sentence example and the response sentence example format shown in FIG. 6, and the emotion extracted from the second user utterance in the part of “(user's emotion)” Is inserted, and a response sentence such as “Have you enjoyed?” Is generated.

  When the second user utterance “It was miserable” is input in step 110, the emotion polarity “negative” is extracted based on the analysis result in step 112 and the emotion polarity dictionary. In step 116, it is determined that the emotion polarities match, and the process proceeds to step 118. In step 118, a response sentence including the nuance of “agreement” is generated. For example, the response sentence example “That's right” is selected from the response sentence example and the response sentence example format shown in FIG.

  Further, for example, if the first user utterance is “desk”, it is determined in step 104 that the “situation” is not included in the first user utterance, and the process proceeds to step 122. In step 122, for example, “Hey, so” is selected from the standard response example shown in FIG.

  Next, in step 124, the response sentence generated or selected in step 118, step 120, and step 122 is converted into an audio signal and output, and the process ends.

  Note that the user utterance that is determined to be input in step 110 is a user utterance that is input after the question text is output in step 108. Therefore, the user utterance is the second user utterance. Is not necessarily emotional. Therefore, if the emotion polarity cannot be extracted from the second user utterance in step 114, the second user utterance is regarded as the first user utterance and the process returns to step 100 or proceeds to step 122. A standard response sentence may be selected and output.

  As described above, according to the response generation device of the first exemplary embodiment, the emotion polarity is estimated from the first user utterance, and the estimated emotion polarity matches the emotion polarity extracted from the second user utterance. In order to determine whether or not the second user utterance has the same content, if the emotional polarities match, a response sentence indicating “agreement” is generated. A response sentence indicating “surprise” is generated. That is, by estimating the emotion polarity represented by the first user utterance, the response can be different between the case where the content of the second user utterance is as expected and the case where it is contrary to the expectation. Conversations.

  In the first embodiment, the case where the emotion polarity represented by the first user utterance is estimated by comparing with the emotion polarity model generated from learning data in advance has been described. However, the present invention is not limited to this method. Absent. For example, a word-emotion polarity dictionary in which emotion polarity is determined for each word is created in advance, and the emotion polarity is obtained by referring to this word-emotion polarity dictionary for each word included in the first user utterance, According to the rules defined based on the number of words whose emotion polarity is “positive” included in the first user utterance, the number of words whose emotion polarity is “negative”, and whether there is a negative expression, etc. The emotion polarity may be estimated. In addition to estimating the emotion polarity directly from the first user utterance, the emotion polarity represented by the first user utterance is estimated, and the emotion polarity is determined with reference to the estimated emotion and the emotion polarity dictionary shown in FIG. You may make it estimate. As in the case of estimation of emotion polarity, a known technique such as estimation by comparing with a plurality of emotion models previously generated for each emotion from learning data can be used for the estimation of emotion.

  Next, a second embodiment will be described. In the second embodiment, when the emotion polarities of the first user utterance and the second user utterance do not match, a response sentence indicating “surprise” is not generated, but a standard response sentence is used. The point of selection is different from the first embodiment. In addition, about the structure and process similar to 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  The configuration of the response generation device 210 according to the second embodiment is the same as the configuration of the response generation device 10 according to the first embodiment shown in FIG. In the response generation device 210 according to the second embodiment, the response generation unit 132 is configured such that the polarity match determination unit 30 determines the emotion polarity estimated by the emotion polarity estimation unit 24 and the emotion polarity extracted by the emotion polarity extraction unit 28. If it is determined that they match, a response sentence including the nuance of “agreement” is generated. If it is determined that they do not match, a standard response sentence is selected. The response sentence generated or selected is converted into an audio signal by the output unit 36 and output from the speaker 14. The generation of the response sentence including the nuance of “agreement” is the same as in the first embodiment, and the selection of the fixed response sentence is the same as the processing of the fixed response unit 34 in the first embodiment. In addition, for example, a response sentence is selected at random from a typical response sentence example as shown in FIG.

  Next, a response generation processing routine in the response generation device 210 of the second exemplary embodiment will be described with reference to FIG. This routine is performed by the CPU executing a response generation program stored in the ROM.

  In step 100, it is determined whether or not the first user utterance has been input from the microphone 12. If the first user utterance has been input, the process proceeds to step 102 to indicate the input first user utterance. The speech signal is recognized as speech to obtain character string information, and morphological analysis is performed on the character string information. If not input, the determination in this step is repeated until input. Here, it is assumed that the first user utterance “I went to an amusement park” is input.

  Next, in step 104, it is determined whether or not “situation” is included in the first user utterance based on the analysis result of the morphological analysis. If “situation” is included, the process proceeds to step 106; otherwise, the process proceeds to step 122. Here, since the verb “go” is included, it is determined that “situation” is included, and the routine proceeds to step 106.

  In step 106, the emotion polarity of the first user utterance is estimated based on the analysis result and a predetermined emotion polarity model. Here, it is assumed that the emotion polarity is estimated as “positive”. Next, in step 108, “How was it?” Is selected and output from a response sentence for asking the user about emotions, for example, the question sentence example shown in FIG.

  Next, in step 110, it is determined whether or not the second user utterance is input from the microphone 12. If the second user utterance is input, the process proceeds to step 112, and the input second user is input. A voice signal indicating an utterance is recognized as voice string information, and morphological analysis is performed on the character string information. If not input, the determination in this step is repeated until input. Here, it is assumed that the second user utterance “It was fun” was input, and that it was misrecognized as “It was sad” in the speech recognition.

  Next, in step 114, the emotion polarity of the second user utterance is extracted based on the analysis result and the emotion polarity dictionary shown in FIG. Here, based on the misrecognized analysis result “I was sad”, the emotion polarity “negative” is extracted with reference to the emotion polarity dictionary.

  Next, in step 116, it is determined whether or not the emotion polarity estimated in step 106 matches the emotion polarity extracted in step 114. If they match, the process proceeds to step 118. If they do not match, the process proceeds to step 122 assuming that there is an error in the analysis result of the first user utterance including voice recognition. Here, since the emotional polarity estimated in step 106 is “positive” and the emotional polarity extracted in step 114 is “negative” and does not match, the result is negative and the process proceeds to step 122.

  In step 122, for example, “Is that so” is selected from the standard response sentence example shown in FIG. 7, and then in step 124, the selected response sentence is converted into an audio signal and output, and the process ends. .

  As described above, according to the response generation device of the second exemplary embodiment, the emotion polarity represented by the first user utterance is estimated, and the estimated emotion polarity and the emotion polarity extracted from the second user utterance are obtained. Determine whether or not they match, and if they do not match, for example, assume that there was an error in speech recognition, and output a standard response sentence such as a response sentence that encourages user speech can do. As a result, it is possible to prevent an inappropriate response sentence from being generated based on the erroneously recognized analysis result, and to allow the dialog to proceed without failure.

  In the first and second embodiments described above, the case where sound is output by a speaker has been described as an example. However, the present invention is not limited to this, and a response sentence may be displayed on a display. Moreover, although the case where audio | voice was input into the microphone from the user was demonstrated to the example, you may make it a user input the text as an input sentence using a keyboard etc.

It is a block diagram which shows schematic structure of the response generation apparatus which concerns on this Embodiment. It is a figure which shows an example of a situation noun dictionary. It is a figure which shows an example of a question sentence example. It is a figure which shows an example of an emotion polarity dictionary. It is a figure which shows an example of the response text example of a response text which shows "agreement", and a response text example format. It is a figure which shows an example of the response sentence example of a response sentence which shows "surprise", and an example of a response sentence example format. It is a figure which shows an example of the example of a response sentence of a fixed form response sentence. It is a flowchart which shows the response production | generation process routine of 1st Embodiment. It is a flowchart which shows the response production | generation process routine of 2nd Embodiment.

Explanation of symbols

10, 210 Response generation device 12 Microphone 14 Speaker 16 Computer 20 Language analysis unit 22 Situation determination unit 24 Emotion polarity estimation unit 26 Question generation unit 28 Emotion polarity extraction unit 30 Polarity match determination units 32 and 132 Response generation unit 34 Fixed response unit 36 Output section

Claims (4)

An input means for inputting an input sentence from the user;
Before outputting the question sentence, the input sentence input by the input means is used as the first input sentence, and the situation of the first input sentence is determined based on the analysis result obtained by analyzing the structure of the first input sentence. A situation determination means for determining whether or not a word or a combination of a word and a word is included,
First control for controlling to output a question sentence prepared in advance for asking the user about emotions when the situation determination means determines that a situation is included in the first input sentence Means,
When the situation determination means determines that a situation is included in the first input sentence, the emotion polarity represented by the first input sentence is estimated based on the analysis result of the first input sentence An estimation means to
For the question sentence controlled and output by the first control means, the structure of the second input sentence was analyzed using the input sentence inputted by the input means from the user as the second input sentence. Extraction means for extracting the emotion polarity of the second input sentence from the analysis result;
When the emotion polarity estimated by the estimation means matches the emotion polarity extracted by the extraction means, a first response sentence prepared in advance is generated and output. Second control means for controlling to generate and output a second response sentence prepared in advance;
A response generation device including:   The first response sentence is a response sentence indicating consent to the second input sentence, and the second response sentence is a response sentence indicating surprise to the second input sentence, to the user The response generation apparatus according to claim 1, wherein the response generation apparatus is a response sentence that prompts input of an input sentence or a response sentence that is a companion.   Third control is performed to output a response sentence that prompts the user to input an input sentence or a response sentence of a conflict when the situation determination means determines that the first input sentence does not include a situation The response generation apparatus according to claim 1, further comprising: Computer
Before outputting the question sentence, based on the analysis result obtained by analyzing the structure of the first input sentence with the input sentence input by the input means for inputting the input sentence from the user as the first input sentence, A situation determination means for determining whether or not the first input sentence includes a word indicating a situation or a combination of a word and a word;
First control for controlling to output a question sentence prepared in advance for asking the user about emotions when the situation determination means determines that a situation is included in the first input sentence Means,
When the situation determination means determines that a situation is included in the first input sentence, the emotion polarity represented by the first input sentence is estimated based on the analysis result of the first input sentence An estimation means to
For the question sentence controlled and output by the first control means, the structure of the second input sentence was analyzed using the input sentence inputted by the input means from the user as the second input sentence. Extraction means for extracting the emotion polarity of the second input sentence from the analysis result;
When the emotion polarity estimated by the estimation means matches the emotion polarity extracted by the extraction means, a first response sentence prepared in advance is generated and output. Second control means for controlling to generate and output a second response sentence prepared in advance;
A response generator to make it function.

Images